Datasets:

chemNLP

/

ncbi-text

like 0

During embryogenesis, hemocytes derive exclusively from head mesoderm at around 2 h after gastrulation (stage 10). From this point of origin, these cells migrate along stereotypical routes to populate the whole embryo by stage 17 (). It has previously been shown that the developmental migration of these cells is dependent on the expression of the VEGF/PDGF ligands Pvf1, -2, and -3 (). The PDGF/VEGF receptor (PVR) is expressed in hemocytes (), and mutant embryos fail to exhibit normal hemocyte migrations, resulting in an accumulation of these cells at their head end (; ). A recent study has demonstrated a role of PVR in controlling anti-apoptotic cell survival of embryonic hemocytes () and suggests that the defect in hemocyte distribution observed in the mutant is largely due to high numbers of hemocytes undergoing apoptosis and becoming engulfed by their neighbors. However, this study also showed that Pvr expression within hemocytes is required for the directed migration of a subset of these cells that enter the extended germ during normal development (), suggesting that this population of hemocytes may well be using Pvf signals as a chemoattractant to guide their migrations. Additionally, ectopic expression of Pvf2 within the embryo has been shown to be sufficient to induce a chemotactic response from embryonic hemocytes (). In addition to migrating along developmental pathways, embryonic hemocytes have been shown to migrate toward a laser-induced wound in a process that resembles the vertebrate inflammatory response. For a hemocyte to chemotax toward a chemotactic source, be it a wound or a guidance cue expressed along developmental migration routes, it has to be able to sense a chemotactic gradient and polarize in alignment with that gradient. Studies using and mammalian neutrophils have demonstrated that the phosphoinositides PtdInsP (PIP) and PtdInsP (PIP) are key signaling molecules that become rapidly and highly polarized in cells that are exposed to a gradient of chemoattractant (; ; ). In these actively chemotaxing cells, phosphoinositide 3-kinases (PI3Ks) rapidly translocate from the cytosol to the membrane at the leading edge of the cell, whereas phosphatase and tensin homologue (PTEN) dissociates from the leading edge and becomes restricted to the sides and the rear. The difference in localization of these two enzymes leads to localized PIP production at the leading edge of the cell (; ). Down- or up-regulation of PIP by deletion of PI3Ks or of PTEN, respectively, results in severely reduced efficiency of chemotaxis (). Though PI3K has been shown to be important for cell motility using these model systems, its role for single-cell chemotaxis in vivo in a multicellular organism has yet to be clarified. has one class I PI3K, Dp110 (), whose role in cell growth control and cell survival has been well characterized (; ; ); however, no role in cell migration and chemotaxis in for this protein has been shown. In this study, we have analyzed the developmental migrations of hemocytes and characterized in detail their migration patterns along the ventral midline. Our quantitative analysis shows that ventral midline hemocytes undergo a rapid lateral migration, during which they are highly polarized. We show that Pvf2 and -3 expression in the central nervous system (CNS), and Pvf2 alone in the dorsal vessel, are essential for directing the migration of hemocytes along these structures, and a decrease in expression of these ligands in the CNS is essential for the normal lateral migration of hemocytes in this region. We have also analyzed the function of PI3K in hemocytes. Using both dominant-negative PI3K–expressing hemocytes and the specific PI3K inhibitory drug LY294002, we show that PI3K is not required for the Pvf-dependent normal dispersal of hemocytes during development but is essential for chemotaxis toward wounds. Additionally, we show that hemocyte chemotaxis toward wounds is dependent on actin polymerization but that PI3K is not required for lamellipodial formation and instead appears to be required to sense a chemotactic gradient from a wound and polarize the hemocyte accordingly. Our results demonstrate that at least two separate mechanisms operate in embryos to direct hemocyte migration and show that although PI3K is crucial for hemocytes to sense a chemotactic gradient from a wound, it is not required to sense the Pvf growth factor signals that coordinate their developmental migrations along the ventral midline and dorsal vessel during embryogenesis. It has previously been shown that in the developing embryo, hemocytes derive exclusively from the head mesoderm and from this origin migrate along invariant pathways to populate the whole embryo by stage 17 (). One major migration route is along the ventral midline, where hemocytes come into close contact with the cells of the CNS midline and the neighboring ventral epidermis. To understand more fully the dynamics of hemocytes during their developmental migrations, we analyzed the migration of these ventral midline hemocytes using live time-lapse imaging of embryos expressing GFP solely in the hemocytes driven by the hemocyte-specific driver (). At stage 12 of development, hemocytes are clustered together at anterior and posterior ends of the midline and express few lamellipodia. Some protrusive structures can be seen on the leading hemocytes as they migrate along the CNS at a rate of 0.4 μm/min and appear to engulf debris along the route (). At stage 16 of development, hemocytes occupy three parallel lines running anterior to posterior along the embryonic ventral midline, as opposed to a single line along the midline at stage 14, with the more laterally placed hemocytes residing in the extracellular space along the lateral margins of the CNS (). Live confocal analysis of hemocytes between these developmental stages reveals that this distribution arises from the rapid lateral migration of a subset of midline hemocytes, which individually leave the midline and occupy these more lateral positions ( and Video 1, available at ). High-magnification imaging of these midline hemocytes at stage 16 reveals that, in contrast to the early stage hemocytes, these cells exhibit large filopodia and lamellipodia, extending 20 μm away from the cell body (). To investigate more thoroughly the pattern of the characteristic lateral migration observed between stages 14 and 16, we used time-lapse confocal microscopy to track the paths taken by individual hemocytes as they performed this movement. Superimposing all hemocyte tracks from individual embryos shows that this movement conforms to a strict segmented pattern where hemocytes leave the ventral midline from spatially reiterated exit points and perform a 90° movement from the midline to arrive at a defined lateral area (). While in the ventral midline, hemocytes aggregate in clusters that correspond to the position of the exit points, where cells initiate their lateral migrations (). Closer analysis of individual tracks shows that before their lateral movement, hemocytes appear to move up and down the ventral midline between the clusters and after lateral migration, to a lesser extent, along the lateral rows (). To further analyze this migration pattern, we quantified the directionality, velocity, and polarity of individual hemocytes while moving within the ventral midline before lateral migration and undergoing the lateral migration. To carry out this quantification, we focused our analysis on the most posterior clusters of hemocytes in each embryo examined (, circled area). The directionality of hemocyte movement was calculated by applying a ratio between the shortest distance possible and the actual distance traversed by hemocytes. As expected from the plotted tracks, the movement during lateral migration is highly directional, showing an mean of 85.6% (±13.1; = 46) directionality (, red track). If the same ratio is applied to the path of the same hemocyte before this lateral migration, while in the ventral midline, the mean directionality observed is much lower (29.6 ± 18.2%; = 35; , blue track). Quantifying the speed of migration, we noted that during the lateral movement, the velocity of migration undergoes a sharp increase, with the mean velocity for lateral migrating hemocytes being 1.8 μm/min (±0.8; = 46), as opposed to 1.1 μm/min (±0.5; = 35) for hemocytes within the midline. Laterally migrating hemocytes not only move faster but also have a more constant speed and rarely stall during the migration, in contrast with hemocytes in the ventral midline clusters, which often pause between periods of intense movement before they migrate laterally or into another cluster (). The increase in speed and directionality of hemocytes during their lateral migration suggests that during this rapid movement, these cells are highly polarized. Actively migrating cells exhibit polarity by extending protrusions in the direction of migration, forming a persistent leading edge with an increased protrusive activity when compared with that of the cell rear. To quantify the polarity of these migrating hemocytes, we artificially divided individual hemocytes into four quadrants (corresponding to anterior lateral, anterior medial, posterior lateral, and posterior medial) and measured the protrusive area in each of these quadrants. We found that during the lateral movement, the combined lateral protrusive area of a migrating hemocyte was, on average, 212 μm (±49; = 6), which is 10 times larger than that of the medial side (mean of 21 μm [±10]; ). This demonstrates that these hemocytes are highly polarized along the medial lateral embryonic axis, displaying a robust and persistent leading edge at their lateral aspect throughout the lateral migration. In contrast, when the same quantitative analysis was performed on hemocytes that remain in the ventral midline, no such persistent polarization was observed. Although cells from the midline are able to extend protrusions similar in dimension to those described for hemocytes migrating laterally, these protrusions rarely remain in the same cell quadrant for >2 min and, consequently, the cell randomly oscillates between different states of polarity (). This oscillation correlates with the cell's seemingly random movement as it patrols a small, defined area within the midline. It has previously been reported that hemocyte developmental migrations are dependent on the expression of the PVR tyrosine kinase (PVR in hemocytes) and that the three PDGF/VEGF ligands—Pvf1, -2, and -3—act redundantly as chemotactic factors, directing the migration of hemocytes throughout the embryo (). However, a recent study has suggested that the main role of Pvf signals during hemocyte development is to function as survival factors, as hemocyte-specific expression of the pan-caspase inhibitor p35 in mutants is sufficient to largely restore the hemocyte defects normally observed in mutant embryos (). Despite these results, evidence still exists that Pvf ligands may be acting as chemoattractants to hemocytes (). To address more thoroughly the role played by Pvf ligands during hemocyte migrations along the ventral midline, we first observed the detailed expression pattern of all three Pvf family members within this region of the embryo. Consistent with previous studies (), we found that Pvf2 and -3, but not Pvf1, were expressed in the embryonic ventral midline. However, a detailed time course showing the expression pattern within the ventral midline reveals that the timing of expression of these two genes is different (). Pvf3 is expressed along the ventral midline at stage 10, when the germ band is fully extended. This expression subsequently decreases and is almost undetectable by stage 14 (). In contrast, Pvf2 expression is absent at stage 10 and is only observed from stage 12 onward. Expression appears strongest at stage 14, and over the next 2 h of development, RNA levels in the CNS decrease in a wave from anterior to posterior such that by stage 15 the ligand is expressed in small, segmentally reiterated points along the posterior region of the ventral nerve cord (). These points of expression correspond to the locations at which hemocytes cluster in the midline at this stage (). We also observed strong expression of Pvf2 in the developing dorsal vessel of stage 14 embryos (), whereas no expression of Pvf3 or -1 was detected in this tissue (). To further investigate the role of the Pvf ligands, we analyzed in detail hemocyte developmental migrations in mutants. As previously described, mutants exhibit a severe hemocyte migration defect in which the cells are unable to migrate from their origin in the head and undergo apoptosis (; ). This defect can be rescued by the hemocyte-specific expression of the pan-caspase inhibitor p35 to prevent apoptosis of these cells (). We observed that these rescued hemocytes not only are unable to infiltrate the germ band as previously shown () but also fail to migrate posteriorly from the head along the CNS and the dorsal vessel (). Because Pvf2 is strongly expressed in both the ventral nerve cord and the dorsal vessel at the same developmental stages that hemocytes are found migrating along these structures, it seemed likely that Pvf2 acts as a chemoattractant to pull hemocytes along these two migration routes. To test this hypothesis, we analyzed hemocyte migration in the mutant line , a homozygous viable [ ] insertion in the Pvf2 gene (). Consistent with a chemotactic role for Pvf2, we found that these mutants display a complete absence of hemocytes along the dorsal vessel but, curiously, showed only a reduction in hemocyte numbers migrating along the ventral midline (). Because Pvf3 is expressed in the ventral nerve cord at earlier stages, we reasoned that these two genes may act redundantly to attract hemocytes along the midline. To test for redundancy, we used RNAi to inactivate Pvf2 and -3 simultaneously. As a control, we first confirmed that injection of double-stranded RNA (dsRNA) against Pvf2 created the same phenotype as the mutant with injected embryos, displaying a lack of hemocytes along the dorsal vessel and reduced numbers of hemocytes on the ventral midline (). Inactivation of Pvf3 alone by RNAi had little effect on hemocyte migration (), but inactivation of both Pvf2 and -3 prevented hemocyte migration along both the dorsal vessel and the ventral nerve cord, mimicking the phenotype observed in the rescued p35-expressing PVR mutant embryos (). The pattern of Pvf2 expression during embryogenesis coincides spatially and temporally with the migration of hemocytes, and our RNAi studies, along with the observation that ectopic expression of Pvf2 can redirect migration (), demonstrate this ligand to be operating as a hemocyte chemoattractant. The processive down-regulation of Pvf2 RNA from anterior to posterior in the ventral midline observed between stages 14 and 15 correlates temporally with the wave of lateral hemocyte movement occurring during wild-type development. Because Pvf2 is acting as a chemoattractant pulling hemocytes toward the midline, it is possible that the decrease in expression of this ligand is responsible for the lateral migration of a subset of hemocytes away from the midline. To test this hypothesis, we overexpressed Pvf2 in the ventral midline using the CNS midline driver (). Double staining of these embryos with the hemocyte-specific anti-croquemort antibody and anti-armadillo showed that the lateral migrations of hemocytes were severely disrupted such that at stage 15 of development, few hemocytes could be seen occupying more lateral positions (), demonstrating that a reduction in Pvf2 expression is necessary for normal lateral migrations. Interestingly, by stage 16, these embryos had recovered the defect, and hemocytes could be seen in the usual three lines running anterior to posterior along the midline (). We have previously shown that embryonic hemocytes rapidly chemotax toward an epithelial wound () in a process resembling the vertebrate inflammatory response. To determine whether Pvf signals play a chemotactic role during this inflammatory response, we made laser ablations to mutants expressing p35. 1 h after wounding, these embryos show a robust wild-type response from the mutant hemocytes, demonstrating that wound chemotaxis is independent of PVR expression (, A and B; Rorth, P., personal communication). To further study the wound chemotactic response, we developed a wounding assay using beads that allow the treatment of the wound region with inhibitory drugs. Such beads are routinely used in embryonic studies for the local application of chemical inhibitors in vivo (). Implanting a bead into a embryo creates an epithelial wound approximately the same diameter as the bead (). As is the case with laser-induced wounds, application of untreated beads to a stage 16 embryo leads to rapid accumulation of hemocytes at the wound site until, by 30 min, they surround the bead (). Like unstimulated midline hemocytes, these cells exhibit large dynamic membrane ruffles and filopodia as they surround and seemingly try to collectively phagocytose the invasive bead ( and Video 2, available at ). These actin-rich protrusive structures are one of the most distinctive features of moving cells and are generally considered to be critical for single-cell migration. To directly test the requirement of cytoskeletal protrusions during hemocyte chemotaxis, we presoaked beads in one of two actin polymerization blocking drugs, Cytochalasin D or Latrunculin B, before bead implantation and examined hemocyte recruitment in embryos 1 h after bead insertion. We found that treatment with either drug blocks this chemotactic response so that drug-treated beads show little or no hemocytes in direct contact with the bead and those hemocytes close to the bead exhibit small or no actin protrusions and are much less motile when compared with their wild-type counterparts (, F and G; and Video 2). In both cases, effects of drug treatment can be seen up to 50 μm away from the bead, as hemocytes that lie further away from the bead exhibit normal dynamic lamellipodia (). We assume that this reflects a drop off in drug concentration as a result of drug diffusion away from the bead. For a hemocyte to chemotax toward a chemotactic source, it has to be able to rapidly sense a chemotactic gradient and polarize accordingly. Because PI3K has been shown in several systems to be a key mediator for cell polarization and directed cell migration (; ; ), we tested the requirement of PI3K for the directed migration of ventral midline hemocytes during development and their chemotaxis toward wounds. Confocal analysis of embryos expressing a dominant-negative form of the PI3K catalytic subunit (), specifically in the hemocytes, showed normal hemocyte distribution at all stages of development, and these cells appeared morphologically indistinguishable from their wild-type counterparts exhibiting dynamic lamellipodia and filopodia (). The same result was observed when the PI3K inhibitor LY294002 was injected into the vitteline space of wild-type embryos during hemocyte migration (unpublished data). To further characterize any possible role for PI3K during these developmental migrations, we performed the same quantitative tests on the ventral midline hemocytes as on wild type and found that -expressing hemocytes migrate along the ventral midline in a pattern identical to that seen in wild-type embryos (unpublished data), demonstrating velocity (while migrating laterally, 1.6 ± 0.5 μm/min [ = 39]), directionality (while in midline, 33 ± 17.5% [ = 28], and while migrating laterally, 89 ± 8.9% [ = 39]), and polarity (mean lateral protrusive area of 250 vs. 16 μm for the medial side) equal to wild-type cells, proving that PI3K is not required for these cells to sense and polarize toward the guidance cues that control their developmental dispersal. Though clearly not required for hemocyte developmental migrations, we wanted to test whether PI3K was required for hemocytes to polarize and chemotax toward a wound site. To test the involvement of PI3K in this process, we made laser ablations in embryos expressing in the hemocytes. In contrast to the wild-type chemotactic response (), the mutant cells failed to chemotax toward the wound site and the wound remained largely undetected by the hemocytes up to 1 h after laser ablation (). The same result was obtained when beads were implanted in these embryos (). To ensure that this phenotype was due to a failure in chemotaxis and not an indirect result of a reduction in hemocyte survival, we counted the number of hemocytes on the ventral side of embryos expressing . We found no significant difference in hemocyte number when compared with wild-type embryos (mean number, 49 ± 9 [ = 17] in and 56 ± 8 [ = 25] in wild type), demonstrating that PI3K plays no significant role in hemocyte survival at this stage of development. This is consistent with previously published data showing that hemocyte-specific expression of does not cause the hemocyte aggregation that is associated with a decrease in cell survival (). To further verify our chemotaxis result, we implanted two beads into embryos in which the hemocytes were expressing GFP but were otherwise wild type. In these experiments, one bead was presoaked in the PI3K inhibitor drug LY294002 () diluted in DMSO before implantation, and the control was presoaked only in DMSO. 1 h after bead implantation, as expected, the control DMSO bead was surrounded by hemocytes, demonstrating that DMSO alone does not block the chemotactic response of the hemocytes. However, no cells chemotaxed toward the bead soaked in PI3K inhibitor (). Time-lapse confocal analysis of hemocytes exposed to LY294002 shows that these cells are actively motile and able to form large, dynamic actin protrusions that look indistinguishable from untreated ventral midline hemocytes that have not been exposed to a wound. However, despite appearing morphologically normal, they are clearly unable to sense the position of the bead and chemotax toward it, even when the bead is implanted only 30 μm away from a large population of hemocytes ( and Video 3, available at ). In this study, we have analyzed in detail the developmental dispersal of hemocytes during embryogenesis. We have found that migration of hemocytes along the dorsal vessel requires a chemotactic Pvf2 signal, and migration along the ventral nerve cord is dependent on and orchestrated by chemoattractant signals from both Pvf2 and -3 expressed in this tissue. It was previously suggested that the migration routes traversed by all hemocytes during development were controlled by all three Pvf ligands operating redundantly as guidance cues (), and our data support a chemotactic role for both Pvf2 and -3. However, a recent study suggests that the primary role of Pvf signaling in embryonic hemocytes is to control anti-apoptotic cell survival (). It is becoming clear that Pvf ligands may well act on hemocytes as both chemoattractants and survival factors. This in itself is not a new idea, and the close relationship between guidance and survival has already been shown to exist in , where the overexpression of the germ cell chemorepellent Wunen causes excessive germ cell death (). It remains to be seen whether all three Pvf ligands can act as both chemoattractants and survival factors or whether each ligand plays a different role, with Pvf2 and -3 acting primarily as chemoattractants and Pvf1 operating as a survival factor. Many obvious parallels exist between the migration of hemocytes along the ventral midline CNS and another developmentally regulated migration in , that of border cell migration. Border cells take ∼6 h to migrate a distance of 100 μm, a speed consistent with that we describe during hemocyte migration along the CNS. Successful border cell migration, like hemocyte migration, requires the expression of the Pvr in the migrating cells () and, just as we see for hemocytes, the chemotactic signals detected by the PVR in the border cells are not transduced through PI3K (). Successful migration of border cells does, however, require Rac signaling and the Rac activator (), the homologue of Dock 180 (). It has been previously shown that hemocyte-specific expression of dominant-negative Rac disrupts all hemocyte developmental migrations, demonstrating that Rac is required for the successful migration of ventral midline hemocytes along the CNS (; ). Given that Pvr couples to the Dock 180 signaling pathway during border cell migration and that Dock 180 has been shown to be involved in the migration of lymphocytes (; ), Mbc/Dock 180 is a potentially important protein for hemocyte migration. Despite the fact that mutant embryos display a grossly normal pattern of hemocyte dispersal (), it would be interesting to look in detail at the migration of these mutant cells along the ventral nerve cord. More work is needed to investigate what other similarities may exist between border cell migration and ventral midline hemocyte migration. During development, only a subset of the hemocytes present in the embryo respond to the midline Pvf expression and migrate along the CNS accordingly. Other cells follow other migratory pathways. What specifies these cells to migrate along the midline? Important studies in border cell migration have shown that the JAK–Stat signaling pathway signaling through the Domeless receptor (Dome) is necessary and sufficient to transform nonmotile epithelial cells into invasive ones (; ; ). Whether a similar signaling mechanism is operating to specify future ventral midline hemoctyes and initiate their migration remains to be seen. We demonstrate that from stage 14 onwards, once hemocytes occupy the entire ventral midline, individual cells begin to rapidly leave the midline and occupy more lateral positions. At this stage of development, hemocytes appear to be highly polarized, exhibiting large lamellipodia at their leading edges and migrating at a speed more than three times faster than their earlier midline migration. We have shown that this lateral movement requires a down-regulation in the attractive signal provided by Pvf2 in the midline, but is this the only driving force for the lateral movement? One possibility is that a different source of chemoattractant exists in the more lateral positions and that once Pvf2 expression is sufficiently down-regulated, this chemoattractant source operates to pull hemocytes laterally. Alternatively, hemocytes may be actively repelled from the midline or from one another, and the lateral migration observed by a subset of these hemocytes is a consequence of these cells attempting to maximize the distance between one another while maintaining contact with the CNS. It remains to be seen which, if any, of these hypotheses is true, but what is certain is that the guidance of hemocytes along the ventral midline of the embryo is not as simple as was first thought, and more studies are required to determine the exact relationship between this subpopulation of hemocytes and the different structures within the CNS as well as the overlying ectodermal cells, any of which could provide either chemoattractants or repellents for the migrating hemocytes to respond to. We have demonstrated a requirement of PI3K for the polarization and active chemotaxis of hemocytes toward an epithelial wound. This is the first demonstration of the role of PI3K for single-cell chemotaxis in and shows a striking correlation with the mechanism of cell chemotaxis used by and mammalian neutrophils (; ; ). In these model systems, class I PI3Ks are activated upon stimulation of G protein–coupled chemoattractant receptors and, once activated, PI3Ks catalyze the production of the phosphoinositides PIP and PIP at the leading edge of the cell. The accumulation of PIP/PIP leads to a rapid and transient recruitment of pleckstrin homology domain–containing proteins, including the serine/threonine kinase Akt/PKB (; ). Akt/PKB itself becomes activated upon recruitment to the membrane and, in , activates the serine/threonine kinase p21-activated kinase a, which eventually leads to the phosphorylation of Myosin II and subsequent polarization of the cytoskeleton (). Evidence also exists to support a role for the PI3K antagonist PTEN in helping to establish and maintain the intercellular PIP gradient required for successful chemotaxis by down-regulating the PIP pathway at the rear of the migrating cell (; ). How much of this signaling pathway is operating in chemotaxing hemocytes remains to be seen. Our current study demonstrates the involvement of PI3K, and previous work has shown that the small GTPase Rac is required for efficient hemocyte chemotaxis toward wounds. In neutrophils, PIP production has been shown to be autocatalytic and to require Rac but not Cdc42 (; ). In the proposed positive feedback loop, it is thought that PIP may stimulate Rac through activation of a specific Rac GEF, which in turn activates PI3K, as well as effectors that mediate lamellipodial protrusion (). Because Rac is absolutely required for hemocyte chemotaxis and lamellipodia formation, it is tempting to speculate that a similar feedback loop may be operating in hemocytes. Further work is required to determine the complex relationships operating among PI3K, Rho family small GTPases, and the actin cytoskeleton that coordinate chemotactic migration in these highly motile cells. The PI3K-dependent mechanism of polarization required for hemocyte chemotaxis toward a wound is extremely fast and perfectly suited for mature, highly motile hemocytes that need to rapidly react to a source of attractive signal, be it a wound, an invading organism, or an apoptotic cell. In contrast, the mechanics to developmentally disperse need not be so rapid, as the aim during development is simply to ensure that hemocytes migrate toward and arrive at their target tissue in a given amount of time and does not require the rapid response to constantly changing environments required for mature hemocytes. The mechanism controlling the developmental migration of hemocytes along the ventral midline is consequently much slower and is dependent on slow-diffusing growth factors of the Pvf family providing short-range guidance information signaling through the receptor tyrosine kinase PVR. These two mechanisms may not be the only ways in which hemocytes are able to chemotax toward an attractive source; indeed, the observation that hemocytes travel different migratory routes in the embryo suggests that they may not all be using the same machinery to polarize and migrate. What does seem to be consistent for both chemotaxis toward developmental signals and toward wounds, like motility in many cell types, is a requirement for Rac signaling and the formation of actin protrusions. The fact that hemocyte migrations within the embryo are strictly regulated and adhere to a stereotyped pattern is important in a developmental context. Throughout embryogenesis, hemocytes carry out important developmental functions within the embryo, such as the engulfment and removal of apoptotic cells () and the laying down of many extracellular matrix molecules, including collagen IV and laminin, that compose the basement membrane surrounding internal organs (). The failure of hemocytes to travel along their normal migratory routes therefore has serious consequences. Such defects have been described in mutants, where a lack of hemocyte migration along the ventral nerve cord results in a failure in CNS condensation (), as well as a disruption in axon patterning (). It is therefore vital for the embryo to ensure that hemocytes arrive at their correct target tissues during development. For this to occur, it is not sufficient to allow these cells to passively disperse throughout the embryo by random migrations; instead, a directed and tightly controlled migration is required. In this study, we have locally applied drugs to embryos using bead implantation. The application of drugs has been a powerful tool in cell culture and in vitro cell motility studies but remains largely unused in Using our bead assay, we will be able to take advantage of the many useful drugs available to block both specific signaling pathways as well as important cytoskeletal processes. Combined with the powerful genetics available in and the relative ease of live imaging in this system, the study of hemocytes provides a powerful model to address the process of cell motility and chemotaxis and will undoubtedly provide a clearer understanding of the regulation and mechanics of single-cell migration in the complex setting of a multicellular organism. For live studies, GFP was expressed in hemocytes using either the hemocyte-specific Gal4 line (; ) or (; ). To visualize both hemocytes and epithelial cells, we used a stock homozygous for () on the second chromosome and homozygous for a recombined chromosome carrying both () and on the third. To drive sufficient expression of both dominant-negative PI3K and GFP in embryonic hemocytes, () flies were crossed to flies, generating + progeny. To visualize hemocyte motility in rescued p35-expressing mutant embryos (), flies were crossed to a stock to generate / + embryos. For Pvf2 overexpression experiments, we crossed flies to (also known as ; ) and performed antibody stainings on the resulting embryos. The () line was used for mutant analysis. Embryos were dechorionated before being transferred to double-sided tape stuck to a slide. Specimens were dehydrated for 6 min before being covered with Volatalef oil 10S. Heparin beads (Sigma-Aldrich) or Affigel blue beads (Bio-Rad Laboratories) were soaked for 3 h in PI3K inhibitor LY294002 (Calbiochem) dissolved in DMSO at a concentration of 100 mM, Latrunculin B (Calbiochem) dissolved in DMSO at a concentration of 10 mM, or Cytochalasin D (Sigma-Aldrich) dissolved in ethanol at a concentration of 10 mM, before being implanted into the embryo using a sharpened tungsten needle. Embryos were subsequently mounted under a coverslip and imaged. Live imaging was performed using a confocal system (LSM510; Carl Zeiss MicroImaging, Inc.) mounted on an Axiovert 100M (Carl Zeiss MicroImaging, Inc.), and the resulting time-lapse series were assembled and analyzed using ImageJ imaging software (NIH). Embryos were mounted as previously described (), and videos were made at room temperature using a Plan-NEOFLUAR 40×/1.3 objective (Carl Zeiss MicroImaging, Inc.). Cell tracking was performed using an ImageJ plugin (Manual Tracking) on maximum projections of eight slices that represent 20 μm of the ventral surface of the embryo. For each time point, the center of the cell body tracked was manually highlighted, and through its coordinates mean velocity was calculated. In situ hybridization of whole-mount embryos was performed with single-stranded digoxigenin-labeled RNA probes and alkaline phophotase immunochemistry. A 755-bp DNA fragment was generated using primers 5′CGAAACGCAAATGGAATTGTAAAGC and 5′CTCCGATTTGGCATCATTGGGTTCC, cloned into pCR-II TOPO vector (Invitrogen) and used as a template for Pvf3 probe production. pVegf27Cb () was used as a template for Pvf2 probe production, and template DNA for Pvf1 probe was a gift from A. Hidalgo (University of Birmingham, Birmingham, UK). Embryos were mounted in 50% glycerol and imaged using an HC PL FLUOTAR 20×/0.5 objective (Leica). Pictures were taken using a camera (DC500; Leica) mounted on a microscope (DM5000B; Leica). For the synthesis of Pvf2 dsRNA, a 688-bp region was amplified from pVegf27Cb () using the primers 5′TCCACATCACGAGAGAAACGAACAC and 5′TTTTGCCATTCTGACGTTTTGACTG. For Pvf3, a region of 498 bp was PCR amplified from pVegf27Ca using primers 5′TCCACATCACGAGAGAAACGAACAC and 5′TTTTGCCATTCTGACGTTTTGACTG. Primer pairs also contained the T7 promoter sequence at their 5′ ends. The PCR products were used as templates for the T7 transcription reactions with the T7 RiboMax large scale production kit (Promega). The dsRNA was dissolved in injection buffer at a final concentration of 2 μg/μl and injected into 0–1-h hold embryos. When both dsRNAs were injected, the concentration of each was kept at 2 μg/μl. Injected embryos were left to develop at 25°C, fixed in 4% formaldehyde before being hand devitellinized, and stained using antibodies as described above. Video 1 shows wild-type hemocyte migration at the embryonic ventral midline. Video 2 shows the effect of Cytochalasin D on hemocyte chemotaxis toward wounds. Video 3 demonstrates the effect of the PI3K inhibitor LY294002 on the same process. Online supplemental material is available at .

Vascular remodeling is a critical part of the pathogenesis of clinically important vascular disorders such as atherosclerosis, restenosis after angioplasty, and saphenous vein graft disease (; ). Despite considerable study, the molecular mechanisms that control vascular smooth muscle cell (VSMC) activities during vascular remodeling are not fully understood. Recent reports linking cadherins to VSMC regulation (; ; ) suggest that these transmembrane adhesion proteins, characterized extensively as major mediators of epithelial cell homeostasis, may also be important in vascular remodeling. Cadherins are involved in Ca-dependent cell–cell adhesion, intracellular junction assembly, and tissue morphogenesis during development (; ; ). Major subdivisions of the large cadherin superfamily include the classical cadherins and the protocadherins (; ; ). The extracellular domains of these proteins share a unique structure, the cadherin motif, which is repeated in tandem in variable numbers. Classical cadherins function as homophilic adhesive molecules, and both extracellular and cytoplasmic domains contribute to this function. Classical cadherin cytoplasmic domains interact with β-catenin and plakoglobin (; ), members of the gene family of transcription factors. This interaction effectively sequesters β-catenin away from the nucleus, limits its transcriptional activity (; ; ), and thus links cadherins to the canonical Wnt signaling pathway, a major determinant of cellular activity during development (; ; ). We identified the protocadherin Fat1 in a screen for molecules expressed differentially after balloon injury of rat carotid arteries. Like classical cadherins, protocadherins have extracellular domains capable of Ca-dependent, homophilic interaction (). Protocadherin cytoplasmic domains, on the other hand, are structurally divergent from those of the classical cadherins, and less is known about their function. Sequestration and inhibition of β-catenin by protocadherins has not been described. Although mammalian Fat1 genes (; ; ) were initially characterized as homologues of the protein Fat (), recent bioinformatics analysis indicates that Fat1 is more closely related to Fat-like (Ftl) (). In , Ftl is expressed apically in luminal tissues such as trachea, salivary glands, proventriculus, and hindgut (). Silencing of results in the collapse of tracheal epithelia, and it has been suggested that Ftl is required for morphogenesis and maintenance of tubular structures of ectodermal origin. Like Fat and Ftl, mammalian Fat1 is remarkable for its very large size (∼4,600 aa). It has a huge extracellular domain that contains 34 cadherin repeats, 5 EGF-like repeats and l laminin A-G motif, a single transmembrane region, and a cytoplasmic tail of ∼400 aa (). Sequences within the Fat1 intracellular domain (Fat1) show limited similarity to β-catenin binding regions of classical cadherins (). Our studies show that Fat1 expression increases after injury of the rat carotid artery, and is positively regulated in cultured VSMCs by several factors that promote cell proliferation and migration. Interestingly, knockdown of Fat1 expression limits VSMC migration, but enhances VSMC growth. This anti-proliferative effect of Fat1 appears to be mediated by Fat1 sequences because expression of a fusion protein containing the Fat1 inhibits cyclin D1 expression and cell growth. Moreover, the Fat1 can interact with β-catenin, prevent its nuclear translocation, and limit its transcriptional activity on both synthetic and native β-catenin–responsive promoters, including that of cyclin D1, a known target of canonical Wnt signaling. These findings point to an integrative role for Fat1 in regulation of critical VSMC activities in which it promotes migration and limits both canonical Wnt signaling and VSMC growth in the remodeling artery. We quantitated mRNA expression by quantitative PCR (qPCR) of cDNA samples from normal and injured rat carotid arteries. Compared with uninjured arteries, mRNA expression was ∼8.5-, 13.0-, and 3.9-fold higher than control at 3, 7, and 14 d after injury, respectively (). To localize Fat1 protein expression in injured arteries, we characterized rabbit antisera raised against a GST-Fat1 immunogen. Immunoblotting of VSMC lysates with one such antiserum, but not preimmune serum, yielded a single high molecular weight band of ∼500 kD, in accord with the predicted size of full-length Fat1 (). Further specificity was demonstrated in RNAi experiments directed against multiple separate targets in the mouse Fat1 sequence (see and ). We then used the antiserum for immunohistochemical studies. As shown in , prominent Fat1 staining appeared in the media 3 d after injury, while at 7 and 14 d after injury Fat1 staining was less evident in the media, but clearly present in the developing neointima. Western analysis of Fat1 expression in the carotid artery injury model, like our qPCR findings, showed a clear induction after injury (unpublished data). To correlate Fat1 expression with the proliferative status of specific cells, we co-stained sections for Fat1 and the proliferation marker PCNA. Although some cells appeared positive for both, we also noted some spatial separation of the signals, particularly evident in areas with limited neointimal formation, which showed prominent Fat1 staining without PCNA (, top right). The latter observation raised the possibility that, despite its overall induction after injury, increased Fat1 expression might have negative effects on VSMC growth in vivo. To identify factors that might contribute to Fat1 induction after arterial injury, we characterized its expression in primary cultured VSMCs. Quiescent rat aortic smooth muscle cells (RASMCs) (time 0 h) were treated with 10% FBS for 2, 6, 12, 18, 24, and 36 h, and the level of Fat1 protein was determined by Western analysis. The Fat1 signal increased strongly between 2 and 12 h and remained elevated through 36 h (). To assess cell cycle status, we also checked cyclin D1 expression in these lysates. Interestingly, Fat1 induction preceded the increase of cyclin D1, a mediator of progression through the G1 phase of the cell cycle (). We then assessed Fat1 expression in response to several factors known to affect the vascular response to injury. Western analyses showed that expression of Fat1 increased in response to Angiotensin II (ATII), basic FGF (bFGF), and PDGF-BB (). Increased Fat1 expression was apparent by 2 h and sustained at high levels from 12 to 36 h after stimulation with each of these factors. Thus, Fat1 expression is regulated consistently and strongly by multiple factors known to promote VSMC growth and migration. Two recent studies have described a role for Fat1 in regulation of epithelial cytoskeletal actin dynamics, planar polarity, and migration, mediated through interactions of the Fat1 cytoplasmic domain with proteins of the Ena/VASP family (; ). Fat1 induction by known VSMC chemotactic factors () suggested that Fat1 might also be involved in VSMC migration. To test this and other potential Fat1 functions, we developed reagents to effectively manipulate Fat1 expression. Transfection of mouse aortic smooth muscle cells (MASMCs) with Fat1 specific small interfering RNAs (siRNAs), but not scrambled or mismatch derivatives, resulted in significantly decreased levels of Fat1 protein (). To isolate and augment signals mediated by the Fat1, we generated a cDNA construct, IL2R-Fat1, in which the entire Fat1 cytoplasmic domain was fused to the extracellular domain and transmembrane region of the interleukin 2 receptor α-chain (IL2R), with or without a COOH-terminal FLAG epitope tag (). Subcellular localization of this fusion protein was tested in 3T3 cells, which do not express detectable Fat1, and A7r5 VSMCs, which express moderate amounts of endogenous Fat1; both transfected 3T3 and A7r5 cells showed an appropriate cell surface signal when stained with anti-FLAG epitope antibody ( and unpublished data). Cell migration in monolayers treated with specific Fat1 siRNA was modestly but significantly decreased compared with control siRNA (), which indicates that Fat1 expression is required for optimal VSMC migration. Surprisingly, we also found decreased migration of VSMCs expressing the IL2R-Fat1 protein in a Transwell assay using FBS as a stimulant in the lower chamber (). We confirmed both expression of Ena/VASP proteins in VSMCs and the ability of the IL2R-Fat1 protein to interact with these signaling intermediates (unpublished data). We surmise that although the IL2R-Fat1 construct may increase intracellular Fat1 signaling, it also dissociates Fat1 extracellular interactions from this intracellular signaling, and thus interferes with directional migration. Altogether, these findings indicate that Fat1 promotes VSMC migration; it is likely that, as described in epithelial cells, interactions with Ena/VASP proteins link Fat1 expression to VSMC cytoskeletal actin reorganization, polarization, and migration. In addition to increased migration, the VSMC response to injury is characterized by cell cycle entry and increased proliferation (). To evaluate how Fat1 induction after injury might affect VSMC growth, we tested the effect of Fat1 knockdown on expression of cyclin D1, a marker of cell cycle activation. Four distinct mouse Fat1 siRNA duplexes attenuated endogenous Fat1 levels in MASMCs; with each duplex, we also observed a significant increase in cyclin D1 expression over control levels ( and unpublished). The similarity of effect achieved by multiple distinct siRNAs argues strongly that increased cyclin D1 expression results from decreased Fat1, and not an off-target effect. The duration of Fat1 inhibition was more than 90% at 2 and 3 d after transfection, with persistent and strong inhibition still apparent after 6 d (). Decreased Fat1 expression corresponded to increased cyclin D1 signal at each time point (2.0–2.5-fold increase of cyclin D1/actin ratio vs. control), suggesting that endogenous Fat1 exerts a tonic inhibitory effect on cyclin D1 expression (). The level of total β-catenin in these cells, by contrast, showed little change. We then examined the effect of Fat1 knockdown on DNA synthesis. Cells were transfected with Fat1 or control siRNA, and then were serum deprived for 48 h before stimulation with 10% FBS and evaluation of BrdU incorporation. In Fat1 knockdown cultures, the fraction of BrdU-positive cells was significantly higher than in control siRNA cells (52 ± 7% vs. 30 ± 8%, P < 0.05) (). These findings indicate that decreased Fat1 expression promotes cell cycle progression and DNA synthesis in VSMCs. Classical cadherins interact with intracellular signaling pathways through their cytoplasmic domains (). To establish cell populations differing primarily in their expression of the Fat1, we transferred the IL2R (without cytoplasmic domain) and IL2R-Fat1 constructs into the GFP-RV retroviral vector (), produced viral supernatants, and transduced A7r5 and primary MASMCs. Additional control cells, denoted RV, were produced using the unmodified GFP-RV vector. Western analysis confirmed IL2R-Fat1 expression in A7r5 and MASMCs (). Interestingly, endogenous cyclin D1 levels were lower in both A7r5 and MASMCs expressing IL2R-Fat1 (). In cell growth assays over 7 d, A7r5 cells expressing IL2R showed no significant change from control RV cells, but decreased cell numbers were evident in the IL2R-Fat1 at all time points after 3 d (). In addition, both A7r5 and MASMCs expressing the IL2R-Fat1 construct showed significantly lower fractions of BrdU-positive nuclei, indicating that this decrease in cell number reflected growth inhibition rather than decreased survival (). In epithelial cells, classical cadherins such as E-cadherin regulate Wnt signaling activity by physically associating with β-catenin at points of cell–cell contact (). The sequences, interacting proteins, and functions of protocadherin cytoplasmic domains are typically thought to be divergent from those of the classical cadherins (), and Fat1 is not regarded as part of the classical cadherin system (). Nevertheless, we found that the Fat1 has growth inhibitory activity, and that expression of cyclin D1, a known target of the canonical Wnt signaling pathway, correlated negatively with Fat1 expression. Together, these findings suggested that growth inhibition by Fat1 might involve β-catenin. In our immunofluorescent analyses of RASMCs (), Fat1 localized to both cell–cell junctions and cellular free edges, while β-catenin was concentrated at sites of cell–cell contact. By two color immunofluorescence analysis, we found areas along cell–cell junctions where the two signals overlapped (). This overlap did not include the cellular free edges, where Fat1 alone was seen (). Junctional β-catenin and Fat1 have been identified in epithelial cells that display apical–basal polarity, but it is thought that the two proteins occupy distinct domains, with β-catenin at apical adherens junctions and Fat1 at basolateral points of cell–cell contact (, ). VSMCs are nonpolarized (), so this model of apical–basal domain specialization may not apply. To test directly if Fat1 and β-catenin can interact at physiologic levels of expression in VSMCs, we immunoprecipitated endogenous Fat1 and tested for recovery of β-catenin. Both this assay and reciprocal coimmunoprecipitations of β-catenin followed by immunoblotting for Fat1 demonstrated interaction of the two proteins (). This finding suggests that the nonpolarized nature of VSMCs allows for protein–protein interactions not found in polarized cell types such as epithelial cells. Further immunoblotting of Fat1 immunoprecipitates with a pan-cadherin antibody did not reveal associated (classical) cadherins that might associate with both Fat1 and β-catenin (unpublished data). To characterize the Fat1–β-catenin interaction further, we used coimmunoprecipitation assays in cotransfected 293T cells to map the sequences required for interaction. We generated a series of constructs bearing deletions within the Fat1 portion of the IL2R-Fat1-3XFLAG (). IL2R-E-cadherin-3XFLAG and IL2R-3XFLAG (containing no Fat1 sequences) constructs served as positive and negative controls, respectively. We confirmed the expression of Myc-tagged β-catenin and FLAG-tagged fusion proteins, and immunoprecipitation of transfected Myc-tagged β-catenin (, bottom panels). Interaction of β-catenin with the IL2R-Fat1-3XFLAG derivatives was assessed by immunoblotting with FLAG antibody (, top). A robust FLAG signal was obtained with the IL2R-Fat1-3XFLAG construct containing the complete Fat1 domain and with derivatives I, III, and V. Weaker signals were seen with constructs II and IV, which lack the FC1 and both FC1 and FC2 domains, respectively. Although these findings based on overexpressed proteins must be interpreted with caution, they suggest that β-catenin interacts with the Fat1 principally through the FC1 domain, but leave open the possibility that the FC2 domain or additional sequences also contribute to the interaction. Interestingly, the E-cadherin–based positive control yielded a comparatively strong band, despite input of substantially less protein. Thus, changes in Fat1 or Fat1 expression affected expression of a β-catenin target gene, cyclin D1, but had little effect on overall β-catenin levels ( and ). Having found evidence for colocalization and interaction of β-catenin and Fat1 in VSMCs, we postulated that Fat1 might be acting like a classical cadherin to affect the subcellular localization and activity of β-catenin. We examined this first using immunocytochemistry. Expression plasmids encoding IL2R or IL2R-Fat1 were introduced into VSMCs, which were subsequently treated with 20 mM LiCl for 6 h to activate Wnt signaling and promote nuclear translocation of β-catenin (). The intensity of nuclear β-catenin staining did not appear to be affected by expression of IL2R (, top, arrows). In contrast, nuclear accumulation of β-catenin appeared decreased in the IL2R-Fat1–expressing cells (, bottom, arrows) as compared with untransfected cells. To assess this effect in a more quantitative way, we determined the distribution of β-catenin in the membrane, cytoplasmic, and nuclear fractions of IL2R-GFP-RV– and IL2R-Fat1-GFP-RV–transduced VSMC cultures treated with LiCl. As shown in , immunoblotting showed a relative decrease in nuclear β-catenin accumulation in cells expressing IL2R-Fat1 as compared with those expressing IL2R (respective nuclear β-catenin/lamin A/C ratios 0.8 [IL2R-Fat1] vs. 1.65 [IL2R]). To assess further the functional significance of the Fat1–β-catenin interaction in VSMCs, we tested the effect of Fat1 overexpression on β-catenin–mediated transcription. A7r5 cells were cotransfected with β-catenin and/or IL2R-Fat1, along with the TCF-luciferase reporter construct Topflash or its negative control, Fopflash (). Topflash reporter activity reflects activation of the canonical Wnt signaling pathway, β-catenin nuclear translocation, and formation of TCF/β-catenin heterodimers; Fopflash contains mutated TCF binding sites and serves as a control for nonspecific activation (). A full-length N-cadherin cDNA and the IL2R–E-cadherin construct were also tested as controls. Specific activation of Topflash by β-catenin was ∼10-fold above basal levels, and the three test constructs all inhibited this activation significantly. Interestingly, the inhibition due to both IL2R-Fat1 (40%) and N-cadherin (55%) was less complete than that resulting from cotransfection of IL2R-E-cadherin, which abolished all β-catenin–mediated transactivation. We also evaluated the effect of decreased Fat1 expression. Immunocytochemistry of LiCl-stimulated MASMCs suggested a relative enhancement of nuclear β-catenin staining in Fat1-depleted cells (). To assess this observation more quantitatively, we transfected MASMCs first with control or Fat1-specific siRNA, and then with the Topflash reporter. As shown in , LiCl-stimulated TCF/β-catenin transcriptional activation was ∼30% higher in Fat1 knockdown cells compared with control. As shown in and , cyclin D1 levels varied inversely with the level of Fat1 expression. The promoter is a known transcriptional target of Wnt signaling and activated TCF/β-catenin complexes (; ), so we postulated that Fat1 might also inhibit the native promoter. VSMCs were cotransfected with β-catenin and/or IL2R-Fat1, along with the promoter–luciferase reporter construct (). N-cadherin and the IL2R–E-cadherin fusion protein were also tested. Most of the β-catenin–mediated activation of the promoter reporter was eliminated by IL2R-Fat1 or N-cadherin expression (). Consistent with the Topflash results, IL2R–E-cadherinIC was more effective, as it decreased promoter activity to a level below baseline. Fat1 is a type I transmembrane protein, and immunofluorescent studies with antiserum specific for Fat1 sequences showed expression at the cell surface, as expected (). We also noted consistent signals in the cell nucleus with this antiserum. This observation, together with a recent report of localization of Fat1 cytoplasmic sequences to the nucleus (), raised the possibility that inhibition of β-catenin by Fat1 might result from a nuclear (transcriptional repressor) function of a cleaved Fat1 fragment, rather than sequestration of β-catenin outside the nucleus. Indeed, incubation without proteinase inhibitors of extracts of A7r5 cells expressing both native Fat1 and the IL2R-Fat1 fusion protein showed the disappearance of these full-length proteins and rapid appearance of a single, relatively stable species of ∼50 kD (). Because the NH terminus of this cleaved product is not yet defined, we designate it as Fat1*; its apparent size in SDS-PAGE suggests that it contains most (if not all) of the ∼400 aa Fat1 domain. Like human Fat1 (), the mouse Fat1 contains a potential NLS (RKMISRKKKR) near its NH terminus. We tested the effect of this sequence on Fat1 localization by immunocytochemical analysis of A7r5 cells transfected with FLAG-tagged expression constructs that retain (Fat1) or exclude (Fat1) the NLS motif. Fat1 localized almost exclusively to the nucleus, whereas Fat1 was apparent in the nucleus and prominent throughout the cytoplasm (). To evaluate these findings in the context of Fat1-mediated VSMC growth inhibition, we tested these Fat1 derivatives for effects on promoter activity. The IL2R-Fat1 fusion protein yielded significant inhibition of β-catenin–mediated promoter activation (); Fat1, but not Fat1, retained this inhibitory effect (). Both Fat1 and Fat1 are present in the nucleus, but the former has a cytoplasmic distribution not shared by Fat1; hence, we attribute this inhibitory effect on β-catenin to the extranuclear presence of Fat1. Fat1 is expressed widely during mouse and rat development (; ), notably in areas with high levels of cellular proliferation. Although in situ hybridization of rat embryos demonstrated expression of mRNA in the developing aortic outflow tract (), the significance of Fat1 in vascular tissues has not been previously explored. We found relatively low expression of Fat1 in normal adult rat carotid arteries, and substantially increased levels during the first few days after injury (). Immunohistochemical analyses () showed prominent Fat1 staining first in the injured arterial media, and subsequently in the neointima, a pattern of expression similar to that of VSMC proliferation in this model (). Interestingly, areas of attenuated neointimal formation showed prominent Fat1 and decreased PCNA staining, providing an initial suggestion that Fat1 might act to limit VSMC proliferation in vivo (). Nevertheless, Fat1 levels in cultured VSMCs increased in response to serum and several factors known to promote VSMC activation and neointimal formation, including ATII (), PDGF-BB (), and bFGF () (). This expression pattern contrasts with that described for N-cadherin, which decreases after stimulation of VSMC with serum or PDGF-BB (), and that of R-cadherin, which decreases substantially in the first few days after injury (). To evaluate how induction of this very large protocadherin might affect the response to vascular injury, we tested the effect of Fat1 on VSMC migration and proliferation, two of the key cellular functions activated in this setting. Both loss of Fat1 expression and expression of the IL2R-Fat1 fusion protein attenuated VSMC migration (). In the context of recent reports regarding Fat1 function in epithelial cells (; ), these findings suggest that increased Fat1 expression facilitates VSMC migration by providing directional cues and stimulating actin cytoskeletal remodeling through its interactions with proteins of the Ena/VASP family. Together with the Fat1 knockdown results, inhibition of migration by the IL2R-Fat1 fusion protein suggests that dissociation of Fat1 extracellular interactions from Fat1-mediated intracellular signaling interferes with directional migration. Despite the induction of Fat1 in the proliferative phase after injury and in response to growth factor stimulation of cultured cells, our results in both loss- and gain-of-function studies ( and ) suggest that Fat1 opposes VSMC proliferation. Loss of growth suppression resulting in imaginal disc overgrowth in led to identification of Fat (), the founding member of the cadherin subfamily that includes mammalian Fat1. Although recent analyses indicate that mammalian Fat1 is more closely related to Ftl () than to Fat, a growth regulatory function has yet to be described for Ftl. Altered growth characteristics were also not identified in mouse −/− neural progenitors and embryonic skin (). Thus, our findings in VSMCs may reflect cell type–specific differences in the expression of cadherins or other protocadherins functionally redundant with Fat1, or differences in the level of β-catenin expression. In either case, the results of Fat1 knockdown studies indicate that in VSMCs, endogenous levels of Fat1 expression are sufficient to limit cyclin D1 expression () and β-catenin–mediated transcription (), whereas our gain-of-function studies () suggest that decreased cyclin D1 expression and cell growth are likely physiologic consequences of Fat1 induction. Cyclin D1, a known TCF/β-catenin target gene (; ), plays a critical role in regulation of G1 phase progression and G1/S cell cycle transition (; ), and the level of its expression is closely controlled. Increased Fat1 expression in response to injury probably acts to slow VSMC proliferation, at least in part by decreasing cyclin D1 expression. Signaling by classical cadherins has been studied extensively, but the mechanisms of protocadherin signaling are not well understood. The intracellular portion of Fat1 shows limited similarity to classical cadherin cytoplasmic domains, with 30 of 137 (22%) residues matching consensus in the FC1 domain and 28 of 84 (33%) residues matching consensus in the FC2 domain (). Although Tanoue and Takeichi described partial colocalization of Fat1 and β-catenin in immortalized epithelial cell lines, they found more β-catenin in apical lateral cell contacts and more Fat1 in basal lateral cell contacts (), and concluded that Fat1 does not participate in the classical cadherin system (). Interestingly, these findings are consistent with the observation that in polarized epithelial cells, complexes forming between adjacent cells vary in composition according to their apical vs. basal position (). Thus, our findings in VSMCs, which are morphologically and biochemically nonpolarized (), may differ because of the lack of apical–basal specialization in this cell type. In immunocytochemical studies, we found that β-catenin and Fat1 colocalized in a junctional pattern at points of contact between VSMCs (); Fat1 staining was also observed at cellular free edges, while β-catenin was not. To our knowledge, a physical interaction between endogenous Fat1 and β-catenin has not been previously demonstrated. We found clear evidence that these proteins interact at physiologic levels of expression. Transfection studies with the IL2R-Fat1 fusion protein indicated that, despite limited similarity to the β-catenin–interacting domains of classical cadherins, the Fat1 domain was sufficient for this interaction (). Although mapping studies suggested that the Fat1 FC1 domain was most important for the β-catenin–Fat1 interaction, deletion of other domains within the Fat1 also decreased the amount of protein coimmunoprecipitation, indicating that sequences both within and outside of the relatively conserved FC1 and FC2 domains may contribute to β-catenin–Fat1 interaction. Interestingly, the FC1 domain corresponds to the area of greatest similarity (54/196 aa identity [27%]) with the Ftl cytoplasmic domain; its role in the β-catenin–Fat1 interaction described here suggests that Ftl may be capable of interaction with armadillo, the homologue of β-catenin. The IL2R-Fat1 chimera allowed us to perform functional analyses without confounding effects attributable to increased expression of the Fat1 extracellular domain. Expression of IL2R-Fat1, but not a control protein lacking the Fat1 domain, decreased nuclear translocation of β-catenin (), and inhibited β-catenin transactivation of both synthetic (Topflash) and native () TCF-dependent promoters (). Although we found evidence of Fat1 cleavage resulting in a Fat1* fragment that may localize to the nucleus (), only a defined Fat1 fragment lacking the NLS (aa 4189–4198) reproduced the inhibitory effect of the IL2R-Fat1 fusion protein. This result suggests that inhibition of β-catenin transcriptional activity is mediated by Fat1 outside the nucleus, and is not due to Fat1 peptides in the nucleus. Thus, it remains to be determined if cleavage and nuclear translocation of Fat1 underlies a specific function, perhaps as a chaperone or transcriptional regulator, or if it is important as a means to inactivate Fat1-mediated inhibition of β-catenin. Our studies to date suggest that the interaction of Fat1 cytoplasmic sequences with β-catenin has consequences for overall regulation of VSMC growth. The underlying mechanism appears similar to that described for classical cadherin-mediated sequestration of β-catenin in epithelial cells (), but in the case of the protocadherin Fat1, this mechanism may be operative only in nonpolarized cells such as VSMCs. Our findings suggest that increased expression of Fat1 after vascular injury facilitates migration and opposes proliferation of VSMCs. The former effect likely involves Fat1 interaction with Ena/VASP proteins, as described in other cell types (; ), whereas the latter effect relies in part on decreased nuclear accumulation of β-catenin (this paper). Interestingly, we found that the Fat1 interaction with and inhibition of β-catenin both appeared less robust than that observed with classical cadherin sequences ( and ), suggesting that Fat1 may be less efficient than the classical cadherins at sequestering β-catenin. Fat1 induction after injury and by growth factors contrasts with the expression pattern of other cadherins found in VSMCs. Together, these observations suggest that Fat1 may guide VSMC migration while remaining relatively permissive of growth in settings when VSMC proliferation is necessary for vascular repair. Ftl is thought to use its exceptionally large extracellular domain to promote epithelial cell separation during formation of tubular organs in embryogenesis (); we speculate that mammalian Fat1, by virtue of its similar structure, may expedite circumferential distribution of VSMCs around the injured artery. Altogether, it is tempting to speculate that Fat1 limits VSMC proliferation while providing directional migration cues important during vascular remodeling, providing an integrative function that may oppose the formation of hyperproliferative cellular clusters. Finally, though expression of Fat1 in human vascular disease has not yet been evaluated, it is possible that loss of Fat1-mediated negative regulation could contribute to VSMC hyperplastic syndromes such as restenosis, transplant arteriopathy, or vein graft disease. All procedures were in accordance with institutional guidelines. The rat carotid artery balloon injury model was implemented as described previously (). In brief, male Sprague-Dawley rats (20 in total, Zivic-Miller) weighing 350–400 g were anesthetized with 40 mg/kg ketamine and 5 mg/kg xylazine. The left common carotid artery was denuded of endothelium and stretched by three passages of a 2F embolectomy catheter according to standard protocols. At 3, 7, and 14 d after injury, animals were reanesthetized and killed, and carotid arteries were harvested and snap-frozen in liquid nitrogen for RNA and protein extraction, or fixed with 4% PFA and processed for paraffin embedding for immunohistochemical analysis. A cDNA fragment identified in differential mRNA display analysis of the rat carotid artery injury model () was cloned, sequenced, and subjected to BLAST analysis, which revealed homology of the sequence fragment with the 3′ end of the rat Fat1 ORF (GenBank/EMBL/DDBJ accession no.). Total RNA was extracted from vascular tissues by homogenization in TRIzol (Invitrogen), treated with DNase I (1 U/μlPromega), and used for first-strand cDNA synthesis. The mRNA levels were quantified in triplicate by qPCR in the Mx3000P Real-Time PCR System with the Brilliant SYBR Green qPCR kit (Stratagene). Rat Fat1 specific primers for qPCR were 5′-CCCCTTCCAACTCTCCCTCA-3′ (forward) and 5′-CAGGCTCTCCCGGGCACTGT-3′ (reverse). PCR cycling conditions included 10 min at 95°C for 1 cycle followed by 45 cycles at 95°C for 30 s, 60°C for 30 s, and 72°C for 60 s. Dissociation curve analysis confirmed that signals corresponded to unique amplicons. Expression levels were normalized by glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA levels for each sample, obtained from parallel assays and analyzed using the comparative ΔΔC method (). Rat carotid arterial sections (5 μm) were incubated overnight with anti-Fat1 antiserum (1:2,000), washed extensively, and incubated with a 1:500 dilution of secondary antibody (biotinylated goat anti–rabbit IgG, DakoCytomation). Slides were incubated with avidin and biotinylated HRP, developed with a peroxidase substrate solution (DakoCytomation), and counterstained with hematoxylin (Fisher Schientific). Specificity of staining was confirmed by omission of the primary antibody. PCNA staining was performed with anti-PCNA (1:100; LabVision), alkaline phosphatase–conjugated goat anti–mouse secondary antibody (1:200), and visualization with BM Purple substrate (Roche). Images were obtained using a microscope (Eclipse E600; Nikon), 40×/NA 0.75 Plan objective, and Coolpix 5400 camera (Nikon). Primary culture RASMCs were prepared as described previously () and maintained in DME (Invitrogen) containing 10% FBS (HyClone), 100 U/ml penicillin, 100 μg/ml streptomycin, and 10 mmol/L Hepes (pH 7.4; Sigma-Aldrich). RASMCs were passaged every 3 to 5 d, and used between 4 and 8 passages from harvest. Primary culture MASMCs were harvested from the aortas of 12-wk-old male Friend virus B mice by enzymatic dissociation, evaluated by immunocytochemical analysis by using α-smooth muscle actin antibody (1:400, Clone 1A4; NeoMarkers), and maintained in DME containing 10% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin. MASMCs were passaged every 2 to 4 d, and used between 4 and 8 passages from harvest. The A7r5 embryonic RASMC, 3T3, and 293T cell lines (American Tissue Type Collection) were cultured in DME containing 10% FBS. ATII was obtained from Sigma-Aldrich, and bFGF and PDGF-BB from Collaborative Biomedical. In stimulation experiments, the cells were made quiescent by incubation in medium containing 0.4% horse serum for 72 h before addition of the FBS or growth factor. Control cultures received an equivalent amount of vehicle. Whole cellular protein was extracted at designed time points. The mouse Fat1 siRNA templates were comprised of 19-bp sense sequences derived from GenBank/EMBL/DDBJ accession no. (position 4881, 5′-GGACCGAAGTCACCAAGTA-3′; position 5126, 5′-GCGACGCATTTAACATTAA-3′; position 6432, 5′-GCATGACACTTTAAATAAA-3′; position 7296; 5′-GTCTGGCAATGATCATAAA-3′) followed by a 9-bp loop sequence, a 19-bp antisense sequence, and a T7 promoter sequence. Control siRNAs included scrambled (GTAACCATAAACAGGCATT) and mismatched (underlined) (GTCTGAATGCATAAA) derivatives of the 7296 sequence, and an unrelated siRNA based on the Renilla luciferase sequence. siRNA was transcribed in vitro using the T7-MEGAshortscript kit (Ambion), and transfected with X-tremeGENE Reagent (Roche) according to manufacturer's recommendations. Fat1 knockdown efficiency was assessed by Western analysis. The mouse Fat1 cDNA was generated by RT-PCR with primers containing HindIII and XbaI sites (underlined) to facilitate cloning: forward 5′-CTCTGCCGGAAGATGATCAGTCGG-3′ and reverse 5′-CACTTCCGTATGCTGCTGGGA. The product was subcloned into the p3XFLAG-CMV-14 expression vector (Aldrich). The IL2R expression construct (a gift of S. LaFlamme, Albany Medical College, Albany, NY; ) was used to construct a chimeric cDNA encoding the IL2R extracellular and transmembrane domains and the Fat1, with or without an in frame 3XFLAG tag (IL2R-Fat1-3XFLAG and IL2R-Fat1, respectively). The IL2R-E-cadherin-3XFLAG construct was produced using a similar strategy. The truncated FLAG-tagged Fat1 constructs, Fat and Fat1, were generated by PCR from the IL2R-Fat1-3XFLAG template using forward primers 5′-CCATGGgcctctgccggaagatgatcagt-3'and 5′-CCATGGGCCAGGCTGAACCTGAAGACAAAC-3′ and the CMV24 reverse primer; the resulting fragments were cloned into pcDNA3.1v5 (Invitrogen). The FLAG-tagged N-cadherin and Myc-tagged β-catenin constructs were gifts from R. Hazan (Albert Einstein College of Medicine, Bronx, NY) and R. Kemler (Max Planck Institute of Immunobiology, Freiburg, Germany), respectively. All constructs were confirmed by sequencing. The retrovirus system we used is based on the IRES-GFP-RV constructs developed by K. Murphy (Washington University, St. Louis, MO) and Phoenix ecotropic packing cells provided by G. Nolan (Stanford University, Stanford, CA). The IL2R-Fat1 cDNA was inserted upstream of the encephalomyocarditis virus internal ribosomal entry sequence (IRES) and green fluorescent protein (GFP) ORF in the GFP-RV vector. A7r5 cells, MASMCs, or RASMCs (5 × 10) were infected with virus-containing supernatant in the presence of 8 μg/ml polybrene. Control cells transduced with virus encoding GFP alone or IL2R and GFP were generated in parallel, and FACS analysis of retroviral transduced cell lines indicated similar levels of GFP expression. Cell migration was assessed by scratch wounding of monolayers and with Transwell 24-well cell culture inserts with 8-μm pores (Costar). For the former, MASMCs transfected with control or Fat1-specific siRNA were grown to confluence, and monolayers were denuded similarly using a 1,000-μl pipette tip. Photomicrographs of the same fields were obtained sequentially at 24 and 30 h after injury using a microscope (TMS; Nikon), Plan 4×/NA 0.13 DL objective, and camera (model 5400; Coolpix), and cellular progress was quantitated by planimetry of the denuded area and converted to distance migrated using NIH Image 1.63 software. For Transwell assays, quiescent cells were harvested, counted, and added (5 × 10/well) to the insert. Culture medium containing 10% FBS as chemotactic agent was added to the lower chamber. After 4 h, nonmigrating cells were removed from upper filter surfaces and the filter was washed, fixed, and stained. We then photographed six randomly selected 200× fields and counted cells that had migrated to the underside of the filter. Cell number was evaluated with the CyQUANT Assay (Molecular Probes). Cells (2 × 10 per well) were plated in 6-well plates in DMEM containing 2% FBS, medium was replaced every other day, and at each time point triplicate wells were washed with PBS and frozen at −80°C. Net sample fluorescence was determined on a Victor 2 plate reader (Wallac) and enumerated by reference to a standard curve. For the BrdU incorporation assay, cells plated on chamber slides (Becton Dickinson) were serum starved (0.4% horse serum) for 48 h and then stimulated with 10% FBS. 10 μM BrdU (Sigma-Aldrich) was added to cells for 6 h before harvest at 24 h. Cells were washed in PBS, fixed in 4% PFA, treated with HCl, and stained sequentially with anti-BrdU antibody (1:200; Abcam) and AlexaFluor 555–conjugated secondary antibody (1:2,000; Molecular Probes). Cells were counterstained with DAPI (Molecular Probes). Signals were visualized by fluorescence microscopy, and the numbers of BrdU-positive and total nuclei per field calculated. Cells were plated on chamber slides 24 h before staining, and then washed with PBS, fixed with PFA, blocked with 3% normal goat serum, and incubated with anti-β-catenin (1:100) and anti-Fat1 (1:1,000) antibodies. Specific staining was identified with goat anti–mouse and chicken anti–rabbit IgG (AlexaFluors, Molecular Probes). Expression of FLAG-tagged proteins was detected using FITC-conjugated anti-FLAG M2 antibody (8 μg/ml, Sigma-Aldrich). After counterstaining with DAPI, samples were mounted (Supermount medium; Biogenex) on glass slides and signals were visualized using an inverted fluorescent microscope (model IX70; Olympus) equipped with 20×/NA 0.4 and 40×/NA 0.6 LWD objectives and standard fluorescent filter sets, a CCD camera (SensiCam ; Cooke), and IPLab software (Scanalytics). Subsequent image processing was performed using Photoshop 7.0 and Illustrator 10.0 (Adobe Systems). Routine control experiments included omission of the primary antibodies. For Wnt pathway activation, cells were treated with LiCl (20 mmol/L) for 6–12 h, and then stained with anti-β-catenin antibody and DAPI nuclear stain. Deletions within the Fat1 portion of the IL2R-Fat1-3XFLAG construct were engineered using the vector XbaI site and introducing NheI restriction sites (QuikChange mutagenesis; Stratagene) in frame at the following positions in the mouse Fat1 aa sequence: 4187, 4244, 4395, and 4497. The sequences between selected pairs of restriction sites were excised, plasmids recircularized, and constructs confirmed by sequencing. Plasmids were introduced into 293T cells using Lipofectamine 2000 (Invitrogen). Whole cell lysates were harvested 24 h after transfection in lysis buffer containing 50 mM Tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 0.5% Nonidet P-40, 0.1% sodium deoxycholate, 1 mM NaVO, 1 mM NaF, with protease inhibitors. Myc-tagged β-catenin was immunoprecipitated by incubating 400 μg of precleared lysate with 2 μg of c-Myc antibody for 2 h at 4°C, followed by incubation with protein G–Agarose (Invitrogen) at 4°C overnight. For immunoprecipitation of endogenous proteins, RASMC whole cell lysates were precleared and then incubated with anti-Fat1 antiserum, anti-β-catenin antibody, or normal rabbit or mouse IgG for 2 h at 4°C, followed by incubation with protein G–Agarose overnight. The beads were washed and immune complexes recovered by boiling in sample buffer. Fat1 and β-catenin were detected by Western analysis, as described above. Membrane, cytoplasmic, and nuclear fractions were prepared using the Compartment Protein Extraction Kit (Chemicon International, Inc.) according to the manufacturer's instructions. Fractionation and loading of proteins was evaluated by Western analysis with anti-lamin A/C antibody (Santa Cruz Biotechnology, Inc.). A7r5 cells growing in DMEM supplemented with 10% FBS were transfected transiently using Lipofectamine 2000 with β-catenin, IL2R-Fat1, Fat1, Fat1, or control expression constructs, along with the TCF wild-type (Topflash) and mutated control (Fopflash) luciferase reporter plasmids (Upstate Biotechnology), or promoter luciferase reporter (a gift from R. Müller, Philipps-Universität, Marburg, Germany; ). MASMCs were transfected by Amaxa electroporation according to the manufacturer's instructions. The total amount of transfected DNA was kept constant. Cell lysates were harvested 24 h after transfection, and luciferase activity was determined using the Glo-lysis buffer system (Promega) and the Victor 2 plate reader. Luciferase activities were normalized to protein levels for each well. The data shown represent transfections repeated at least three times each. Experiments were repeated at least three times. Data are presented as mean ± SEM. Comparisons between two groups were analyzed by test, and comparisons between three or more groups were assessed by analysis of variance (ANOVA) with a Bonferroni/Dunn post hoc test. Significance was accepted for values of P < 0.05.

Classical cadherins interact homophilically with cadherins of neighboring cells to form adherens junctions, which serve both as mechanical linkages between cells and as signaling hubs that relay information from the extracellular environment. Epithelial cadherin, or E-cadherin, is thought to be a tumor suppressor molecule largely because it is frequently down-regulated in carcinomas (; ; ). E-cadherin has also been shown to directly suppress metastasis in the late stages of tumor progression using a transgenic mouse model (). Loss of contact inhibition of proliferation is a hallmark of cancer cells lacking E-cadherin, and transfection of E-cadherin into several such cancer cell lines causes a decrease in proliferation (; ; ). Despite the abundance of literature supporting an antiproliferative role for E-cadherin, there is also evidence that E-cadherin is associated with increased cell proliferation. In colon carcinomas, proliferation is associated with the localization of E-cadherin to the cell periphery (). Ovarian cancers up-regulate E-cadherin, the suppression of which inhibits their proliferation (; ). In nontumorigenic contexts, E-cadherin levels are maintained in proliferating tissues (). In fact, loss of E-cadherin in these physiological settings does not lead to uncontrolled growth, but instead prevents proliferation and causes tissue degeneration during development (), in lactating mammary glands (), and in hair follicles (). Thus, the effects of E-cadherin on proliferation appear to be multifaceted, dependent on context, and poorly defined. Cross talk between cell–cell and cell–substrate interactions may contribute to the effects of cadherins on proliferation. The introduction of E-cadherin into cells cultured on a nonadhesive surface not only decreases proliferation but also causes cells to aggregate into large clusters (). When cultured on an adhesive substrate, cells expressing E-cadherin exhibit increased cell attachment to the substrate when compared with their nonexpressing counterparts (; ). Because such cadherin-induced changes in aggregation or adhesion to ECM can directly affect cell proliferation, the adhesive context in which cadherin engagement is manipulated may contribute to the different proliferative responses that have been observed. In studies of VE-cadherin, which is the major cadherin in endothelial cells, paradoxical effects on proliferation appear to depend on cross talk with cellular adhesion to the ECM. Engagement of VE-cadherin causes growth arrest with increasing cell densities, in part, by causing cells to decrease their adhesion and spreading against the underlying substrate (). In a setting where cell spreading is held constant, engagement of VE-cadherin causes an increase in proliferation (). It appears that various adhesive contexts need to be explored to fully appreciate the mechanisms by which cadherins regulate proliferation. E-cadherin engagement influences several intracellular signaling pathways that are involved in the regulation of proliferation, including the canonical Wnt pathway, receptor tyrosine kinases, and Rho GTPase signaling (). Signaling to Rho GTPases has been of particular interest because of their involvement in regulating the stability of junctions and associated cytoskeletal structures (; ). Specifically, E-cadherin activation of Rac1 has been observed by several groups (; ), and appears to lead to actin recruitment and physical strengthening of adherens junctions (; ). Rac1 is also involved in regulating progression through the G phase of the cell cycle (; ) by modulating p21 levels and cyclin D transcription (; ). However, because Rac1 activity appears to provide different functions in response to different stimuli (; ), Rac1 signaling induced by E-cadherin engagement may not be related to Rac1 signaling in proliferative regulation. Indeed, a link from E-cadherin engagement to proliferation through Rac1 has not been previously reported. We examined the effects of E-cadherin engagement on proliferation of normal rat kidney epithelial cells (NRK-52E) and nontumorigenic human mammary epithelial cells (MCF-10A) under a variety of adhesive contexts. Limited degrees of cell–cell contact, which were introduced at intermediate cell seeding densities or by forming pairs or small clusters of cells, stimulated cell proliferation, but further increasing cell–cell contact by seeding to confluence inhibited proliferation. The proliferative stimulus was mediated by E-cadherin engagement and coordinated through Rac1 and p120-catenin, whereas the cell–cell contact inhibition of proliferation was driven by a decrease in cell adhesion and spreading on the underlying ECM. These findings demonstrate that cell–cell contact can either enhance or inhibit proliferation via distinct mechanisms, and suggest a novel pathway by which E-cadherin can locally modulate tissue growth in contexts such as development, tissue mass homeostasis, and wound healing. To explore the role of cell–cell contact in the proliferation of epithelial cells, we varied the degree to which cells contacted neighboring cells by seeding at different densities. We G-synchronized NRK-52E cells (Fig. S1, available at ) and seeded them from densities at which cells were completely isolated from each other (2 × 10 cells/cm) to confluence (4 × 10 cells/cm) overnight in the presence of BrdU, and analyzed for entry into S phase (). At confluence, cell proliferation was low. Decreasing the plating density from confluence increased the percentage of cells entering S phase, with maximal proliferation at intermediate seeding densities (2–6 × 10 cells/cm). Interestingly, further decreasing cell density so that cells did not contact their neighbors caused an unexpected decrease in proliferation. We also noted that cells at the lowest seeding density were often rounded in shape, whereas cells at intermediate densities appeared to be well spread. Directly measuring the spread area of cells revealed that the degree of cell spreading against the substrate was also biphasic with seeding density, and correlated with the levels of proliferation (). These findings were confirmed in a second epithelial cell type, MCF-10A nontumorigenic human mammary epithelial cells. Progressively increasing MCF-10A cell–cell contact by seeding G-synchronized cells (Fig. S1) from sparse (10 cells/cm) to confluent (2 × 10 cells/cm) densities also resulted in the density-dependent biphasic proliferative and cell- spreading response, with peak levels at intermediate densities (2 × 10 cells/cm), and a precipitous inhibition of proliferation and spreading at high seeding densities (). It has previously been observed that increasing the degree of cell spreading increases proliferation (; ). Because changes in cell spreading correlated with changes in proliferation in both NRK-52E and MCF-10A cells that were seeded at different densities, we examined whether changes in cell spreading were required for contact-induced proliferation. To control cell spreading, we seeded cells onto substrates patterned with microscale agarose wells of varying sizes (). The microwells were fabricated with walls of nonadhesive agarose on top of a glass substrate that was coated with ECM protein. Cells were seeded onto substrates with bowtie-shaped microwells such that two cells would settle into each bowtie. Each cell in the pair adhered to the base of the well and spread to fill half the well, making contact with each other through the center of the bowtie (). These contacts were stable over time, as the microwells prevented cells from migrating apart. As a control, single cells were seeded into triangular-shaped wells with areas equal to one half of a bowtie. We found that given the same degree of cell spreading, pairs of cells proliferated at a dramatically higher rate compared with single cells for both NRK-52E and MCF-10A cells (). This effect was observed for several different microwell sizes, demonstrating that the changes in cell spreading induced by cell–cell contact were not necessary for contact-dependent up-regulation of proliferation. The ability for cell–cell contact to induce proliferation independent of changes in cell spreading against the underlying ECM suggested that contact-induced proliferation may be a more general phenomenon that is not specific to the two-dimensional culture context. To address this possibility, we examined whether the biphasic proliferative response to cell–cell contact also occurs in three-dimensional culture. MCF-10A cells were seeded at a density of 5–10 × 10 cells/cm within collagen gels overnight. Under these conditions, both single, isolated cells, as well as clusters with varying numbers of cells, developed. Immunostaining revealed no striking differences in individual cell morphology or their junctions among the different sized clusters (). To examine the proliferation of cells within the clusters, we seeded G-synchronized cells into collagen gels in the presence of BrdU overnight. Single cells proliferated at a low rate, whereas cells within small clusters (two to five cells per cluster) exhibited high rates of proliferation. Proliferation was progressively inhibited when cells were aggregated in increasingly larger clusters (). Interestingly, we observed that proliferation in the large clusters was predominantly limited to cells on the surface of the cluster where fewer cell–cell contacts are formed, and that most cells within the interior of the cluster remained quiescent. We then examined whether cadherins were involved in cell–cell contact–mediated changes in proliferation. We constructed an adenovirus containing a mutant of E-cadherin lacking the β-catenin–binding domain (Ad-EΔ), which has previously been shown to act as a dominant negative by blocking E-cadherin–mediated intercellular adhesion (; ). Immunostaining of E-cadherin in Ad-EΔ–infected cells confirmed the loss of cadherin localization at the cell–cell junctions in both NRK-52E and MCF-10A cells (). Infection with Ad-EΔ eliminated the contact-induced peak in proliferation seen at intermediate densities, when compared with Ad-GFP–infected control cells in both cell lines (). Interestingly, expression of EΔ did not affect proliferation of cells seeded at very high densities, which have many cell–cell contacts, suggesting that E-cadherin is not required for the reduced levels of proliferation at confluence observed in this setting. We also examined whether expression of EΔ blocked the proliferation stimulated by cell–cell contact within microwell cultures in both epithelial cell types. Ad-EΔ reduced the proliferation of pairs of cells to the levels of single cells that were spread to the same degree (). We confirmed these results by using a blocking antibody against E-cadherin that prevented E-cadherin engagement in MCF-10A cells (). Inhibition of cadherin engagement with the blocking antibody abrogated the contact-induced peak in proliferation at intermediate seeding densities and the increase in proliferation of pairs of cells compared with single cells (). Similarly, knockdown of E-cadherin expression using siRNA also eliminated contact-induced proliferation (). Together, these data suggest that E-cadherin is required for stimulation of proliferation induced by cell–cell contact. Interestingly, in all three methods of eliminating E-cadherin engagement and in both cell lines, E-cadherin engagement appears not to be required for the cell–cell contact–induced proliferation arrest at high cell densities. An alternative possibility is that cell–cell contact nonspecifically crowds cells to spread less against the underlying substrate, and this decrease in cell–ECM interaction arrests cells. To address this, G-synchronized NRK-52E and MCF-10A cells were seeded overnight into microwells of different sizes, such that single cells attached in each microwell, and analyzed for S phase entry (). In both cell lines, the micropatterned islands decreased proliferation with decreased cell spreading, even in the absence of cell–cell contact (). This inhibition of proliferation on micropatterns was not affected by infection of Ad-EΔ, confirming that E-cadherin is not involved in this regulation of proliferation by cell spreading, and that Ad-EΔ does not nonspecifically disrupt proliferation in these cells (Fig. S2, available at ). These data suggest that cell–cell contact inhibits proliferation by decreasing cell spreading, and stimulates proliferation through E-cadherin engagement. Although the cadherin-blocking studies demonstrated that E-cadherin was required for the stimulation of proliferation observed at intermediate densities and in pairs of cells in bowtie-shaped microwells, it was unclear whether cadherins were inducing proliferation through juxtacrine influences or by acting as receptors themselves. To explore this further, we engaged cadherins of single, isolated, patterned MCF-10A cells using beads coated with a chimera of the ectodomain of human E-cadherin fused to the immunoglobulin Fc domain (hE-Fc; ). In both unspread (300 μm) and spread (750 μm) conditions, cells that were bound to hE-Fc–coated beads exhibited higher proliferation compared with cells that were bound to protein A–coated control beads (). These data demonstrate that the engagement of E-cadherin alone can stimulate proliferation independently of juxtacrine influences. We next explored the role of Rac1 as a potential downstream mediator of the E-cadherin–induced proliferation because cells at intermediate densities exhibited morphological characteristics of high Rac1 activity, such as increased cell spreading (). First, we examined the timing of Rac1 activation with respect to contact formation. G-synchronized MCF-10A cells were seeded at an intermediate density (2 × 10 cells/cm) and assayed for the formation of cell–cell contact by immunofluorescence staining and for Rac1 activity by pulldown assay over the course of 24 h. Initial formation of cadherin-containing contacts occurred at 4 h after seeding. By 8 h, most cells had formed contacts with neighbors, and the intensity of staining at junctions continued to increase over the subsequent 16 h (). Relative Rac1 activity was initially low, gradually increased to a peak at ∼8 h after seeding, and then decreased to baseline levels by 24 h (). The correlation between E-cadherin staining and Rac1 activity at 8 h after seeding supported the possibility that Rac1 was activated by cell–cell contact. To directly examine the role of cell–cell contact on Rac1 activity levels, we seeded G-synchronized MCF-10A cells at high (2 × 10 cells/cm), intermediate (2 × 10 cells/cm), and low (2 × 10 cells/cm) densities and assayed for Rac1 activity 8 h after seeding. Rac1 activity was twofold higher at intermediate cell densities when compared with either low or high cell densities (). Phalloidin staining of MCF-10A cells 8 h after seeding at different densities also demonstrated that cells at the intermediate density exhibited increased membrane ruffling, a phenotype of elevated Rac1 activity (). To determine whether the Rac1 activity induced by cell–cell contact was involved in the proliferative response, G-synchronized MCF-10A cells were infected with an adenovirus containing dominant-negative Rac1 (Ad-RacN17) and examined for S phase entry at different seeding densities. Infection of Ad-RacN17, like Ad-EΔ, eliminated the peak in proliferation at intermediate densities, and did not affect proliferation at high and low densities (). In the microwell system, where changes in cell spreading were prevented, expression of RacN17 reduced the proliferation of pairs of cells to that of single cells (). Immunostaining for E-cadherin in Ad-RacN17–treated cells confirmed that dominant-negative Rac1 did not affect localization of E-cadherin at the cell–cell junction (). This inhibitory effect appeared to be specific to a Rac1-mediated pathway, as inhibition of RhoA signaling through its effector Rho kinase by exposure to 50 μM Y27632 did not inhibit proliferation at any density, but, interestingly, increased proliferation at the low densities (). Y27632 also had a stimulatory effect on proliferation of single and pairs of cells patterned in microwells, and did not appear to mediate these effects by altering E-cadherin localization to the cell–cell contacts (unpublished data). Y27632 has been shown to activate Rac1 and cyclin D signaling in fibroblasts (). Supporting this possibility, infection of cells with Ad-RacN17 inhibited the increase in proliferation with Y27632 at low densities (unpublished data). These data demonstrated that cell–cell contact–induced proliferation is mediated through Rac1, and not RhoA. Our findings indicated that E-cadherin and Rac1 activity are both required for the proliferation induced by cell–cell contact, but the causal relationship between E-cadherin engagement and Rac1 activity remained unclear. To address whether E-cadherin is responsible for Rac1 activation at intermediate densities, we infected G-synchronized MCF-10A cells with Ad-EΔ or Ad-GFP, and assayed for Rac1 activity 8 h after seeding. The increase in Rac1 activity observed at intermediate densities was abrogated by Ad-EΔ (), but baseline Rac1 activity at high and low densities were not significantly affected. Phalloidin-stained cells at intermediate densities revealed that Ad-EΔ diminished the previously observed membrane ruffling (). Furthermore, the peak of cell spreading at intermediate densities was abolished by Ad-EΔ (). These data confirm that Rac1 and its functional effects on cytoskeletal processes lie downstream of E-cadherin at intermediate seeding densities, and suggest that E-cadherin stimulates proliferation through a Rac1-mediated pathway. Because the Rac1 effector Pak has been implicated in Rac1-mediated proliferation, we explored its role in E-cadherin–mediated proliferation. Similar to Ad-EΔ– and Ad-RacN17–infected cells, MCF-10A cells infected with an adenovirus that expressed a kinase-dead mutant of Pak (Ad-PakR299) also lost the biphasic proliferative response (Fig. S3 A, available at ). However, this full-length dominant-negative Pak has been shown to interact with and inactivate Rac1. Expressing the more specific Pak-PID mutant, in contrast, showed no effect on the proliferative response to cell density (Fig. S3 B). Lastly, endogenous total Pak levels and Pak phosphorylation monotonically decreased with cell-seeding density (Fig. S3 C). Together, these data suggest that Pak signaling is not involved in the cadherin- and Rac1-induced proliferation. Cadherin engagement has been observed to stimulate Rho GTPases, in part by binding p120-catenin (p120) and abrogating the ability of p120 to inhibit Rho (). To explore the role of p120 in E-cadherin–induced proliferation via Rac1, we generated MCF-10A cell lines stably expressing p120-siRNA or empty vector alone as a control using the pRetroSuper retroviral system (). Cells with p120-siRNA expressed <30% of control levels of p120 (). G-synchronized cells were seeded at different densities and assayed for proliferation. p120 knockdown abolished the biphasic proliferative response to cell seeding density, and cells exhibited a higher level of proliferation at the lowest seeding densities, as compared with control cells (). To examine whether Rac1 was involved in this up-regulation of proliferation, we seeded control and siRNA-treated cells at 9 × 10 cells/cm, and measured Rac1 activity. Cells lacking p120 exhibited increased Rac1 activity (). Furthermore, Rac1 was required for the proliferation observed in siRNA-treated cells, as Ad-RacN17 abolished the knockdown-induced proliferation (). We next examined whether reexpression of p120 could inhibit the Rac activity observed in knockdown cells. siRNA- treated cells were infected with a second retroviral vector (LZRS) expressing the murine form of the p120 found in MCF-10A cells (isoform 3A). Knockdown cells stably expressing murine p120 had significantly lower Rac activity when compared with knockdown cells expressing the empty second retroviral vector (). Furthermore, reexpression of p120 suppressed the high proliferative levels observed in knockdown cells (). This inhibitory effect did not require the binding of p120 to cadherins because expression of a mutant p120, which lacks ARM repeat 1 and therefore cannot bind to cadherins (p120Δ; ), also exhibited low levels of Rac1 activity and proliferation at intermediate seeding densities. These data suggest that p120 may be involved in cadherin-induced Rac1 signaling and proliferation, and support a model in which p120 normally suppresses Rac1 and proliferation until engagement of E-cadherin sequesters p120, disinhibiting Rac1 activity. E-cadherin engagement at cell–cell contacts has a known function in the suppression of proliferation, which is best described in the context of tumorigenesis (). We observed a biphasic proliferative response to cell–cell contact. Blocking E-cadherin engagement abrogated the elevated proliferation levels observed at intermediate seeding densities, but not the inhibition of proliferation observed at confluence. Instead, mimicking the reduction in cell adhesion and spreading induced at confluence by culturing isolated cells on micropatterned substrates resulted in proliferation arrest, suggesting that decreased cell spreading may be responsible for contact inhibition in these cells. These findings contrast with previous conclusions that transformed cells exhibit uncontrolled growth after down-regulation of E-cadherin (), possibly because the tools to uncouple the confounding effects of cell–cell contact, cadherin engagement, and cell spreading have only recently become available. Nonetheless, reconciling responses between tumorigenic and nontumorigenic cells may be inappropriate because cancerous cell lines may have already lost their adhesion-regulated controls (). The finding that degrees of contact may differentially regulate cells through numerous mechanisms highlights the need for better approaches to tease out the various environmental cues that may be affected by cell–cell adhesion. In propagating a proliferative signal, cadherins may act directly as receptors that cause intracellular signaling, or they may function primarily to bring cells into contact with each other to signal via other juxtacrine receptors. For example, E-cadherin has been shown to initiate signaling in an EGF receptor–dependent manner (; ). Using E-cadherin-Fc–coated beads to ligate E-cadherin, we found that the engagement of E-cadherin alone is sufficient to stimulate proliferation. Although these findings do not eliminate the possibility that juxtacrine signals can also contribute to the cadherin-mediated proliferative response, they add to the growing body of evidence that cadherins can provide direct, functionally relevant signaling beyond their structural role. We demonstrate that E-cadherin–activated Rac1 and downstream effects on cell spreading and membrane ruffling only occurred with limited cell–cell contact. Several mechanisms may be responsible for the activation of Rac1 in these limited cell–cell contact settings. First, the dynamics of the cadherin contacts in a cell with only a few bordering cells may be distinct from those in a cell within a confluent monolayer. As in studies of integrin activation of Rho GTPases (; ), such receptor dynamics may be important for cadherin activation of Rac1. Second, the mechanism for up- and down-regulation of Rac1 signaling may be distinct; for example, E-cadherin engagement at intermediate densities might activate Rac1 and the decrease in cell spreading at high densities might inhibit Rac1. This biphasic response demonstrates how multiple inputs are likely integrated by the Rac1 signaling pathway to produce a decisive response within the cell. E-cadherin activation of Rac1 appears to involve p120. p120 has also been implicated in the regulation of other Rho GTPases by cadherins. In the case of RhoA, cadherin binding of p120 appears to compete with the ability of p120 to inhibit RhoA signaling (). Although the role of p120 in mediating E-cadherin–induced Rac1 activity has been less well characterized, our findings suggest that p120 may function analogously, whereby E-cadherin engagement shifts p120 between cadherin-bound and Rac-inhibitory roles. This model is also consistent with reports that Rac1 activation by E-cadherin engagement is inhibited by a mutation that prevents p120 from binding to E-cadherin (). E-cadherin–mediated Rac1 activity stimulated proliferation. Rac1 activity has been shown to regulate cell cycle progression via MAPK signaling (), as well as the NFκB pathway (). The Rac1 effector Pak has also been shown to activate numerous mitogenic pathways (; ). However, our data suggest no role for Pak in E-cadherin–mediated proliferation via Rac1. A viable alternative is that Rac1 signaling may feed back to affect cell–matrix interactions through changes in actin and integrin dynamics ()—a possibility that will require further study. Although Rac1-induced proliferation was evident before the current study, it has been unclear what physiologic situation might invoke the Rac1 proliferative pathway. Our results now suggest that E-cadherin engagement provides a physiologic stimulus for Rac1-mediated proliferation (). Major differences are revealed when comparing cellular response to cadherin engagement among different cell types. E-cadherin engagement in epithelial cells stimulated proliferation via Rac1. In endothelial cells, engagement of VE-cadherin induces RhoA signaling (), and cell– cell contact in fibroblasts stimulates a contractile response (). Interestingly, both endothelial cells and fibroblasts have been previously reported to exhibit a RhoA- and tension-dependent pathway to proliferation (; ; ), whereas we demonstrate that epithelial cells proliferate through a RhoA-independent and Rac1-dependent mechanism. In fibroblasts, Rac1 also induces cyclin D expression, but this pathway is cryptic under normal conditions, and is revealed only upon artificial inhibition of RhoA or Rho kinase (). Together, these data suggest that the links between specific cadherin subtypes, Rho GTPase signaling, and cell proliferation may vary for different cell types. Subtle shifts in such regulatory pathways could have far-reaching consequences on the canonical relationships between multicellular organization, cell structure, and cell function. Although cadherins were originally discovered to serve as a mechanical linkage between adjacent cells, it has become apparent that these receptors also regulate cell function via biochemical signaling pathways. The biphasic increase in Rac1 activity, cell ruffling and spreading, and proliferation via cell–cell contact observed might be important in several physiological contexts. During both development and adult tissue homeostasis, the link between cell–cell contact, Rac1, and proliferation may be in place to ensure that cells at the edges of epithelial sheets or masses ruffle, spread, and proliferate, whereas those fully constrained within these structures remain quiescent. In the context of loosely associated cells coming together to form new tissue, this system also would encourage tissue growth and rearrangement only when enough cells of the same type are associated with each other, but not when single cells are mislocalized or when cells have formed a sufficient mass. Thus, the ability of cells to sense varying degrees of cell–cell contact through this biphasic cadherin–Rac1 pathway may provide a key element in focusing cellular activity to the appropriate coordinates within a multicellular tissue, and underscores the importance of the numerous transduction mechanisms that are regulated by cadherins and used by cells to navigate their complex, structured microenvironment. NRK-52E and MCF-10A cells were obtained from the American Type Culture Collection and cultured according to their recommendations. Phoenix cells (a gift from A. Reynolds, Vanderbilt University, Nashville, TN) were cultured as previously described (). hE-Fc–producing CHO cells (a gift from A. Yap, University of Queensland, Brisbane, Australia) were cultured as previously described (). Reagents were obtained as follows: anti–E-cadherin (36 [BD Biosciences]; HECD-1 and SHE78-7 [Zymed Laboratories]; DECMA-1 [Sigma-Aldrich]); anti-p120 (98; BD Biosciences); anti–α-tubulin (Sigma-Aldrich); anti-GAPDH (Ambion); anti-Pak1/2/3 and anti–phospho-Pak1(Thr423)/Pak2(Thr402) (Cell Signaling Technology); and Y27632 (Calbiochem). Myc-tagged RacN17 adenovirus, GFP-tagged PakR299, and GFP-tagged Pak-PID adenoviruses were gifts from A. Ridley (University College London, London, UK), W. Gerthoffer (University of Nevada, Reno, NV), and J. Chernoff (Fox Chase Cancer Center, Philadelphia, PA) and V. Weaver (University of Pennsylvania, Philadelphia, PA), respectively. Cells were G-synchronized by replacing growth medium with starvation medium (1% serum for NRK-52E or 0% serum for MCF-10A) for 24 h. Synchronization was confirmed in >90% of cells in G/G by FACS analysis of propidium iodide–stained (Invitrogen) cells (Fig. S1, A and C). To determine a time point for proliferation assays, cells were seeded onto 25 μg/ml fibronectin- or 50 μg/ml collagen-coated glass substrates, pulsed with BrdU, and analyzed for BrdU incorporation (GE Healthcare). In all proliferation experiments, cells were cultured in the presence of BrdU for a time period during which cells had entered S phase, but had not begun mitosis (as determined by examination of mitotic figures); 18 h for NRK-52E and 22 h for MCF-10A (Fig. S1, B and D), and then fixed and assayed for BrdU incorporation. Unless otherwise noted, at least 300 cells were examined across a minimum of three experiments for all conditions reported. Collagen gels (2.5 mg/ml) were generated with a solution of acidic collagen (BD Biosciences), sodium bicarbonate (Sigma-Aldrich), Hepes buffer and 10× M199 (both from Invitrogen), which was neutralized with sodium hydroxide (Sigma-Aldrich). Cells were pelleted and resuspended within the collagen solution, and incubated at 37°C until the collagen solidified. Full serum media was added on top of the gel and proliferation was assayed as described in the previous section. Microwell substrates were prepared as previously described (). In brief, stamps of polydimethylsiloxane (Dow Corning), which were cast from photolithographically generated master patterns, were treated with UV/ozone for 5 min before use. A 0.6% agarose/40% ethanol solution in water was flowed between the stamp sealed against a glass slide. Upon stamp removal, substrates were coated with fibronectin or collagen for at least 1 h. Cells were seeded and assayed for proliferation as described in the Proliferation assays section. hE-Fc was purified from conditioned media of CHO cells stably expressing a secreted human E-cadherin fused to the Fc region of IgG, as previously described (), and was used at 100 μg/ml in 0.1% BSA for binding to protein A-coated latex beads (Bangs Laboratories, Inc.). Beads were applied to cells 2 h after seeding, and cells were fixed and analyzed for proliferation as described in the Proliferation assays section. Images of fixed samples were acquired at room temperature using an epifluorescence microscope (model TE200; Nikon) equipped with Plan Fluor 10×, 0.3 NA, and Plan Apo 60×, 1.4 NA, oil immersion lenses, Spot camera and software (Diagnostic Instruments), or an epifluorescence microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with 40× Plan-Neofluar, 1.3 NA, oil immersion, 63× Plan-Apochromat, 1.4 NA, oil immersion objectives, an Axiocam camera, and Axiovision software. For measurements of projected cell area, cells were outlined in 10× phase-contrast images and analyzed using Spot software. For immunostaining, cells were fixed in 1:1 methanol/acetone for 20 min (for E-cadherin) or formaldehyde followed by 0.05% Triton X-100 (for p120-catenin), blocked with goat serum (Invitrogen) in PBS, and incubated in primary and Alexa Fluor 594– or 488–conjugated secondary antibodies (Invitrogen). Apotome and AxioVision software (Carl Zeiss MicroImaging, Inc.) were used to capture images in three-dimensional cultures. Some image levels were adjusted using Photoshop (Adobe). The cDNA fragment encoding human E-cadherin lacking 105 bps at the COOH-terminus (the β-catenin–binding domain) was amplified by PCR from hEcad/pcDNA3 vector (a gift from C.J. Gottardi and B.M. Gumbiner, University of Virginia, Charlottesville, VA) using 5′ (5′-GAGGCGGCCGCACCATGGGCCCTTGGAGCCGC-3′) and 3′ (5′-GAGCTCGAGTCAGGAGCTCAGACTAGCAGC-3′) oligonucleotide primers. Recombinant adenoviruses encoding human E-cadherin bicistronic to GFP were prepared using the AdEasy XL system (Stratagene) as previously described (). GTP-loaded Rac1 was measured using a commercially available kit (Upstate Biotechnology) as previously described (). Pak1-PBD beads were used as supplied and also made using GST-tagged recombinant Pak1-PBD produced in BL21 cells containing the pGEX-PBD vector (a gift from L. Romer, Johns Hopkins University, Baltimore, MD). Protein levels were determined by Western blot, detected with HRP-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories), developed using ECL substrate (Pierce Chemical Co.), and quantified using Versadoc imaging system (Bio-Rad Laboratories). siRNA against E-cadherin (a gift from R. Assoian, University of Pennsylvania, Philadelphia, PA) was transfected using Lipofectamine 2000 (Invitrogen) 24 h after seeding MCF-10A cells at 5 × 10 cells/cm. Cells were G-synchronized and seeded onto appropriate substrates, and proliferation was assayed as described in the Proliferation assays section. Retroviral supernatants were produced in Phoenix cell packaging line, as previously described (). Stable lines of MCF-10A cells expressing p120-siRNA were generated with pRetroSuper containing p120-siRNA. Cells were selected and maintained in 4 μg/ml puromycin. To generate cell lines reexpressing murine p120, some of both control and RNAi cell lines were infected with a second retrovirus (LZRS) containing p120 or p120Δ, and selected and maintained in 800 μg/ml G418. Empty vector controls were used in all cases. All retroviral reagents were gifts from A. Reynolds. Fig. S1 shows the synchronization and proliferation profiles for NRK-52E and MCF-10A cells. Fig. S2 shows that E-cadherin is not required for inhibition of proliferation caused by reduced cell spreading in single-patterned cells. Fig. S3 shows that the Rac1 effector Pak is not involved in cell–cell contact–mediated proliferation. Online supplemental material is available at .

Endocytosis, a process characterized by the internalization of extracellular materials and membrane proteins via vesicular intermediates, plays many roles in regulating cell–cell signaling pathways. In addition to the well-established role of attenuating signaling activity by clearing active receptor molecules from the cell surface, endocytosis has been proposed to facilitate signaling by transporting active receptor molecules to sites where downstream effectors are localized (; ; ). A novel role of endocytosis has recently been proposed for the Notch signaling cascade, in which the internalization of the ligand facilitates activation of the receptor (; ), although the exact mechanism of this critical event remains elusive. The Notch pathway is a signaling module that is highly conserved in all metazoans and has been implicated in a variety of developmental processes (). How Notch transduces signals from the plasma membrane and affects gene regulation has been extensively analyzed in , as well as several other model systems. It is now apparent that proteolytic processing of the Notch receptor is tightly associated with its ability to transduce signals (; ). Notch is first cleaved during its transit through the biosynthetic pathway, thereby reaching the cell surface as a heterodimer of Notch extracellular domain (NECD) and a membrane-tethered intracellular domain (; ). The binding of Notch to its ligand induces two additional cleavage events, releasing a signaling-competent Notch intracellular domain fragment from the plasma membrane (; ; ). Notch intracellular domain then translocates into the nucleus and regulates gene expression by acting as a transcriptional coactivator (; ). Endocytosis appears to play a key role in regulating the activity of the Notch pathway. The importance of vesicular trafficking in Notch signaling was first noticed when mutations in dynamin, a GTPase required for the detachment of vesicles from plasma membrane (; ), was found to produce a Notch-like phenotype (). Clonal analysis suggested that in Notch signaling, dynamin function is required in both signal-sending and signal-receiving cells (), suggesting that endocytosis impinges on the pathway at two independent steps. Although the role of endocytosis in signal-receiving cells is less clear, the internalization of ligand for the Notch receptor in the signal-sending cells appears to be a key event in activating the Notch cascade (). In , there are two known Notch ligands, Delta (Dl) and Serrate (Ser), members of the Dl, Ser, and Lag-2 protein family (DSL). Both Dl and Ser appear to use an ubiquitin-mediated endocytic pathway to activate Notch receptors (; ; ; ). The covalent addition of ubiquitin to polypeptides, besides being a tag for proteasome-mediated protein degradation, can serve as a sorting signal for membrane protein internalization (; ). The ubiquitination of Dl and Ser for subsequent internalization is mediated by () and (), which encode two structurally unrelated E3 ubiquitin ligases (; ; ; ; ; ). Although Neur and dMib regulate distinct Notch-dependent processes, they appear to be interchangeable in mediating the ubiquitination and internalization of the DSL ligand (; ; ; ). Another critical component of this process is (), the homologue of epsin (). contains an ubiquitin-interacting motif (; ), as well as motifs that bind to clathrin and other classes of adaptors (). Thus, it is thought that lqf functions as a cargo-specific clathrin adaptor, capable of recognizing and sequestering monoubiquitinated DSL ligand into clathrin-coated vesicles (CCVs; , ; ), although an alternative function for epsin in nonclathrin endocytosis has been proposed (; ). Although a requirement of ligand endocytosis for Notch activation seems clear, the mechanism of how the internalization of the DSL ligand in the signal-sending cells promotes the proteolytic processing of Notch in the neighboring signal-receiving cells remains poorly understood. One set of models proposed that the internalization of Notch bound DSL ligand could either clear NECD from the extracellular space or generate physical force to dissociate NECD from the membrane-tethered intracellular domain, allowing the subsequent cleavage processing to occur (). Alternatively, it has been suggested that endocytosis is required to transport DSL ligand to subcellular compartments, where the ligand is rendered signaling competent before being recycled back to the cell surface (; ). Because, at present, the analysis of the roles of DSL endocytosis in Notch signaling relies on those mutations disrupting the assembly of cargo-containing CCVs, it is difficult to distinguish whether it is the internalization by itself or the transit of Dl through specific endocytic compartments that is critical for Notch activation. To better understand the mechanism of this critical process, the effects of additional endocytic mutations in Notch signaling need to be assessed. The clathrin coats of newly formed CCVs need to be dissociated so the vesicles can fuse with target organelles and the released clathrin triskelions can be reutilized for subsequent rounds of endocytosis. Hsc70, a constitutively expressed member of the Hsp70 chaperone family, has been implicated in promoting the release of clathrin triskelions and other coat proteins from CCVs in vitro (; ; ). In addition to Hsc70, another important factor in the clathrin uncoating reaction is thought to be auxilin, which contains clathrin binding domains, as well as a J-domain (; ). The J-domain, a conserved motif shared by members of the DnaJ protein family, can bind to Hsp70 family proteins and stimulate their low intrinsic ATPase activity (). Thus, auxilin is thought to function as a cofactor in the uncoating reaction by recruiting ATP bound Hsc70 proteins to CCVs (; ). In support of this, inhibition of auxilin function in vivo using yeast mutants, RNAi, or injection of interfering peptides can disrupt clathrin function (; ; ). Recent biochemical analysis suggests that auxilin participates in other steps of the CCV cycle, in addition to clathrin coat disassembly (). Still, it is unclear what the relevant endocytic cargo of auxilin may be under physiological conditions or whether auxilin has any role in regulating cell–cell signaling in metazoan systems. To further understand the roles of endocytosis in cell signaling during animal development, we sought to generate loss-of-function mutations in auxilin from an F complementation screen in . From this screen, we isolated six loss-of-function mutations in . In support of previous biochemical data, we find that auxilin interacts genetically with Hsc70 and clathrin. In addition, the location of the genetic lesion in one of our alleles suggests that the putative lipid binding tensin domain plays a role in regulating clathrin function. The mutations also interact specifically with and disrupt several Notch-mediated processes, suggesting that auxilin participates in an endocytic event critical for regulating the Notch cascade. Indeed, our analysis suggests that auxilin is required for internalization of the Dl proteins that are critical for activating the Notch receptor. The genome contains a single homologue (; hereafter referred to as ) located at the base of the third chromosome right arm (82A1). Conceptual translation of the ORF reveals a polypeptide of 1,165 amino acids, with an NH-terminal kinase domain, followed by a tensin-related domain, a clathrin binding domain, and a COOH-terminal DnaJ domain (). The presence of this NH-terminal kinase domain suggests that dAux is structurally more similar to the ubiquitously expressed cyclin G–associated kinase (; ) than the neuronal cell-specific bovine auxilin (). Indeed, as with cyclin G–associated kinase, dAux appears to be ubiquitously expressed throughout embryonic development, although higher levels of dAux expression are detected in embryonic Garland cell primordium and larval Garland cells (). To understand the role of auxilin under physiological conditions, we set out to isolate loss-of-function mutations in using an F noncomplementation screen with two deletions, (81F6-82A5) and (82A1-E7; FlyBase). Because of the cytological location of , we reasoned that loss-of-function mutations in should fail to complement by lethality but complement (). Using these criteria, we isolated one mutation in from ∼1,600 chemically mutagenized third chromosomes. Sequencing analysis of this mutant revealed a single nucleotide change, which alters the Ile at position 670 to a Lys in the tensin-related domain (). animals could survive until adulthood, suggesting that is a partial loss-of-function allele. Furthermore, the emergence of homozygous mutant adults suggests that there are no other recessive lethal mutations on that chromosome. . and , have been characterized molecularly and contain nonsense mutations at amino acids Trp328 and -1150, respectively (). However, animals homozygous for these stronger mutations die before the larval stages, precluding the phenotypic analysis of the larval tissues. exhibited several morphological defects, including rough eyes, extra bristles, missing wing veins, and male and female sterility. mutants were grossly disorganized, with patches of brown necrotic tissues on the surface (). mutants showed temperature dependence, as the eye roughness was significantly milder for mutants raised at 18°C () than those grown at 25°C (). At a higher temperature, such as 29°C, the homozygous mutant state was completely lethal. and are lethal over at 25°C, mutant animals of heteroallelic combinations for / and could occasionally survive until adulthood when raised at 18°C. at 25°C (), suggesting that the increase in the severity of eye defects correlates with the decrease in dAux activity. mutants using the upstream activating sequence (UAS)–GAL4 expression system (). (). mutants contained supernumerary vibrissae (, compare G and H) and, frequently, extra anterior sternopleural bristles. Occasionally, extra bristles were also detected on the notum or scutellum of mutant animals. Furthermore, at a low frequency, some mutant animals had wings with incompletely formed posterior crossveins, and absent wing vein material at the posterior wing vein margin (unpublished data). / and / raised at 18°C (, compare I and J). mutants resembled phenotypes, suggesting a link between the role of dAux in endocytosis and Notch signaling. interacts with and in vivo. We expressed a dominant-negative form of () in the eye using the driver (). Expression of this ATP hydrolysis-defective Hsc70-4 caused a rough eye (), presumably the result of defective endocytosis on developmental signaling pathways. intensified the rough eye phenotype, indicating genetic interaction of and in vivo (). To test whether dAux interacts with clathrin, we expressed the clathrin light chain (Clc) fused to GFP (; ) using in a homozygous background. Although expression of under the control of had no detectable effect on eye development or viability in wild-type animals, the expression of greatly reduced the viability of dAux mutants. In rare escapers, the eyes of /+ / were rougher, and there was dramatic enhancement of the wing phenotypes, including severe notching, wing vein thickening, and ectopic vein formation (, compare C and D). interacts genetically with and in vivo. To understand the role of dAux in vesicular trafficking, we examined the subcellular localization of dAux proteins, using a fluorescently tagged dAux fusion (). Both dAux-mRFP and Clc-GFP were expressed in larval Garland cells using Act5c-GAL4. Although intense vesicular Clc staining was seen around the cell periphery, the vesicular dAux-mRFP appeared more centrally localized and showed little overlap with (). interacts with the Clc in vivo, no apparent difference in Clc-GFP pattern was detected between wild type and homozygous mutant Garland cells by confocal microscopy (unpublished data). To further understand the roles of during development, we investigated the cause of the rough eye phenotype by tangential sectioning of the adult retina. In wild-type eyes, rhabdomeres of the eight photoreceptors are organized in a stereotypical manner in lattices of ommatidia (). mutant retina showed that regular arrays of ommatidia were disrupted. Furthermore, supernumerary photoreceptor cells were detected in 39.67% ( = 510) of mutant clusters (). The identities of these extra photoreceptors could be either outer (cells with large rhabdomeres; 21.42%) or inner (cells with small rhabdomeres; 18.25%), suggesting that is not affecting the determination of a particular photoreceptor cell fate. To understand the origin of these extra photoreceptors, we stained eye imaginal discs, the monolayer epithelia that give rise to adult eyes, with αElav antibody, which labels the nuclei of neuronal photoreceptor cells (). Organized clusters of eight normal Elav-positive cells were seen in wild-type eye discs (, inset). tissues, and there were supernumerary Elav-positive cells in some of the clusters (, inset). . / . The number of Elav-positive cells appeared to increase in eye discs mutant for stronger alleles (Fig. S1, available at ), suggesting that dAux activity is critical in controlling formation of the proper number of photoreceptor cells. To analyze this disruption in the organization of ommatidia arrays, we stained the eye discs with αBoss antibody, which specifically labels the apical surfaces of the R8 cells, the first photoreceptor specified in each cluster. mutant discs varied greatly, and clusters with multiple Boss-positive cells were occasionally detected (, inset). suggests a disruption in this process. The Notch signaling cascade participates in the formation of neuronal tissues during embryonic development. To determine whether dAux has a role in the specification of neuronal cell fate during embryogenesis, we inhibited dAux function using dsRNA injection and stained the injected embryos with αElav antibodies. 73% of embryos injected with dsRNA exhibited a strong neurogenic phenotype, with transformation of nearly all epidermis to neural tissues (). In contrast, injection of buffer () or dsRNA (not depicted) did not affect neural patterning, suggesting that the phenotype of embryos injected with dsRNA was specific. A quantitative summary of the phenotypes exhibited by injected embryos is tabulated in . This RNAi data, along with other dAux phenotypes, suggests that dAux acts in the general regulation of neuronal development. homozygous mutant females with wild-type males were stained with αElav antibodies. Elav staining of embryos derived from wild-type parents revealed highly organized central and peripheral nervous systems with characteristic numbers of neuropositive cells (). In contrast, embryos maternally deficient but zygotically heterozygous for showed mild hypertrophy in the ventral nerve cord, disorganization and slight reduction in the peripheral nervous system, and abnormal body morphology (). mutant adults. phenotypes of supernumerary photoreceptors, bristles, and embryonic neuronal cells are all reminiscent of those exhibited by mutants deficient in Notch signaling. To test for dAux participation in Notch pathway regulation, we asked whether the mutations interact with mutant alleles of . 36.36% ( = 44) of heterozygous , a null allele of , exhibits a haploinsufficient phenotype of wing “notching” at the posterior margins (). , increased the severity of both the penetrance (100%; = 36) and the phenotype (), indicating that interacts genetically with . represents a loss-of-function allele, a similar increase in the penetrance and the severity of the phenotype was seen with other alleles of , as well as the deletion that removes the entire locus. In addition to wing development, Notch is involved in many developmental decisions during retina formation, and overexpression of a full-length (), using the driver, causes a rough eye (), presumably disrupting some of these Notch-dependent processes. are normal, disruption of one copy of the gene strongly reduced the eye size and worsened the rough eye of (), suggesting that dAux has a role in regulating the Notch-mediated signaling pathway. To test the specificity of this interaction, we overexpressed a full-length () under the control of . EGF signaling is involved in the differentiation of all cell types during retina development (), and as for , overexpression of with causes a rough eye (). had no effect on the rough eye phenotype caused by (). This suggests that dAux does not participate in all signaling pathways required for eye development but that it may specifically regulate signaling activity of the Notch cascade. To ask which cells require functional dAux for activation of the Notch pathway, we performed mosaic analysis in the adult retina. We reasoned that if dAux is required in the signal-sending cells, mutant clusters near the border of the clone would be less affected than those near the center of the clone because they can receive signals from the nearby wild-type cells. On the other hand, if dAux is required in the signal-receiving cells, mutant clusters should exhibit defects regardless of their location in the clone. , marked by the absence of the gene, were generated by γ-ray irradiation. Tangential sections through these clones showed that, consistent with the homozygous mutant eyes, some mutant clusters contained supernumerary R-cells. Mutant clusters near the border of the clone were mostly wild type, whereas the mutant clusters in the center of the clones exhibited a stronger mutant phenotype (). Thus, it appears that dAux acts noncell autonomously in Notch activation. and a truncated form of Notch, which mimics the signal-competent Notch fragment after proteolytic cleavage (; ). Consistent with its role in lateral inhibition, expression of this activated form of in the eye discs using the driver greatly reduced the number of Elav-positive cells, indicating a strong inhibition of photoreceptor recruitment (). eye discs (). However, in mutant eye discs that also expressed the activated form of Notch, the number of Elav-positive cells was reduced (), indicating that activated is epistatic to . This suggests that dAux acts upstream of the generation of this signal-competent fragment of Notch. Recent evidence indicates that internalization of the DSL ligand may play a critical role in regulating Notch activity. Because our data suggest that dAux acts noncell autonomously and upstream of activated Notch in Notch signaling, we suspected that Dl endocytosis could be regulated by dAux. , and / mutant animals were dissected and stained with antibody raised against the extracellular domain of Dl. In wild-type eye discs, evenly spaced Dl staining was first seen in cells that are recruited to form photoreceptor clusters behind the morphogenetic furrow (). The Dl staining in these cells appeared to overlap with cortical phalloidin (not depicted), suggesting that most Dl protein initially localized at or proximal to the plasma membrane. In more mature clusters located in the posterior region of the eye disc, the Dl staining became more vesicular. These Dl-positive structures showed little or no colocalization with the Clc, Rab11, or Grasp65, which are markers for clathrin-coated structures, recycling endosomes, and Golgi, respectively, but overlapped moderately with Rab5 and extensively with Rab7, suggesting that most of the internalized Dl proteins are in the early and late endosomes (unpublished data). To ensure that our determination of Dl subcellular localization was not influenced by fixatives or histological techniques, we generated , a functional Dl chimera with an mRFP () fused to the intracellular COOH terminus of the Notch ligand, placed under the control of UAS regulatory element (). As with the antibody staining of endogenous Dl, Dl-mRFP in live eye discs also appeared vesicular and exhibited overlap with Rab5 and -7 endosomal markers (Fig. S2, available at ). In homozygous mutant eye discs, Dl staining appeared more excessive and disorganized in cells behind the furrow, reflecting the defects in photoreceptor specification (). Although Dl- and Rab7-positive vesicular structures could still be detected in mature clusters in the more posterior region of the mutant eye discs (not depicted), there appeared to be an increase in the peripheral staining of Dl in the first few rows of cells immediately posterior to the furrow (). This suggests that Dl internalization in these cells, where Notch signaling is thought to participate in the proper spacing of photoreceptor cell clusters, was disrupted. / eye discs showed even more severe disruption of Dl localization, with excessive, peripheral Dl staining extended from the region behind the furrow to the posterior edge of the disc, consistent with a greater decrease in dAux activity in heteroallelic animals (). mutants could result from either a block in Dl internalization and degradation or an increase in Dl gene expression. Because an increase in Dl expression in larval eye discs mutant for was previously reported (), it seemed likely that the level of Dl expression would be affected in mutant cells. To determine if this was the case, we monitored the transcriptional activities of the endogenous Dl promoter by measuring the β-gal activities from a Dl enhancer trap line (). eye discs (unpublished data), suggesting that the apparent increase in Dl staining was not due to elevated Dl transcription. To understand the physiological roles of J-domain–containing proteins during metazoan development, we isolated and characterized mutants in . mutation interacts genetically with and the . The in vivo link between auxilin and Hsc70 is further strengthened by the observation that a nonsense mutation () near the very COOH terminus, where the J-domain is located, can strongly disrupt dAux function. These genetic observations are in agreement with in vivo analyses of auxilin function from other systems, which showed that clathrin function was disrupted in auxilin-deficient cells (; ; ). suggest a relevance of the tensin-related domain, a putative lipid binding domain, in clathrin-mediated endocytosis, despite the fact that it does not appear to be required for catalyzing the dissociation of clathrin triskelions from CCVs in vitro (; ). It has been suggested that, in addition to disassembling clathrin coats, auxilin participates in the dynamin-mediated constriction during CCV formation (). However, our subcellular localization analysis did not reveal dAux proteins colocalizing with clathrin at the cell periphery. Instead, most auxilin proteins appear to be associated with intracellular structures, in regions devoid of clathrin staining. This lack of overlap between dAux and Clc seems more consistent with the notion that auxilin is required for the dissociation of clathrin coats from CCVs under physiological conditions. Our analysis of clearly suggests that auxilin plays an important role in the Notch cascade in multiple Notch-dependent processes. Supportive evidence comes from the strong genetic interactions between dAux and Notch and the phenotypic similarities ranging from eye and wing development to neural development during embryogenesis. has no dominant effect on the phenotype caused by the overexpression of EGFR. Together, these observations argue that dAux acts specifically as a general component in the Notch cascade. Analysis from several groups has suggested that ligand internalization is a key event for Notch activation. tissues and other genetic data further support this notion. The distribution of phenotypically mutant clusters in a genotypically mutant clone suggests that dAux acts noncell autonomously. In addition, the epistasis analysis places dAux function upstream of an activated form of Notch. to those reported for (; ) and (), we suspect that dAux functions along with neur and lqf in the ubiquitin-dependent endocytic pathway in the signal-sending cells. The identification of dAux as a critical factor in Notch ligand endocytosis has strong implications on the mechanism of Notch activation. Unlike neur and lqf, which are postulated to tag and sequester cargos into vesicles, auxilin is thought be involved in disassembly of clathrin coats. Thus, the revelation of dAux as another component in this pathway suggests that Dl-containing endocytic vesicles need to proceed past the clathrin uncoating step to activate Notch. One possible mechanism is that recycling of Dl is a prerequisite to form signaling-competent Dl-containing exosomes (), although the presence of these structures under physiological conditions remains to be demonstrated. Alternatively, it may be that, as previously proposed, the DSL ligand is not signaling competent before endocytosis but is “activated” during transit through recycling compartments. Indeed, the transit through Rab11-positive recycling endosomes has been suggested as a critical step for Dl activity (). However, although Dl appears to colocalize extensively with coalesced perinuclear Rab11-positive structures in the sensory organ precursor cells (), our analysis found little spatial overlap between Rab11 and Dl in cells near the furrow. One possible explanation for this apparent difference is that the transit of Dl through Rab11-positive structures in the eye disc cells occurs more transiently, therefore evading detection by immunostaining at a steady state. Another explanation for the relevance of ligand endocytosis hypothesizes that Dl internalization causes a mechanical stress on the Notch receptors, which then induces subsequent cleavages. A variation of this model proposes that the objective of Dl internalization is to remove the NECD fragment from the intercellular space so proteolytic processing can occur. If auxilin is solely involved in clathrin-coat disassembly, it will be difficult to reconcile our data with these two models because the internalization of Dl into CCVs, the presumed force-generating event, should have already been completed in mutants. All fly crosses were performed at 25°C in standard laboratory conditions unless otherwise specified. To screen for loss-of-function alleles, males were mutagenized with 25 mM ethyl methane sulfonate (Sigma-Aldrich) and mass mated with virgins. Progeny were then individually mated with flies, and those that failed to complement the deletion were recovered and maintained over or balancers. Using these criteria, 11 lines were isolated from ∼1,600 crosses. They were then mated with flies, and those that complemented were characterized further. Of the 11 lines, 3 complemented . For the genetic interaction and epistasis analysis, (), (), UAS-Notch, (), (), , , and were used. To label subcellular structures, (), (), (this study), (this study), and () were used. For mosaic analysis, flies were mated with . First instar larvae were then irradiated with a 1,000 rad γ-ray (Gammacell 220), and flies containing mosaic clones were identified by the presence of w patches on the adult retina. Immunostaining of eye discs and tangential sections of adult retina were performed according to . Embryos were aged to stage 15 after injection and fixed as described previously (). Rat α-Elav 7E8A10 (Developmental Studies Hybridoma Bank), mouse α-Boss, and mouse α-Dl C594.9B (Developmental Studies Hybridoma Bank) were used at 1:100, 1:3,000, and 1:100 dilutions, respectively. Alexa Fluor 568 phalloidin (Invitrogen) and fluorescently conjugated secondary antibodies were used according to the manufacturer's instructions. Fluorescently labeled samples were mounted in VECTASHIELD Mounting Medium (Vector Laboratories). Light micrographs and fluorescent images were acquired at 25°C with 20× (0.5) and 40× (0.75) lenses on a microscope (BX61; Olympus) equipped with a camera (DP70; Olympus) and DP Manager software. All confocal microscopy images were acquired at 25°C with 20× (0.5) and 60× (1.25) lenses using a confocal microscope (OPTIPHOT-2 [Nikon]; MRC1024 system [Bio-Rad Laboratories]) and LaserSharp 3.0 software (Bio-Rad Laboratories). Images were 3D reconstructed in Volocity (Improvision) and then processed in Photoshop (Adobe) to adjust γ levels and image size. Adult wings were dissected and mounted in Gary's magic mountant. Scanning electron microscopy was performed as previously described with a scanning electron microscope (JSM-840 [JEOL]; ). To construct UAS-dAux, an EcoRI–XhoI fragment containing the entire auxilin (CG1107) ORF was excised from GH26574 (Research Genetics) and subcloned into pUAST. To construct UAS-dAux-mRFP, a 0.7-kb COOH-terminal portion (containing the internal SpeI site) of was PCR amplified and cloned into pHFK-mRFP-KB as an EcoRI–XhoI fragment. After sequencing verification, the NH-terminal half was inserted as a 2.9-kb EcoRI–SpeI fragment, and the entire fusion was then subcloned as an EcoRI–NotI fragment into pUAST. To construct UAS-Dl-mRFP, PCR-amplified mRFP was fused the to the COOH terminus of Dl (pBS-Dl, DGRC) as an NdeI–HindIII fragment, and the resulting Dl-mRFP was cloned into pUAST an EcoRI–XhoI fragment. To construct UAS-GFP-Rab11, Rab11 coding sequence was amplified by PCR and subcloned as a ClaI–BamHI fragment into the COOH terminus of GFP in pHFK-GFP-RC. To construct UAS-dGrasp65-GFP, dGrasp65 coding sequence was amplified by PCR and subcloned as an EcoRI–KpnI fragment into the NH terminus of GFP in pHFK-GFP-KB. The resulting GFP-Rab11 and dGrasp65-GFP fusions were verified by sequencing and subcloned as NotI fragments into pUAST, respectively. Transgenic flies carrying these constructs were generated by P-element–mediated transformation as previously described (). Synthesis of dsRNA was done according to . PCR primers bearing T7 promoter sequence at the 5′ ends were used to amplify 610- and 599-bp fragments of (5′-TCGAGTCGACGTACAAGACG-3′ and 5′-GCCTGATACAACCGCATTTT-3′) and (5′-GTCAGTGGAGAGGGTGAAGG-3′ and 5′-CCCAGCAGCTGTTACAAACTC-3′) coding sequence, respectively. In vitro transcription of the PCR fragments produced dsRNAs, which were then prepared as 5-μM aliquots for injection as described previously (). RNAi injections into embryos were done as previously described (). Fig. S1 shows that the severity of photoreceptor recruitment defects correlates with the decrease in dAux function. Fig. S2 shows that Dl proteins are localized in Rab5- or Rab7-positive vesicular structures in wild-type eye discs. Online supplemental material is available at .

Polarized cells have an uneven distribution of active ATPases and they position their mitochondria so that ATP is produced close to where it is needed. Transportation of mitochondria to meet local energy needs is especially critical in neurons, where the site of mitochondrial production in the cell body () can be centimeters away from a growth cone or synapse with high local ATP demand. Mitochondria move along both microtubules and actin, using microtubule-based molecular motors for long distance movements (). They are among the most abundant and most mobile membrane-bound organelles and, thus, are a major cargo for microtubule motors. Conventional kinesin moves mitochondria to the plus ends of microtubules, while dynein moves them toward the minus ends (; ). Mutations in kinesin motors disrupt organelle transport causing mitochondria to bunch up in the axon or cell body, leading to neuronal dysfunction (). Mitochondria are distributed in cells with exquisite fine-tuning of both their location and number, and their transport is likely to be a very well-regulated process. Their distribution varies in response to multiple regulatory cues such as energy requirements, growth factors, or the membrane potential of the mitochondria (; ; ). For example, observed increased anterograde transport of mitochondria to active versus inactive growth cones, while showed that mitochondria accumulate at local sites of nerve growth factor application. These experiments left two major questions: How do mitochondria connect to molecular motors? How is their movement by molecular motors controlled? A key finding for addressing these two questions came from recent genetic screens in . Defects in axonal transport were lethal at the embryonic or larval stages in previous screens, hampering the identification of proteins involved in axonal transport of mitochondria in . However, Stowers and colleagues created mosaic flies whose eyes were homozygous for a mutant allele while the rest of the body was heterozygous (). Here, mutant flies were viable, but blind due to a loss of nerve excitation in the eye. Two independent screens performed by Stowers et al. and Guo et al. used this system to identify two distinct components important for transport; milton (), which coimmunoprecipitated with kinesin heavy chains, and miro (), an integral mitochondrial membrane protein. Mutations in either of the genes appeared to abolish anterograde mitochondrial transport. The present paper by links these two results by showing both that kinesin, milton, and miro work together in anterograde transport and that milton attaches kinesin to mitochondria through miro. #text xref italic #text It is very likely that miro is not only an adaptor for milton, but is also a critical regulator of kinesin-dependent mitochondrial transport. Potential mechanisms of regulation of transport by the miro–milton complex are shown in . Miro is a GTPase with both two GTP-binding domains and two EF hand domains that can potentially bind calcium. This means that either GTPase activity or calcium binding can regulate miro's conformation and, therefore, its ability to recruit milton or arrange the milton–kinesin complex at the surface of mitochondria. The existence of several splice variants of milton with different kinesin and miro binding properties implies that there might be several populations of mitochondria with different transport properties. None of these potential regulatory mechanisms has yet been tested, but some are very likely to occur. In addition to recruiting kinesin via milton, miro may have other important mitochondrial functions. Miro is present in yeast, whereas milton is not, and it is known to play a role in maintaining normal mitochondrial morphology. Furthermore, yeast use actin rather than microtubules for mitochondrial transport, so the function of miro in yeast is clearly different. Could miro be a more general mitochondrial adaptor that binds to other motile complexes (such as Arp2/3, myosin V, or dynein) besides milton? Recent observations of mitochondria transport in fly neurons in vivo demonstrated that there are two populations of mitochondria; one moves predominantly anterogradely while the other moves retrogradely (). Could it be that the GTPase or calcium “switch” on miro toggles between these two states or between microtubule- and actin-based transport? It will take time to determine the exact role of miro in motor-based mitochondrial transport, but in the short term it is reasonable to ask whether kinesin is bound to retrogradely transported neuronal mitochondria. If not, could it be dissociated by GTP hydrolysis of miro? This is an exciting and important area for further study because miro is likely to be the key universal adaptor and regulator for mitochondrial transport. #text

Ubiquitylation is a key posttranslational regulator of protein activity, stability, and/or localization. That the addition of this highly conserved 76–amino acid moiety can result in such a range of outcomes is, in part, a result of the variety of ways ubiquitin can be covalently attached to the substrate protein. Ubiquitylation is the formation of an isopeptide bond between a substrate lysine and the COOH group of the COOH-terminal glycine of ubiquitin. The addition of a single ubiquitin is termed monoubiquitylation or, if several substrate lysines are modified in this manner, multiubiquitylation. However, because ubiquitin itself has seven lysines, which can also be subjected to isopeptide formation, a polyubiquitin chain can form. Polyubiquitylation can have several forms, depending on which lysine is used. All lysines (K6, 11, 27, 29, 33, 48, and 63) can be subjected to chain formation, but the most common linkages are K48 and K63. K48 polyubiquitylation is a very efficient tag for marking proteins for degradation at the proteasome, whereas K63 chains are involved in nonproteasomal functions, including protein trafficking and DNA repair (for review see ). Mono- and multiubiquitylation are major regulators of protein trafficking in both the exocytic and endocytic pathways. Ubiquitin can affect protein trafficking in two ways: either by direct attachment to the cargo protein (cis-regulation) or modification of the protein-trafficking machinery (trans-regulation; ). Ubiquitylation of cargo proteins at the plasma membrane can be sufficient to induce endocytosis, and it also acts as a sorting signal at the TGN and the multivesicular body (MVB; ). Furthermore, ubiquitin may participate in vesicle targeting, as several proteins involved in vesicle docking and fusion bind ubiquitin (). At the plasma membrane, the ubiquitin signal is recognized by epsin and Eps15, and, at the MVB, it is recognized by a complex containing hepatocyte growth factor–regulated tyrosine kinase substrate (Hrs) and signal-transducing adaptor molecule (STAM). Subsequent interactions with protein complexes ESCRT (endosomal sorting complex required for transport) I, II, and III result in the targeting of ubiquitylated cargo into internal vesicles to be degraded at the vacuole/lysosome (; ). Many of the protein-trafficking machinery components are themselves ubiquitylated without being subject to proteasomal degradation, raising the possibility of the regulation of cargo-trafficking machinery interaction by reciprocal ubiquitylation status (; ). In addition to conjugation, substrate ubiquitylation status may also be regulated by ubiquitin cleavage, which is performed by deubiquitylating enzymes (DUBs). The human genome encodes for ∼80 DUBs, which can be divided into five classes on the basis of differences in the catalytic domain (; ). Four classes of DUBs—the ubiquitin COOH-terminal hydrolases (UCHs), ubiquitin-specific proteases (USPs), Machado-Joseph disease protein domain proteases, and ovarian tumor proteases—are cysteine proteases, whereas the JAMM motif proteases are metalloproteases. Although DUB function does not neatly segregate along class lines, a few generalizations can be made. The substrates for UCHs tend to be small peptides (20–30 amino acids), and UCHs primarily function in the recycling of ubiquitin. The USPs are the largest group, with ∼55 members in humans (). They contain a characteristic catalytic core, which is defined by a small number of motifs flanked by large NH- and/or COOH-terminal extensions. Except for the catalytic core motifs, which only extend over ∼100 amino acids, there is no other homology between USP family members. The divergent NH- and COOH-terminal extensions are proposed to impart substrate specificity. Along with the substrate-specific E3 ligases, USPs are the only classes of ubiquitin-modifying enzymes to have expanded significantly throughout evolution (). Information on the other three classes of DUBs is fairly sparse and relatively recent, preventing generalizations; however, individual members of the ovarian tumor protease and JAMM classes are involved in the regulation of protein trafficking (as detailed below). Protein trafficking is regulated by DUBs in several ways, such as maintaining cellular levels of free ubiquitin, antagonizing the degradation of trafficking proteins in the ubiquitin–proteasome system, and regulating nonproteasome-dependent functions of monoubiquitin or multiple monoubiquitin signals. Secreted and membrane proteins enter the exocytic pathway at the ER. The ubiquitin pathway is a major regulator of protein quality at this juncture, as misfolded proteins inserted into the ER are very rapidly recognized, ubiquitylated, and degraded by proteasomes in the cytoplasm (). Whereas some common characteristics of misfolded proteins may be recognized by the ubiquitylation machinery, it has recently been shown that one DUB, USP4, plays a very substrate-specific role at the ER (). USP4 associates with the cytosolic COOH terminus of the A-adenosine receptor, a Gs-coupled receptor (). This regulates the quality control of A-adenosine receptor, facilitating the passage of the receptor through the ER and Golgi and resulting in increased A-adenosine receptor at the plasma membrane. Several controls showed that this increase was caused by facilitated transport through the exocytic pathway and not by recycling from the endocytic pathway (). This interaction was specific, as another DUB, USP14, could not substitute for USP4, and USP4 had no effect on the trafficking of other G protein–coupled receptors. As >50% of the A-adenosine receptor is degraded at the ER, the cis-regulatory function of USP4 has the potential to be a significant and specific regulator of A-adenosine receptor function at the plasma membrane. The regulation of traffic between the ER and cis-Golgi is also affected by DUB activity, but this time it involves trans-regulation. In yeast, the DUB Ubp3 participates in the stabilization of two trafficking proteins, Sec23, a COPII subunit protein, and β′-COP, a COPI subunit, that are required for ER to Golgi transport and Golgi to ER retrograde transport, respectively (; ,). In this role, Ubp3 has an essential cofactor, Bre5, and deletion mutants of either result in defects in the bidirectional transport between the ER and Golgi. Ubp3 appears to be essential for cleavage of the isopeptide bond between Gly76 of ubiquitin and the ubiquitylated lysine residue of the Sec23 and β′-COP substrates. Accumulation of monoubiquitylated forms of Sec23 and β′-COP and an increased turnover of these proteins are seen in both ubp3Δ mutants and bre5Δ mutants. The enzyme–substrate relationship between Ubp3 and Sec23 was further confirmed by demonstrating direct protein–protein interaction and the inability of a catalytically inactive mutant to eliminate the monoubiquitylated form of Sec23 in ubp3Δ cells (). Notably, the monoubiquitylated Sec23 did not dissociate as freely as wild type from the ER once bound, and it failed to interact with another component of the COPII complex (). Thus, ubiquitylation leads to the down-regulation of Sec23 not only by leading to its degradation but also through altering its biochemical properties such that it can no longer perform its function in ER to Golgi transport. Data concerning whether such relationships have been conserved by mammalian cells is preliminary but promising. Both Ubp3 and its cofactor Bre5 have human homologues (UBP10 and G3BP1/2 [Ras–GTPase-activating protein SH3 domain–binding protein], respectively). The ability of G3BP to modulate the deubiquitylating activity of UBP10 has been demonstrated in vitro on artificial substrates (), and a yeast two-hybrid assay has been used to demonstrate protein interaction between USP10 and bovine β′-COP, whereas no interaction was observed between USP10 and five other components of the COPI complex (). Thus, it appears likely that the functional relationship between a specific USP-type DUB and protein transport between the Golgi and ER is evolutionally conserved. DUBs also influence the exocytic pathway in a more general manner by regulating the dynamics of organelle reassembly. The VCIP135 DUB is an essential cofactor for p97–p47-mediated Golgi and ER reassembly (; ), a process that is required after cell division. For Golgi reassembly, although the role of VCIP135 is dependent on its DUB activity, it acts independently of the proteasome. However, a ubiquitin mutant that cannot bind p97–p47 was found to inhibit Golgi reassembly when added to the system before disassembly (). It may be inferred from this that a ubiquitylation-dependent interaction between p97–p47 and an unknown protein, which becomes ubiquitylated before Golgi disassembly, is required for Golgi reassembly. This as yet unidentified protein is clearly a candidate for a VCIP135 substrate. Regardless of substrate identity, VCIP135's involvement in Golgi and ER reassembly demonstrates that a DUB can functionally participate in membrane fusion events, which are key components of membrane protein-trafficking events. The endocytic pathway begins with the internalization of plasma membrane proteins, which is followed by multiple sorting events at the early/recycling endosome and the late endosome/MVB compartments. These ultimately either return the protein to the plasma membrane or deliver it to the lysosome for degradation. Most, if not all, plasma membrane proteins in yeast are endocytosed in a ubiquitin-dependent manner, indicating that ubiquitylation is an ancient signal for trafficking (). In mammalian cells, the best-studied examples of ubiquitin-dependent endocytosis involve ligand-activated receptor tyrosine kinases. Both E3 ubiquitin ligases and DUBs regulate endocytic traffic both in cis and trans. Epsin is a trafficking accessory molecule involved in both clathrin-mediated endocytosis () and the internalization of ubiquitylated cargo in a clathrin-independent manner (; ; ). That epsin may be regulated by a DUB was first indicated by genetic studies in , where mutations in the liquid facets (, epsin) gene were dominant enhancers of the eye defect observed in mutants of the DUB fat facets (), a USP-type DUB (). Two experiments support the enzyme substrate relationship between and . First, the overexpression of replaced the requirement of during eye development, as would be expected if faf's role was to stabilize lqf (). Also, the deubiquitylation of lqf by wild-type faf but not a catalytically inactive mutant stabilized lqf (). Consequently, it has been concluded that in , faf opposes the ubiquitin–proteasome-mediated degradation of lqf. Subsequently, it was found that lqf is essential for directing the endocytosis and subcellular localization of Delta in Delta/Notch signal-sending cells such that Delta can be activated (). Perturbation of this function is the basis for the eye phenotype of faf-null and lqf-null mutants (). The relationship between the homologues of lqf (epsin1) and faf (FAM/USP9X) is conserved in higher vertebrates but is likely to have functional differences. Epsin1 does not appear to be polyubiquitylated and degraded by the proteasome but is instead monoubiquitylated (). The levels of monoubiquitylated epsin1 decrease simultaneously with a global decrease in ubiquitylated proteins upon calcium-induced depolarization in rat synaptosomes or stimulation of calcium signaling in nonneuronal cell types (). Loss of monoubiquitylated epsin1 was specifically prevented by siRNA-mediated FAM/USP9X knockdown. It was proposed that the ubiquitylation of epsin may prevent its interaction with several binding partners, including lipids, AP-2, and clathrin; therefore, FAM-mediated deubiquitylation activates epsin (). Curiously, epsin is not the only link between the FAM/USP9X DUB and membrane protein trafficking. FAM colocalizes with markers for several protein-trafficking compartments, including the TGN and late endosomes, and there is strong circumstantial evidence linking FAM/USP9X to trafficking of the E-cadherin–β-catenin complex in epithelia (). Additionally, doublecortin, an essential neural protein that associates with microtubules and clathrin adaptor proteins AP-1 and AP-2 (), is a binding partner but not a substrate of FAM (). Other DUBs may regulate the endocytic traffic of specific cargo at the plasma membrane. In mammalian cells, activated G protein–coupled receptors associate with β-arrestin, and both proteins are ubiquitylated before internalization of the complex. In this instance, ubiquitylation of the β-arrestin adaptor protein rather than the receptor itself is required for receptor internalization (). Different G protein–coupled receptors show different recycling kinetics that correlated with the ubiquitylation status of the associated β-arrestin. Furthermore, overexpression of a β-arrestin–ubiquitin fusion, such that β-arrestin could not be deubiquitylated, slowed the recycling kinetics of fast recycling receptors and also led to enhanced receptor internalization and degradation (). This study strongly suggests that fast recycling receptors recruit DUBs to act on β-arrestin. In yeast, the soluble form of Ubp1 may also function at the plasma membrane or sorting endosome to deubiquitylate an as yet unidentified component of the protein-trafficking machinery and recycle ubiquitylated proteins. This was based on the observation that the overexpression of soluble Ubp1 disrupted the lysosomal trafficking of the ATP-binding cassette transporter protein Ste6 as well as the α-factor receptor Ste2 even though Ubp1 did not alter the ubiquitylation status of either cargo (). However, Ubp1 deubiquitylating activity was required as a catalytically inactive form and had no effect on either plasma membrane protein (). A subsequent critical junction for protein sorting in the endocytic pathway occurs at the endosomes. Precisely how protein sorting at the endosomes is regulated is far from clear, except to say that it involves the interdependent interactions between individual components of large multiprotein complexes. Ubiquitin influences endosomal sorting at three levels: if cargos remain ubiquitylated, they are ultimately fated for degradation in the lysosome; many of the accessory proteins are ubiquitylated; and they may also contain motifs that bind ubiquitin or ubiquitin-like domains on cargo or other accessory proteins. Delineating a precise role for DUBs is difficult, as for any protein in these processes, but several recent studies have made significant progress in understanding the role of two: USP8 (UBPY) and AMSH (associated molecule with the SH3 domain of STAM), a JAMM-class DUB. Knockdown of USP8 levels using siRNA significantly inhibits the down-regulation of ligand-activated growth receptors such as EGF receptor (EGFR) and Met (; ; ). Conversely, AMSH negatively regulates EGFR down-regulation and is proposed to recycle the EGFR at the sorting endosome (; ). The molecular mechanism underlying these opposing effects is hinted at by the observation that USP8 and AMSH bind a central SH3 domain of STAM proteins in a mutually exclusive manner (). STAM and its constitutive binding partner Hrs participate in recognizing ubiquitylated cargo on early endosomes (), leading to the sorting of such cargo into internal vesicles at the MVB (). Overexpression of a STAM2A mutant lacking its SH3 domain interfered with the lysosomal degradation of PDGF and its receptor (; ). A proportion of USP8 and AMSH colocalize with STAM at endosomes (; ). The functional relevance of the interaction between USP8, AMSH, and STAM in the endocytic trafficking of activated growth factors is supported by other observations, including that catalytically inactive AMSH, a potential “substrate trap” mutant, resulted in the accumulation of ubiquitin on endosomes, an increased association with STAM, and the generation of a minor product consistent with ubiquitylated STAM (). Further interactions between AMSH and other trafficking accessory proteins such as clathrin heavy chain and a component of ESCRT III indicate that AMSH is “deeply integrated as a hub protein within the MVB-sorting protein interaction network” (). It was also observed that the associations between AMSH and trafficking machinery was reinforced by the simultaneous binding of STAM, with the subsequent activation of AMSH coupled to its association with the MVB-sorting machinery (). The role of USP8 in facilitating the passage of EGFR and Met to the lysosome has been supported by the observation that USP8 can deubiquitylate monoubiquitylated growth factor receptors as well as act on both K48- and K63-linked ubiquitin chains in vitro (). USP8 is recruited to endosomes upon EGF stimulation but shows no association with endosomes in starved cells in contrast to AMSH (). Interestingly, when USP8 is depleted, STAM becomes destabilized, which is a process dependent on the proteasome (). Therefore, part of USP8's function may be to maintain STAM levels. However, it is proposed that USP8 might regulate multiple components in the endocytic pathway, such as the growth factor receptors themselves. Precisely where in the endocytic pathway a DUB might deubiquitylate a receptor is critical, as it could result in opposite effects. If USP8 deubiquitylates EGFR at the MVB, this facilitates EGFR's progression toward degradation in the lysosome and, thus, aids receptor down-regulation (; ). However, it has also been suggested that USP8 might be active at the sorting endosome, in which case the deubiquitylated receptor is recycled to the plasma membrane and is ready for another round of signaling (). It is not clear why the data of showed an opposite effect on EGFR by USP8, but it is supported by their observations that upon EGF stimulation, USP8 directly binds EGFR. Another DUB UCH37 has been suggested to deubiquitylate activated type I TGF-β receptor, thereby preventing its down-regulation, but it was not shown whether this occurs at the plasma membrane or at the sorting endosome (; ). In the ubiquitin–proteasomal system, coupling of DUBs to the proteasome is necessary for ubiquitin recycling (), and one yeast DUB has been shown to play an analogous role at a late stage of the ubiquitin–lysosome system. Doa4 (UBP4) was one of the first identified yeast DUBs, and it was noted that in doa4Δ mutant cells, many substrates of the ubiquitin–proteasome system were stabilized as a result of a depletion of ubiquitin (; ). Interestingly, a screen for genetic suppressors of the doa4 phenotype identified members of the vacuolar protein-sorting pathway and not components of the proteasomal system (). Further examination of these mutants led to the conclusion that Doa4 acts at the late endosome/prevacuolar compartment to recover ubiquitin from membrane proteins before their sorting into internal vesicles and subsequent lysosomal degradation (; ; ). A vertebrate homologue of Doa4 has not been identified, although deubiquitylation of the specific cargo protein EGFR has been shown to occur before its lysosomal degradation (). This deubiquitylation event was inhibited by the proteasomal inhibitor lactacystin, which delayed but did not prevent the lysosomal degradation of EGFR, leading to the intriguing possibility that in vertebrates, the deubiquitylating activity of the proteasome might fulfill the role played by Doa4 in yeast. The study of ubiquitylation and membrane protein trafficking is an exciting and relatively recent field. Although the study of DUBs has lagged behind that of ubiquitin ligases, it is already apparent that DUBs play key and varied roles in protein trafficking. Through participation in membrane protein-trafficking events, DUBs regulate protein localization and stability, membrane fusion, signaling pathways, and developmental events. The current deficiencies in our knowledge are many and mostly self-evident, but two major obstacles need to be overcome for significant progress to be made. The first concerns the observation that many DUBs having multiple subcellular locations and substrates/binding partners. The second challenge comes from the observation that many DUBs and E3 ligases are complexed together and often regulate both the activity and stability of themselves and each other as well as common substrates (). The possible permutations and combinations inherent in these interactions make drawing simple linear models from over- or underexpression studies nearly impossible. These complex, reciprocal regulatory networks are reminiscent of other cell signaling pathways, and so an appreciation of what has been learned from phosphorylation signaling pathways will be useful in answering the “where to from here?” question. The regulation of kinases and phosphatases in time and space determine cell signaling dynamics (), so for the study of DUBs in protein trafficking, it will be critical to include techniques such as fluorescence resonance energy transfer to determine exactly which subcellular pools of DUBs and substrates are interacting. It will also be important to molecularly dissect the DUBs to identify individual localization signals and/or specific substrate-binding sites so that more precise questions can be addressed. The interaction between DUBs and E3 ligases echoes a universal motif found in cellular networks in which kinases, phosphatases, guanine nucleotide exchange factor, and GTPase-activating proteins are in complexes regulating the same substrate (). Far from representing a futile cycle, such arrangements can provide ultrasensitivity to signaling pathways and are modulated by controlling the localization and recruitment of the different enzymes to the complex (). Clearly, defining DUB–E3 ligase pairs and the signals that recruit them to specific points of protein trafficking will represent a major step forward along the pathway, for although ubiquitylation may mark the beginning of a protein's journey, it is not over until DUBs signal the final destination.

The nuclear envelope (NE) of eukaryotic cells consists of the inner nuclear membrane (INM) and outer nuclear membrane, as well as nuclear pore complexes (NPCs), which span both membranes and mediate transport processes. In metazoa, the NE breaks down before mitosis and is reformed after chromosome segregation. This reassembly of the NE starts in late anaphase with a rapid accumulation of membranes around chromatin. In living cells, this membrane recruitment happens within minutes, whereas the subsequent expansion and maturation of the NE takes at least 1 h (). In cell-free extract systems, such as the egg extract, NE assembly can be reconstituted in vitro (). Similar to the situation in vivo, membrane vesicles attach to sperm chromatin within minutes, followed by a much longer phase of NE maturation (for reviews see ; ). Binding of membranes is independent of energy or cytosol and is not restricted to defined regions on chromatin. However, one important prerequisite for membrane recruitment in this system is the decondensation of chromatin. This is mediated by nucleoplasmin, which is a protein that removes basic proteins and protamines from sperm chromatin and allows the deposition of histones (Fig. S1, available at ; ). Decondensation probably exposes binding sites for membrane vesicles, which in turn efficiently accumulate on chromatin. Hence, at the onset of NE assembly, the accessibility of chromatin-binding sites and a dramatic change in the affinity of membranes for chromatin are critical. However, it is unclear what mediates this initial interaction between membranes and chromatin. In vitro studies demonstrated that specific populations of membrane vesicles exist that bind to chromatin and function in NE assembly (; ). The affinity of membranes for chromatin is thought to depend on transmembrane proteins and is modulated by mitotic phosphorylation (; ). Two nuclear transmembrane proteins that directly bind chromatin in vitro, lamin B receptor (LBR) and lamina-associated polypeptide 2β (Lap2β), have been identified. There is also evidence that, at least in some systems, LBR can target membranes to chromatin (; ), but there is no evidence that the depletion of either protein would affect NE assembly. In contrast, much less is known about the nature of the binding sites on chromatin. Both LBR and Lap2β interact with chromatin proteins (HP1 and BAF, respectively), but they also bind to naked DNA (; ). LBR has a higher affinity for DNA than for chromatin proteins (). BAF interacts with other integral membrane proteins of the NE, including emerin and MAN1, which contain the so-called LEM domain (for review see ). However, there is no evidence that HP1, BAF, or histones are directly involved in membrane recruitment during NE assembly. On the other hand, a direct test for the involvement of DNA is difficult to perform, as chromatin templates are destroyed upon the removal of DNA (; unpublished data). In previous NE assembly studies that used protein-free DNA, membrane binding was only investigated after the DNA was converted into chromatin (; ). In this analysis, we address directly whether NE precursor membranes interact with DNA and provide evidence that membrane–DNA interactions are critical during NE assembly. In the first experiment, we tested whether DNA could compete with chromatin for binding of membranes during NE assembly. sperm chromatin was incubated with boiled cytosol to allow initial decondensation and then transferred to cytosol containing membranes and plasmid DNA as a competitor. We found that at early time points (after 10 min) almost no vesicles were recruited to chromatin in the presence of competitor DNA (, row 3) and that this effect on vesicle recruitment was not dependent on the presence of cytosol (, rows 5 and 6). After 2 h, control reactions showed normally shaped nuclei with fully decondensed chromatin and a smooth membrane staining (, row 1). No such structures were detectable in samples containing competitor DNA (, row 2). Although membrane vesicles were attached to chromatin, they did not form a smooth NE, and the chromatin did not fully decondense. To test whether plasmid DNA, indeed, competed with chromatin for membranes, we added more membranes, cytosol, or buffer to the reactions. Only additional membranes could rescue the inhibition by competitor DNA to allow normal closed nuclear formation (, row 4, and B, rows 3–5). To investigate whether plasmid DNA has an unspecific, inhibitory effect on membranes, we added the competitor DNA at different time points after the initiation of nuclear assembly. The inhibitory effect of DNA depended on its presence early on in assembly (). The number of nuclei formed returned to control levels when DNA was added 30 min after the initiation of assembly (), at a time when the NE still has to expand substantially. This indicates that DNA does not generally affect NE assembly. Instead, it seems to interfere with specific membranes that attach to chromatin early in the assembly process. The data suggest that DNA competes with chromatin for the binding of these membranes. To investigate the basis for inhibition in more detail, we incubated membrane vesicles purified from egg extracts with protein-free DNA that was immobilized on magnetic beads. Note that because all known chromatin proteins and assembly factors are soluble, this treatment should detect direct DNA interactions, rather than those that depend on chromatin assembly. The beads were removed, and the remaining vesicles were transferred to cytosol to analyze their ability to form a NE around sperm chromatin. The quantity of membranes added from the control and DNA-depleted samples was normalized by protein content. In control reactions, membranes were efficiently targeted to chromatin after 10 min, and after 120 min normal nuclei had formed (, middle row). The chromatin was fully decondensed and NPCs had assembled. In contrast, membranes that were passed over the DNA column only rarely formed normal nuclei (, top row; for quantitation). After depletion over DNA beads, almost no membranes were detectable on chromatin after 10 min. After 120 min, the chromatin was associated with membrane vesicles and still condensed. In addition, very little NPC immunofluorescence signal was detectable. A punctate staining was observed, which was only marginally stronger than the background in control reactions without membranes (, bottom row). Hence, although membranes eventually accumulated on chromatin, this recruitment was significantly delayed and did not yield a functional NE. These effects were not observed when membranes were passed over a column to which the negatively charged polymer heparin sulfate was attached (unpublished data), suggesting that the DNA column was not just acting as a nonspecific negatively charged ion exchanger. The fusion capacity of the membranes not removed by the DNA column was not significantly affected, as they still efficiently formed an ER-like network () on coverslips (). This indicates that the negative effects shown in are specific for NE formation. To test whether transmembrane proteins with a known affinity for DNA were depleted in the inactive supernatants, we analyzed the membranes from three separate depletion experiments by Western blot (). eggs contain a specific isoform of Lap2β, which is named Lap2ω (), and a variable amount of this protein and of LBR were removed. However, in most cases there was still a considerable quantity of both proteins left in the unbound fraction. This suggests either that a modest reduction in these transmembrane proteins is sufficient to block NE assembly or that unknown membrane proteins required for NE formation are efficiently depleted by the DNA column (see the following paragraph). We conclude that the DNA column removed vesicles with an affinity for DNA and that these membranes are among those normally targeted to chromatin and required during NE assembly. Depletion of these vesicles blocks NE formation. The remaining membranes show a much lower affinity for chromatin and are unable to form a functional NE. The affinity of vesicles for DNA is probably mediated by proteins associated with them. Indeed, pure liposomes neither bind to chromatin nor to DNA beads (unpublished data). Potential mediators of membrane binding to DNA are transmembrane proteins. Given the topology of the double NE membrane, these transmembrane proteins should be localized to the INM once the NE is assembled. In addition, the only parts of the proteins that can directly contact chromatin are their cytosolic (and later nucleoplasmic) domains. To test whether cytosolic regions of nuclear transmembrane proteins bind to chromatin in our in vitro system, we expressed four such domains recombinantly: the LBR NH terminus, the human Lap2β NH terminus, the human MAN1 COOH terminus, and the NH terminus of BC08, a novel potential NE transmembrane protein (unpublished data). As these protein fragments are all very basic (all have an isoelectric point [pI] of approximately nine), we included control proteins with different pI values in our analysis. The proteins were incubated with chromatin and analyzed by immunofluorescence. All transmembrane protein domains showed chromatin binding, although with different affinities (). Interestingly, we observed a correlation between the pI value of a protein and its ability to bind to chromatin; two neutral control proteins, maltose-binding protein and the nucleoporin Nup43, showed no binding, whereas two basic (ribosomal) proteins, RS10 and RS7, accumulated on chromatin to different extents. The binding properties of the proteins to chromatin correlated well with their affinity for DNA, as they showed identical binding patterns to DNA beads (). We conclude that the positive charge of these proteins confers affinity for chromatin, presumably via interactions with DNA. However, the different signals observed among the tested proteins suggest that basic charge may not be the only critical binding determinant. Based on these results we investigated whether INM proteins are enriched in positively charged cytosolic domains. To date, 14 mammalian transmembrane proteins that localize to the INM () are known, and at least nine are conserved in . Using computer programs we predicted their cytosolic domains and calculated the corresponding pI value. For nine human proteins the membrane topology was either known or could be predicted unambiguously. Seven of these proteins were found to contain a basic cytosolic domain (pI > 8.5) larger than 100 amino acids, and this result was conserved in mouse and (). Hence, at least half of the known INM transmembrane proteins show this characteristic, including the proteins that were detectable early on chromatin during NE assembly, such as LBR, Lap2β, and pom121 (). As 14 proteins are a rather limited dataset, we investigated a list of 67 potential NE transmembrane proteins that was published by . By analyzing these proteins (that should at least be enriched in INM proteins) we found that 46% of the proteins contain a basic, cytosolic domain that is longer than 100 amino acids. To test whether having such a domain is common to transmembrane proteins in general or whether these domains are enriched in the NE, we analyzed 150 transmembrane proteins that were not localizing to the NE, but to the ER or Golgi. A large basic domain was present in only 4% of these proteins. We conclude that proteins of the INM are enriched in long, basic, cytosolic domains, which is a characteristic that is not prevalent in the proteins of other endomembrane systems (). As a large and basic domain confers affinity for chromatin (), the data suggest that interaction with DNA could be a general mechanism by which transmembrane NE proteins mediate membrane recruitment on chromatin. However, these results do not mean that cytosolic domains have to be basic to bind DNA. The cytosolic domain of the egg-specific Lap2ω is not basic (), yet the protein shows a high affinity for DNA (see the following paragraph). The cytosolic domains of several INM proteins directly bind DNA (). To test the involvement of these domains in membrane binding to DNA, we performed a competition experiment. DNA beads were preincubated with recombinant proteins before they were added to membranes. The DNA-bound INM proteins were analyzed by Western blot. The NH termini of both human Lap2β and LBR efficiently inhibited vesicle recruitment to the beads (), as almost no bound Lap2ω or pom121 were detected. This efficient competition by the INM protein fragments suggests that these proteins are directly involved in recruiting vesicles to DNA. Interestingly, membranes bound normally to DNA that was preincubated with core histones (), suggesting that INM proteins might have a different mode of binding to DNA than histones. As an alternative approach to test direct binding of INM proteins to DNA, we incubated DNA beads with membranes and UV cross-linked the samples. After several rounds of stringent washes, the cross-linked proteins were analyzed (, lane 1). In the cross-linked sample, Lap2ω and the transmembrane nucleoporin NDC1 () were detected, whereas the late-recruited transmembrane nucleoporin gp210 was absent. The efficiency of UV cross-linking is very low. Using histone cross-linking, we calculated that only roughly 1% of the proteins were cross-linked to the beads under these conditions (unpublished data). Given the sensitivity of detection of our antibodies, this meant that we could not extend our analysis to additional INM proteins, as the signal in the cross-linked samples was, as predicted, under the detection limit. Nevertheless, the data strongly support the hypothesis that several INM proteins directly mediate binding of membrane vesicles to DNA. To identify more DNA-binding INM proteins, we solubilized membranes with a mild detergent and passed the mixture of proteins over the DNA column. The bead-bound material was analyzed by SDS-PAGE. The general protein pattern showed that several bands were specifically enriched in the DNA-bound fraction (). Analysis by Western blot revealed that several NE proteins bound to the DNA beads (); LBR, Lap2ω, NDC1, and pom121 were strongly enriched in the bound fraction. In contrast, neither gp210 nor Sec61a, which is an ER control protein, showed a comparable enrichment in the bound fraction. Collectively, our data indicate that DNA can act as a binding site for membranes that are essential for NE formation. This binding can be mediated by transmembrane proteins integrated in NE-precursor membranes. We suggest that binding of transmembrane proteins to DNA is at least part of the mechanism for the rapid and highly efficient recruitment of membranes to chromatin in late anaphase, which is when NE assembly begins. As nuclear transmembrane proteins disperse throughout the ER during mitosis (), the redundancy in chromatin binding of multiple INM proteins that is suggested by our data could help collect transmembrane NE proteins at the right place when the mitotic spindle disassembles and chromatin needs to be rapidly enclosed. The redundancy of this mechanism could also account for the finding that in vivo knockdown studies with single nuclear transmembrane proteins that contact specific chromatin proteins did not result in an inhibition of NE assembly in a variety of systems (; ). We did not analyze nontransmembrane proteins such as lamins. The role of lamins in the early steps of NE assembly is not clarified (for reviews see ; ), and we cannot exclude that lamins contribute to the initial contacts of membranes to chromatin via binding to DNA. As a major chromatin component, DNA is present everywhere in chromatin, and is not limiting. Accordingly, the INM proteins LBR, Lap2ω, and pom121 are all uniformly distributed on chromatin during the first minutes of in vitro NE assembly (Fig. S2, available at ; ). Although some proteins of the INM bind to specific regions of chromatin during NE assembly in somatic cells (), the molecular basis for their nonuniform distribution is currently unknown. In a living cell, the presence of the disassembling spindle may affect the membrane-targeting process because of membrane–spindle interaction and differential accessibility of chromatin. As the NE matures, these first membrane–chromatin contacts could be the basis from which more specific interactions between INM proteins and chromatin are formed. These involve chromatin proteins that might compete with free DNA for binding to nuclear transmembrane proteins, such as the LEM domain proteins and the nuclear lamins. egg cytosol, membranes, and demembranated sperm heads were prepared as previously described (), except that the cytosol was centrifuged for an additional 12 min at 16,000 to remove residual membranes. For nuclear assembly reactions, 10 μl cytosol was mixed with 0.3 μl sperm chromatin (3,000 sperm heads/μl) and incubated for 10 min at 20°C to allow chromatin decondensation. Subsequently, 2 μl of the mixture was added to 10 μl cytosol containing 0.3 μl of membranes, 20 mg/ml glycogen, an ATP regenerating system, and, where indicated, DNA (a 5-kb pBluescript-based plasmid) at 15 μg/ml. The negative effects of DNA on NE assembly were overcome by adding 1 μl of membranes to the reaction. At the time points indicated, the membranes were stained with the lipid dye DilC (Invitrogen) and the reactions were fixed with 2% formaldehyde/0.5% glutaraldehyde. DNA was stained with DAPI, and the samples were spun through a 30% sucrose cushion onto coverslips. For the ER fusion assay (), 10 μl cytosol containing an ATP generating system was mixed with DNA-depleted or control membranes and small amounts of fluorescently labeled (DiOC; Invitrogen) membranes. 1 μl of the reactions was transferred to a microscope slide, incubated for 90 min at 20°C, and analyzed by confocal microscopy. For the membrane-depletion experiments, 1 μl of membranes was incubated with 3–9 μl of DNA (or empty) beads in 8 μl S250 buffer (10 mM Hepes, pH 7.5, 50 mM KCl, 2.5 mM MgCl, and 250 mM sucrose) for 15 min at 20°C. The beads were removed with a magnet, washed, and processed for SDS-PAGE. The supernatants were equalized for protein concentration and volume and, subsequently, added to NE assembly reactions or ER fusion assays, as described in the previous section. For the competition experiments, 4 μl DNA beads were incubated with 4 μg of recombinant proteins (see next section) or core histones (Roche) and 100 μg BSA in 10 μl S250 buffer. After 10 min, 2 μl of membranes were added in 20 μl of buffer, and the binding was stopped after 15 min by washing the beads in buffer and processing them for SDS-PAGE. For the DNA cross-linking assay, membranes were first floated in a sucrose gradient, in accordance with the study by . 100 μl of the two lightest membrane fractions were incubated with 15 μl DNA beads. After binding, the samples were irradiated on ice in a UV Stratalinker (Stratagene) at 0.6 J/cm. The beads were washed repeatedly with 2 M NaCl and 1% Triton X-100 and spotted on a membrane. The dot blot was then processed with antibodies specifically recognizing the proteins of interest. After UV cross-linking and washing, empty beads did not yield signals. The efficiency of UV cross-linking was determined using core histones. Floated membranes were solubilized in 500 μl PBS with 1% octylglucopyranoside (Calbiochem) and 0.5 M NaCl for 10 min at 4°C. Insoluble material was removed by centrifugation for 10 min at 280,000 . The supernatant was incubated with 20 μl DNA beads for 15 min at 20°C. The beads were removed, washed (so empty beads did not detectably bind proteins), and analyzed by SDS-PAGE. Proteins were expressed from pQE plasmids (QIAGEN). The proteins had an NH-terminal His tag and were purified using Ni-NTA agarose by standard protocols. Maltose-binding protein from was Alexa Fluor 488–labeled and used in this form (a gift from K. Ribbeck, European Molecular Biology Laboratory, Heidelberg, Germany). For chromatin-binding assays, 2 μg of recombinant proteins were added to decondensed sperm chromatin either in 10 μl cytosol or S250 buffer supplemented with 10 mg/ml BSA. After 20 min at 20°C, the reaction was fixed in 4% formaldehyde, spun on a coverslip, and processed for immunofluorescence using the monoclonal RGS-His antibody (QIAGEN). Alternatively, the proteins were incubated with DNA beads in 40 μl S250 buffer containing 100 μg BSA. The proteins used were LBR nt (amino acids 4–210), the novel protein B08 (amino acids 1–77, available from GenBank/EMBL/DDBJ under accession no. ), RS10, and RS7 (both full length), which were all from ; and human MAN1 (amino acids 672–911), Lap2β (amino acids 1–410), and Nup 43 (full length). For the generation of a polyclonal antiserum against LBR, we used the NH-terminal fragment corresponding to amino acids 4–210. Antibodies against pom121, gp210, and NDC1 were previously described (; ). Antibodies against Lap2β (also recognizing Lap2ω) were a gift of G. Krohne (University of Würzburg, Würzburg, Germany). Antibodies against canine Sec61a (also recognizing the homologue) were a gift of B. Dobberstein (University of Heidelberg, Heidelberg, Germany). NPCs were visualized by mAb 414 (BAbCO). An anti–mouse antibody labeled with Alexa Fluor 546 (Invitrogen) was used for immunofluorescence. Where unknown, the membrane topology of transmembrane proteins was determined using the TMHMM server at and PSORT II at . The pI values of cytosolic domains were calculated at . Sequences of 150 transmembrane proteins not localizing to the NE were obtained from the mouse subcellular localization database at . Only cytosolic domains larger than 100 amino acids were included in the analysis, as caused by the “positive outside rule” () there are generally short cytosolic sequences in transmembrane proteins that are more positively charged than their lumenal counterparts. Fig. S1 shows the chromatin structure of decondensed sperm chromatin and assembled nuclei, which were analyzed by micrococcal nuclease digests. Fig. S2 shows the localization of LBR and Lap2ω on chromatin after 10 min of in vitro nuclear assembly. Online supplemental material is available at .

The nuclear envelope (NE) divides eukaryotic cells into a nuclear and a cytoplasmic compartment. It comprises two lipid bilayers, the inner and the outer nuclear membrane. Local fusions between both membranes create giant aqueous channels (nuclear pores), through which all nucleocytoplasmic exchange proceeds. These pores are embedded into elaborate protein structures of eightfold rotational symmetry called the nuclear pore complexes (NPCs; ; ; ; ). NPCs are built by multiple copies of ∼30 different nucleoporins (Nups), which form the central, proteinaceous NPC structure and thus maintain the very special membrane topology and curvature at the pore. In addition, they create a selective permeability barrier that controls the fluxes of material through the central channel. Membrane-integral Nups anchor NPCs within the nuclear membrane. Two of them, gp210 and POM121, were previously identified in vertebrate NPCs. gp210 forms homodimers and possesses a cleavable signal sequence, an ∼200-kD luminal domain, followed by a stop-transfer sequence that serves as membrane anchor and a short cytoplasmic tail (; ; ; ). gp210 is evolutionary well conserved and is found in metazoans, such as vertebrates (), insects (), or nematodes (), in several protozoa, such as , and in plants (). Nevertheless, fungi apparently lost the gp210 gene, and it has been reported that several cell types of mouse do not express the gp210 protein (; ). It therefore appears that, at least under some circumstances, cells can bypass the requirement for gp210. POM121 () is less conserved than gp210 and is found only in vertebrates. This may indicate that NPC assembly and maintenance does not necessarily require a membrane-anchored POM121 orthologue. POM121 shows a topology opposite to that of gp210. It comprises an NH-terminal signal anchor and an ∼120-kD COOH-terminal domain that faces the NPC channel (). The membrane anchor, however, is not required for assembling POM121 into NPCs. Instead, the central POM121 domain is necessary and sufficient for incorporation into NPCs (). The COOH-terminal part contains FG repeats and therefore might contribute to the permeability barrier of NPCs. Considering these constraints, mitochondria appeared to be the ideal platform for an ectopic POM121 bait. As a control, we first fused EGFP alone behind the NH-terminal part of TOM20 (residues 1–70), which comprises an anchor for insertion into the outer mitochondrial membrane, followed by a very hydrophilic segment that prevents mistargeting to the endoplasmic reticulum (). The resulting fusion (“Mito-GFP”) localized correctly to the outside of mitochondria, as judged by colocalization of the GFP signal with mitotracker-stained mitochondria (unpublished data; see the following paragraph). The Mito-GFP signal showed no overlap with any NPC marker (, top). However, when the central POM121 domain (lacking membrane anchor and FG repeats) was added as a third module to the Mito-GFP fusion, mitochondria efficiently recruited NPC constituents ( and ). This was evident from stains with mAb414, which recognizes several FG-repeat Nups () or with the Impβ fragment that detects FG- and GLFG-repeat Nups (). A more detailed analysis revealed that not all Nups were attracted to the ectopic POM121 sites. In particular, Nups from the nuclear NPC side, namely, Nup50, Nup153, and TPR (translocated promoter region; ; ; ), remained absent from the structures (), possibly because they are actively imported into nuclei and thereby depleted from the cytoplasm and, thus, not available at the ectopic assembly sites. The ectopic POM121 fragment most efficiently recruited the centrally located Nup62 (; ) and Nup205 (). At later times of expression, we also observed recruitment of Nup98 () and even of Nup358/RanBP2 () from the cytoplasmic NPC filaments (; ). The large distance (>20 nm) between the central Nups and the anchoring site for Nup358 within bona fide NPCs suggests that POM121 not only attracts its nearest neighbors but might even be sufficient to initiate the assembly of large substructures of the NPC. So far, POM121 and gp210 are the only known membrane-spanning Nups in vertebrates. If they constituted the primary anchors of NPCs within the pore membrane, they would be expected to be essential for the assembly of bona fide NPCs. To test this, we used RNAi against the POM121 mRNA (). The depletion of POM121 was efficient, but it did not produce any obvious phenotype (). Pore recruitment of the Nup107-160 complex (; ) and of mAb414-reactive Nups, including Nup62 and Nup358, remained normal. Assembly of gp210 (Fig. S1, available at ) and Nup153 (not depicted) into NPCs was not affected either. Nuclei of POM121-depleted cells also efficiently accumulated an IBB–GFP fusion protein (Fig. S1), indicating that their NE was sufficiently intact to prevent an uncontrolled leakage of the previously imported reporter to the cytoplasm. In addition, POM121-depleted cells remained viable and divided at a rate similar to that of cells that were not transfected with the anti-POM121 siRNA duplexes (unpublished data). Finally, light-microscopical analysis of POM121-deficient cells in the various cell cycle stages did not reveal any signs of aberrant progression through mitosis or cytokinesis ( and not depicted). This unexpected lack of phenotype could be explained formally by a catalytic action of the depleted factor, whereby minute residual amounts suffice for normal function. POM121, however, possesses no known enzymatic activity and associates so stably with NPCs that the half-time for its dissociation from NPCs is longer than a typical cell cycle (). It is therefore hard to imagine how a POM121 molecule from one NPC could possibly catalyze the assembly of another NPC that is devoid of POM121. The knockdown could reduce the POM121 signal at NPCs at least 20-fold without affecting the NPC localization of other Nups (). After such depletion, some NPCs still gave a faint POM121 signal, whereas others appeared POM121 negative (Fig. S2, available at ). Therefore, a substantial proportion of NPCs and the vast majority of asymmetric NPC units assembled independently of POM121 under these conditions, suggesting that POM121 is not essential for the process. In contrast, it has been proposed that POM121 is essential for assembling an NPC-perforated NE from components of the egg extract (). This reported requirement for POM121 might be specific for very fast embryonic cell cycles during amphibian embryogenesis. For somatic mammalian cells, however, our RNAi data suggest that POM121 is either not limiting or even fully dispensable for the formation of the NE and NPCs. Phylogenetic data also argue against a unique and indispensable role for POM121 in NPC/NE assembly. Neither yeasts, insects, nematodes, nor, indeed, any nonvertebrate eukaryote contains a recognizable, membrane-anchored POM121 orthologue, yet they have functional NPCs and an intact NE. gp210 is far better conserved than POM121. It probably already existed in the earliest eukaryotes, and it might, therefore, be less dispensable for NPC assembly. However, mouse gp210 is absent from many mesenchymal cell types of kidney, teeth, and lung as well as from several epithelial and fibroblast cell lines (; ). To determine whether this represents a mouse-specific phenomenon or applies to other mammals as well, we raised antibodies against human gp210, analyzed several human cell lines, and observed that human primary fibroblasts also lack gp210 (). Having found that NPCs of human fibroblasts naturally operate without gp210, we studied the consequences of gp210 depletion from HeLa cells, a cell type that normally expresses this protein. RNAi against gp210 mRNA could drastically reduce gp210 levels at NPCs. Yet, other Nups assembled normally into NPCs (), as judged by immunofluorescence detecting the Nup107-160 subcomplex and the FG Nups. Mammalian genomes encode a second gp210 paralogue (gp210L) that shares 41% sequence identity with the canonical gp210 (unpublished data). We can rule out the possibility that gp210L compensates for any loss of gp210 function because gp210L is expressed in neither HeLa cells nor fibroblasts (unpublished data). Collectively, this body of data suggests that gp210 paralogues are not essential for the NPC assembly process in mammalian cells. This conclusion is in line with the observations that gp210 appears dispensable for nuclear assembly in the egg extract system () and that NPCs in gp210-deficient cells appear structurally intact at the EM level and stain positive for FG-repeat Nups (). In simultaneous knockdowns, POM121 and gp210 could be depleted from HeLa cells to <5–10% of their original levels (). Nonetheless, the assembly of other Nups into NPCs appeared undisturbed and the cells remained viable. Likewise, no defects in NPC assembly became apparent when POM121 was depleted from primary human fibroblasts (), which are already devoid of gp210 (). An anchorage of NPCs within the NE without a membrane-integral protein is unlikely. The assembly of functional NPCs in cells devoid of gp210 and depleted of POM121 therefore predicts the existence of at least one additional human membrane-integral Nup, which promotes NPC biogenesis also in the absence of POM121 and gp210. Indeed, in the accompanying article, we describe the metazoan orthologue to yeast Ndc1p as a novel, six times membrane–spanning constituent of animal NPCs and demonstrate that it crucially contributes to the NPC assembly process (). We also report in the accompanying article that none of the membrane-integral Nups, not even NDC1, are universally required for NPC formation (). Crucial cellular functions are often backed by multiple players, and this recurring theme apparently also applies to the membrane anchorage of NPCs. This yields the experimental problem that such redundancies easily obscure the function of a given protein. POM121 is a good example of this: even though it is apparently nonessential for NPC formation and maintenance, we could demonstrate that the central POM121 domain can recruit other Nups to an ectopic assembly site and thus appears capable of initiating at least some steps of the NPC assembly process. Antibodies were newly raised in rabbits against human POM121 and gp210. Antibodies against human Nup50, Nup62, Nup96, Nup98, Nup107, Nup153, Nup205, Nup358, and TPR have been described (). All polyclonal antibodies were affinity purified on the respective antigen columns. mAb414 was obtained from Eurogentec and the mAb against Nup88 from BD Biosciences. Primary human neonatal fibroblasts (Hs27 cells) were obtained from the European Collection of Cell Cultures and cultivated for no more than 20 passages in Dulbecco's modified Eagle's medium high glucose, supplemented with 10% FCS, 100 U/ml penicillin, and 100 μg/ml streptomycin. Human HeLa cells (European Collection of Cell Cultures) were maintained in Dulbecco's modified Eagle's medium low glucose, supplemented with 10% FCS, 1× nonessential amino acids (Sigma-Aldrich), 100 U/ml penicillin, and 100 μg/ml streptomycin. Transfection of cultured human cells with siRNAs was performed essentially as described earlier (). Annealed siRNAs were purchased from Dharmacon Research. Antisense strands were complementary to the following nucleotide positions of the respective open reading frames: POM121, POM121, and gp210. DNA transfections were performed with Fugene6 (Roche) according to the manufacturer's instructions. Expression of the Mito-EGFP and the Mito-EGFP-POM121 fusion was driven by a doxycycline-regulatable promoter system (pRevTRE2; CLONTECH Laboratories, Inc.) at 10 ng/ml doxycycline for 24–60 h. For immunofluorescence, cultured cells were washed briefly with PBS; fixed for 4 min in 3% paraformaldehyde; freshly dissolved in PBS; washed in PBS followed by PBS + 50 mM NHCl (5 min); permeabilized with 0.25% Triton X-100 in PBS; and blocked for at least 30 min in 1% BSA, 10% goat serum, and 0.1% Triton X-100. Primary antibodies were applied for 60 min in blocking buffer. Nonbound antibodies were washed off with PBS. Alexa-labeled secondary antibodies were obtained from Invitrogen and used at 1:250 dilution. The secondary antibodies and the DNA stain Hoechst 33342 were applied for 30–60 min in blocking solution, followed by extensive washing and mounting in Vectashield (Vector Laboratories). For anti-Nup205 stains, conditions were modified as previously described (). Confocal microscopy was performed with a laser-scanning microscope (SP2; Leica) using the 405-, 488-, 561-, or 633-nm laser lines for excitation. All pictures were taken with PlanApo oil objectives (100× NA 1.4 and 63× NA 1.32; Leica); for scans with the 405-nm laser, λ blue objectives were used. Figures were assembled in Photoshop or Illustrator (Adobe). Fig. S1 shows an intact NE in POM121-depleted cells. Fig. S2 shows distribution of residual POM121 after depletion by RNAi. Online supplemental material is available at .

A quarter of all conceived human embryos are aneuploid, i.e., they either have too many or too few chromosomes (). The consequences of such chromosomal abnormalities are profound, affecting not only fertility, but also triggering spontaneous miscarriages. A few abnormal karyotypes are compatible with human life, including Down's (trisomy 21), Turner (a single X chromosome), and Klinefelter's (XXY) syndromes, but are also associated with developmental disabilities of variable penetrance. Analysis of human sperm and eggs has revealed that aneuploidy affecting embryos is primarily caused by an error-prone meiotic chromosome segregation mechanism in oocytes. Whereas ∼1–2% of human sperm have an abnormal chromosomal content (the same level of aneuploidy is recorded in mouse haploid germ cells, including oocytes), an astonishing 20–25% of the human oocytes are aneuploid (). The cause of this high error rate for the meiotic process in human female germ cells is unclear. Meiosis is a specialized cell division process that generates genetically distinct haploid cells through a process that involves one DNA replication step followed by two cell divisions (; ). The newly replicated sister chromatids are bound by cohesin complex proteins that ensure that cohesion between sister chromatids is retained at the first cell division, but lost at the second meiotic division (). Each pair of cohesin-bound sister chromatids constitutes a chromosome, which subsequently becomes connected with its homologous partner at the zygotene to pachytene stages of prophase I in a process called synapsis. The synaptic process is promoted by the formation of a large number of DNA double-stranded breaks (DSBs) that are generated by the topoisomerase II–related transesterase SPO11 (). The repair of a subset of the DSBs results in crossovers between the homologous chromosomes and, ultimately, in chiasmata, providing essential physical links between the chromosomes (; ; ). Synapsis is also dependent on a conserved proteinaceous structure called the synaptonemal complex (SC). The SC is composed of two axial elements (AEs) and a large number of individual transverse filaments that connect the AEs along their entire length. In addition, a central element has been defined at the center of the transverse filament structure (; ). In mammalian male and female germ cells, several different meiosis-specific proteins have been defined as components of the SC, including the AE proteins SC protein 2 (SYCP2) and 3 (; ; ) and the transverse filament protein SYCP1 (). The AE proteins SYCP2 and -3 are found at the interchromatid domains of the sister chromatids, which is where they jointly form axial cores together with the cohesin complex proteins. Several different error surveillance systems (checkpoints) have been characterized in meiotic cells (; ; ). A failure to repair DSBs that is caused by inactivation of DNA repair/recombination proteins such as DMC1, MSH4, and MSH5, or DNA damage checkpoint proteins such as ATM, will activate a DNA damage checkpoint that results in female germ cell death at early postnatal development (). The mismatch repair protein MLH1 takes part in the conversion of crossovers into chiasmata at a late stage of the recombination pathway (; ; ). Surprisingly, inactivation of this protein in murine germ cells does not activate a DNA damage checkpoint. Instead, in mouse oocytes that are deficient for , the resulting achiasmatic mutant germ cells cannot establish a proper meiotic spindle and are eliminated at the metaphase I stage by the spindle checkpoint (). The absence of SYCP3 results in decompaction of the meiotic chromosome axis, premature loss of cohesin complexes from the meiotic chromosome axis, and irregular interruptions of the synaptic process as defined by SYCP1 (; ). mice are fertile, although one-third of their offspring die in utero at an early stage of embryonic development as a result of aneuploidy (). We investigated the nature of the chromosomal errors introduced by the absence of SYCP3 and how these errors evade the meiotic quality assurance systems, thereby generating aneuploid offspring. Our results illustrate the importance of the axial element of the synaptonemal complex for efficient repair of recombination events. The absence of SYCP3 results in a complete elimination of male spermatocytes at the zygotene–pachytene transition of prophase I (). To investigate how the elimination of SYCP3 affects the oocyte maturation process, ovarian morphology and oocyte numbers were analyzed in pre- and postnatal animals from embryonic day (E) 16.5 to 8 d postpartum (dpp). animals were either stained with hematoxylin and eosin or immunostained using antibodies against germ cell nuclear antigen (GCNA) or c-kit. GCNA and c-kit specifically stain the nuclei and the cytoplasm of oocytes, respectively (; ). Histomorphometric analysis revealed no difference in the relative numbers of oocytes at E16.5, E18.5, or at birth when ovaries from wild-type or females were compared ( and ). females. A majority of the oocytes in 1-dpp mice are found in small clusters, called germ cell cysts, which are seen in both wild-type and mutant ovaries (). ovary, which was not seen in the wild-type ovary (, C and D, and ). ovary was further accentuated 4 dpp (, E and F; and ). ovaries in 2- and 4-dpp mice, compared with wild type ( and ). It is likely that the loss of primordial oocytes is attributable to the rapid elimination of the germ cell cysts seen during early postnatal development, as the primordial oocytes develop from these cysts. We found that the collective loss of germ cell cysts and primordial oocytes in the mutant ovary amounts to 34% in 2-dpp mice and 52% in 4-dpp mice (). A further reduction in primordial follicle number occurs at 8 dpp ( and ), suggesting that primordial follicles are also susceptible to elimination within the mutant ovary environment. ovary gives rise to both primary and secondary follicles in numbers that closely match the wild-type situation. We conclude that loss of SYCP3 function results in a drastic loss of germ cell cysts and primordial follicles during early postnatal development. females is reduced, with ∼66% compared with wild-type, but no further depletion occurs relative to wild type, giving rise to an oocyte reservoir in 8–12-wk-old females that can sustain normal levels of fertility at this age (). The loss of germ cell cysts and primordial follicles in the ovary of the mutant females could be caused by an apoptotic process that is introduced by the absence of SYCP3. ovary, compared with the wild-type counterpart (Fig. S1, available at ). females are predominantly localized at the cortex area (at the outer edge of the ovary sections), suggesting that the affected cells correspond to the germ cell cysts that are preferentially lost in the mutant background. The number of TUNEL-positive cells is low, relative to the total number of oocytes that are lost in the mutant ovary. This is most likely because of the transient nature of the TUNEL staining, which is also described in another mouse model monitoring female germ cell death during early postnatal development (). females suggests the involvement of the DNA damage checkpoint, which is known to become active at an early stage of postnatal development in oocytes (). Therefore, we monitored the progression of DNA repair of DSBs in zygotene to diplotene mutant oocytes. oocytes are described in Fig. S1, Materials and methods, and . In brief, both early zygotene and zygotene oocytes were derived from E16.5 embryos, whereas pachytene and diplotene oocytes were derived from E18.5 or E19.5 embryos. Formation of the axial cores was monitored by STAG3 staining, synapsis (transverse filament formation) was monitored by SYCP1 staining, and centromere morphology was monitored by CREST staining. Introduction of DSBs in meiotic DNA at the leptotene stage of prophase I results in the phosphorylation of H2AX (generating a modified form called γH2AX). γH2AX appears at leptotene in chromatin regions throughout the nucleus and generally form large, cloud-like patterns, suggesting that the majority of the affected H2AX molecules are found in chromatin loops that project out from the axial cores of the chromosomes (; ). Subsequent repair of DSBs results in the disappearance of most of the γH2AX signal at the pachytene stage. A second and independent wave of γH2AX staining appears in late zygotene and pachytene cells, which are associated specifically with the asynapsed axial cores of the meiotic chromosomes (; ). We found that γH2AX immunostaining of wild-type early zygotene oocytes revealed dispersed, cloud-like signals throughout the nucleus (), whereas only a few patches of γH2AX signals associated with the remaining asynaptic axial cores were observed in pachytene nuclei ().We also consistently observed a few residual γH2AX patches in diplotene nuclei, the nature of which is not clear (). oocytes at the early zygotene stage and noted that it was indistinguishable from the pattern observed in wild-type cells at this stage (). This suggests that neither SPO11-derived DSB formation nor phosphorylation of H2AX is dependent on SYCP3 expression. oocytes revealed that a majority of these cells retained a strong, cloud-like nuclear γH2AX signal, which is similar to the pattern seen in early zygotene cells (, and J). Strong γH2AX staining in late meiotic cells could reflect inefficient DNA repair or residual asynapsis. oocytes, however, is very different from the γH2AX pattern seen in meiotic cells with asynapsed chromosomes (; ). oocytes suggests that the DNA repair process is impaired in the mutant cells. oocytes, the spatial and temporal distribution of several DNA repair/recombination proteins were studied during meiosis. The RecA homologues RAD51 and DMC1 take part in heteroduplex formation during meiosis (). RAD51 and DMC1 foci are formed along the AEs in wild-type meiotic cells and are observed on both asynapsed and synapsed parts of the SC, but these foci disappear during pachytene (). oocytes, as seen in wild-type cells (). pachytene and diplotene oocytes, compared with wild-type cells (, and L). oocytes were found in the nuclear space next to the SCs (labeled by SYCP1), but an increased number of foci associated with the SC were also seen (Fig. S2, available at ). RPA is a single-stranded DNA–binding protein that promotes DNA DSB repair (; ). RPA foci appear on the SCs later than RAD51/DMC1 foci and have been suggested to be part of an antirecombination protein complex that prevents the formation of superfluous reciprocal recombination events (). oocytes (). oocytes did not decrease, as in wild-type oocytes at this stage (). diplotene oocytes, despite an almost complete loss of such foci in wild-type oocytes at the same stage (). MSH4 is a MutS homologue that promotes homologous alignment and crossover formation (). MSH4 foci overlap with RPA foci, but appear slightly later during meiosis in mouse germ cells (). oocytes (unpublished data). oocytes. MLH identifies the sites of meiotic exchanges along the pachytene chromosomes and is critical for chiasma formation (; ; ; ). It has been shown that the number of MLH1 foci varies according to the length of the SC (). This correlation, however, does not apply to SYCP3-deficient meiotic chromosomes. pachytene oocytes (). oocytes at postpachytene meiotic stages to reveal if the temporal distribution of this protein is affected in the absence of SYCP3. We found that MLH1 foci persisted into the diplotene stage in wild-type oocytes, as previously described (), but that these MLH1 foci disappeared at the end of diplotene (, C and D, G and H, and K and L). oocytes, and that many of these foci also remained until the very late diplotene stage (, A and B, E and F, and I and J). No MLH1 foci were observed in oocytes after birth (unpublished data). Many of the residual MLH1 foci that were retained at the very late diplotene stage were not associated with SYCP1 staining (). The most likely explanation for this is that the time course of the desynaptic process (i.e., the removal of SYCP1) is not affected in mutant oocytes (), whereas MLH1 foci persist for longer in these cells. The residual MLH1 foci, however, remain in association with the meiotic chromosome axis, as shown by the association of these foci with STAG3-staining regions in late diplotene mutant oocytes (). We noted that mutant oocytes that retained a large number of residual MLH1 foci also displayed a strong nuclear γH2AX staining (), suggesting that the impaired repair process affects DNA associated with the chromosome axis, as well as with DNA loops that project out from this axis. In summary, we found that the absence of SYCP3 results in a persistent γH2AX pattern during meiotic prophase and in a delayed removal of RAD51/DMC1, RPA, MSH4, and MLH1 from meiotic chromosomes. oocytes. oocytes at different stages of meiosis, using antibodies that detect γH2AX, RAD51/DMC1, RPA, MSH4, and MLH1, consistently identified two groups of cells, where one group retained less staining than the other (). late diplotene oocytes displayed weak γH2AX staining. oocytes at late diplotene also contained relatively few RAD51/DMC1 and MLH1 foci (, , and ). oocytes is impaired, not blocked, giving rise to a spectrum of mutant cells with different levels of damage. A decreased efficiency of the DNA repair/recombination process should have consequences for completion of the crossing-over process between the homologous chromosomes. A failure to form or maintain chiasmata between the homologous chromosomes during meiosis will result in premature chromosome separation, giving rise to two separately labeled univalents (achiasmatic) chromosomes. To investigate this, chromosome-specific probes were labeled and used in FISH experiments. We found, as expected, that the chromosome-specific probes (19, 17, 12, 2, 1, and X) labeled single individual chromosome structures in wild-type oocytes (). oocytes that contained univalent chromosomes (, and ). oocytes at 2 dpp varied considerably (). The difference in univalency rate for the six different chromosomes in the analyzed mutant mouse oocytes is best explained in the context of their reported mean chiasmata frequency (; ; ). Chromosomes 1 and 2 display an almost twofold higher mean chiasmata frequency than reported for chromosomes 12, 17, and 19, strongly suggesting that homologous chromosomes connected with two or more chiasmata are more likely to retain at least one chiasma even in the absence of SYCP3. oocytes contain multiple univalent chromosomes (). oocytes give rise to achiasmatic chromosomes and results in a sharp increase in the number of oocytes that contain univalent chromosomes at 2 dpp. To follow the fate of the mutant oocytes during follicle formation, we studied the same six chromosomes in oocytes derived from females at 8 dpp by using FISH (). We found that the percentage of mutant oocytes at 8 dpp that contained univalents for the analyzed chromosomes was reduced considerably compared with 2 dpp. For example, the percentage of oocytes having a univalent chromosome 19 at 2 dpp was reduced from 13 to ∼6% at 8 dpp. ovary, our results suggest that oocytes that contain univalent chromosomes are preferentially eliminated during postnatal development. Using the combined statistics derived for the six chromosomes analyzed in , we estimate that although ∼36% of the oocytes at 8 dpp retain univalent chromosomes, ∼75% of the oocytes at 2 dpp contain univalent chromosomes. Our results therefore show that more than half of the oocytes that contain univalent chromosomes at 2 dpp are eliminated as the germ cell cysts mature into primordial follicles. It has been proposed that structural changes in the organization of the axial cores of meiotic chromosomes could affect the maturation of DSBs into crossovers (). We show that loss of SYCP3 impairs both the DNA DSB repair process and the formation of crossovers between homologous chromosomes. oocytes. oocytes, including the temporal appearance of γH2AX, the recruitment of RAD51/DMC1, RPA, MSH4, and MLH1 to DNA DSBs, or the time course of meiotic markers such as STAG3, SYCP1, and CREST. oocytes. Similarly, RAD51/DMC1, RPA, MSH4, and MLH1 foci persisted for an extended time period at the late meiotic stages in the mutant oocytes. Such patterns were not observed in wild-type oocytes and likely reflect a failure to complete recombination within the temporal window provided by meiotic prophase I. oocytes contain univalents at 2 dpp. It has been shown that the inactivation of proteins that participate in the repair of meiotic DNA DSBs activates a DNA damage checkpoint during early postnatal development, resulting in the complete elimination of affected oocytes (). oocytes are eliminated beginning at 2 dpp. females. Together, these results suggest that the absence of SYCP3 activates a DNA damage checkpoint in oocytes. oocytes (). oocytes were approximately the same (), the number of chiasmata at the MI stage was reduced in oocytes compared with wild-type oocytes. Loss of MLH1 from meiotic chromosomes in wild-type meiotic cells normally precedes the removal of the SYCP1 protein, suggesting that the crossing-over process is completed in the context of an intact SC (; ). oocytes do not colocalize with residual SYCP1 staining. oocytes affects the efficiency of the remaining MLH1 recombination complexes and that a subset of these fails to complete the crossing-over process. A failure to establish chiasmata between homologous chromosomes could also be caused by an impaired positive genetic-interference mechanism (; ). This mechanism ensures that crossovers are correctly distributed between chromosomes. A partially inactivated interference mechanism could lead to an unregulated distribution of a fixed number of chiasmata and result in a loss of obligatory chiasmata, thereby generating achiasmatic chromosomes. It has been proposed that the SC ensures a high level of interference (; , ; ). We have studied if SYCP3 is required for interference by monitoring the number of MLH1 foci, which is a cytological marker for chiasmata distribution along SYCP1-labeled meiotic chromosomes (; ; ; ). deficiency increases the length of the meiotic chromosome axes by twofold and introduces irregular gaps in SYCP1 staining along the axes (the meiotic axis in the SYCP1-negative gaps cannot be traced with certainty, as antisera against cohesin complex proteins such as STAG3 only weakly stain these regions; ; ). oocytes lack associated MLH1 foci. pachytene oocytes that displayed relatively intact SYCP1-labeled meiotic chromosomes and monitored the frequency of such structures, which had two MLH foci associated to them. Analysis of 24 oocytes and 31 wild-type oocytes produced a very similar result, where both groups showed an average of ∼3.5 intact SYCP1-labeled structures each having two MLH1 foci per cell. oocytes. oocytes where the TF structure as labeled by SYCP1 is severely fragmented, making it impossible to trace the meiotic chromosome axis, we cannot analyze if the MLH1 distribution pattern is affected. oocytes contain asynapsed configurations of chromosome 19; unpublished data), excluding this as an important mechanism to explain the univalency statistics observed at 2 dpp in mutant oocytes. Our experiments show that loss of SYCP3 affects the efficiency of the DNA repair/recombination process. oocytes is impaired, not blocked. We provide two sets of evidence for this; we found that ∼34% of the oocyte pool remains at 8 dpp and of those that remain only approximately one-third contain univalent chromosomes. oocytes, strongly suggesting that loss of SYCP3 generates a temporal spectrum of recombination intermediates. mouse model is the effectiveness with which it contributes to the formation of aneuploid offspring (). oocytes that contain univalent chromosomes can bypass the DNA damage checkpoint at early postnatal development. A similar situation has been noted in mice that are deficient for MLH1 (; Edelman et al., 1996). In these mice the final crossovers are not completed, giving rise to the formation of achiasmatic chromosomes; however, the DNA damage checkpoint does not become activated. In sharp contrast to -deficient oocytes (), however, -deficient oocytes that contain univalent chromosomes also bypass the spindle checkpoint at the first meiotic cell division and give rise to aneuploid offspring (). Our results for -deficient oocytes are in agreement with studies of human oocytes that suggest that a reduced level of recombination is linked to an increase in aneuploidy (). Interestingly, it has been observed that γH2AX signals are more slowly removed during meiosis in human oocytes compared with sperm, suggesting that progression of DSB repair is slower in oocytes (). oocytes does not occur until the diplotene stage. spermatocytes are already eliminated at the zygotene/pachytene stage of meiosis (). A similar temporal difference in the loss of damaged male and female germ cells has been noted for a large number of gene deficiencies (). We propose that the relative incidence of aneuploidy observed for male and female gametes can be partly explained by a temporal difference in the activation of the DNA damage checkpoint during meiosis. In cases where a mutation generates a temporal spectrum of recombination deficiencies, the timing of the activation of the DNA damage checkpoint becomes crucial. The late activation of the female DNA damage checkpoint during meiosis, relative to the temporal activation of the same checkpoint in male germ cells, provides additional time for the formation of advanced recombination intermediates that can no longer be detected by this checkpoint in oocytes. This increases the risk that such recombination intermediates will contribute to the formation of univalent chromosomes. Derivation of the knockout mice has been previously described (). females to generate offspring. males and females were then mated to produce (wild-type) and mice. To detect the pregnancy, two females were caged with one male after 16:00 (4:00 pm). The vaginal plugs were examined daily between 8:00 and 9:00 (am). The day that the plug was found was defined as E0.5. For ovary collection at embryonic stages, pregnant mice were killed at E16.5, E17.5, and E18.5. For ovary collection at postnatal stages, the pups were killed after birth at days 1, 2, 4, and 8, which were referred to as 1, 2, 4, and 8 dpp. Ovaries from adult mice were also collected at 8 wk. Ovaries were fixed in 4% paraformaldehyde for 4 h before paraffin embedding. The entire ovary embedded in the paraffin was sequentially sectioned at 5 μm. Every tenth section was stained either by hematoxylin and eosin or immunostained for GCNA, which is a germ cell marker (), or c-kit, which is an oocyte marker (). These sections were then used for estimation of oocyte numbers. In embryonic and newborn ovaries, oocytes can be clearly distinguished from somatic cells by GCNA staining. Immunohistochemistry was performed with a rat anti–GCNA-1 (a gift from G.C. Enders, University of Kansas Medical Center, Kansas City, KS) and a polyclonal rabbit anti–c-kit (PC34; Oncogene Research Products), using the Vectastain Elite ABC kit (SK 4100; Vector Laboratories), according to the manufacturer's instructions. The peroxidase substrate DAB (DakoCytomation) was used to visualize the immunostaining reaction and hematoxylin was used for counterstaining. For the postnatal mice ovaries, primordial and primary follicles were defined by their morphology and by c-kit immunostaining. Oocyte counts were first determined individually for germ cell cysts (germ cells that were not individually separated by stromal cells), primordial follicles (small oocytes surrounded by a few flattened pregranulosa cells), primary follicles (oocytes with a visible nucleolus surrounded by a single layer of cuboidal granulosa cells, ranging from five to nine cells), and secondary follicles (an oocyte with a visible nucleolus surrounded by two layers of cuboidal granulosa cells made up of more than eight granulosa cells). Only follicles with a visible nucleus were counted to avoid double counting. The total oocyte numbers for each ovary were summarized from different follicle stages by using five sections/ovary (6 sections/ovary in 8-dpp mice and 15 sections/ovary in 8 wk-old mice). Three to seven ovaries per genotype (null and wild-type mice were from the same litter) were included in each group. Apoptotic cells in paraffin-embedded sections of ovaries were identified using a TUNEL staining kit (Seriologicals Corp.), following the manufacturer's instructions. The sections were counterstained with methyl green. Every tenth section from the same ovary used for oocyte counting was also used for TUNEL staining. The relative number of apoptotic cell was summarized from five sections/ovary for each study group, with the exception of six sections taken from ovaries derived from 8-dpp mice. Statistical calculations of oocyte numbers were performed by one-way analysis of variance, using the SigmaStat program (SPSS, Inc.). P ≤ 0.05 indicates a significant difference. oocytes were obtained using a “dry-down” technique () from ovaries at E16.5 (early and later zygotene oocyte), E18.5, and E19.5 (pachytene and diplotene oocytes). oocytes. Staging of oocytes was performed using several markers, including SYCP1, STAG3, and CREST, as well as DAPI (Fig. S2; ). In early zygotene, synapsis has started and short SYCP1 fibers are visible, but centromeres are not yet paired (around 40 CREST foci). In zygotene, up to 50% of the AEs take part in synapsis and centromere pairing has been initiated. In late zygotene, 50–80% of the AEs take part in synapsis and most centromeres are paired (generating ∼20 CREST foci). In late pachytene and early diplotene, a majority of the AEs are synapsed, and if some bivalents have desynapsed they appear to repel each other; most centromeres are still paired. In late diplotene, most of the SYCP1 fibers have disappeared, and the number of CREST foci varies between 20 and 40. Mutant oocytes were assigned a stage when the aforementioned criteria were fulfilled. Primary antibodies and dilutions used were guinea pig anti-SYCP1 and anti-STAG3 at 1:200 (), human anti-CREST at 1:1,000, mouse anti-γH2AX (Upstate Biotechnology) at 1:100, rabbit anti-DMC1/RAD51 at 1:100 and rabbit anti-RPA at 1:500 (gifts from P. Moens, York University, Toronto, Canada), rabbit anti–human MSH4 (a gift from C. Her, Washington State University, Pullman, WA) at 1:100, and mouse anti–human MLH1 (BD Biosciences). All primary antibody incubations were performed overnight at 4°C or 37°C. Secondary antibodies were swine anti–rabbit conjugated to FITC (DakoCytomation) at 1:100, donkey anti–guinea pig conjugated to Cy3 (Jackson ImmunoResearch Laboratories) at 1:1,000, goat anti–human conjugated to Cy5 (GE Healthcare), goat anti–mouse Alexa Fluor 488 (Invitrogen) at 1:1,000, and goat anti–rabbit Alexa Fluor 350 (Invitrogen) at 1:500. All secondary antibodies were incubated for 1 h at room temperature. DNA was stained with DAPI. Slides were mounted with antifade medium before being analyzed. Slides were viewed at room temperature using fluorescence microscopes (DMRA2 and DMRXA; Leica) and 100× objectives (Leica) with an aperture of 1.4 providing epifluorescence. Images were captured with a digital charge-coupled device camera (model C4742-95; Hamamatsu) and the Openlab 3.1.4 software (Improvision). Images were processed using Photoshop version 9 (Adobe). Oocytes were obtained from 2- and 8-dpp female mice ovaries. To increase the yields of oocytes from 8-dpp mice, ovaries were initially incubated for 30 min at 37°C with collagenase and DNase (). The cells were isolated by pipetting and fixed by using 1% paraformaldehyde and 0.15% Triton X-100. Oocytes were detected by GCNA staining. The oocytes were also distinguished from somatic cells on the basis of their size, the dispersed nature of their chromatin, and a characteristic congregation of centromeres at several distinct locations within the nucleus (). After immunostaining, the slides were washed and air dried, and then denatured in 70% formamide and 2× SSC at 70°C for 2–4 min. Hybridization with specific chromosome probes was performed for 40 h at 37°C. The Cy3-labeled chromosomal probes (Chrombios GmbH) were used to identify chromosomes 1, 2, 12, 17, 19, and X in the oocyte by using FISH. Double- and triple-color FISH probes were labeled with Chr19-Cy3, Chr17-Cy5, and Chr12 (or ChrX)-DEAC. The washing step followed the manufacturer's protocols (Chrombios GmbH). DAPI was used as a DNA counterstain, and slides were mounted with antifade before analysis. Fig. ovary. Fig. S2 shows the classification of zygotene and diplotene stage meiotic cells. Online supplemental material is available at .

Meiosis, which is a process unique to germ cells, consists of two successive rounds of cell divisions without intervening DNA replication, and, thus, produces haploid germ cells to cope with genome doubling at fertilization. One hallmark of meiosis is the assembly and disassembly of a proteinaceous tripartite structure, the synaptonemal complex (SC), during the prophase of meiosis I (; , ; ). The SC consists of two lateral elements (LE) and a central element (CE). During the leptotene stage of prophase I, axial elements (AE) are formed along chromosomal cores between sister chromatids. During the subsequent zygotene stage, the AEs of two homologous chromosomes become connected by transverse filaments (TF) in a process referred to as synapsis. Because TFs overlap in the center to form a CE, AEs, TFs, and the CE constitute the tripartite SC. In the context of SCs, AEs are called LEs. At the pachytene stage, synapsis occurs along the entire length of homologous chromosomes, except for XY chromosomes in mammals. During the diplotene stage, SCs disassemble and homologous chromosomes are separated, except at regions of crossover, which are known as chiasmata (). Studies of meiosis-specific proteins in model organisms (e.g., Zip1, Red1, Hop1, and Mek1) have provided insights into the functions of SC (). In budding yeast, the Red1 protein localizes to AEs and is required for the formation of AEs (). Hop1 interacts with, and localizes to the same sites as, Red1, and the localization of Hop1 to AE is dependent on Red1 (; ; ). Additionally, genetic studies have shown that the stoichiometry of Red1 and Hop1 is critical for AE assembly (; ). Mek1 is a meiosis-specific serine/threonine kinase involved with the SC assembly, and it interacts with Red1 physically and genetically. Phosphorylation of Red1 is Mek1-dependent and is required for Red1– Hop1 heterooligomer formation, and, thus, is critical for the formation of the SC (; ). Zip1 is a major component of TFs in yeast and is required for chromosomal synapsis (). Components of TFs have been identified in (CG) and (SYP-1 and -2; ; ; ). A common feature of TF components in these diverse organisms is that they are coiled coil proteins (). Although extensive ultrastructural studies of SCs have been performed in mammalian species such as rat and hamster (), isolation of SCs have identified the key components of mammalian SCs, including SC proteins 1, 2, and 3 (SYCP1, -2, and -3; ; ; ; ; ). SYCP1 contains coiled coils and is a major component of TFs. Despite their similar functions in meiosis, SYCP1 bears no apparent sequence similarity with TF components in other organisms, such as yeast Zip1, fly CG, and nematode SYP-1 and -2, other than them all being coiled coil proteins. Recently, two SYCP1-interacting proteins (SYCE1 and CESC1) have been reported to localize exclusively to the CEs (). In contrast to SYCP1, SYCP2 and -3 are structural components of AEs/LEs. Although SYCP2 bears limited homology with yeast Red1 over a short region, SYCP3 does not appear to have a yeast sequence homologue (). Recently, genes encoding the TF component SYCP1 and one of the AE components, SYCP3, have been disrupted in mice by gene targeting (; ). In mutant mice, normal AEs are formed; homologous chromosomes align with each other, but do not undergo synapsis (). Thus, the meiotic defects in the -deficient mice reflect the functional implication of SYCP1 as a TF component. knockout mice have been characterized in detail in several studies (, ; ; , ; ; ). spermatocytes, AEs are not formed and the other known AE component, SYCP2, fails to localize to axial chromosomal cores (; ). Thus, it was concluded that SYCP3 is a main determinant of AEs/LEs and that SYCP2 plays a role in shaping the in vivo structure of AEs/LEs (). Despite these extensive studies, the role of SYCP2 in the formation of AEs/LEs remains largely unknown, especially the question of whether SYCP2 is required for the incorporation of SYCP3 into AEs/LEs. To investigate the functions of SYCP2 in meiosis, we have cloned the full-length cDNA sequence for the mouse gene. We report the generation of mutant mice by gene targeting and the characterization of the essential role of SYCP2 in SC assembly and chromosomal synapsis in males. Our findings reveal novel insights into the molecular mechanisms underlying mammalian SC assembly. We demonstrate that SYCP2 is a primary determinant of AEs/LEs and is required for the incorporation of SYCP3 into the SC. We previously identified as a mouse germ cell–specific gene in our cDNA subtraction screen (). We obtained the composite full-length cDNA sequence (5 kb) by screening a testis cDNA library. The mouse SYCP2 protein (1,500 aa) shares 63 and 88% sequence identity with human and rat orthologues, respectively (; ). One striking feature of SYCP2 is the presence of a short coiled coil region near its COOH terminus (residues 1,379–1,433 in the mouse SYCP2), which is conserved in rat and human SYCP2 proteins. Antibodies were generated against the mouse SYCP2. Western blot analysis shows that SYCP2 migrates with an apparent molecular mass of 190 kD (). These antibodies were also characterized by the immunostaining of spread nuclei of spermatocytes and double immunostaining with a previously described anti-SYCP2 antibody to confirm that they are specific to SYCP2 (Fig. S1, available at ; ). It has previously been established that SYCP2 interacts with SYCP3 in the yeast two-hybrid assay (). To define the SYCP3-binding domain, we generated NH-terminal truncations of SYCP2 and tested each of them for interaction with the full-length SYCP3. Our results show that the COOH-terminal 310-aa region of SYCP2 (residues 1,191–1,500) is necessary for its interaction with SYCP3 (). Additionally, the deletion of an internal region (residues 1,346–1,476) in SYCP2, including the coiled coil domain, abolishes its interaction with SYCP3 (). This interaction is further supported by results from in vitro GST pulldown experiments. Although the COOH-terminal SYCP2 polypeptide (residues 1,191–1,500) binds to GST-SYCP3, it fails to interact with GST-SYCP3 when the coiled coil–containing region (residues 1,346–1,476) is deleted (). Collectively, these data demonstrate that SYCP2 and -3 are able to form heterodimers or oligomers in vitro. To elucidate the function of in meiosis, we generated mutant mice by homologous recombination in embryonic stem (ES) cells. The mouse gene consists of 44 exons and spans a 70-kb genomic region. Exons 39–43 encode the SYCP2 region from residues 1,346–1,476, including the coiled coil domain, which is required for binding to SYCP3 (). In our targeting construct, the 1.9-kb genomic region harboring exons 39–43 is replaced with a floxed neomycin selection marker (). Therefore, deletion of the essential coiled coil domain is expected to disrupt the gene. To address whether the mutant allele is transcribed, RT-PCR was performed on testis RNA from both wild-type and homozygous mutant ( ) mice with primers residing within exons 38 and 44, respectively. As expected, the mutant allele is, indeed, transcribed in testes (). Sequencing of the mutant RT-PCR product shows that splicing occurs from exons 38–44 without causing a frame shift. and testes (). The truncated SYCP2 protein is referred to as SYCP2t. Our GST pulldown experiment shows that SYCP2 interacts with SYCP3 in vitro (). To address whether these two proteins are associated with each other in vivo, we performed coimmunoprecipitation experiments with soluble nuclear fractions of testicular protein extracts (). SYCP3 was completely immunoprecipitated with anti-SYCP3 antibody. In the wild type, SYCP2 immunoprecipitated with SYCP3. In contrast, SYCP2t was not coimmunoprecipitated, but rather stayed in the immunoprecipitated supernatant. Reciprocal immunoprecipitation experiments confirmed the association of SYCP2 and -3 (unpublished data). This set of experiments shows that SYCP2 is associated with SYCP3 in vivo and that the coiled coil region of SYCP2 is necessary for its binding to SYCP3. mice appear to be healthy and normal in size. mice is sexually dimorphic. males are sterile, but females are subfertile. testes (). However, both heterozygous ( ) males and females are fertile, suggesting the absence of dominant-negative effects by SYCP2t. mice yields a normal Mendelian ratio (27:78:35) of wild-type, , and offspring. testes is 70% less than that of the wild type. testes are significantly smaller in diameter than those in wild type (). Seminiferous tubules of wild-type testes contain a full spectrum of spermatogenic cells, including spermatogonia, spermatocytes, and spermatids (). testes exhibit complete meiotic arrest in spermatogenesis (). In testes, spermatogenic cells develop into zygotene-like spermatocytes, but fail to differentiate into normal pachytene spermatocytes (). seminiferous tubules. testis. Type I tubules contain 2–3 layers of zygotene-like spermatocytes (). In type II tubules, zygotene-like spermatocytes are absent, but a few layers of heavily eosin-stained cells are present and might correspond to apoptotic cells (). Type III tubules are characterized by a single layer of spermatogonia/Sertoli cells (). mice are empty (unpublished data). Collectively, these studies demonstrate that SYCP2 is required for meiosis and spermatogenesis in males. In type II tubules, germ cells are heavily eosin-stained, with increased chromatin density, suggesting that they undergo apoptosis (). A TUNEL assay shows the presence of many apoptotic cells in certain tubules, which are likely to correspond to type II tubules (). In contrast, no apoptotic cells are present in type I and III tubules (). Apoptosis of germ cells in type II tubules is also observed by electron microscopy (unpublished data). testis might reflect coordinated differentiation of germ cells in a given tubule. The presence of three types of tubules could be explained as follows. In testis, spermatogenesis proceeds from spermatogonia, to the leptotene stage, to the zygotene stage, resulting in the accumulation of zygotene-like spermatocytes in type I tubules. Subsequently, these spermatocytes fail to differentiate into pachytene spermatocytes and undergo apoptosis in type II tubules. Eventually, all apoptotic spermatocytes are eliminated in type III tubules. mice, suggesting that apoptotic spermatocytes might be reabsorbed in the seminiferous tubules. spermatocytes, we performed immunostaining on spread spermatocytes with anti-SYCP1 antibodies and CREST antiserum (Kolas et al., 2005). In normal pachytene spermatocytes, homologous chromosomes are fully paired, and SYCP1 is localized to synapsed regions (). In testes, no pachytene spermatocytes are present, which is consistent with the histological analysis (). In spermatocytes, SYCP1 is present in several short fibers (), suggesting a failure in homologous chromosome synapsis. CREST antiserum stains centromeres; therefore, it is used to determine the synaptic process in meiosis, along with DAPI staining of nuclei (). During the leptotene stage, 40 centromeres (CREST foci) are expected. As synapsis proceeds, the number of CREST foci decreases. In pachytene spermatocytes, no more than 21 CREST foci are expected (; ). The number of CREST foci in zygotene spermatocytes ranges from 21 to 40. We analyzed >100 spermatocytes that showed positive SYCP1 staining. spermatocytes had 36 CREST foci/nucleus on average (36.0 ± 3.3; = 104). No spermatocytes with >40 CREST foci were observed (), suggesting that sister chromatid cohesion at the centromeric regions is not affected. Interestingly, many CREST foci were present in pairs (). spermatocytes, we performed immunostaining with anti-SYCP3 antibodies and CREST antiserum. testes, as determined by Western blotting (). spermatocytes, SYCP3 accumulates as several large aggregates in the nucleus, but fails to localize to axial chromosome cores (, arrows). In addition, SYCP3 forms small nuclear foci (, arrowhead). Therefore, we conclude that SYCP2 is essential for the incorporation of SYCP3 into AEs/LEs. In spermatocytes, SYCP3 is not localized to axial chromosomal cores and, thus, is not suitable for analysis of AE formation (). spermatocytes. Silver nitrate stains AEs and paired LEs. spermatocytes are stained with silver nitrate (; ). Silver-stained SCs are abundant in wild type (). spermatocytes after examining >100 spread nuclei (). spermatocytes reveals the presence of CE-like structures with chromatins aligned to form SC-like structures (). spermatocytes lack typical LEs that are electron dense in wild-type spermatocytes (). spermatocytes. Cohesin complexes connect sister chromatids during mitosis and meiosis (). Cohesin proteins are required for assembly of AEs/LEs in diverse organisms. STAG3 is a mammalian meiosis-specific cohesin (). STAG3 apparently localizes to axial chromosomal cores (cohesin complexes) in wild-type spermatocytes (Fig. S2, available at ). In spermatocytes, STAG3 still localizes to long fibers (; and Fig. S2), which correspond to cohesin complexes formed along common cores of sister chromatids, suggesting that SYCP2 is not required for sister chromatid cohesion. spermatocytes (). -spread nuclei (). First, SYCP2t colocalizes to thick fibers with SYCP1, which are continuous, but variable in length (). Second, SYCP2t is observed in fine fibers, where SYCP1 is absent. Close examination reveals that these fine fibers are not uniform in staining and appear as bead-on-a-string arrays (). This staining is likely to reflect authentic SYCP2t localization rather than nonspecific background because it is observed with polyclonal antibodies from two rabbits and two guinea pigs. In addition, the same localization pattern is obtained using antibodies raised against a different region of SYCP2 (Fig. S3, available at ; ). spermatocytes (). spermatocytes (), suggesting that SYCP2t is associated with axial chromosomal cores. We next examined the association of SC proteins with chromatin using different fractions of testicular extracts: cytoplasmic, nuclear, and chromatin (). During the preparation of nuclear extracts, nuclear proteins are bound to the chromatin pellet in various degrees. As expected, the core histones, such as histone H3, are tightly bound to chromatin. testes. In contrast, SYCP2 and -3 behave differently. In wild-type, most SYCP2 is tightly bound to chromatin (, lane 4). Although SYCP2t localizes to axial chromosomal cores, the majority of SYCP2t can be extracted from chromatin (, lane 7), suggesting that its association with chromatin is significantly weakened by the deletion of the coiled coil domain. testes, is associated with chromatin (), which is consistent with the immunolocalization data (). females and wild-type littermates (10 mice/genotype) were mated with wild-type males for 2 mo. The mice were checked daily, and the number of offspring was recorded. males, females produced viable offspring. females displayed a dramatic decrease in litter size, producing on average 4.4 offspring per litter (4.4 ± 1.5; = 16 litters). In comparison, wild-type littermates generated 7.8 offspring per litter (7.8 ± 1.7; = 16 litters; P < 0.0001). ovaries revealed no apparent defects in follicular development (unpublished data). females at 17.5 d postcoitum (dpc) and found that although mutant oocytes exhibit homologous chromosome alignments, as determined by SYCP1 staining, full chromosome synapsis is interrupted by the presence of prominent axial gaps in the SC (). oocytes (). In these oocytes, SYCP3 was found in several large nuclear aggregates (, arrows). Interestingly, remnant SYCP3 staining was also detected as prominent nuclear foci that were consistently associated with chromosome ends at the pachytene stage of meiosis (, arrowheads). oocytes revealed that about half of the SYCP3 foci are associated with CREST signals (unpublished data). oocytes. oocytes, which is consistent with the formation of homologous chromosome synapsis (). xref #text A lambda phage cDNA library was screened essentially as previously described (). In brief, lambda phage lysates were prepared from 24 subpools (∼80,000 clones each) of the mouse testis cDNA library (BD Biosciences) and used as PCR templates. -positive subpools were identified by PCR with -specific primers chosen from previously obtained partial sequences (). 5′ and 3′ cDNA fragments were amplified separately from positive subpools by PCR, in which one -specific primer and one vector primer were used. PCR products were sequenced. The composite cDNA sequence has been deposited in GenBank under accession no. . The cDNA fragment corresponding to residues 1,255–1,500 was cloned into the pQE-30 vector (QIAGEN). The 6× His-SYCP2 fusion protein was expressed in M15 bacteria, purified with Ni-NTA resin, and eluted in 8 M urea. Two rabbits and two guinea pigs were immunized with the recombinant SYCP2 protein (Cocalico Biologicals, Inc.). The anti-SYCP2 antiserum (serum 1918 and GP21) was used for Western blot (1:500) and immunofluorescence (1:100). Various truncated fragments were cloned in the pACT2 vector (BD Biosciences). Mouse coding region was amplified from bulk testis cDNAs by PCR, cloned into the pAS2-1 vector (BD Biosciences), and sequenced. Interaction between SYCP3 and various SYCP2 proteins were assayed by cotransformation into the reporter yeast strain Y190, followed by standard β-galactosidase filter assay. Full-length SYCP3 was cloned into pGEX4T-1, expressed as a GST fusion protein in bacteria, and affinity purified. SYCP2 fragments were produced in the presence of [S]methionine using the TNT in vitro transcription and translation kit (Promega). GST pulldown assays were performed as previously described (). In the targeting construct, the 1.9-kb genomic DNA harboring exons 39–43 was replaced with a floxed neomycin selection cassette (). The two homologous arms (2 and 2.1 kb) were amplified by PCR with high-fidelity DNA polymerase from a -containing bacterial artificial chromosome clone (RP23-160K5). The thymidine kinase–negative selection marker was cloned adjacent to the right arm. The V6.5 ES cells were electroporated with the linearized targeting construct and were cultured in the presence of 350 μg/ml G418 and 2 μM ganciclovir. 48 double-resistant ES cell clones were recovered and screened for homologous recombination events by long-distance PCR. Five clones were produced by homologous recombination via both arms. Two clones (A4 and C6) were injected into B6C3F1 blastocysts (Taconic). No difference in phenotypes was observed between mice derived from these two ES clones. All of the studies were performed with mice from clone C6. All of the offspring were genotyped by PCR. The wild-type allele was assayed by PCR (400 bp) with the primers AGATGAGGGCATATCACCGA and TAAGCACACTCACCA-TCTCC. The PCR product (300 bp) of the mutant allele was amplified by PCR with the primers GCATGTTATCAACCTTATCCCT and CCTACCGGTGGATGTGGAATGTGTG. RT-PCR for splicing assays was performed with the following primers, which were located in exons 38 and 44: TCTGTTCCTAAGGACTGGCA and TACAAGCTGCATTTGGAGTCA. All of the experiments involving mice were approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania. mice were fixed in 10% (vol/vol) neutral buffered formalin (Fisher Scientific) and embedded in paraffin. 8-μm-thick testis sections were cut and used for TUNEL assay. TUNEL assays were performed with the ApopTag peroxidase in situ Apoptosis Detection kit according to the manufacturer's instructions (CHEMICON International, Inc.). Samples were counterstained briefly in 0.5% (wt/vol) methyl green and visualized on a microscope (Axioskop 40; Carl Zeiss MicroImaging, Inc.). For histology, testes were fixed in Bouin's solution, embedded in paraffin, sectioned, and stained with hematoxylin and eosin. Surface spread of spermatocyte nuclei was performed as previously described (; Kolas et al., 2005). females were mated with males. Vaginal copulatory plugs were checked the next morning. Fetal oocytes were collected for analysis at 17.5 dpc. The primary antibodies used for immunofluorescence were as follows: anti-SYCP1 (a gift from P. Moens and B. Spyropoulos [York University, Toronto, Ontario, Canada] and C. Höög [Karolinska Institute, Stockholm, Sweden]; ; ; ), anti-SYCP2 serum 493 (1:400; a gift from C. Heyting, Wageningen University, Wageningen, Netherlands; ), anti-SYCP3 (1:500; a gift from S. Chuma, Kyoto University, Kyoto, Japan; ), anti-STAG3 (1:500; a gift from J.L. Barbero, Centro Nacional de Biotecnologia, Madrid, Spain; ), FITC-conjugated anti-γH2AX (1:500; Upstate Biotechnology), and CREST antiserum (1:5,000; a gift from B.R. Brinkley, Baylor College of Medicine, Houston, TX). Tissue sections were visualized under an Axioskop 40 microscope. Images were captured with a digital camera (Evolution QEi; MediaCybernetics) and processed with ImagePro software (Phase 3 Imaging Systems) and Photoshop (Adobe). EM was performed at the Biomedical Imaging Core facility at the University of Pennsylvania, as previously described (). In brief, 21-d-old testes were fixed in 2.5% glutaraldehyde and 2% paraformaldehyde for 4 h, and then postfixed in 1% osmium tetroxide for 1 h. The specimens were dehydrated in ethanol, transferred to propylene oxide, and embedded in EM-Bed 812 medium (Electron Microscopy Sciences). The specimens were polymerized at 68°C for 48 h. Ultrathin sections were cut with a diamond knife, mounted on single-hole grids, stained with bismuth solution, and examined with an electron microscope (Tecnai-T12; FEI). Digital images were captured with a charge-coupled device camera (Gatan, Inc.). Fig. S1 shows the immunostaining of SCs with our anti-SYCP2 serum to demonstrate the specificity of this antibody. Fig. spermatocytes. Fig. spermatocytes with two different anti-SYCP2 antibodies (). Online supplemental material is available at .

Nuclear pore complexes (NPCs) permit the exchange of metabolites and macromolecules between the nuclear compartment and the cytoplasm. They are embedded in the nuclear envelope (NE) and belong to the largest macromolecular assemblies of the cell. There are two modes of NPC assembly (; ). The first pathway leads to the insertion of NPCs into a closed NE. It represents the only pathway of NPC formation in lower eukaryotes, and it allows the interphase cells of higher eukaryotes to double their NPC number between two mitoses (). The “open mitotic mode” is a pathway that is only used in higher eukaryotic cells, in which NPCs and NEs are disassembled during mitosis. The resulting soluble Nup subcomplexes and vesicular or reticulate membrane structures then reassemble upon mitotic exit, reforming an NPC-perforated NE around chromatin (; ; ; ). The open mitotic mode is characterized by a synchronous assembly of the entire NPC population of a cell. It has been widely studied in cell culture systems (; ; ) and in an in vitro system based on egg extracts (; ; ; ). Although most of the NPC structure might self-assemble through interactions between individual nucleoporins (Nups), assembly factors probably assist in this process. Importin β, for example, appears to act as a RanGTPase-regulated chaperone, which initially shields certain Nup complexes and releases them in proximity to chromatin (; ; ). The actual pores within the NE can be considered products of local fusion between the inner nuclear membrane (INM) and the outer nuclear membrane (ONM). It is still unclear which mechanisms create them, but two scenarios can be envisaged as to how the special structure of the pore membrane forms during exit from an open mitosis. First, vesicles could fuse around preassembled, chromatin-attached NPC scaffolds and thereby create the pore membrane before, or concomitantly with, the closure of the NE. Alternatively, the assembly of NPCs in telophase could follow principles similar to those in interphase, i.e., the double membrane of the NE could form first and, subsequently, be perforated by a local fusion between INM and ONM. How new NPCs are inserted into a closed NE is still unclear, but, again, two strategies can be envisaged. First, a preexisting NPC could grow and then split into two daughter pores (). Intermediates of such a mechanism should be NPCs of higher than the standard eightfold rotational symmetry. Indeed, NPCs with a rotational symmetry of up to 10-fold have been detected (). However, there is no evidence for 16-fold symmetrical intermediates, as predicted for a presplitting NPC or, indeed, for any other plausible combination of pre- and postsplitting symmetry. Furthermore, such pore splitting would also require a membrane fusion event, namely, between opposing sides of the parental pore membrane. In view of the massive NPC structure, the inaccessibility of the lipid bilayers at the pore membrane, and the wide diameter of the pore channel, it is difficult to imagine how a fusion could possibly occur at such a position. Therefore, it appears more likely that a true de novo insertion of NPCs into the NE occurs. Indeed, experiments using the NPC assembly inhibitor BAPTA indicate that such an insertion does not require preexisting NPCs (). A de novo insertion of NPCs into the NE must include a local fusion between INM and ONM to yield the actual pore. How this fusion comes about is still unknown. One complication is that INM and ONM are held ∼20–25 nm apart; hence, the fusion machinery needs to bring them into a sufficiently close proximity to allow membrane fusion to occur. A second complication is that the pore-forming fusion occurs at the luminal faces of INM and ONM. Therefore, it must use factors other than the classical fusion machineries of the secretory pathway, which catalyze membrane fusions through the cytoplasmic sides of the target membranes. In analogy to membrane fusion events mediated by SNAREs or viral fusion proteins (), however, it appears likely that integral membrane proteins play a critical role. Possibly, these integral fusion factors remain stably associated with mature NPCs as membrane-integral Nups. Membrane-integral Nups probably fulfill several additional functions, e.g., the recruitment of other Nups to assembly sites at the nuclear membrane, as (static) anchors of (mature) NPCs within the NE, as part of the rigid NPC structure, and, if equipped with the Nup-typical phenylalanine-glycine (FG)–rich repeats, as constituents of the permeability barrier of nuclear pores. Given the striking conservation of general NPC architecture, it would be very surprising if the integration of yeast and animal NPCs into the NE traced back to different evolutionary origins. Nevertheless, thus far it appeared that NPCs from yeast and vertebrates are equipped with completely different sets of membrane-integral Nups. POM121 and gp210 (; ) have, so far, been the only known membrane-integral constituents of vertebrate NPCs, but they are both absent from fungi. The yeast possesses three membrane-integral Nups: Pom152p, Pom34p, and Ndc1p (referred to as Cut11p in ). Pom152p and Pom34p are not essential, and they lack obvious orthologues in higher eukaryotes (; ). In contrast, Ndc1p is essential (; ; ; ). It is, however, not only a Nup but also a constituent of spindle pole bodies (SPBs), which are the NE-embedded form of centrosomes that is typical of yeast. Nuclear pore and SPB membrane exhibit analogous topological features. Nevertheless, NPCs and SPBs represent distinct structures, apparently sharing just a single component, which is Ndc1p (). Ndc1p is required for inserting newly formed SPBs into the NE, and this function is clearly essential for the viability of yeast (; ). So far, however, a role for Ndc1p in NPC biogenesis is only indicated by genetic interactions with Nic96p and by the mutant, which, at the nonpermissive temperature, fails to properly incorporate Nup49p into otherwise functional NPCs (). It is still unclear if the function of Ndc1p in NPC biogenesis goes beyond anchoring individual Nups to the NPC scaffold. However, if it had a fundamental function in NPC biogenesis, then it should be conserved across eukaryotic kingdoms and should also be present in those eukaryotes that have an open mitosis and, thus, lack NE-embedded SPBs. In the accompanying study (see Stavru et al. on p. 477 of this issue), we report the observation that functional mammalian NPCs can assemble in cells that are devoid of gp210 and severely depleted of POM121. This suggested that, to date, at least one crucial membrane-integral Nup of mammals must have escaped detection. We confirm this assumption and demonstrate that metazoan NPCs contain an additional constituent, which is orthologous to yeast Ndc1p. Depletion of human NDC1 (hNDC1) from HeLa cells causes severe NPC-assembly defects. Loss of NDC1 function in also causes severe phenotypes, but it is not ultimately lethal. This leads to the conclusion that none of the membrane-integral Nups is essential under all conditions for NPC biogenesis, and points to an extreme flexibility and robustness of the NPC assembly process. In search of the missing component, we reasoned that a membrane-integral constituent of yeast NPCs might have an as yet unidentified orthologue in higher eukaryotes. Searches with Pom152p or Pom34p did not yield any convincing hits. However, we found Ndc1p orthologues not only in other ascomycetous fungi (e.g., , , , and ) but also in basidiomycetes (e.g., , ) and viridiplantae (e.g., , , , , and ), as well as in nematodes (e.g., ), insects (e.g., ), cnidarians (e.g., ), tunicates (e.g., ), amphibians (e.g., ), fish (e.g., ), birds, and mammals ( and Fig. S1, available at ). NDC1 was, thus, an excellent candidate for constituting a widely conserved membrane anchor of NPCs. hNDC1 was previously identified in a proteomics screen as NE transmembrane protein 3 (Net3; ). To determine its intracellular localization at a higher resolution, we expressed NH- and COOH-terminal GFP fusions of hNDC1 in HeLa cells (Fig. S2, available at ). At low or moderate expression levels, a clear colocalization with NPCs was observed, suggesting that NDC1 is also a Nup in human cells. To localize the endogenous hNDC1, we raised antibodies against two regions of the protein and used them for immunofluorescence on HeLa cells. For both sets of antibodies, a clear NPC staining was evident ( and ). In , we used either mAb414, which recognizes several FG repeat Nups (), or the fluorescently labeled Impβ fragment () to decorate NPCs, and we observed conspicuous colocalization with the hNDC1 signal. As already mentioned, we identified Ndc1 orthologues in numerous other eukaryotes and, hence, wanted to know if localization at NPCs represents a general feature of NDC1 family members. Therefore, we raised antibodies against NDC1, against the more widely expressed isoform of the two paralogues (variant 1; ), and against NDC-1. Again, colocalization with the respective nuclear pore markers, i.e., Nup62, TPR, or mAb414, was observed (). NDC1 is, thus, a widely conserved constituent of NPCs. hNDC1 clearly behaves like an integral membrane protein; it fractionates with membranes and withstands membrane extraction at pH 12.0 (unpublished data). The number and orientation of the transmembrane segments (TMSs) determine which parts of hNDC1 are exposed to the cytoplasmic/NPC side of the membrane and, hence, are available for interaction with other Nups. Therefore, we decided to resolve its topology. An in silico analysis was used to generate a topology model (see Materials and methods), which was subsequently tested experimentally. The model predicted six putative TMSs and cytoplasmic exposure for the NH and COOH termini, as well as for loops 2 and 4, whereas loops 1, 3, and 5 are predicted to face the lumen of the NE ( and ). In agreement with the model, we found the COOH-terminal domain (NDC1), as well as the extreme NH and COOH termini, to be accessible for antibodies from the cytoplasmic side of the membrane (for data and experimental description see and Fig. S3, available at ). In contrast, antibodies against loop 5 recognized their epitope only when the internal membranes had been solubilized by Triton X-100 (). This is expected, if loop 5 is located in the lumen of ER or NE. In a second set of experiments, we introduced N-glycosylation sites (NGSs) into loops 1, 3, or 5. Indeed, we observed the selective glycosylation of these sites, when in vitro translation was performed in the presence of RER membranes (). As this modification occurs only in the RER lumen (), one can conclude that loops 1, 3, and 5 are indeed luminal. This experiment also indicates that NDC1 is initially integrated into RER membranes before its assembly into NPCs. Such intermediates in the RER can indeed be detected microscopically, when GFP-tagged hNDC1 is overexpressed (Figs. S2 and S3). The experimental data, thus, support the topology model, at least for the human member of the NDC1 family. Its ∼45-kD COOH-terminal domain (NDC1), which includes the most conserved part of this protein (), is therefore entirely extraluminal and available for interactions with other Nups. In the next step of our analysis, we used postembedding immunogold EM to localize the conserved COOH-terminal domain of hNDC1. The antibodies gave a very specific labeling along the NE, with >95% of the gold decorating NPCs (see representative EM images in ). The positions of the gold labels are consistent with the assumption that the COOH-terminal domain resides within the body of the NPC proper and is part of the NPC scaffold. Sequence analysis of the conserved COOH-terminal domain of NDC1 predicted several consensus phosphorylation sites for mitotic kinases. Because mitotic phosphorylation plays a key role in disassembling NPCs (), it was tempting to assume that a mitotic modification of NDC1 might contribute to this disassembly process. Such modifications often change the mobility of protein species in SDS gels, and, indeed, immunoblots revealed a prominent slow-migrating NDC1 species that was specific for HeLa cells arrested in M phase (Fig. S4, available at ). We are currently investigating the nature of these modifications and their positions within the NDC1 sequence. Having established that all tested metazoan NDC1 proteins are constituents of NPCs, we wanted to elucidate the consequences of a loss of NDC1 function. For this we used the RNAi approach to knockdown hNDC1 in human cell lines (). The four different siRNA duplexes that were efficient (see Materials and methods) all gave a similar phenotype ( and not depicted). The reduction of NDC1 correlated with a proportional loss of the NPC signal for the mAb414-reactive FG repeat Nups, or for Nup88, which anchors Nup214 and Nup358 to NPCs (). Therefore, the assembly defects caused by the NDC1 depletion might not be restricted to the vicinity of the pore membrane, but extend to NPC structures that are distant from the pore membrane. To address the question of whether eukaryotes can assemble at least rudimentary NPCs without NDC1, we switched the model organism and analyzed NDC-1 genetically in . Database searches pointed to the −− strain (), generated by S. Mitani at the Japanese deletion consortium. The strain carries a mutation at the B0240.4 locus, which is predicted to disrupt the ORF of ceNDC-1 just after the second membrane-spanning segment. If expressed, the resulting deletion would still comprise ∼25% of the protein sequence, but would lack all parts of the protein that are conserved and potentially exposed toward the NPC. The mutant strain has so far been propagated only in the heterozygous form because the phenotype of the homozygous mutant is so severe that it was initially listed as sterile or lethal. However, we were able to detect rare cases of homozygous mutant worms that not only developed until adulthood but also produced a few offspring. The homozygous mutant −− genotype was confirmed by single-worm PCR (not depicted), as well as by immunoblots showing that the ceNDC-1 protein is, indeed, absent in homozygous mutant worms (). We have now maintained these homozygotes for >15 generations and can therefore exclude the possibility that their survival is only attributable to a maternal ceNDC-1 mRNA pool inherited from a heterozygous progenitor. Immunofluorescence also confirmed the absence of the ceNDC-1 protein from the mutant worms (unpublished data). In addition, it revealed a significantly reduced mAb414 signal of the NE, as compared with wild-type worms. However, this staining for FG repeat–containing Nups was not completely lost, indicating that rudimentary NPCs can assemble and persist in the absence of ceNDC-1 (). Consistent with the assumption that these rudimentary NPCs are functionally impaired, we observed a very high embryonic and larval mortality rate for the homozygote −− mutant. The few surviving individuals developed very slowly until adulthood, and most of them remained sterile. These phenotypes culminated in a strongly reduced brood size ( and Fig. S5, available at ). The surviving homozygous animals displayed additional pleiotropic phenotypes (unpublished data), such as the “clear” phenotype, which indicates the failure of properly developing internal structures and organs. Homozygous adults were smaller than the heterozygote −− mutants or wild-type worms. This also held true for the eggs and embryos. To prove that these phenotypes were indeed the consequences of the mutation, in the −− homozygous background we generated a transgenic worm that expresses NDC-1∷GFP from the endogenous promoter. The NDC-1∷GFP fusion protein was detectable by anti–ceNDC-1 antibodies () and localized to NPCs (). Because we had introduced the NDC-1∷GFP fusion in the form of an extrachromosomal array, which typically gives a mosaic expression, the GFP signal was not observed in all cells of the embryos. Nevertheless, expression of the transgene complemented the ceNDC-1 loss-of-function phenotypes and dramatically improved fertility of the homozygous −− mutant (Fig. S5 and ). The phenotypes of the −− strain, therefore, are caused by the ceNDC-1 gene disruption and are not the consequences of secondary mutations. SPBs and NPCs are distinct structures; therefore, Ndc1p must cooperate with distinct sets of components to create either nuclear or SPB pores (). The SPB- and NPC-relevant interactions of Ndc1p even appear to be in competition, as indicated by the observation that the deletion of the membrane-integral Nup POM152 suppresses SPB assembly defects that are caused by certain Ndc1p mutations (). The extreme sensitivity of yeast cells to any change in Ndc1p dosage () indicates how delicate the equilibria in these interactions might be. We identify metazoan orthologues to Ndc1p and show that they constitute an integral component of NPCs in mammals, amphibians, insects, and nematodes. Metazoan NDC1 is presumably fully dedicated to its function at NPCs because metazoa lack NE-embedded SPBs. NDC1 is now the third known membrane-integral Nup in vertebrate NPCs, and its presence may be one possible explanation as to why NPCs can still form in the virtual absence of POM121 and gp210, which are the other two integral constituents (Stavru et al., 2006). The crucial contribution of NDC1 to the NPC assembly process is indicated by the severe NPC biogenesis defects that occurred when the protein was either depleted by RNAi or when the ORF had been disrupted genetically. The biogenesis of NPCs is a very elaborate process. It requires not only the self-assembly of ∼700 individual polypeptide chains (representing multiple copies of the ∼30 different Nups) into a single giant protein complex but also a local fusion between INM and ONM to create the actual pore, as well as the implantation of the NPC scaffold into this pore. NPCs are essential structures, and their failure to assemble would be lethal. Therefore, it is not surprising that the assembly process is robust and fault tolerant. This resistance toward disturbances becomes particularly apparent in the fact that more than half of the yeast Nups can be singly deleted without causing deleterious defects (). For most deletions of such nonessential Nups, however, synthetic–lethal interactions with loss-of-function alleles of other Nups have been found (; ; ). This illustrates that many of the crucial protein–protein interactions are backed by more than one player. Such inherent flexibility probably contributes greatly to the intrinsic fault tolerance of the NPC assembly process. Based on experiments in the egg extract system, a different explanation for the fidelity of assembling an NPC-perforated NE has been given, namely a surveillance of the process by a POM121-dependent checkpoint system (). For several reasons, we view this concept with some caution. Bona fide checkpoints allow active intervention into those cellular processes that could result in uncorrectable errors (). The mitotic spindle checkpoint, for example, reduces the probability of an uncorrectable aneuploidy by delaying sister chromatid segregation until each of the chromosomes is properly attached to the mitotic spindle. Nuclei enclosed by a pore-free membrane, however, are not uncorrectable dead-end products. Instead, NPCs can still be integrated into them at later time points (). In addition, the great number of NPCs, which become embedded into an NE, should make the nuclear assembly process tolerant against occasional failures to assemble individual NPCs. Considering further that all crucial checkpoints, such as the DNA damage and mitotic spindle checkpoints, are disabled during the early cell cycles in the developing embryo (), we find it hard to understand why an NPC assembly checkpoint should be kept in operation. Finally, we observed that POM121-depleted human cells formed functional NPCs and showed no uncoupling between NE formation and NPC assembly (Stavru et al., 2006). Consistent with the concept of redundancy and robustness, NPCs appear not to rely on just a single anchor within the NE. Instead, they typically contain several membrane-integral Nups (e.g., three different ones in either yeast or mammals). Genomic data indicate that two of them, gp210 and NDC1, are evolutionary conserved (). The fact that both are found in metazoans as well as in plants, clearly suggests that they evolved before the unikont/bikont bifurcation, which is considered as the oldest time point of a major evolutionary diversification of known eukaryotes (). Primordial NPCs were therefore probably equipped with both gp210 and NDC1. However, it appears that some lineages (e.g., all fungi) lost gp210, whereas other lineages (e.g., or other protozoa) lost NDC1 from their genomes. This brings us to the unexpected conclusion that none of the integral Nups is—generally and in all cellular settings—essential for NPC assembly and function. This also explains why the nematode can live, although miserably, in the absence of NDC1, why many mammalian cell types, such as fibroblasts, assemble fully functional NPCs without gp210, and why POM121 can be depleted from human cells without deleterious defects. Studies in viral systems and in the secretory pathway have clearly established that a controlled membrane fusion requires energy (). In all of the cases characterized so far, it is conformational energy stored in fusion-promoting proteins that forces the opposing lipid bilayers to such a short distance that they can eventually coalesce. Viral fusion proteins can release their conformational energy only once, and such a single-use fusion factor would be sufficient to explain NPC biogenesis in yeast. In higher eukaryotes with open mitosis, however, the situation is more complex. NE and NPC disassemble here once per cell cycle and, subsequently, reassemble from the existing membrane and protein components. Of course, the still elusive fusion factor could be degraded and resynthesized during every cell cycle and, in this case, it might not remain associated with mature NPCs. This would explain why no such activity has been found so far. Otherwise, multiple cycles of NPC formation and disassembly would require an additional recycling machinery, which converts the fusion factor from a postfusion to a prefusion conformation. Such recycling machinery would have to reside in the lumen of the NE, and it will be very interesting to see whether it exists or whether nuclear pore formation relies on “disposable” fusion proteins. Ndc1p orthologues were identified by BLAST from public databases. cDNAs comprising the coding regions of human, mouse, , and NDC1 were obtained from the German Resource Center for Genome Research or amplified from total RNA by RT-PCR. Coding regions were verified by DNA sequencing. Multiple alignments were performed with the ClustalW algorithm (). Membrane-spanning segments were predicted by combining the results of different algorithms and the hydrophobicity profiles of the respective sequences. The multiple alignments of predicted TMSs were manually corrected. The orientation of the TMSs was predicted from the constraints (a) that the cytoplasmically flanking region of a membrane anchor is typically more positively charged than the luminally flanking one () and (b) that adjacent TMSs must have opposite orientation. The complete coding sequences of NDC1 were given the following accession numbers (available from GenBank/EMBL/DDBJ): , ; , ; (variant 1), ; and , . Antibodies were newly raised in rabbits or guinea pigs against the following protein fragments: hNDC1 (anti–loop 5), hNDC1 (anti–COOH-terminal domain), dmNDC1, ceNDC-1, hPOM121, human gp210, xNDC1, and dmTPR. Antibodies against human Nup62, Nup358, Nup96, and Nup107 () have been described earlier. All polyclonal antibodies were affinity purified on their respective antigen columns. The mAb against p62 was also previously described (). mAb414 was obtained from Eurogentec, and the mAb against Nup88 was obtained from BD Bioscience. The antibody against lamin was a gift from G. Krohne (Biozentrum, Universität Würzburg, Würzburg, Germany). XL-177 cells were cultivated in 65% Leibovitz' L-15 (Sigma-Aldrich) supplemented with -glutamine, 15% FCS, 100 U/ml penicillin, and 100 μg/ml streptomycin. S2 cells were obtained from the American Type Culture Collection (ATCC CRL-1963) and cultivated in serum-free medium supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin, and -glutamine, according to the manufacturer's instructions. Worm cultures were maintained using standard techniques (). The heterozygous −− strain was obtained from S. Mitani at the Japanese deletion consortium (Tokyo Women's Medical University School of Medicine, Tokyo, Japan). To select for homozygous −− mutants, sick-looking hermaphrodites were singled out on plates. Most animals died without generating any offspring. The remaining animals had few offspring. The animals were selfed for at least 15 generations. Single-worm PCRs were performed on homozygous and heterozygous animals to confirm the homozygosity of the mutants. The NDC-1∷GFP fusion was created by cloning the genomic copy of , including the putative promoter region into pPD95.81 (). This reporter construct (20 ng/μl) was coinjected with 80 ng/μl pRF4 () into the gonads of homozygous worms. Extrachromosomal arrays were selected by the phenotype, and the ability of NDC-1∷GFP to rescue the −− phenotype. NDC-1∷GFP was visualized with an epifluorescence microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.). Images were collected with an Axiocam (Carl Zeiss MicroImaging, Inc.) and the contrast was adjusted with Photoshop (Adobe). Pictures of worm plates were taken with a digital camera (Coolpix; Nikon) mounted onto the binocular. Transfection of cultured human cells with siRNAs was carried out essentially as previously described (). Annealed siRNAs were purchased from Dharmacon. Antisense strands were complementary to nucleotides 1,915–1,935, 405–425, or 1,569–1,596 of the hNDC1 ORF. In addition, we performed RNAi with stealth siRNAs (Invitrogen), whose sense strand modification is thought to reduce nonspecific effects. Its antisense strand was complementary to nucleotides 1,085–1,109 of the hNDC1 ORF. For each of these four siRNAs, we observed the same correlation between hNDC1 knockdown and depletion of mAb414-reactive Nups from the NE. shows results with the stealth oligo duplex. DNA transfections for expression of EGFP-tagged NDC1 were performed with Fugene6 (Roche), according to the manufacturer's instructions. Cells were analyzed 24–60 h after transfection. For immunofluorescence, cultured cells were washed briefly with PBS, fixed for 4 min in 3% paraformaldehyde that was freshly dissolved in PBS, washed in PBS, quenched with 50 mM NHCl in PBS for 5 min, permeabilized with 0.25% Triton X-100 in PBS, and blocked for at least 30 min in 1% BSA, 10% goat serum, and 0.1% Triton X-100. Primary antibodies were applied for 60 min in blocking buffer. Nonbound antibodies were washed off with PBS. Alexa Fluor–labeled secondary antibodies were purchased from Invitrogen. The secondary antibodies and the DNA stain Hoechst 33342 were applied for 30–60 min in blocking solution, followed by extensive washing and mounting in Vectashield (Vector Laboratories). Immunofluorescence on embryos was performed after freeze-fracturing and formaldehyde fixation. For in vitro translation, the hNDC1 coding region was cloned downstream of a T7 promoter. The following NGSs were inserted: SSNGTS after residue 155 of hNDC1 (NGS-loop 3), SSNGTS after residue 253 (NGS-loop 5). The glycosylation site SSNGTS inserted after residue 61 in loop 1 was not glycosylated, probably because it is too close to the membrane (). However, when extended to the sequence TSGSGNSSNGSGT, it was glycosylated to 70–80%. Fig. S1 shows the evolutionary conservation of NDC1. Fig. S2 shows EGFP-tagged hNDC1 is targeted to NPCs. Fig. S3 shows the membrane topology of hNDC1 probed by epitope tagging. Figure S4 shows that hNDC1 is heavily modified during mitosis. Fig. S5 shows that NDC-1∷GFP rescues the high mortality and sterility phenotype of the −− allele. Online supplemental material is available at .

Peroxisomes are small membrane-bound organelles that function in cellular metabolism in diverse ways, including the β- and α-oxidation of fatty acids, the oxidation of bile acids and cholesterol, and conversion of hydrogen peroxide to nontoxic forms. The peroxisome's importance in lipid metabolism and defense against oxidative stress explains why defects in peroxisome biogenesis underlie several severe inherited diseases known as peroxisomal biogenesis disorders, including Zellweger syndrome, infantile Refsum disease, and neonatal adrenoleukodystrophy (). Genetic and proteomic studies in yeast and mammalian cell systems have led to the identification of up to 32 proteins (collectively called peroxins or PEX) involved in peroxisome biogenesis. Of these peroxins, three in mammalian cells—PEX3, PEX16, and PEX19—and two in yeast cells—Pex3p and Pex19p—are specifically involved in peroxisomal membrane protein (PMP) import (; for review see ). When any of these proteins are absent or mutated in cells, peroxisomes disappear. PEX19, a farnesylated protein found in both cytosol and peroxisomes, binds nascent PMPs in the cytoplasm and targets them to the peroxisomal membrane (). PEX3, an integral membrane protein, acts as a docking receptor for incoming complexes of PEX19 and its PMP cargoes (). PEX16, an integral membrane protein absent in most yeast, is thought to serve as a receptor for PEX3 or as a component of the membrane translocator (; ). Despite knowledge of the essential components involved in peroxisome biogenesis, the origin of peroxisomes has been controversial. The long-standing view is that peroxisomes are semiautonomous organelles, like mitochondria and chloroplasts, which multiply strictly by growth and division (; for review see ). This view is based on the premise that peroxisomal proteins (both matrix and membrane associated) are synthesized on free ribosomes and are imported directly into peroxisomes from the cytoplasm (). However, unlike mitochondria and chloroplasts, peroxisomes can disappear from a cell and be regenerated de novo. For example, cells defective in PEX3, PEX16, or PEX19 do not have any detectable peroxisomes; yet, upon introduction of the wild-type version of the missing or mutated gene, peroxisomes are quickly regenerated (; ). This capacity of peroxisomes to regenerate, together with other observations related to the intracellular sorting pathways of PMPs (see the following paragraph), has led to an alternative view of peroxisomal biogenesis, in which other organelles—specifically, the ER—participate in the formation and maintenance of peroxisomal membranes (for reviews see ; ; ). Several lines of experimental support for an ER-dependent mode of peroxisome biogenesis, especially in yeast and plants, have been obtained over the past few years. For example, in the yeast cell type , Pex16p and Pex2p are -glycosylated (a modification only occurring in the ER), suggesting they pass through the ER en route to peroxisomes (). In plant cells, both Pex16p and cottonseed peroxisomal ascorbate peroxidase localize to a distinct subdomain of the ER called peroxisomal ER in addition to being found in peroxisomes (; ). Finally, in cells, peroxisomes reappear by outgrowth of Pex3-GFP–containing structures from the ER during complementation of ΔPex3p mutants with Pex3p-GFP (; ). In addition, Pex3p that is targeted to the ER by an attached signal sequence is routed to peroxisomes (). In yeast and plant cells, therefore, the ER seems to play a direct role in delivering lipid and protein components to peroxisomes. Whether the ER plays a similar role in peroxisome biogenesis in mammalian cells remains unclear. The only evidence suggesting this comes from studies in mouse dendritic cells, in which PEX13 and PMP70 have been reported in reticular structures apparently connected to smooth ER (). Other studies have provided results that are inconsistent with an ER role. Exogenously expressed PEX3 or PEX19, for example, are not localized to the ER in mammalian cells, even when they are overexpressed in peroxisome-deficient cells (). Because PEX3 does not directly target to the ER in mammalian cells or in plant cells (), the findings in related to Pex3p trafficking (and potentially its conclusions regarding the role of the ER in peroxisome biogenesis) may not be generalizable to all organisms. Indeed, peroxisome biogenesis in mammalian cells is widely assumed to occur primarily by fission of preexisting peroxisomes with any de novo pathway, either ER dependent or independent (for reviews see ; ), occurring only under unusual conditions in mutated cells. The role of the ER in the biogenesis of mammalian peroxisomes would therefore seem to be limited to providing membrane components (e.g., lipids) rather than to providing a platform for the outgrowth of new peroxisomes, as observed in cells (; ; ). These types of concerns have led us to directly investigate the pathway for peroxisome biogenesis in mammalian cells. Toward this end, we have used monomeric and photoactivated versions of GFP linked to the essential membrane peroxin, PEX16, to address whether specific peroxisomal membrane components in mammalian cells are normally derived from the ER or whether this occurs only when preexisting peroxisomes are missing. We furthermore have used a novel photo-chase strategy highlighting old and new peroxisomes to address whether new peroxisomes in mammalian cells form primarily by growth and division of preexisting peroxisomes or by the maturation of new peroxisomes derived from the ER. Our findings provide the first direct evidence in mammalian cells that the ER plays a central role in both the origin and maintenance of peroxisomes. To characterize the dynamic distribution of human PEX16, the COOH terminus of PEX16 was tagged with monomeric versions of various fluorescent proteins (GFP, photoactivatable GFP [PAGFP], or Venus). Two lines of evidence suggested that all of the resulting chimeras targeted and functioned properly when expressed in mammalian cells. First, when PEX16 tagged with GFP (PEX16-GFP) was expressed in COS-7 cells, complete colocalization was observed between PEX16-GFP and a coexpressed peroxisomal reporter molecule consisting of the red fluorescent protein (RFP) tagged to type 1 peroxisomal matrix targeting signal, SKL-COOH (RFP-SKL; ). Second, in cells from the human fibroblast cell line GM06231 lacking peroxisomes because of a mutated PEX16 gene, introduction of PEX16-GFP led to the appearance of new peroxisomes (), indicating PEX16-GFP can complement PEX16 function. COS-7 cells expressing PEX16-GFP at higher levels, achieved by increasing the time between cell transfection and imaging (from 15 to 24 h; Fig. S1, available at ), were examined to determine whether PEX16-GFP changed its distribution once the machinery involved in PEX16 sorting to peroxisomes became limited by PEX16 overexpression. To identify peroxisomes as well as ER in these cells, the peroxisomal marker RFP-SKL and the ER marker ssRFP-KDEL were individually coexpressed with PEX16-GFP. The distribution of PEX16-GFP in these cells included the ER as well as peroxisomes (). Hence, the membrane localization of PEX16-GFP is not restricted to peroxisomes but includes the ER under conditions of high PEX16-GFP expression. Expression of PEX16-GFP at low levels in the peroxisome biogenesis disorder (PBD) 399-T1 human fibroblast cell line that lacks PEX19 and in which peroxisomes are absent () showed the chimera residing exclusively in the ER, with RFP-SKL diffusely distributed throughout the cytosol (). A similar ER pattern was observed for PEX16-GFP expressed in a PEX3 mutant human fibroblast cell line (PBD400) that, similar to PBD399-T1 cells, lacks peroxisomes (; unpublished data). When peroxisomes are absent, therefore, PEX16-GFP targets to the ER and not to the cytosol. Cell fractionation and immunoblot analysis of PEX16-GFP–transformed COS-7 cells performed 24 h after transfection to allow for protein overexpression revealed that PEX16-GFP resided only in the membrane (or nonsoluble fraction), in contrast to GFP expressed alone, which was primarily in the soluble fraction (). Hence, PEX16-GFP does not appear to ever reside in the cytosol. Peroxisomes that were observed in cells having high PEX16-GFP expression were closely aligned with the ER ( [box] and D), suggesting that peroxisomes and the ER are intimately associated. Repetitive photobleaching (or fluorescence loss in photobleaching [FLIP]) of PEX16-GFP fluorescence in a small, centralized area of the ER in these cells (, red box) resulted in PEX16-GFP fluorescence being lost throughout the ER without affecting PEX16-GFP fluorescence in surrounding peroxisomes (). Molecules of PEX16-GFP can thus freely diffuse throughout the ER, whereas they are retained within individual peroxisomes. To investigate whether the ER localization of PEX16 represented an intermediate in the pathway for delivery of PEX16 to peroxisomes, we developed a photo/pulse-chase–labeling assay using PEX16 attached to PAGFP (PEX16-PAGFP). PAGFP is undetected until “activated” by high-energy light, whereupon it becomes brightly fluorescent. Activated PAGFP molecules remain fluorescent over time, whereas PAGFP molecules that have not been photoactivated (including newly synthesized and newly folded forms) stay invisible (). The photo/pulse-chase assay was performed in COS-7 cells coexpressing PEX16-PAGFP and RFP-SKL 24 h after transfection, as outlined in . Initially, a small region of interest (ROI; , red box) containing only ER (blue) and no peroxisomes (red) was repeatedly irradiated over 30 min with 413-nm light (pre-PA). Because PEX16-GFP diffuses freely throughout the ER, most PEX16-PAGFP molecules in the ER should become photoactivated (, green) under this treatment. An image of the cell was collected sequentially in the 543-nm channel to visualize peroxisomes containing RFP-SKL immediately before and after each photoactivation to ensure no peroxisomes moved into the ROI and that no PEX16-PAGFP molecules within peroxisomes become photoactivated. After photoactivation in this manner for 30 min, images of PEX16-PAGFP at 488-nm fluorescence and RFP-SKL at 543-nm fluorescence were acquired (, Post-PA t = 0 min). The cell was then incubated for 5 h (chase) before acquiring another set of images (, Post-PA t = 5 h) to assess whether fluorescent PEX16-PAGFP molecules had redistributed to RFP-SKL–containing peroxisomes. As shown in , before photoactivation of the ER pool of PEX16-PAGFP, GFP fluorescence (excited at 488 nm) was negligible, whereas the fluorescence attributable to RFP-SKL (543 nm) was readily visible and localized primarily to individual (punctate) peroxisomes (, Pre-PA). Upon repeated photoactivation of the small ROI in the lower left part of the cell, PEX16-PAGFP fluorescence became visible in the ER and in a few puncta that partially overlapped with RFP-SKL, suggesting they were peroxisomes (, Post-PA t = 0 min, arrows). After the 5-h chase period, significantly more PEX16-PAGFP fluorescence was localized in peroxisomes (, Post-PA t = 5 h, arrows). Quantification of peroxisomes containing both PEX16-PAGFP and RFP-SKL over the 5-h chase period in five independent experiments revealed that 10–40% of all peroxisomes in these cells became labeled with photoactivated PEX16-PAGFP (unpublished data). Because PEX16-PAGFP molecules were pulse-labeled in the ER and later appeared in peroxisomes, the data suggested that PEX16-PAGFP undergoes specific transport from the ER to peroxisomes. To investigate what molecular features of PEX16 were necessary for it to pass from the ER to peroxisomes, we used a PAGFP-tagged variant of PEX16 in which the NH-terminal membrane peroxisome–targeting sequence (residues 66–81; -RKELRKKLPVSLSQQK-; ) was deleted. Expression of this construct (delPEX16-PAGFP) in COS-7 cells resulted in only an ER pattern of localization with no peroxisome labeling (Fig. S2, available at ). In the photo-chase assay, photoactivation of the ER pool of a cell expressing delPEX16-PAGFP resulted in the fluorescence attributable to delPEX16-PAGFP never redistributing to peroxisomes and remaining within the ER (). The small structures containing photoactivated delPEX16-PAGFP seen in the juxtanuclear region of these cells (, Post-PA t = 0 min and Post-PA t = 5 h) presumably represented compacted ER cisternae, as their fluorescence was diminished upon repeated photobleaching of a small area of ER in delPEX16-GFP–expressing cells (Fig. S2). Thus, delivery of PEX16 from ER to peroxisomes is dependent on the membrane peroxisome–targeting sequence found within PEX16. Nascent polypeptides containing an NH-terminal signal sequence are bound by the signal recognition particle in the cytoplasm and transferred to the ER before the remainder of their mRNA is translated (). By attaching such a sequence to the NH terminus of PEX16, we reasoned that we could force newly synthesized PEX16 proteins to be cotranslationally inserted into the ER before they targeted elsewhere in the cell (such as to peroxisomes). With such a construct, we could then address whether PEX16 could target to peroxisomes after being cotranslationally synthesized in the ER. We appended residues 14–90 from the well-defined type I signal anchor sequence of leader peptidase (designated as sa; ; ) to the NH terminus of PEX16-GFP (yielding saPEX16-GFP) to preserve the native (N-C; ) membrane topology of PEX16 (). Evidence that sa functioned properly as a signal sequence for targeting of proteins to the ER was demonstrated by appending it to GFP alone (producing saGFP) and expressing the construct in COS-7 cells treated with brefeldin A to block secretory transport out of the ER. In these cells, saGFP accumulated in the ER (), whereas untagged GFP expressed in COS-7 cells was distributed diffusely throughout the cytosol and nucleus (). To test whether saPEX16-GFP could target to peroxisomes after biosynthesis in the ER, we examined COS-7 cells coexpressing saPEX16-GFP and ssRFP-KDEL 15 h after transfection. A large pool of saPEX16-GFP could be seen colocalized with ssRFP-KDEL in the ER (). Differential permeabilization and antibody binding experiments demonstrated that saPEX16-GFP in the ER maintained the same C topology as that of PEX16-GFP (Fig. S3, available at ). Repetitive photobleaching (i.e., FLIP) of saPEX16-GFP fluorescence in an area of ER abolished saPEX16-GFP fluorescence throughout the ER and revealed a pool of saPEX16-GFP fluorescence in peroxisomes (; note the localization of saPEX16-GFP– in RFP-SKL–containing peroxisomes in ). Therefore, saPEX16-GFP can undergo transport to peroxisomes after being synthesized in the ER. When the cells were imaged at earlier times after transfection (8 h) to examine saPEX16-GFP localization at lower expression levels, saPEX16-GFP was exclusively colocalized with RFP-SKL in peroxisomes (). A possible explanation for this is that saPEX16-GFP moves efficiently from ER to peroxisomes until excess saPEX16-GFP saturates the machinery for targeting to peroxisomes. Consistent with this, time-lapse imaging of cells expressing saPEX16-GFP over a 10-h period (during which saPEX16-GFP expression levels went from low to high) revealed newly synthesized saPEX16-GFP accumulating in peroxisomes before also accumulating in the ER (). To determine whether saPEX16-GFP could complement PEX16 function and give rise to new peroxisomes, cells from the human GM06231 cell line lacking peroxisomes were transfected with RFP-SKL, PEX16-GFP and RFP-SKL, or saPEX16-GFP and RFP-SKL. The distribution of GFP and/or RFP fluorescence in these cells was then examined 48 h after transfection (). In cells transfected with RFP-SKL alone, the peroxisomal reporter was distributed diffusely in the cytosol and no fluorescence at 488 nm was detected, indicating that peroxisomes were indeed absent in these cells (). In contrast, in cells cotransfected either with PEX16-GFP and RFP-SKL () or with saPEX16-GFP and RFP-SKL (), numerous punctate and globular peroxisomal structures containing both sets of expressed proteins were observed (, arrows). Because the signal anchor sequence on saPEX16-GFP forced it to be inserted into the ER membrane before delivery to other membranes, these results indicated that a pathway from the ER involving PEX16 was sufficient to support peroxisome production de novo. Interestingly, some of the punctate structures in the cells coexpressing the GFP-tagged PEX16 and RFP-SKL molecules contained only PEX16- or saPEX16-GFP. These structures may represent so-called early or nascent peroxisomes (; ) that have not yet begun importing lumenal peroxisomal proteins after complementation. There are two ways in which wild-type nascent PEX16 can target to the ER: by posttranslational targeting, in which PEX16 is synthesized on free ribosomes in the cytoplasm and then is posttranslationally targeted to and inserted into ER membranes, or by cotranslational targeting, in which ribosomes containing NH-terminal PEX16-nascent chains are first targeted to the ER and then PEX16 is cotranslationally inserted into ER membranes. To distinguish between these two possibilities, we used an in vitro binding assay in which PEX16 was synthesized using a rabbit reticulocyte lysate in the presences of S-methionine. To assay for cotranslational targeting, the translation reaction was performed in the presence of ER microsomes. To assay for posttranslational targeting, ER microsomes were added to the reaction after the protein was first fully translated and further protein translation was inhibited by the addition of cycloheximide. Preprolactin (PPL) and cytochrome b-glyc (Cb-glyc; cytochrome b with a glycosylation site in its luminal domain) were used as appropriate cotranslation and posttranslation controls, respectively (; ). As shown in , PEX16 pelleted more readily with ER microsomes during cotranslational targeting (39%) than during posttranslational targeting (18%) or during translation without microsomes (10%). These results were similar to those observed for the cotranslational targeting of PPL; i.e., significantly more PPL pelleted with microsomes with its signal sequence cleaved during co-T2 targeting compared to post-T2 targeting ( [PPL, compare lane 3 with 6 and 9] and B). In contrast, Cb-glyc exhibited binding to ER microsomes both when microsomes were added during the translation reaction and when microsomes were added after translation (, Cb-glyc), as expected for this posttranslationally targeted protein (). Furthermore, the actual integration of Cb-glyc into ER microsomes was demonstrated by its glycosylation, which shifted it to a higher molecular mass in the pellet fraction (, lanes 3 and 6). The increase of PEX16 in the pellet fraction during posttranslational targeting experiments compared with minus membrane experiments () may be due to nonspecific interactions of PEX16's hydrophobic transmembrane domains with microsomes rather than to PEX16's actual integration into the lipid bilayer of the ER. To determine whether this was the case, we engineered a glycosylation site at the putative luminal domain of PEX16 (PEX16-glyc; N-X-S starting at residue 161; ). We reasoned that if any PEX16-glyc molecules in the assay were glycosylated, they must have been specifically integrated into ER microsomes. As shown in , a significant proportion of PEX16-glyc molecules underwent glycosylation during cotranslational targeting, whereas none did so during posttranslational targeting. The data thus confirmed that PEX16 undergoes cotranslational insertion into the ER under in vitro conditions. It was previously hypothesized that PEX16 acts as part of the machinery involved in recruiting PMPs to membranes (; ). If so, overexpressed PEX16 that is localized to the ER should cause other PMPs, such as PEX3 and PMP34, to retarget to ER membranes. To test this prediction, we constructed chimeras of PEX3 and PMP34 fused to the GFP or Cerulean blue fluorescent protein () and examined their subcellular location in the presence or absence of overexpressed PEX16 fused to the Venus fluorescent protein (PEX16-Venus; ). Neither PEX3- nor PMP34-GFP was targeted to the ER when expressed in cells lacking peroxisomes (Fig. S4, available at ) or in cells containing peroxisomes (). Indeed, PEX3-GFP was colocalized exclusively with RFP-SKL in peroxisomes at low expression levels () and accumulated in mitochondria at higher expression levels (). PMP34-Cerulean was also localized to peroxisomes at low expression levels () but accumulated in the cytoplasm at higher expression levels (). Importantly, when PEX16-Venus was coexpressed with either PEX3- or PMP34-Cerulean, both PMPs colocalized with PEX16-Venus in the ER (). When a small area of the ER was repeatedly photobleached to remove ER fluorescence in cells coexpressing the PMPs and PEX16-Venus, both PMPs were observed in peroxisomes (unpublished data). Similar colocalizations of PEX16-Venus and PEX3-Cerulean in the ER were observed in the PBD399-T1 fibroblast cells lacking peroxisomes (Fig. S4), indicating that the recruitment of PEX3-Cerulean by PEX16-Venus to the ER also occurred in other cells. Hence, PEX16 appears to function in the recruitment of PEX3 and PMP34, and possibly other PMPs, to membranes. Recent studies in cells have suggested that de novo biogenesis of peroxisomes occurs by direct outgrowth of peroxisomal structures from the ER (; ; ). To test whether a similar mechanism occurs in mammalian cells, we devised an assay to directly visualize the formation of peroxisomes in wild-type mammalian cells (). The assay involved two photoactivation events separated by 24 h that permitted old and new peroxisomal components to be differentiated in living cells. We reasoned that daughter peroxisomes formed by division of preexisting peroxisomes should all contain peroxisomal components from their mother peroxisomes. However, peroxisomes formed by a de novo pathway from the ER should contain only newly synthesized components. Therefore, by distinguishing recently synthesized peroxisomal components from older peroxisomal components, we could distinguish peroxisomes formed de novo from those formed by fission. NRK cells were transiently transfected with PAGFP fused to SKL (PAGFP-SKL) to mark preexisting peroxisomes and with PEX16-Cerulean to monitor peroxisomes before (pre-PA) and after (post-1st PA t = 0 h) photoactivation of PAGFP-SKL (). All PAGFP-SKL molecules in the peroxisomes of the cell were initially photoactivated using 413-nm laser light, and an image was collected (post-1st PA t = 0 h). The cell was then incubated at 37°C for 24 h to allow newly synthesized, nonphotoactivated PAGFP-SKL molecules to accumulate. The cell was then fixed to prevent peroxisomes from moving, and another image was collected (post-1st PA t = 24 h). Immediately thereafter, the cell was photoactivated a second time (re-PA) to highlight all newly synthesized (previously “invisible”) PAGFP-SKL molecules, and a final image was collected (post-2nd PA). The image acquired after the second PA (post-2nd PA) was compared with the previous image (i.e., post-1st PA t = 24 h) to determine whether the PAGFP-SKL molecules that were synthesized during the 24 h were only localized to previously fluorescent peroxisomes or if they were localized to both previously fluorescent peroxisomes and nonfluorescent nascent peroxisomes within in the cell. Before photoactivation, no fluorescence attributable to PAGFP-SKL was observed in a cell coexpressing PAGFP-SKL and PEX16-Cerulean (, pre-PA; and not depicted). However, upon photoactivation, punctate peroxisomes containing PAGFP-SKL became brightly fluorescent (, post-1st PA t = 0 h). Imaging of the same cell 24 h later revealed a minimal loss of fluorescent signal in the cell (, post-1st PA t = 24 h). Indeed, quantification of the PAGFP-SKL fluorescence in different cells throughout the 24-h period revealed that the level of fluorescence remained virtually constant (). This suggested that PAGFP-SKL proteins were long-lived and that repeated low-light imaging did not lead to their fluorescence being photobleached. The number of fluorescent peroxisomes measured at the start and end of this imaging period only slightly increased (∼8%), presumably because of peroxisomes undergoing fission (, fission). After the second photoactivation, the fluorescence associated with already fluorescent peroxisomes increased significantly (, compare post-1st PA t = 24 h and post-2nd PA). Once formed, therefore, peroxisomes continue to import newly synthesized PAGFP-SKL. Notably, fluorescence also appeared in peroxisomes that were not previously fluorescent (i.e., peroxisomes that did not contain any PAGFP-SKL molecules highlighted during the first photoactivation step; [post-2nd PA] and C [higher magnification, arrows]). These “new” peroxisomes were not the result of preexisting peroxisomes that had lost their fluorescent signal by either degradation or photobleaching after the initial photoactivation, as the total fluorescent signal from photoactivated PAGFP-SKL did not diminish over the 24-h chase period (). Rather, they appeared to represent peroxisomes formed de novo during the 24-h chase period. Quantification of the number of newly appearing peroxisomes after the second photoactivation event was determined by calculating the difference in the number of peroxisomes before and after the second photoactivation. A mean of 30 ± 10 new peroxisomes per cell was found based on results from eight independent experiments (). This represented an ∼20% increase in the total number of peroxisomes in the cell over the 24-h period and was much greater than the slight increase in peroxisomes observed after the first photoactivation based on a test ( = 2.42; P < 0.05). Therefore, a de novo pathway appeared to play a significant role in the formation of peroxisomes in these cells. In this study, we provide evidence based on live cell imaging approaches that peroxisomes in mammalian cells can arise de novo from the ER. Using PAGFP and in cellula pulse-chase analysis of PAGFP-labeled PEX16 (an early event peroxin), we demonstrate that PEX16 traffics normally from the ER to peroxisomes and that peroxisomes arise from the ER in wild-type cells and not only in mutant cells lacking peroxisomes. We further show that PEX16 is incorporated into ER-derived microsomes cotranslationally and that PEX16 is capable of recruiting other PMPs to membranes. Finally, by visualizing the production of peroxisomes within a cell over time using PAGFP, we show that most new peroxisomes are derived de novo, with possibly only a minor fraction arising from fission of preexisting organelles. Our findings complement the growing body of evidence from yeast cells, including and , and plant cells indicating that peroxisomes originate from the ER and are not semiautonomous organelles like mitochondria or chloroplasts (; ; ; ; ; ). In addition to solidifying this new concept, our results help clarify the role of PEX16 in promoting peroxisome outgrowth from the ER and introduce a new pulse-chase assay using PAGFP in living cells for examining peroxisome biogenesis under various physiological conditions. COS-7 and NRK cells were obtained from the American Type Culture Collection. The immortalized PEX19-deficient human skin fibroblast cell line PBD399-T1 and the PEX16-deficient human skin fibroblast cell line GM06231 were gifts from S.J. Gould (John Hopkins University School of Medicine, Baltimore, MD) and P.A. Walton (University of Western Ontario, London, Canada), respectively. All cells were cultured in DME (Biosource International) supplemented with 10% heat-inactivated fetal bovine serum (Biosource International) and 2 mM glutamine (Invitrogen) at 37°C in a humidified atmosphere containing 5% CO. The construction of pPEX16-GFP was described previously (). pPEX16-PAGFP and -Venus were generated by excising the ORF from pPEX16-GFP with BglII and SalI and ligating the resulting fragment into the same sites in either pmPAGFP-N1 () or pVenus-N1 (). pdelPEX16-GFP encoding the ORF lacking its mPTS (residues 66–81; -RKELRKKLPVSLSQQK-; ) was generated using QuikChange PCR site-directed mutagenesis (Stratagene) with pPEX16-GFP as template DNA and the complementary synthetic oligonucleotides Fp 5′-GCCGCTCAATGACGGGATCCTAAAGCTGCTGACATGGCTGAGCG-3′ and Rp 5′-CGCTCAGCCATGTCAGCAGCTTTAGGATCCCGTCATTGAGCGGC-3′. The primers Fp 5′-GTGGTACCATGGAGAAGCTGCGG-3′ and Rp 5′-GTCGACTCAGCCCCAACTGTAG-3′ were used to amplify the ORF, which was then ligated into NcoI–SalI–digested pSPUTK () to yield psUTK-PEX16. psUTK-PEX16-glyc was constructed by site-directed mutagenesis of two sites in the ORF using the primers Fp 5′-CACAGCCCTGGCAACCGCAGTCCTACGTGGGG-3′ and Rp 5′-CCCCACGTAGGACTGCGGTTGCCAGGGCTGTG-3′. The bolded nucleotides represent mutations that resulted in two amino acid substitutions, H162R and E163S, as well as the introduction of an N-linked glycosylation consensus site (-N-X-S-) at residues 161–163, along with a novel PvuI restriction site for the convenient assessment of mutant clones. pSA1 was a gift from R.S Hedge. psaPEX16-GFP was constructed by amplifying the sequence encoding amino acid residues 14–90 in SA1 from pSA1 with the primers Fp 5′-GATGCTAGCGATGGCAATATGTTTGCCCTG-3′ and Rp 5′-GTCAGCTCTCTGGTAGATAGGATC-3′. The resulting PCR products were then digested with NheI and BglII and ligated into NheI–BglII–digested pPEX16-GFP. pPMP34 was a gift from R.J.A. Wanders (University of Amsterdam, Amsterdam, Netherlands; ). The ORF was amplified from pPMP34 with the primers Fp 5′-GCTGAATTCCACCATGGCTTCCGTGCTGTCCT-3′ and Rp 5′-TTCGGATCCCGGTGTTGGTGTGCACGCTTCAGCC-3′, PCR products were digested with EcoRI and BamHI and ligated into equivalent sites within either pmGFP-N1 or pmCeruleanBlue-N1 to yield pPMP34-GFP and -Cerulean, respectively. The pPEX3-GFP and -Cerulean were generated by amplifying the ORF in pCMV-SPORT-PEX3 (MGC-9125; American Type Culture Collection) using the PCR and oligonucleotides Fp 5′-GAAGATCTGCCACCATGCTGAGGTCTGTATGGAATT-3′ and Rp 5′-AAAAGTCGACTTCTCCAGTTGCTGAGGGGTAC-3′. The resulting products were digested with BglII and SalI and then ligated into equivalent sites in either pmGFP-N1 and pCeruleanBlue-N1. pmRFP- and pPAGFP-SKL were generated using the oligonucleotides Fp 5′-CGGGATCCACCGGTCGCCACCATG-3′ and Rp 5′-CAGCGGCCGCTTAAAGCTTGGAGGCGCCGGTGGAGTGGCG-3′ and either pmRFP-N1 or pmPAGFP-N1 as template DNA. The resulting products were digested with BamHI and NotI and then ligated into pmRFP-N1. All plasmids were confirmed by automated sequencing. The plasmid pssRFP-KDEL (encoding for an RFP molecule containing a signal sequence of PPL to target it to the ER lumen and a KDEL sequence to retain it there) was a gift from E.L. Snapp (Albert Einstein College of Medicine, Bronx, NY). In preparation for transient transfections, cells were grown in four-well, Lab-Tek chambered coverglasses to 70–80% confluency. Transient transfections of COS-7 and NRK cells were performed using FuGENE 6 according to the manufacturer's instructions (Roche Molecular Biochemicals). The Lipofectamine 2000 transfection reagent (Invitrogen) was used for PBD399-T1 and G062351 cells as described by the manufacturer. All transfected cells were incubated at 37°C in 5% CO for 8–24 h before microscopic imaging, unless otherwise noted. For photo/pulse-chase experiments () photoactivations of PAGFP were performed using a 413-nm laser (Coherent Enterprise II) at full power in a rectangular ROI. The ROI was selected in an area of the transformed cell that was free of peroxisomes as indicated by RFP-SKL. Photoactivations were repeated once every minute for a period of 30 min. The de novo peroxiosme biogenesis assay () was performed on NRK cells transfected with pPAGFP-SKL and pPEX16-Cerulean 12–15 h before photoactivation. The entire cell expressing PEX16-Cerulean was photoactivated using two iterations of a 413-nm laser line at full power in a rectangular ROI. Three z section images of 1 μm thickness were collected immediately before and after the photoactivation. 24 h after photoactivation, cells were fixed with 3.7% formaldehyde in phosphate-buffered saline for 15 min at room temperature before another round of photoactivation. Again, three z section images of the cell were collected immediately before and after the second photoactivation. Z sections were complied into one projection using the computer software Lview (Carl Zeiss MicroImaging, Inc.). ImageJ software (NIH) was used to calculate the number of peroxisomes in each cell. In brief, images were converted into threshold images that were then analyzed with the ImageJ “analyze particle” macro. In vitro transcription and translation reactions and the membrane binding assay were performed as described previously (). Cotranslational targeting reactions were performed in the presence of 1 equivalence of ER microsomes (1 equivalence = 1 fmol of signal recognition particle receptor α-subunit (). Posttranslational reactions were stopped by adding cycloheximide to a final concentration of 20 μg/ml before adding 1 equivalence of ER microsomes. 1 h after incubation at 24°C, ER microsomes were separated from both cotranslational and posttranslational reactions by subjecting each reaction to centrifugation over a 0.5-M sucrose cushion. Equal amounts of each fraction were analyzed by SDS-PAGE using a Tris-Tricine buffer system (), and radioactive proteins were visualized and quantified using a phosphorimager (Typhoon; GE Healthcare). Fig. S1 shows the relative expression of GFP-SKL, PEX16-GFP, and saPEX16-GFP over time. Fig. S2 shows the cellular localization of delPEX16-GFP before and after FLIP. Fig. S3 the differential cell permeabilization assay of cells expressing ssGFP-KDEL, PEX16-GFP, and saPEX16-GFP. Fig. S4 shows cellular localization of PEX3- and PMP34-GFP and the colocalization of PEX3-Cerulean with PEX16-Venus in PBD399-T1 cells. Online supplemental material is available at .

The product of the p53 gene is capable of inducing transcription activation or repression of a variety of genes, which in turn trigger cell cycle arrest and promote apoptosis, differentiation, or senescence (). p53 levels and activity are regulated by numerous stress-induced posttranslational modifications that converge on two distinct domains of the protein (). A kinase cascade phosphorylates several serine residues in the p53 NH-terminal region. Among these, phosphorylation of serine 15, 20, and 46 and threonine 18 are of most significance, as they act by promoting p53 stabilization (; ), by preventing nuclear export (), or by favoring p53 recruitment on specific sets of promoters (). A second pathway targets the p53 COOH terminus and involves the activity of at least two classes of acetyltransferases, p300/CBP and p300/CBP-associated factor (PCAF; ; ). p300/CBP acetylates lysines located in the last COOH-terminal portion of p53, specifically, lysines 370, 372, 373, and 382 (). PCAF has been linked to acetylation of a single residue, lysine 320, located within a flexible linker domain nearby the oligomerization domain, which also contains a nuclear localization signal (; ). These initial studies demonstrated that all these acetylation events lead to enhancement of p53 DNA binding activity in vitro. More recently, acetylation has been shown to stimulate the p53–DNA interaction in vivo as well (), in agreement with the original proposal that acetylation relieves the inhibitory role exerted by the COOH terminus on p53 DNA binding capacity (; ). However, thus far, the biological significance of individual acetylation of each of these lysine clusters is unclear. The proapoptotic activity is the most ancestral function of p53, but additional and more complex activities have clearly appeared during evolution. This is argued based on the properties of the p53 protein of , , which differs from its human counterpart because of its exclusive role in promoting apoptosis. Similarly, the homologue of p53, , displays predominantly proapoptotic activity (). The evolution of p53 toward cell cycle regulatory functions in complex multicellular organisms may reflect the need of cells to deal with various forms of stress in a more advantageous fashion and to mount adaptive responses that preserve the life of tissues with limited proliferative potential. In fact, in mammalian cells, the effects of p53 on cellular growth are pleiotropic and exhibit cell type specificities, and the p53-induced cell cycle arrest, although prolonged, is often reversible (). This reversibility is presumably important in conditions of “repairable” cellular damage, for resumption of proliferation when apoptosis can be avoided. Yet, the molecular basis for p53's ability to commit cells toward these different outcomes is still an object of intense investigation. Given the large repertoire of p53-responsive genes, one possibility is that a particular combination of target genes, activated or repressed, may determine whether cells will survive or die after engagement of p53 activity. Because cells are continuously exposed to genotoxic signals of different nature and intensity, they must have elaborated ways that allow p53 to interact with specific classes of genes but not with others, ultimately leading to adaptation and differential susceptibility to stress. How, then, does p53 select such combination patterns? Important determinants of selectivity are likely the accessibility of chromatin and the structural characteristics of the DNA consensus sequences within p53-responsive elements, which show a surprisingly high degree of variation (; ). When several p53-response elements were analyzed for their ability to be activated in a yeast-based assay that eliminates the influence of chromatin on transcription, an unexpected 1,000-fold difference in p53 transactivation ability was detected depending on the central sequence present within each response element. In addition, tumor-derived p53 mutants that have altered conformation compared with the wild-type (WT) protein exhibit distinct promoter specificities and retain their ability to transactivate certain promoters but not others. Thus, the intrinsic DNA binding affinity of p53, together with conformational changes, may contribute to differential gene activation. Posttranslational modifications may play an important role in modifying both p53 conformation and p53 affinity for its downstream targets. In this study, we have characterized the functional and biochemical properties of p53 mutants mimicking constitutive acetylation as well as of truly acetylated p53. We show that two distinct acetylable clusters at position 373 (K373) and 320 (K320) regulate p53 activity in quite a different fashion. A model is proposed to explain how site-specific acetylation operates and cross-talks with other events, such as additional posttranslational modifications and protein–protein interactions, to modify p53 affinity for different classes of genes during stress signals, leading to cell survival or death. We began our studies by asking how acetylation of different sites is regulated during stress signals. For these experiments we used a p53-null, human lung carcinoma cell line, H1299, where the expression of p53 was reconstituted via a tetracycline-inducible vector. The kinetics of acetylation of K320 and K373 were then studied in cells exposed to the radiomimetic DNA-alkylating agents adozelesin and bizelesin or to the topoisomerase inhibitor etoposide. Adozelesin and bizelesin are members of the cyclopropylpyrroloindole (CPI) family of DNA minor groove alkylating agents endowed with anti-tumor activity (). Adozelesin is a monofunctional CPI, whereas bizelesin is a CPI dimer that can alkylate adenines on one or both DNA strands, forming double-stranded DNA cross-links that render it more cytotoxic than adozelesin. Thus, these drugs allowed us to assess whether different types of DNA damage trigger distinct acetylation events. In cells treated with adozelesin, which arrested predominantly in G1 and did not undergo apoptosis (), acetylation of K373 was no longer visible after 4–12 h of treatment, whereas acetylation of K320 remained detectable for 24 h after the addition of the drug (). In contrast, treatment with bizelesin or etoposide resulted in robust and sustained levels of acetylation of K373 and induced an initial arrest in G2/M followed by apoptosis ( and not depicted). These results provided evidence that the kinetics of K320 and K373 acetylation can be dissected depending on the type of damage. To test whether the amount of DNA damage also influences the extent and sites of acetylation, we treated cells expressing p53 with increasing doses of adozelesin or bizelesin (). In this case, we observed a dose-dependent enrichment of the levels of K373 acetylation, whereas acetylation of K320 was only modestly enhanced. Thus, various acetylation sites act as a “sensor” system for the type and extent of DNA damage and may differentially regulate the ability of p53 to induce cell cycle arrest or apoptosis. To gain more insights into the function of acetylation, we created gain-of-function p53 acetylation mutants. Glutamine in place of lysine has been shown to mimic the effects of a constitutive acetylation in the case of histones and of p53 (), probably because glutamine is a neutral amino acid, which, like -acetyl-lysine, has an amide group that can function as a hydrogen bond donor or acceptor. Thus, we constructed mutants harboring lysine to glutamine substitutions at position 320 (Q320), a triple mutant at position 370/372/373 (Q373), or at all these positions (DM). Native p53 and its derivative acetylation mutants were expressed in the p53-null cell line, H1299, via a tetracycline-regulated promoter (see and Fig. S1 C, available at , for typical expression levels of these p53 proteins). Cells were treated with different DNA-damaging agents, and their cell cycle distribution was assessed (). In untreated cells, expression of native p53 produced an arrest, especially at the G1 phase, accompanied by a reduction of cells transiting throughout S phase, as expected. The cell cycle profiles of cells harboring p53 acetylation mutants were different in several ways: a higher percentage of cells expressing p53DM arrested in G1, whereas cells harboring p53Q373 arrested markedly in G2/M, suggesting that these residues predominantly influence the activity of p53 on the G2 checkpoint. To rule out the possibility that these differences may be due to unique clonal characteristics of these p53 cell lines, similar experiments were performed by using either additional individual clones or polyclonal mixtures. The analysis of the growth characteristics and viability of these cells produced substantially similar results (Fig. S1, A and B). Results presented thus far are consistent with the possibility that acetylation of residues around position 373 activates the apoptotic pathway, whereas acetylation of K320 suppresses cell death. We then hypothesized that acetylation of K320 could function to disable or delay the apoptotic program elicited by p53, thus allowing cells to resume proliferation in conditions of moderate DNA damage once p53 signaling is extinguished. To test this, cells expressing native p53 or p53 acetylation mutants were pulsed with low concentrations of adozelesin for a short period of time in the presence of tetracycline. p53 expression was shut off by removing tetracycline from the media 24 h after drug treatment, and cell growth was monitored for several days thereafter (). In these conditions, untreated and drug-treated cells expressing p53Q373 displayed a significant loss of viability, whereas cells expressing p53Q320 were able to resume proliferation. In addition, cells expressing p53DM again displayed a phenotype intermediate between that exhibited by p53Q320 and -Q373. These results argue against a simple model in which the number of acetylated lysines controls p53-apoptotic activity in a dose-dependent manner. Rather, they suggest that the effects of acetylation are strictly position specific and place acetylation of K320 as a central event in favor of cell survival. Posttranslational modifications regulate p53's interaction with other cellular proteins. We next asked whether acetylation influences the association with factors known to bind p53, particularly to coactivators and corepressors. p53 protein complexes isolated from native or p53 acetylation mutant–expressing cells were subjected to immunoblot analysis with antibodies recognizing p300, PCAF, HDAC1, mSin3, and p53 itself. HDAC1 and mSin3 are components of a corepressor complex, which plays an important role in the ability of p53 to silence transcription of critical antiapoptotic promoters (). As shown in , the p53Q373 mutant coprecipitated a significantly higher amount of p300 compared with native p53 and p53Q320 and, conversely, p53Q320 was more efficient in interacting with PCAF, indicating that each acetylation cluster specifically modifies the interaction of p53 with distinct types of histone acetyltransferases. Further, p53Q373 coprecipitated higher amounts of HDAC1 than native p53 or p53Q320 (). To further substantiate these findings, we studied the binding pattern of endogenously, “truly” acetylated p53 after treatment with etoposide. Consistent with data obtained with the acetylation mimics, acetylated p53 detected with the anti–acetyl-K373–specific antibody coprecipitated a significant amount of HDAC1 and SIRT1 (). Interestingly, these deacetylase–p53 complexes were detectable for at least 12 h after treatment, indicating that acetylation of K373 can stabilize the interaction of p53 with deacetylase–corepressor complexes. Similarly, in A549 cells treated with adozelesin, truly acetylated p53 at K373 displayed a significantly stronger affinity for p300 compared with acetyl-K320 p53 (, compare lanes 1 and 2). These data further support the notion that glutamine in place of lysine mimics the effects of constitutive acetylation and imply that acetylation of different sites selectively modifies the affinity of p53 for different types of chromatin-remodeling enzymes. p53 conveys complex cellular responses that are due in part to its ability to function as a transactivator and as a transrepressor. To understand how acetylation operates, we performed microarray analysis on cells expressing native p53, p53Q320, and p53Q373 (Table S1, available at ). Approximately 40,000 probe sets were examined by using high-density oligonucleotides arrays at 12 h after addition of tetracycline to the media. For this analysis, we applied a present call noise filter (one in six arrays), fold-change thresholds (>2), and a p-value of <0.05 as our initial dataset for preliminary analysis and gene identification, leading to further consideration of ∼27,000 transcripts. To more specifically identify genes regulated in an acetylation-dependent manner, genes found in the array platform derived from cells expressing native p53 served as background, so that p53Q373 or -Q320 gene-expression patterns were directly compared with native p53. With this analysis, we determined that both p53Q320 and -Q373 lead to an increase of a similar number of transcripts, whereas p53Q320 repressed 955 genes, which was significantly less than the 1,576 transcripts repressed by p53Q373. Thus, p53Q373 is a stronger repressor than p53Q320, consistent with its ability to interact more efficiently with HDAC1 and SIRT1/mSir2 (). Further, in agreement with the observation that p53Q373 sensitizes cells to apoptosis, many of the genes found activated by this mutant promote cell death (i.e., APAF1, caspase 6, pig3, pig11, AMID, PCBP4, and IGFBP3), and a significant number of those repressed promote survival (i.e., survivin, API5, BIRC3, and IL31RA). The behavior of p53Q320 was, instead, opposite. For example, several of the proapoptotic genes activated by p53Q373 were repressed by p53Q320 (i.e., APAF1 and pig11) relatively to native p53 and, vice versa, some antiapoptotic genes repressed by p53Q373 were activated by p53Q320 (i.e., survivin, TRAF2, AATF, and BIRC4). Numerous other genes were also conversely expressed between p53Q320 and -Q373 (e.g., cyclin B1; Table S1, bold). To gain a further unbiased analysis of this phenomenon, a “heat map” was generated, which demonstrated differential, if not contrasting, gene-expression patterns between p53Q320 and -Q373 (). To validate the gene changes identified with the microarray platform, semiquantitative PCR was also performed on various relevant apoptotic- and growth arrest–related genes, some of which are shown in . This approach confirmed that the regulation of previously known p53 targets, such as pig11 and APAF1, or of new p53-regulated genes like Jagged2, was significantly different depending on the site of the mutation. Although we cannot completely exclude the possibility that at least some of these changes are an epiphenomenon, reflecting events secondary to p53 activation, many of the relevant genes differentially modulated by p53Q320 and -Q373 are well-known primary targets of p53, for example, APAF1, pig11, survivin, caspase 6, mdm2, and cyclin B1. In addition, a large number of newly identified genes that we found up-regulated in the microarray platform contain p53 binding elements (), thus suggesting that they might be direct p53 targets. In addition, this approach indicated that p53Q373, unlike p53Q320, predominantly controls the expression of genes that regulate apoptosis. Data based on the microarray analysis suggested that acetylation of K320 or K373 may lead to direct p53 binding onto different types of promoters. However, previous studies showed that simultaneous acetylation of multiple COOH-terminal lysines enhances the DNA binding activity of p53 on typical DNA binding elements (; ). In light of our results, we decided to investigate how each of the acetylation clusters individually influences p53–promoter interactions. The DNA binding properties of acetylation mutants and of truly acetylated p53 were studied in vivo in chromatin immunoprecipitation (ChIP) assays. To assess the influence of chromatin on the ability of p53 to interact with its DNA consensus sites, we also performed electrophoretic mobility shift assays (EMSAs) with in vitro–purified p53 (Fig. S2, available at ). A first observation arising from these experiments was that native p53 bound to the endogenous promoter of p21/WAF with significantly higher affinity compared with the PIG3, BAX, or p53AIP1 promoters (). Results obtained with EMSAs essentially recapitulated those seen with the ChIP assays and indicated that p53 possesses substantially different intrinsic affinity for its DNA binding elements. Noticeably, both p53Q320 () and its truly acetylated counterpart immunoprecipitated with the anti–acetyl-320 antibody (), bound to the high-affinity p21/WAF promoter more efficiently than native p53 or p53Q373. Conversely, p53Q373, as well as endogenously acetylated p53 at position K373, interacted better with the low-affinity proapoptotic promoters of BAX (), PIG3, and p53AIP1 (), compared with other p53 proteins. It is important to note that both the EMSAs and ChIP assays were performed in conditions in which p53 levels were equalized (). Thus, these differences reflect acetylation-mediated changes in the intrinsic p53's DNA binding affinity. Furthermore, the different promoter binding pattern of p53 acetylation mutants resulted in coherent changes in the expression levels of corresponding encoded products, as demonstrated by the different protein levels of Bax and p21/WAF in these cell lines (). Based on these data, we infer that acetylation of K373 enhances the interaction of p53, particularly with low-affinity proapoptotic promoters. The aforementioned results offer a plausible molecular basis for the proapoptotic activity exhibited by p53Q373; however, they do not completely explain the biological effects of the mutant at position 320. Because K320 is located in a previously identified nuclear localization signal (), we suspected that acetylation of this residue might affect the subcellular localization of p53. In fact, as shown in , the p53 acetylation mimic at position K320 was localized predominantly in the cytoplasm, whereas p53Q373 was almost exclusively nuclear (). The presence of the mutation at position 320 also partially placed p53Q373 off the nucleus, as argued by the cytoplasmic localization of p53DM. In addition, when lysines were replaced with arginine rather than with glutamine, both R320 and R373 mutants localized in the nucleus (), indicating that neutralization of the positive charge of lysine interferes with nuclear accumulation of p53. Surprisingly, however, the increased cytoplasmic concentration of p53Q320 and -DM is not sustained by an increase in nuclear import but rather by an accelerated nuclear export. This is argued because treatment with leptomycin B, which blocks the activity of the exportin protein Crm1 (), resulted in net nuclear accumulation of both mutants (). To further validate these results, two additional sets of experiments were performed. First, we studied the subcellular distribution of truly acetylated p53 in H1299 cells treated with adozelesin, which enhances the levels of K320 acetylation (). Treatment with adozelesin increased the cytoplasmic fraction of p53, and such fraction reacted with the anti–acetyl-320 antibody. Second, we determined the localization of p53 after overexpression of PCAF or of p300, which acetylate K320 and K370/372/373, respectively. Overexpression of PCAF resulted in increased cytoplasmic levels of p53 (), in contrast to overexpression of p300, where p53 remained substantially nuclear. The observation that K320 acetylation might place p53 off the nucleus by accelerating nuclear export led us to investigate the possible cross-talk between acetylation and phosphorylation of serine 15 (S15), which has been implicated in nuclear accumulation (). As shown in , S15 phosphorylation was almost undetectable in cells expressing p53Q320, whereas the p53Q373 mutant was hyperphosphorylated at this residue compared with other p53 proteins, suggesting that acetylation and phosphorylation might be functionally linked. To assess whether other specific combination patterns of acetylation and phosphorylation do exist, we examined the phosphorylation state of S392 and S46. S392 phosphorylation increases the stability and the nuclear localization of p53 (), whereas phosphorylation of S46 is specifically involved in mediating the interaction of p53 with the promoter of the proapoptotic gene p53AIP1 (). We found that p53Q373, but not p53Q320, was hyperphosphorylated at both sites. As a whole, our results imply that individual acetylation of lysines 320 and 373 of p53 has general, yet differential, effects on the intrinsic DNA binding activity and on the ability to interact with cellular proteins and influences other key posttranslational modifications, such as phosphorylation. We were interested in understanding, at a mechanistic level, how these acetylation events may affect so many important p53 functions. We hypothesized that modifications of K320 or K373 induce conformational changes that in turn influence multiple protein–protein interactions. To test this, we assayed the ability of p53Q320 and -Q373 to interact with antibodies recognizing epitopes located within spatially distant portions of p53 (). We used the p53 antibody DO-1, which binds to amino acids 20–25 in the NH terminus (B-1); the PAb421 antibody, which recognizes a COOH-terminal epitope located between amino acids 372–381 (B-2); and the PAb240 antibody, a typical conformational antibody that recognizes residues 213–217 within the DNA binding domain (B-3). This latter epitope becomes exposed when p53 is in an inactive mutant conformation or when WT p53 is bound to DNA (). To assess the accessibility of these epitopes in the absence of other changes, native or acetylation mutant forms of p53 were purified in vitro to near homogeneity from insect cells infected with recombinant baculoviruses, and each protein was immunoprecipitated with limiting or saturating amounts of antibodies. The purity of these preparations is shown in Fig. S3 (available at ). The results of these experiments were striking. Indeed, all three epitopes were more accessible by their specific antibody in the case of p53Q320. The increased accessibility of the PAb240 epitope can particularly explain the lesser binding of this mutant to low-affinity promoters, as exposure of these residues indicates a partially denatured conformation. In contrast, p53Q373 completely disrupts reactivity with the PAb421, probably because the PAb421 epitope spans within this region. Importantly, no significant differences were observed when p53 proteins were immunoprecipitated with a goat polyclonal antibody (N-19; B-4) or were subjected to direct immunoblot after p53 had been denatured in SDS-PAGE (C-2). Thus, in vitro, mutations of K320 and K373 impart conformational changes that modify the accessibility of epitopes located within the NH terminus, the central portion, or the COOH terminus of p53. Likewise, in vivo acetylation-mediated changes in the availability of these sites might explain differential interactions with cellular factors, including coactivators and corepressors, or kinases and phosphatases that in turn influence transcription and phosphorylation, respectively. Our results are consistent with a model whereby acetylation of lysines 320 and 373 acts as a finely tuned sensor-effector system that enables p53 to coordinate gene-expression patterns in response to DNA damage (). We have demonstrated that various p53 acetylation mutants, as well as truly acetylated p53 (), possess different intrinsic affinities for their downstream promoters and that acetylation of residues around position K373 is necessary, particularly for binding to low-affinity proapoptotic promoters, such as BAX, both in vitro and in vivo ( and Fig. S2). It is important to note that although the expression levels of p53Q373 were consistently twofold higher than those of other p53 proteins ( and Fig. S1) in conditions in which p53 levels were predetermined and equalized, such as in EMSAs (Fig. S2), in ChIP assays (), and in the phosphorylation studies (), p53Q373 interacted more strongly with proapoptotic promoters and was hyperphosphorylated compared with WT p53 or p53Q320. Thus, these changes reflect qualitative differences in the behavior of this mutant. In light of these data, p53Q373-mediated activation of proapoptotic promoters might depend on a combination of effects, specifically, a direct enhancement of the intrinsic DNA binding affinity, hyperphosphorylation of strategic residues, and an increase in p53 nuclear retention that allows occupancy of “hard-to-reach” and low-affinity promoters. In contrast, we found that p53Q320 interacted efficiently with the high-affinity p21/WAF promoter but to a lesser extent with the promoters of BAX, p53AIP1, and PIG3. This correlated with a PAb240+ conformation, indicative of a partially denatured configuration, as well as with decreased phosphorylation of residues that favor nuclear retention of p53. We interpret these data to conclude that the effects of K320 acetylation rely, at least in part, on its accelerated nuclear export ( and ) and different conformation (), both of which could prevent p53 from reaching nuclear concentrations and a configuration necessary to efficiently saturate low- affinity promoters. We have further shown that K373 acetylation stabilizes the interaction of p53 with p300, HDAC1, and SIRT1/Sir2 (). These molecular features were linked to an increase in the total number of repressed genes by p53Q373 compared with p53Q320 and with qualitative differences in the pattern of activated genes (Table S1). Expression of p53Q320, which interacts with PCAF but has significantly less affinity for p300 (), appeared instead to correlate with a lower threshold of activation and with a distinct pattern of gene expression. In addition, an unbiased “heat map” of genes regulated by the two acetylation mutants clearly demonstrated differential, if not contrasting, gene-expression patterns between p53Q320 and -Q373 (). Collectively, our results imply that each of the various clusters of acetylation regulates p53 distribution in different cell compartments and chromatin domains and acts as an interaction platform for various types of chromatin-remodeling enzymes to activate specific sets of genes. The aforementioned differences in the biochemical properties of p53 acetylation mutants are physiologically relevant. Indeed, cells expressing p53Q320 were capable of promoting cell survival and resumption of proliferation in conditions of moderate DNA damage, unlike those harboring p53Q373, which were irreversibly committed toward apoptosis (). Thus, in view of the DNA binding properties of the corresponding mutants, it seems legitimate to speculate that acetylation of K320 enables cells to tune the expression of various p53 target genes in a way that enhances the ability of cells to survive if the damage does not accumulate to catastrophic levels. While this work was in progress, a report showed that p53 activates a cell cycle checkpoint that responds to glucose availability and promotes survival (). In keeping with our findings, these observations lead to the important conclusion that the cell cycle and death pathway are under the control of a subpopulations of p53 with very distinct biochemical properties. As noted previously, the p53 protein of () differs from its human counterpart because of its exclusive role in promoting apoptosis (; ). Accordingly, unlike mammalian cells, does not possess some of the pathways needed for recovery after p53 activation, as judged by the absence of genes encoding for key negative regulators of p53 (e.g., ). Significantly, our analysis of the degree of conservation of the different p53 acetylation clusters between mammalian and p53 () showed that possesses a COOH-terminal domain with homology to the stretch of amino acids containing acetylated residues around position 373 of human p53, as noted by others (), but no detectable K320 homology. Thus, regulation of p53 activity via K320 acetylation might have evolved in higher eukaryotes to suppress the apoptotic program and to allow recovery after damage. We postulate that this reflects the necessity of complex multicellular organisms to spare the life of tissues with limited proliferation potential. In strong support of this interpretation, we have now shown that K320 acetylation is particularly favored in neuronal cells after injury, where it regulates the expression of genes specifically involved in neuronal survival and regeneration (unpublished data). It is becoming clear that the COOH-terminal region of p53 is a target of numerous additional posttranslational modifications, such as phosphorylation, ubiquitination, sumoylation, and methylation (). The similarity between the posttranslational modifications that exist in the histone tail and in the p53 COOH-terminal domain is intriguing. In the case of histones, it has been proposed that different patterns of such modifications, acting alone or in combination on one or more tails, establish a “histone code” that is deciphered by other proteins to coordinate downstream events (). In this model, the recruitment of nucleosome-modifying enzymes and of effector molecules by such “histone cassettes” will execute the code by changing local chromatin structure. For p53, the challenge for further studies will be to understand whether specific combinations of posttranslational modifications do exist that modify and expand the ability of this protein to regulate gene expression in a manner similar to that proposed for the histone tail. Experiments presented here suggest that mutations of each of the two lysine clusters is associated at the very least with a different phosphorylation pattern in the NH-terminal region of p53, with a different ability to interact with coactivators and corepressors, and with a different gene-expression profile. Moreover, in conditions whereby the activity of each acetylation mutant was studied independently, such as in the H1299 cell lines, we were able to trace an effect of each acetylation cluster on apoptosis or on the G1 and G2 checkpoints (). This finding, together with biochemical studies on the phosphorylation state and interaction profile of p53 acetylation mutants, supports the idea that the biological activities of p53 may be dictated by multiple cassettes, each interacting with specific effectors and containing well-defined combination patterns of posttranslational modifications. This interpretation might explain why so many posttranslational modifications target p53 (). It is well known that certain organs, such as the central nervous system, kidney, heart, liver, and lung, are naturally radio resistant, and there is good evidence that p53 is an important determinant of tissue-specific radio sensitivity (). We have shown that in a lung tumor cell line, the extent of K320 or K373 acetylation is qualitatively and quantitatively influenced by the type and extent of DNA damage. It is also possible that the extent of acetylation of each of these sites is determined in a tissue-specific manner by the local availability of acetylases or deacetylases. Based on the fact that expression of p53Q320 allows for resumption of proliferation after DNA damage, we predict that strategies aimed at enhancing acetylation of K320 could foster the development of compounds that protect peripheral tissues from toxicity during the course of chemotherapy. H1299 and A549 lung carcinoma cells were grown in DME supplemented with 10% FCS and 2 mM -glutamine. The p53 mutants used for the tetracycline-inducible system were constructed by using a Pfu-polymerase–based, site-directed mutagenesis (Stratagene), followed by cloning of the amplified cDNAs into the pCDNA/TO4 vector (Invitrogen). The primers used for mutagenesis were as follows: p53Q320, 5′-CCCCAGCCAAAGCAGAAACCACTGGATGGAGAA and p53Q373, 5′-AGCCACCTGCAGTCCCAACAGGGTCAGTCTACC. p53DM was created with two consecutive rounds of site-directed mutagenesis. A Flag-encoding sequence was fused in frame to the NH terminus of p53 to facilitate immunodetection of the protein. Transfections were performed by using either Lipofectamine (Invitrogen) or FuGENE 6 (for coverslips; Roche). Full-length PCAF and HA-tagged p300 were expressed from pCDNA vectors. Cells were plated at 20–40% confluence 12–18 h before transfection. Cells were exposed to the transfection reagent for ∼16 h, washed twice with PBS, and refed with complete medium for an additional 24 h. Adozelesin and bizelesin were provided by T. Beerman (Rowell Park Cancer Institute, Buffalo, NY). ChIP assays were performed as described elsewhere (). The primers used for amplification of p53-responsive elements were as follows: p21, 5′-TCACCATTCCCCTACCCCATGCTGCTC and 3′-AAGTTTGCAACCATGCACTTGAATGTG; BAX, 5′-AGCTCATGCCTGTAATCCCAGCGCT and 3′-AAATAGCATGCTTCCAGGCAGGACGT; P53AIP1, 5′-AGCTGAGCTCAAATGCTGAC and 3′-CCAAGTTCTCTGCTTTC; and PIG3, 5′-CAGGACTGTCAGGAGGAGGCGAGTGATAAG and 3′-GTGCGATTCTAGCTCTCACTTCAAGGAGAG. p53 was purified from ∼10 dishes of SF21 cells infected with recombinant baculoviruses expressing native or acetylation mutant p53. Cells were harvested and lysed in extraction buffer (20 mM Hepes, pH 7.5, 0.5 M KCl, 0.4 mM EDTA, 0.2% NP-40, 10 mM β-mercaptoethanol, 0.1 mM PMSF, and 10 μg/ml pepstatin), incubated on ice for 20 min, and centrifuged at 14,000 for 30 min at 4°C. p53 was purified by using a anti-Flag immunoaffinity column (Sigma-Aldrich) in the presence of 0.5 M KCl, extensively washed, and eluted with the Flag peptide. EMSAs were performed in a 30 μl total volume that contained buffer A (5× 100 mM Hepes, pH 7.9, 125 mM KCl, 0.5 mM EDTA, 50% glycerol, and 10 mM MgCl2); buffer B (10× 10 mM spermidine, 40 mM DTT, 1.2% NP-40, and 2 mg/ml BSA); 10 ng of double-stranded poly(d[I-C]); 50 ng of labeled oligonucleotide; and 50, 75, or 150 ng of p53. Reactions were incubated with the probe at room temperature for 20–40 min and run on native 6% polyacrylamide gels, which were run at room temperature until the xylen-cyanol blue reached 6 cm from the bottom of the gel. The oligonucleotides used were as follows: p21/WAF, 5′-TCTGGCCATCAGGAACATGTCCCAACATGTTGAGCTCTGG and 3′CCAGAGCTCAACATGTTGGGACATGTTCCTGATGGCCAGA; Gadd45, 5′TCTGTGGTACAGAACATGTCTAAGCATGCTGGGGACTGCC and 3′-GGCAGTCCCCAGCATGCTTAGACATGTTCTGTACCACAGA; and Bax, 5′AATTCGGCTACCTCACAAGTTAGAGACAAGCCTGGGCGTGGGCTATATTGTAGCGAAT and 3′-ATTCGCTACAATATAGCCCACGCCCAGGCTTGTCTCTAACTTGTGAGGTAGCCGAATT. Preparation of cell extracts and immunoprecipitations were performed as previously described (). Antibodies used in this study were for p53 (FL393 and N-19 [Santa Cruz Biotechnology, Inc.] and Ab-1 and Ab-6 [Calbiochem]), S15-, S46-, and S392-phospho-p53 (Cell Signaling Technology), Acetyl-320-, Acetyl-373-, and Acetyl-373/382-p53 (Upstate Biotechnology), PCAF (E8; Santa Cruz Biotechnology, Inc.), p300 (N-15; Santa Cruz Biotechnology, Inc.), p21 (WAF1/Ab-5; Calbiochem), mSin3 (AK-11; Santa Cruz Biotechnology, Inc.), Sir2 (7342; Abcam), Bax (anti-Bax/NT; Upstate Biotechnology), and actin (I19; Santa Cruz Biotechnology, Inc.). Proteins were detected by using a chemiluminescence-based system (Pierce Chemical Co.) according to the manufacturer's instructions. The H1299 cells were plated on glass coverslips, fixed in 4% paraformaldehyde, and permeabilized by addition of 0.1% Triton X-100/PBS solution. Cells were stained using antibodies directed against p53 (FL393), PCAF (E-8), or p300 (N-15) and then stained for DNA content (DAPI; Invitrogen). The appropriate fluorescent secondary antibodies were obtained from Invitrogen. Where indicated, cells were treated with 5 ng/ml leptomycin B (Sigma-Aldrich) for 24 h. Cells were visualized using a microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) with either a 40× or 60× objective. Images were acquired using a camera (Axio; Carl Zeiss MicroImaging, Inc.). Analysis of different Z stacks was performed using the AxioVision 3.0 software (Carl Zeiss MicroImaging, Inc.). Expression profiling was performed as described previously and fulfilled all stringent quality-control measures as detailed previously (). We used two normalization processes: one for chip–chip comparisons (scaling factors) and one for gene–gene comparisons (normalization to the mean of the naive signal intensities for each gene). The scaling factor determinations were done using default algorithms (MAS 5.0; Affymetrix, Inc.) with a target intensity of chip sector fluorescence to 800. We have recently shown that the use of MAS 5.0 signal intensity values, together with a present call noise filter achieves an excellent signal/noise balance relative to other probe set analysis methods (dChip; robust multichip average [RMA]; ). Data analyses were limited to probe sets that showed one or more “present” (P “calls”) in the six GeneChip profiles in our complete dataset. Data were analyzed and visualized using the GeneSpring software (Silicon Genetics). Initial data analysis also included a fold-change filter of >2 increase or decrease relative to WT p53 (MAS 5.0). Functional classification was performed using DAVID software (). For hierarchy and clustering analysis, the arrays were analyzed using the R statistics package () and the Affymetrix, Inc. library () of the Bioconductor software package. Expression values were determined using the RMA algorithm using the RMA function in the Affymetrix, Inc. library at its default settings. Genes that exhibited an expression change >2 relative to the control (WT p53) for at least one of replicates and one of the mutations were selected for further analysis. These genes were then clustered using hierarchical clustering with “complete” agglomeration, and each cluster was further analyzed based on the known function of the genes contained in the cluster. Fig. S1 demonstrates that mixed populations of H1299–WT p53, p53Q320, p53Q373, or p53DM clones display similar cell cycle profiles and viability compared with a single clone population after p53 protein induction. In Fig. S2, EMSAs are used to study the DNA binding activity of p53 acetylation mimics in the absence of chromatin. Finally, Fig. S3 demonstrates that the p53 protein purified from baculovirus-infected incest cells has a high level of purity. Table S1 depicts apoptotic and cell cycle–related genes influenced in H1299 cells by the expression of the p53Q320 and -Q373 acetyl-mimics. Online supplemental material is available at .

Mitochondrial localization and transport ensure the proper inheritance of mitochondria upon cell division (; ) and position mitochondria where energy demands or oxygen supplies are greatest (). It is likely that the concentration of local cytoplasmic Ca also depends on mitochondrial Ca uptake (; ). Consequently, mitochondria accumulate in subcellular regions with high metabolic requirements and high Ca influx () and redistribute in response to changes in the local energy state (; ). The transport of mitochondria is particularly vital in neurons because of their extended processes, and the disruption of mitochondrial transport is correlated with neurodegenerative disease (). The mechanisms of mitochondrial transport differ between species and can require actin, microtubule attachment, or kinesins (; ; ). In metazoans, mitochondrial motility involves both actin- and microtubule-dependent mechanisms (; ; ; ). In particular, plus end–directed movement involves conventional kinesin (kinesin-1) motors (; ; ), although kinesin-3 motors are also implicated (; ). Little is known about how mitochondrial kinesin is regulated or coupled to the organelle (; ; ; ; ; ; ). We recently identified a novel protein called milton, which is required for mitochondrial transport within photoreceptors (). Mitochondria were absent from () photoreceptor axons, but were normally distributed and appeared to be functional in their cell bodies. Although devoid of mitochondria, their axons and synapses were otherwise surprisingly normal in their general architecture, possessing microtubules, synaptic vesicles, and active zone specializations. Thus, the transport defect was selective for mitochondria (; ). The mechanism of milton's action was unknown, but milton was associated with mitochondria and coimmunoprecipitated with kinesin heavy chain (KHC) in extracts of fly heads (). The mammalian homologues milton 1 and 2, which are also called -linked -acetylglucosamine–interacting protein 106 (OIP106) and γ-aminobutyric acid A receptor–interacting factor-1 (GRIF-1), also colocalize with mitochondria and coimmunoprecipitate with KIF5B, which is a mammalian homologue of KHC (; ; ; ). Therefore, we have suggested that milton acts as an adaptor or regulator of the mitochondrial anterograde motor. We demonstrate a protein apparatus that recruits kinesin to mitochondria and thereby permits anterograde movement. Milton, which interacts with both KHC and the mitochondrial protein miro, is essential in this apparatus. In contrast, kinesin light chain (KLC) is dispensable for mitochondrial transport in axons. Flies that are homozygous for die as second instar larvae, and transcripts are broadly expressed in these flies, suggesting a wider role for milton than its reported function in photoreceptors (). Therefore, we expressed GFP fused to a mitochondrial-import signal (mitoGFP; ) in neurons of the central nervous system to examine mitochondrial distribution in first instar larvae that are homozygous for which is a null allele. The segmental nerves that connect the central nervous system to the body wall of the larva contain motor and sensory axons, and thereby provide the clearest structures in which to image axonal mitochondria. In control larvae, numerous mitochondria were present in these axons. However, in larvae, axonal mitochondria were absent (). This defect is selective for mitochondria, as indicated by the continued presence of immunoreactivity for KHC, which is likely to transport many cargoes (; ), and the synaptic vesicle marker synaptotagmin (). We also assayed mitochondrial distribution in the ventral nerve cord, which consists of two central neuropil regions that run the length of the cord and are surrounded by a cortex of cell bodies (). The neuropil regions contain axons, dendrites, and pre- and postsynaptic endings. In control larvae, mitoGFP was present in the cell bodies, but was most abundant in the neuropil, reflecting the increased concentration of mitochondria at the synapses. In the mutant, the mitoGFP pattern was reversed, with little GFP remaining in the neuropil. Synaptotagmin localization was unchanged. Thus, the selective loss of mitochondria from axons and synapses is not restricted to photoreceptors. Moreover, milton probably also mediates mitochondrial transport in dendrites because at least half of the mitochondria of the neuropil are expected to derive from postsynaptic elements. The mitochondrial transport defect in mutants, and the in vivo association between milton and KHC, suggests that milton is an adaptor that links KHC to mitochondria. To test this hypothesis, we transfected cDNAs encoding milton () and myc-tagged rat KIF5B (myc-KHC; ), either alone or together, into COS7 cells. milton and its mammalian homologues function identically in all of our assays (Fig. S2, available at ; see Mammalian milton homologues); therefore, we have used the rat kinesin in these assays. Transfected alone, milton immunoreactivity was located exclusively on mitochondria (). In untransfected cells, the endogenous KHC was typically cytoplasmic, although in some cells KHC was also observed on mitochondria. Upon transfection with milton, the KHC became highly enriched on the mitochondria, with little remaining detectable elsewhere in the cell (). The phenomenon was more dramatic when rat myc-KHC was overexpressed in these cells; myc-KHC was largely cytoplasmic, and, in some highly expressing cells, colocalized with microtubules (), as previously observed (). However, when cells were cotransfected with both milton and myc-KHC, KHC was overwhelmingly located on mitochondria, colocalizing precisely with both MitoTracker and milton (64 out of 66 cells; ). In addition, mitochondrial distribution was altered by the cotransfection of milton and KHC. When milton alone was highly expressed, the mitochondria were clustered near the nucleus in 90% of the transfected cells (), which is a phenomenon encountered in only 3% of control cells. At lower expression levels, the mitochondria remained distributed as in untransfected cells (not depicted). Overall, milton expression caused less clustering of mitochondria in COS7 cells than had previously been observed in human embryonic kidney 293T (HEK293T) cells, in which all of the mitochondria become localized in an aggregate near the microtubule-organizing center (). In contrast, coexpression of KHC and milton caused many, though not all, mitochondria to reside at the cell margin and to form clumps at the tips of cell processes. This redistribution was not caused by a general change in the cytoskeleton. Neither KHC nor milton, transfected singly or in combination, altered the arrangement of microtubules in these cells (Fig. S1, available at ). The microtubules are chiefly oriented with plus ends toward the periphery of the cell; thus, the redistribution of mitochondria suggests that milton has recruited and activated coexpressed KHC and thereby caused a plus end–directed shift of many of the mitochondria. Together with the biochemical association of milton and KHC in and the mitochondrial localization of milton (), the ability of milton to recruit KHC to mitochondria offers direct support for the hypothesis that milton is a mitochondria-specific adaptor protein for the kinesin-1 family. To further characterize the interaction between milton and myc-KHC, we mapped the region of milton that is required for their association by cotransfecting HEK293T cells and assaying the ability of portions of milton to coprecipitate with full-length myc-KHC. Milton comprises 1,116 amino acids with no recognizable structural motifs, except for a long, predicted coiled coil domain (residues 140–380) that contains a high degree of amino acid identity with the equivalent regions in mammalian milton homologues (). Milton 1–450 was sufficient to associate with myc-KHC. Milton 1–750 also coimmunoprecipitated with KHC, but the COOH-terminal domain of milton (Flag-tagged milton 750–1,116 and 847–1,116) did not (). We also mapped the interaction by looking for colocalization in transfected COS7 cells, whose larger cytoplasmic volume made colocalization easier to score than in HEK293T cells. Milton 1–450 colocalized with myc-KHC in aggregates, but these were not on mitochondria. In contrast, Flag-milton 750–1,116 did localize to mitochondria, but in its presence myc-KHC remained cytoplasmic (). Thus, the associations of milton with KHC and mitochondria are separable, and the 1–450 region of milton is sufficient for the interaction with KHC. We determined the region of KHC necessary for its association with milton by cotransfecting milton with each of three truncated myc-KHC constructs and then immunoprecipitating with anti-milton mAb 5A124. Milton associated with both full-length KHC and KHC lacking the last 64 amino acids of the tail domain (myc-KHC 1–891), but not with KHC lacking the entire tail domain (myc-KHC 1–810) or a larger deletion (myc-KHC 1–682; ). Therefore, the KHC tail region 810–891, but not 892–955, was necessary for associating with milton. Consistent with this finding, we observed that myc-KHC 1–891, but not 1–682, could be recruited to mitochondria by milton overexpression (). The kinesin-1 family, including KHC and mammalian KIF5s, are generally considered to be tetramers composed of two KHCs and two KLCs. The deletion of amino acids 810–891 of KHC diminishes KLC binding (), and because endogenous KLC was present in the aforementioned experiments, milton might interact either directly with KHC or via KLC. Therefore, rat HA-tagged KLC1 (isoform C; ) was transfected into COS7 cells both alone and in combination with KHC and milton (). Alone, or when transfected with milton, KLC was cytoplasmic (; ). Even when milton, myc-KHC, and HA-KLC were coexpressed, KLC was invariably cytoplasmic and not located on mitochondria (). Moreover, in these cells, myc-KHC was also always cytoplasmic (). Thus, KLC expression inhibited the recruitment of KHC to the mitochondria by milton. We demonstrated by coimmunoprecipitation that KLC similarly inhibited the interaction of milton and KHC; HA-KLC expression prevented the coprecipitation of myc-KHC and milton (). Although not significantly homologous, the KHC-binding regions of milton and KLC are both predicted coiled coils and, therefore, may have similar and competing interactions for a binding site on KHC. These observations strongly suggested that KLC is not a part of the milton–KHC complex and that milton replaces KLC when associating KHC with mitochondria; therefore, we tested this hypothesis in vivo by biochemical and genetic means. As we previously observed, KHC immunoprecipitated with milton from homogenates of fly heads (). In these same precipitates, however, KLC was not detected. To increase the sensitivity of the assay, we immunoprecipitated milton from flies overexpressing myc-KLC (). Again, KHC coimmunoprecipitated with milton, but KLC did not (). Similarly, the immunoprecipitation of myc-KLC with anti-myc brought down KHC, but milton remained in the supernatant. Immunoprecipitation with antibodies to KHC, however, brought down both myc-KLC and milton. Thus, KHC appeared to form separate complexes either with milton or with KLC; the latter are likely to be more plentiful because of the greater abundance of cargoes requiring KLC. Because the milton–KHC complex did not contain KLC, we hypothesize that milton substitutes for KHC in mitochondrial transport. To further test the hypothesis that mitochondrial transport was KLC independent, we examined mutants lacking , which is the unique KLC gene in (). In contrast to larvae (), the peripheral nerves of homozygous larvae had abundant mitochondria (not depicted). Maternally contributed KLC, however, might account for the absence of a phenotype. Therefore, as a more rigorous test, we made eye clones homozygous null for () in a heterozygous background using the EGUF/ method (). Loss of did not prevent the differentiation and viability of photoreceptors, although eyes were somewhat small and roughened, and their axons were frequently disordered and sometimes short and defasciculated. Nevertheless, axonal mitochondria were numerous, just as in controls, in third instar larvae () and in adults (not depicted). Thus, axonal transport of mitochondria can occur in the absence of KLC. In contrast, the axons of -null photoreceptors lacked mitochondria (). Although, thus far, only a single milton cDNA; which is hereafter called milton-A, has been described () and used in this paper, protein and RNA analysis suggested greater complexity. Using cDNAs and ESTs from the Berkeley Genome Project, we found that milton splice variants can produce at least four distinct protein products with divergent NH termini. After differing 5′ ends, the transcripts converge in exon 9, at amino acid 129 of milton-A (). Anti-milton P1–152 antibody, which was raised against amino acids 1–152 of milton-A, recognized a single band on a Western blot of extract, but anti-milton mAb 5A124, which was raised against a domain present in all the predicted splice variants (milton-A 908–1,055), recognized multiple bands (). These observations are consistent with expression of splice variants in vivo. The 129 amino acids of the NH terminus of milton-A are replaced by 136 amino acids in milton-B and 269 amino acids in milton-C (). These domains have little homology to either milton-A, or to one another. In milton-D, the NH terminus domain is replaced by an untranslated region, such that a translation start site corresponding to Met 138 of milton-A is predicted within exon 9. All of the predicted variants contain the predicted coiled coil domain. The National Center for Biotechnology Information database of GenBank mRNAs and ESTs for human milton 1/OIP106 predict alternative NH termini that converge at the same point as the variants, but no alternative NH termini are predicted for mammalian milton 2/GRIF-1 as yet. Do all of the splice variants function equally in recruiting KHC to mitochondria? Each variant localized to mitochondria when transfected into COS7 cells (). They differed, however, in their interactions with KHC. Milton-D recruited myc-KHC to the mitochondria and coimmunoprecipitated with it (). Thus, milton's KHC association domain is contained within the sequences that are shared by milton-A and the shorter milton-D. Together with our earlier data (), these results indicate that the KHC association domain resides within the region corresponding to 138–450 of milton-A. Because this sequence is also present in milton-B and -C, these isoforms were likewise expected to associate with KHC and recruit it to mitochondria when equivalent amounts were expressed. Surprisingly, this was only true for milton-B (); milton-C neither recruited cotransfected myc-KHC to mitochondria nor coimmunoprecipitated with myc-KHC (). Milton's mammalian homologues function similarly to milton, localizing to mitochondria, coprecipitating with KHC, and recruiting KHC to mitochondria (Fig. S2; ). Therefore, we examined the distribution of endogenous miltons in mammalian cells by means of the P1–152 antiserum, which could recognize milton 1 and 2 because of the high conservation of the epitope (milton-A 1–152). In rat cerebellar granule neurons, endogenous milton colocalized with the mitochondrial marker cytochrome oxidase (Fig. S2 E). In COS7 cells, endogenous milton was also observed on mitochondria (Fig. S2 F). Thus, milton and its mammalian homologues are likely to play equivalent roles in mitochondrial transport. Despite its mitochondrial localization, milton has neither a mitochondrial import sequence nor a transmembrane domain. To examine how milton associates with mitochondria, we expressed partially deleted forms of milton. Milton's COOH terminus (expressed as either amino acids 847–1,116 or 750–1,116) had a mitochondrial distribution in COS7 cells (), although some remained cytoplasmic in highly expressing cells. In contrast, milton 1–450 was primarily nuclear at low expression levels () or filled the cytoplasm in higher expressing cells (not depicted). Milton 1–750 was also cytoplasmic, although it was occasionally enriched near the nucleus ( and ). Thus, the COOH terminus of milton must contain a domain that is sufficient to be targeted to mitochondria. A mitochondrial protein that might interact with milton was identified in a catalog of yeast two-hybrid interactions of proteins (). This protein, miro, contains two GTPase domains, a pair of EF hands, and a COOH-terminal transmembrane domain, and localizes to mitochondria (). When mutated, both human miro () and the yeast orthologue, Gem1p (), alter the subcellular localization of mitochondria in a manner reminiscent of milton overexpression (). Mutations in were recently isolated (; ) and found to lack axonal mitochondria (). Thus, miro is likely to be another essential component of the machinery for mitochondrial transport, and, therefore, we examined its relationship with milton. We have confirmed the interaction of miro and milton that was predicted by the two-hybrid screen by coimmunoprecipitation. HEK293T cells were transfected with milton-A or -D and either miro that was tagged with the T7 epitope or a control T7-tagged protein. Both milton isoforms were found in anti-T7 immunoprecipitates only when T7-miro was coexpressed (). When expressed in COS7 cells, miro invariably localized to mitochondria and induced a redistribution of mitochondria into aggregates (). This aggregation was more severe when milton was coexpressed, but was not accompanied by a change in microtubule structure (Fig. S1). Because miro has a transmembrane domain, we hypothesized that miro might be important for the mitochondrial localization of milton and that miro lacking the transmembrane domain (miroΔTM; amino acids 1–574) might have dominant-negative effects. Unlike full-length miro, miroΔTM, when expressed in COS7 cells, was diffusely distributed throughout the cytoplasm and was also nuclear when very highly expressed. Moreover, miroΔTM did not alter mitochondrial distribution (). However, miroΔTM could still bind to milton, as indicated by their coprecipitation when cotransfected (). In contrast to the strictly mitochondrial localization of milton when expressed with full-length miro, milton was displaced from mitochondria in most cells by expression of miroΔTM (). Thus, miro appears to be important for the association of milton with mitochondria, perhaps by serving as a receptor for milton on the mitochondrial surface. Consistent with this hypothesis, a truncated milton (1–750), which did not associate with mitochondria when expressed alone (), was recruited to mitochondria when miro was overexpressed in COS7 cells (). Milton 1–750 was also able to coimmunoprecipitate with miroΔTM (). Notably this milton construct does not contain the mitochondrial association domain we identified in the COOH terminus; it is therefore likely that milton associates with mitochondria through at least two regions: residues 847–1,116 bind to an unidentified protein and 1–750 bind to miro (). #text Mitochondria were visualized in using transgenic stocks containing mitoGFP () and placed under the control of a UAS promoter. Expression in selective tissues was driven by D42-Gal4, which is expressed in a subset of neurons (), or ey-Gal4, which is expressed in photoreceptors (). To visualize mitochondria in culture, cells were incubated with 100–300 nM MitoTracker orange (Invitrogen) for 15 min. Milton-A deletion constructs were constructed as follows: milton-A 1–450 and milton-A 1–750 were made from full-length milton-A in pCMV Tag1 () that was partially digested by Sal1; milton-A 608–1,116, 750–1,116, 608–942, and 847–1,116 were made by PCR with 5′ primers containing a BamHI site and 3′ primers with HindIII sites and were cloned into pCMV Tag1 (Stratagene) with in-frame NH terminus Flag tags. Milton-B (LD33316), -C (LD28289), and -D (AT08952 and AT28977, which differ only in the 5′ untranslated region) were obtained from the Berkeley Genome Project (University of California, Berkeley, CA), and their PCR-amplified NH termini were substituted for that of milton-A in pCMV Tag1 milton-A. The NH termini of these clones were amplified by PCR with 5′ primers containing a BamHI site and a 3′ primer with an XhoI site and cloned into BamHI–XhoI–digested pCMV Tag1 milton-A. pCMV Tag1 milton-C was made similarly, but using NotI at the 5′ instead of BamHI. Human xpress-OIP106 (milton 1) and rat xpress-GRIF-1 (milton 2) were provided by G. Hart (Johns Hopkins University, Baltimore, MD; ). Rat myc-KHC, myc-KHC1-682, myc-KHC1-810, myc-KHC1-891, and HA-KLC constructs were provided by K. Verhey (University of Michigan, Ann Arbor, MI; ). An EST, RE01164, corresponding to miro (CG5410-PE), was obtained from the Berkeley Genome Project and cloned between the BamHI and NotI sites of pA1T7 (), thereby placing a T7 epitope tag at the NH terminus. MiroΔTM was generated by digesting the miro construct with EcoRI and NotI, and then ligating an oligo-encoding a stop codon between these sites. COS7 and HEK293T cells were cultured in DME supplemented with 10% FCS, -glutamine, and penicillin/streptomycin. Rat cerebellar neurons were cultured as previously described (). Cells were transfected with calcium phosphate and immunostained 24–36 h later. In all cotransfection experiments, 1:1 ratios of DNA were used, except for miroΔTM, which was transfected in a 250-fold excess. Immunocytochemistry was performed as previously described () and used either anti-milton mAbs 2A108, 4A75, or 5A124, or anti-milton antiserum P1–152. Other primary antibodies used in this study include: chick anti-myc (Invitrogen), 9E10 (Santa Cruz Biotechnology, Inc.), chick anti-HA (GTS, Inc.), anti-Xpress (Invitrogen), goat anti-T7 (Bethyl Laboratories, Inc.), mouse anti-T7 (Novagen), anti-kinesin (AKIN01; Cytoskeleton, Inc.), anti-cytochrome oxidase (BD Biosciences), rabbit anti-KLC (a gift from J. Gindhart, University of Richmond, Richmond, VA; ), mouse anti-HSP60 (Stressgen Bioreagents), and rabbit anti-HA (Novus Biologicals, Inc.). The following fluorescently tagged reagents were used: goat anti–mouse Alexa Fluor 488, goat anti–chick Alexa Fluor 647, donkey anti–mouse Alexa Fluor 647, donkey anti–goat Alexa Fluor 633, and donkey anti–mouse Alexa Fluor 647 (all from Invitrogen), and goat anti–mouse Cy3, horseradish peroxidase Cy5, and donkey anti–rabbit FITC (Jackson ImmunoResearch Laboratories). Cells were lysed in 5 mM EDTA, 300 mM NaCl, and 50 mM Tris-HCl, pH 7.5, and a protease inhibitor cocktail set III (Calbiochem) was used at 1:1,000, 0.1 mg/ml PMSF (Sigma-Aldrich), and 1% Triton X-100 when precipitating milton with KHC, KLC, or 0.5% Triton X-100 for miro and milton experiments. Lysates were precleared with irrelevant antibodies and protein A, incubated with anti-milton antibodies mAb 9E10 or anti-T7, and protein A–Sepharose beads (GE Healthcare) for 2–3 h at 4°C. Immunoprecipitates were separated by SDS-PAGE and transferred to nitrocellulose membranes. For immunodetection, anti-milton mAb 5A124 and anti-milton mAb 2A108 were used at 1:40; rabbit anti-KLC was used at 1:100; mAb 9E10, rabbit anti-HA, and rabbit anti-KHC were used at 1:1,000; and donkey anti–rabbit HRP and goat anti–mouse HRP were used at 1:10,000 (Jackson ImmunoResearch Laboratories, Inc.) and and (). and ( null; provided by L. Goldstein, University of California, San Diego, CA; ). and (A. Pilling and W. Saxton, Indiana University, Bloomington, Indiana). ;; (III) and ; (; Bloomington Stock Center). The -null mutant, was recombined onto chromosome. Recombinants were confirmed by PCR analysis and by lethality complementation analysis with alleles ( , and were provided by J. Gindhart). photoreceptors and and . Fig. S1 shows that microtubules are not disrupted in transfected cells. Fig. S2 shows that mammalian miltons colocalize with mitochondria and recruit KHC to the mitochondria. Online supplemental material is available at .

Sarcomere assembly is a process orchestrated by the sequential expression of structural and signaling proteins, which ultimately leads to the formation of mature myofibrils. It involves the exchange of nonmuscle myosin IIB for muscle myosin II, the incorporation of titin and titin-binding proteins into the nascent myofibril, the lateral alignment of sarcomeric proteins, and the fusion of α-actinin–rich Z bodies into Z bands (). During maturation of the sarcomere, titin's NH terminus is localized in Z bodies, and muscle myosin II is aligned along the developing myofibril, presumably in a titin-dependent process (). Mature myofibrils are characterized by the alignment of muscle myosin II filaments to form A bands and the fusion of Z bodies to form Z bands (; ; ). They contain a continuous elastic filament system along the myofibril with titin molecules overlapping at the Z disc and M band, which has been regarded as a molecular ruler or blueprint for sarcomere assembly (; ; ; ; ). In addition to its structural role in myofibrillogenesis, titin's M-line region has been implicated in sarcomere assembly through the titin kinase domain and its in vitro substrate titin cap (T-cap), which is also known as telethonin. Activation of the titin kinase and phosphorylation of T-cap in differentiating myocytes has been hypothesized to be involved in reorganization of the cytoskeleton during myofibrillogenesis (). So far, no suitable animal or tissue culture model was available to test this hypothesis. We have successfully used the Cre-lox recombination system to excise titin's M-line exons (MExs) 1 and 2 in striated muscle and demonstrated their importance in both skeletal and cardiac muscle (; ). Loss of titin's M line leads to impaired stability of the muscle fiber with the disassembly of existing sarcomeres. This results in reduced cardiac output followed by a failure to thrive and lethality dependent on the onset and level of Cre expression. The conditional knockout approach enabled the generation of adult animals to study titin's function in the mature heart and skeletal muscle, but expression kinetics of the Cre recombinase transgene preclude the analysis of titin's role in sarcomere assembly during early embryonic development. To distinguish a role in sarcomere assembly from a role in stabilizing preexisting sarcomeres and to address potential nonmuscle functions, we have converted our conditional M-line titin knockout into a complete knockout using germline recombination. In this study, we show that titin's M-line region is dispensable for initial sarcomere assembly, including the correct localization of M-band proteins, but that it is required to fortify the sarcomere structure and for lateral growth. Although the titin M line–deficient hearts start to contract and loop properly, wall thickness and trabeculation are reduced from embryonic day (E) 9.5 followed by apoptosis secondary to the reduced cardiac output. Monitoring the localization and embryonic expression of M-band proteins and proposed substrates of the titin kinase, we were able to attribute the sarcomere disassembly to titin's structural functions. Unlike in the adult knockout sarcomere, kinase-deficient titin does not integrate into the A band and, thus, fails to form a continuous filament system. The failure to cross-link myomesin and titin results in increased mobility of titin's COOH terminus. These structural changes lead to reduced stability and ultimately to disassembly of the sarcomere. The titin M-line region is critical for the maintenance of sarcomere structure and function in adult muscle (). To address its role in early cardiac development and possible nonmuscle functions, we have used germline expression of the Cre recombinase under control of the protamine promoter to convert our conditional knockout model into a constitutive titin M line–deficient animal (). After recombination, the mutant titin allele is transmitted through the male germline and leads to the expression of a titin protein that is deficient in the titin kinase region encoded by titin's MEx1 and 2 (). In addition to the kinase domain, this region contains binding sites for the ubiquitin ligase MuRF-1 (muscle-specific RING finger protein 1), signaling proteins such as calmodulin and FHL2 (four and a half LIM-only protein 2), and the M-band protein myomesin (a detailed M-band protein map indicating the deletion is provided in ). Although heterozygous knockout animals are fertile and do not display any phenotypic abnormalities, homozygous knockouts die in midgestation ( and ). Knockout, heterozygous, and wild-type animals are present at the appropriate Mendelian ratios and express titin isoforms of the expected sizes. Unlike in embryonic development, the adult M line–deficient titin is expressed at lower levels than the wild-type isoform (compare adult with embryonic heterozygous animals; ). Expression of the wild-type and truncated titin is restricted to the heart and somites in the developing embryo (), as shown by in situ hybridization. At E9.5, there is no titin MEx1 expression detectable in knockout hearts. Compared with the wild-type and heterozygous animals, head and body size are reduced. The symmetrical body shape is unlike the shrunken head phenotype observed in the mutant, which maps to the titin locus (). We followed the embryonic phenotype from E9, when knockout embryos are of similar size as their littermates and display proper development. This is not only reflected in identical cardiac morphology but also in appropriate initiation and maintenance of the heartbeat in both knockout and wild-type embryos (; and Figs. S1 and S2, available at ). Because embryonic development at the organ, cellular, and ultrastructural level did not differ between wild-type and heterozygous animals, we restricted all further comparisons to wild-type and homozygous knockout animals. At E10, the mutant embryo appears normal (including the ratio of heart to body size) but small for its age. The heart undergoes proper looping, but trabeculation and wall thickness are reduced (Fig. S2). By day 11, atrophy results in instability of the ventricular wall with pericardial effusion (). At this stage, the developmental delay is reflected in the decreased number of somites in M-line knockout embryos (). The reduced cardiac wall thickness and asystole (loss of cardiac activity) from E9.5, together with the failure to thrive, led us to investigate apoptosis in muscle and nonmuscle tissues as a potential mechanism. At E9.5, when we first see a genotype-dependent size difference, the level of apoptosis is comparable between knockout and wild-type animals (Fig. S3, available at ). At E10.5, apoptosis is significantly increased in knockout animals, particularly around the peritoneal cavity in the abdominal part of the trunk (Fig. S3, bottom). Because nonmuscle knockout cells divide and differentiate normally until cardiac function is affected, there is no primary defect in nonmuscle cells that would interfere with the cell cycle or cause apoptosis. The increased apoptosis from E9.5 in knockout animals affects all tissues. It is not increased in cardiac versus extracardiac tissue (abdominal cavity, limbs, and brain), which would also indicate a secondary change and does not imply a nonmuscle function of titin's M-line region such as an anti-apoptotic effect. Titin's kinase domain has been proposed to be required for sarcomere assembly. As a mechanism, it has been suggested that phosphorylation of the titin kinase substrate T-cap, which transiently localizes to the M band during development, coordinates the assembly of Z-line and M-band lattices during myofibrillogenesis (). To investigate titin's structural functions, we have used electron microscopy to follow sarcomere assembly, growth, and disassembly. At E9, assembled sarcomeres were detected in the knockout, and no structural alteration compared with wild-type sarcomeres was apparent (). SDS-PAGE revealed the exclusive expression of kinase-deficient titin in our knockout animals (, lane 3). Therefore, our ultrastructural data suggest that the titin kinase domain is not required for sarcomere assembly. From E9.5, knockout myofibrils fail to grow laterally (quantification in ), but Z-disc and M-band structure are maintained through E10. Thereafter, knockout sarcomeres disassemble, and, at E11, only a few filaments in disarray remain. The embryonic heart does not reveal M bands as an electron-dense area (), so electron microscopy is blind to early changes in the developing cardiac M band. Thus, although titin M line–deficient and wild-type sarcomeres are indistinguishable at the ultrastructural level early in development, their structure and molecular composition could already be affected. Accordingly, we followed the integration of titin into the sarcomere using antibodies directed against titin's I-band region proximal to the Z disc (N2B) and its COOH terminus (M8/M9) as well as an anti–α-actinin antibody as a marker for the Z disc (; also see the localization of titin epitopes in Fig. S2). In wild-type animals, titin molecules overlap at the Z disc and M band to form a continuous filament system (, compare A with B; ; ; ). In knockout animals, titin's NH terminus is incorporated into the Z disc, which results in close proximity of α-actinin and the titin N2B region at E9.5. Upon disassembly, Z bodies distribute from regular striation (, open arrowheads) to random patches of variable size (, closed arrowheads). This sign of disintegration of the Z bodies is complemented by the diffuse staining for titin's N2B region, which is now separate from α-actinin. Unlike the NH terminus of kinase-deficient titin, its M-line region is not integrated into the developing sarcomere at all (). Even at E9.5, there is no accumulation of titin's M8/M9 epitope between Z discs. The absence of a distinct M8/M9 staining could result either from the mislocalization of M8/M9 or from the expression of a truncated protein, which does not only lack the kinase region but fails to include the M8/M9 domains. We used Western blot analysis of embryonic hearts derived from wild-type and heterozygous animals to confirm proper expression of the truncated M-line region (). Because the M8/M9 epitope is included in the wild-type and knockout protein of heterozygous animals, the loss of M-line staining is indeed the result of the partial integration of kinase region–deficient titin into the sarcomere at the Z disc only. Next, we used immunofluorescence staining to follow M-band assembly and the fate of myomesin in the absence of titin's myomesin-binding site (see ). The periodic myomesin staining indicates that even in the absence of titin's M line, myomesin is incorporated into the sarcomere (). It localizes properly between Z discs, as demonstrated by costaining with the Z-disc protein α-actinin. Nevertheless, in knockout animals, myomesin staining is more diffuse. This might reflect the reduced number of binding sites in the absence of titin's kinase region because myomesin levels are not up-regulated in knockout versus wild-type animals (). In both our conditional titin knockout animals (; ) and the conventional knockout described in this study, titin kinase-deficient sarcomeres disassemble. Nevertheless, structure is preserved better in the adult than in the developing sarcomere with titin's M line integrated properly, even in the absence of titin's kinase region. Titin's sarcomeric functions are in part inherent in the protein (spacer/spring) and in part relayed through protein–protein interactions. Thus, we used expression analysis of both titin and its binding proteins at various stages of development to help discriminate their roles in the embryonic and adult heart. Overall mRNA levels of titin and its binding proteins are reduced in embryonic development and increase by up to two orders of magnitude from E9.5 to adulthood. To monitor titin expression, we used TaqMan probes that distinguish the Z disc, kinase region, M line, and an internal region of titin (). Adult titin isoforms lack various internal exons of the Ig and PEVK regions (; ). The novex-3 isoform, which encodes a truncated 700-kD protein, does not even include the M-line region (). Correlating titin's Z-disc and M-line expression allows us to determine the amount of full-length versus novex-3 titin. Increased embryonic M-line RNA levels indicate a higher ratio of full-length titin compared with the truncated novex-3 isoform in the developing embryo from E9.5 (). In early development, we see a reduced expression of the heart-specific titin N2B region, which would imply altered elastic properties of the titin filament system. Various titin-binding proteins have been proposed to act as a substrate for the titin kinase and regulate embryonic sarcomere assembly. Of these, T-cap, Sqstm1 (sequestosome 1), and Nbr1 (neighbor of BRCA1 gene 1) are expressed at <20% of adult levels in the embryonic heart (). Combining protein and RNA data, MuRF-1, myomesin, and calmodulin are the only proteins expressed in significant amounts at the time when the knockout phenotype develops (; and Fig. S2). This would argue against a role of the proposed kinase substrates T-cap, Sqstm1, and Nbr1 in the development of the phenotype. Because of its size, it has been notoriously difficult to study titin's signaling and structural functions in vivo. Although multiple titin-deficient animals and cell lines have been generated (; ; ; ; ; ), it has not been possible so far to address titin's role in early sarcomere assembly. In this study, we present a novel animal model to investigate titin in the developing sarcomere. We have generated an internal homozygous deletion in the titin gene, which excises titin's kinase region (MEx1 and 2), allowing us to study its role in sarcomere assembly and address its potential functions in nonmuscle cells. The transgenic mice with loxP sites flanking titin MEx1 and 2 have been described previously (). They were converted from a conditional to a complete knockout using protamine-Cre transgenic mice (The Jackson Laboratory; ). Male double heterozygotes (protamine-Cre/wild type; titin MEx1/2) were backcrossed to 129/SvEms-+Ter?/J to obtain a clean colony of heterozygous knockouts devoid of the Cre transgene (TiMEx1/2). Template DNA was prepared from yolk sac or tail according to standard procedures, and recombination of the titin locus was monitored by PCR (primers PL1 and PL4). Lox and wild-type loci were typed using primers PL1 and PL2. All primers used have been described previously (). Timed matings were set up between heterozygotes (MEx1/2). The morning of vaginal plug detection was regarded as day 0.5 after conception. Embryos were harvested at E8.5, 9.0, 9.5, 10.0, 10.5, 11.0, and 11.5. All experiments involving animals were performed according to institutional and National Institutes of Health (NIH) Using Animals in Intramural Research guidelines. SDS lysates for titin gels were prepared by homogenization of pooled dissected hearts from at least three different litters in 0.5 M Tris-HCl, pH 6.8, followed by DNase digestion for 30 min at 37°C and lysis in a buffer containing 8 M urea, 2 M thiourea, 3% SDS, 75 mM DTT, 0.05 M Tris-HCl, pH 8.6, and 0.03% bromphenolblue. Titin isoforms were separated using an SDS-agarose gel electrophoresis system followed by staining with Coomassie () or Sypro Ruby (Invitrogen). Whole-mount RNA in situ hybridization for the exon preceding MEx1 and for MEx1 was performed using digoxigenin-labeled antisense riboprobes. RNA labeling of linearized plasmid templates was accomplished with digoxigenin-UTP by in vitro transcription with T7 polymerase according to the manufacturer's specifications (Roche Diagnostics). After fixation, embryos were washed in PBS plus 0.15% Tween 20, dehydrated, bleached with 25% HO in MeOH, and rehydrated in an ascending/descending series of methanol (25, 50, and 75% in PBS–Tween 20 and 100% methanol). The tissue was permeabilized with 20 μg/ml proteinase K and refixed with 4% PFA and 0.2% glutaraldehyde for 20 min at room temperature. Embryos were prehybridized in 50% formamide, 5× SSC (0.75 M NaCl and 0.75 M sodium citrate, pH 7.0), 0.1 mg/ml single-stranded DNA (Sigma-Aldrich), 40 μg/ml heparin, 50 μg/ml tRNA, 0.15% Tween 20, and 60 mM citric acid. Hybridization of the digoxigenin-labeled probe was performed overnight at 65°C. Embryos were washed in solution 1 (50% formamide, 5× SSC, and 0.15% Tween 20) at 65°C twice for 40 min, washed in solution 2 (10 mM Tris, pH 7.5, 50 mM NaCl, and 0.15% Tween 20) three times for 15 min, and washed in solution 3 (50% formamide, 2× SSC, and 0.15% Tween 20) three times for 60 min. Embryos were incubated with 0.75 U antidigoxigenin antibody, Fab fragments (Roche Diagnostics) in 5% goat serum in TBS–Tween 20 overnight at 4°C, and washed with TBS–Tween 20 for at least 8 h at room temperature. Embryos were stained with NBT/BCIP solution (350 and 175 μg/ml, respectively; Roche Diagnostics). Images were taken after refixation with 4% PFA and clearance with 80% glycerol. Embryos were dissected followed by overnight fixation in 4% PFA, equilibration with 30% sucrose in PBS, and embedding in Tissue Tek (optimal cutting temperature compound; Vogel). 5-μm sagittal sections were air dried, fixed with 4% PFA, blocked, and permeabilized in 0.3% Triton X-100, 0.2% BSA, and 10% normal goat serum for 60 min. Cryosections were incubated overnight with primary antibodies at 4°C followed by fluorescent-conjugated secondary antibodies (AlexaFluor488 goat anti–rabbit [Invitrogen] and Cy3 goat anti–mouse [Jackson ImmunoResearch Laboratories]). Primary antibodies were used at the following dilutions: 1:500 monoclonal antisarcomeric α-actinin (Sigma-Aldrich), 1:100 monoclonal anti-titin T3 (gift from D.O. Fürst, Universität Bonn, Bonn, Germany; ), 1:10,000 polyclonal antimyomesin EH (gift from E. Ehler and J.-C. Perriard, ETH Zürich; ), and rabbit polyclonal antibodies 1:1,000 anti–titin-Z1/Z2, 1:200 titin-N2B, and 1:500 anti–titin-M8/M9 (both were gifts from S. Labeit, Universitätsklinikum Mannheim; ). Stained tissue was mounted with fluorescent mounting medium (DakoCytomation) and analyzed at room temperature using Immersol 518N (Carl Zeiss MicroImaging, Inc.) on a confocal scanning laser microscope (LSM5 Pascal with software version 3.0 SP2; Carl Zeiss MicroImaging, Inc.) with a plan-Neofluar 100× 1.3 NA lens (Carl Zeiss MicroImaging, Inc.). Images were assembled using Photoshop 9.0 and Corel Draw 12.0. For TUNEL assay (terminal deoxynucleotidyltransferase-mediated dUTP-biotin nick end labeling), cryosections of mouse embryos were processed using the in situ cell death detection kit (Roche Diagnostics) according to the protocol supplied by the manufacturer. In brief, sections were fixed in 4% PFA and permeabilized with 0.1% Triton X-100 in 0.1% sodium citrate for 2 min on ice. After washing, slides were incubated with TdT terminal transferase and fluorescein-dUTP for nick end labeling. Sections were counterstained with 1:500 antisarcomeric α-actinin (Sigma-Aldrich). Stained tissue was analyzed using a microscope (BX51; Olympus), a CCD camera (Visitron Systems 7.4 Slider; Diagnostic Instruments), and MetaMorph software version 6.2r2 (Universal Imaging Corp.). Embryos for ultrastructural analysis of the sarcomere assembly were dissected and fixed with 3% formaldehyde in 0.2 M Hepes, pH 7.4, for 30 min followed by immersion in 8% formaldehyde/0.1% glutaraldehyde in 0.2 M Hepes, pH 7.4, overnight. Embryos were postfixed with 1% OsO for 2 h, dehydrated in a graded ethanol series and propylene oxide, and embedded in Poly/Bed 812 (Polysciences, Inc.). Ultrathin sections (70 nm) were contrasted with uranyl acetate and lead citrate and were examined with an electron microscope (model 910; Carl Zeiss MicroImaging, Inc.). Digital images were taken with a 1k × 1k high speed slow scan CCD camera (Proscan). The diameter of ∼25 sarcomeres per embryo was measured on a range of ultrathin sections taken from two embryos per genotype and day of gestation using the analySIS 3.2 software (Soft Imaging System). Thermal cycling conditions were as follows: 50°C for 2 min, 95°C for 10 min followed by 59 cycles of 95°C for 10 s, and 60°C for 1 min. Data were collected and analyzed with the Sequence Detection System 2.1 software (Applied Biosystems). The comparative CT Method (ΔΔC Method) was used as described in the User Bulletin 2: ABI PRISM 7700 Sequence Detection System. All results are expressed as means ± SD. An unpaired two-tailed test was performed to assess differences between two groups. A P value of <0.05 was considered significant. Fig. S1 shows the morphology and contractile function of titin-deficient hearts. Fig. S2 provides the location of antibody epitopes and primer-binding sites and shows expression changes in titin-deficient and wild-type animals. Fig. S3 shows increased apoptosis in knockout embryos that are secondary to the cardiac phenotype. Online supplemental material is available at .

Activation of phosphatidylinositol (PI)-3-kinase and small GTPases at the cell membrane controls various intracellular events, including cell–cell junction formation (), extension of the leading edge (), and infectious processes of pathogens (). Activation of PI-3-OH kinase and subsequent PIP production activates Rac through PIP binding to Rac guanine nucleotide exchange factors (; ). PIP is a lipid component of the cell membrane, and Rac is integrated into the membrane via lipid modification (). Therefore, the site of action downstream of PIP and Rac should be the membrane. WAVE2 belongs to the Wiskott-Aldrich syndrome protein (WASP) family of proteins, which activate the actin-related protein (Arp) 2/3 complex to stimulate signal-induced actin polymerization. Five WASP family proteins have been identified, and WAVE1, WAVE2, and neural WASP (N-WASP) are expressed ubiquitously (; ). Studies of cells from WAVE1, WAVE2, and N-WASP knockout mice have shown that WAVE2 is the protein that activates the Arp2/3 complex downstream of the small GTPase Rac (; ; ). We have reported that WAVE2 is a PIP-binding protein, although PIP alone was not involved in regulating the ability of WAVE2 to activate the Arp2/3 complex (). A large protein complex has been proposed to suppress WAVE1 activity by trans-inhibition in which a trimeric protein complex, including PIR121/Sra1, Nap1, and Abi, binds to WAVE1 and suppresses WAVE1 activity (). WAVE2 also forms a large protein complex including HSPC300, Abi1, Nap1, and Sra1/PIR121 (; ; ; ; ). Studies of cultured cells have shown that Sra1, Nap1, and Abi are involved in stabilizing WAVE2; the knockdown of Abi1, Nap1, or Sra1 results in decreased amounts of all of the proteins in the WAVE2 complex (; ; ). Among proteins in the WAVE2 complex, Abi1 and HSPC300 bind directly to the NH-terminal WAVE homology domain (WHD) of WAVE2. WHD-mediated association with Abi1/2 also contributes to the localization of WAVE2 at the leading edge of lamellipodia (). However, purified HSPC300, Abi1, Nap1, and Sra1/PIR121 do not suppress the activity of WAVE2 purified from a baculovirus system (; ). Therefore, trans-inhibition does not appear to occur with WAVE2 purified from baculovirus. The activity of the native WAVE2 complex has yet to be examined. A WAVE2-binding protein that is not included in the aforementioned protein complex is IRSp53/BAIAP2/Bap2α (). The Src homology 3 (SH3) domain of IRSp53 binds to WAVE2, and the NH-terminal Rac-binding (RCB) domain (residues 1–228) binds to Rac (; ). Thus, IRSp53 might be the link between Rac and WAVE2 that is involved in lamellipodium formation. The NH-terminal region of IRSp53 (residues 1–250), including the RCB domain, is termed the IRSp53/missing in metastasis homology domain (IMD). The IMD possesses actin filament bundling activity, and the overexpression of IRSp53 induces microspike/filopodium formation (; ; ). Furthermore, Cdc42 does not bind to the RCB domain but binds to the Cdc42–Rac interactive binding motif between the RCB and SH3 domains (; ). These findings have not been reconciled, and the role of IRSp53 in actin cytoskeletal reorganization remains to be clarified. In this study, we investigated the activity of the WAVE2 complex purified from various cellular preparations and examined the contributions of WAVE2-binding proteins to the regulation of WAVE2 downstream of Rac. To purify the WAVE2 complex from cultured cells, we stably expressed WAVE2 tagged with FLAG in A431 cells. The amount of WAVE2 in the FLAG-WAVE2–expressing cell line was approximately twice that of control vector–transfected cells (Fig. S1 A, available at ). This cell line and the control cell line were used throughout this study. WAVE2 in FLAG-WAVE2–expressing cells was localized similar to WAVE2 in control A431 cells (Fig. S1 A). A significant portion of endogenous WAVE2 localized at cell–cell junctions and lamellipodia (). In cells with decreased Abi1 expression by RNAi, the amount of WAVE2 was significantly decreased in FLAG-WAVE2–expressing cells, indicating that tagged WAVE2 under stable expression behaved similarly to endogenous WAVE2 (). WAVE2 tagged with GFP did not localize properly in the absence of Abi1, but it localized in a manner similar to endogenous WAVE2 in the presence of Abi1 as reported previously, indicating that Abi1 is essential for the localization of WAVE2 (; ; ). We previously identified IRSp53 as a WAVE2-binding protein. The presence of IRSp53 splice variants was indicated by the presence of IRSp53 bands at two positions on the Western blot. IRSp53 was also localized at cell–cell junctions and lamellipodia (). In cells with decreased IRSp53 expression, the amount of WAVE2, Abi1, Nap1, or Sra1 in whole cell lysates was not altered (), and WAVE2 remained at the cell periphery (). The amount of Abi in the WAVE2 protein complex was not significantly altered by IRSp53 knockdown (Fig. S1 B), as indicated by immunoprecipitation with anti-FLAG antibody. In cells subjected to IRSp53 RNAi, lamellipodium formation in response to EGF treatment or expression of constitutively active (CA) Rac (Rac CA) was significantly decreased ( and ). Time-lapse analysis confirmed the decrease in lamellipodium formation, especially the speed of extension at the leading edge of cells treated with IRSp53 RNAi ( and Video 1, available at ). The reduction in speed of the extension of cells treated with Abi1 RNAi was more drastic than of cells treated with IRSp53 RNAi ( and Video 1). A similar protein, insulin receptor tyrosine kinase substrate (IRTKS), was also found in A431 cells by RT-PCR analysis (Fig. S1 B). RNAi for IRTKS was performed in A431 cells, and specific decreases in mRNA were observed for IRTKS (Fig. S1 B). However, the reduction of IRTKS did not affect ruffle formation (). We then examined whether IRSp53 is essential for ruffle formation in fibroblasts. Mouse embryonic fibroblasts (MEFs) from WAVE2 knockout mice were used. The expression of full-length (wild type) WAVE2 restored ruffle formation upon the expression of Rac CA, but the expression of ΔPR WAVE2 with a deleted proline-rich region (IRSp53-binding region) or of ΔWHD WAVE2 with a deleted WHD region (Abi1-binding region) did not (Fig. S2, available at ). The A5 mutant of WAVE2 (defective in binding to PIP; ) also failed to restore ruffle formation (Fig. S2). Importantly, ΔPR WAVE2 localized at the cell periphery, whereas ΔWHD WAVE2 did not. Consistent with the localization of WAVE2 at the cell periphery in IRSp53 RNAi A431 cells (), IRSp53 was not essential for the localization of WAVE2. In MEFs, ruffle formation was severely impaired by Abi1 RNAi (Fig. S2 and Video 2). IRSp53 RNAi caused a decrease of ruffle formation and extension speed of the leading edge, indicating the involvement of IRSp53 in ruffle formation in fibroblasts (Fig. S2 and Video 2). Because IRSp53 binds to both Rac and Cdc42, we examined the interaction between IRSp53 and WAVE2 upon the expression of CA mutants. In the presence of Cdc42 CA, the interaction was significantly decreased (). In contrast, the interaction between IRSp53 and WAVE2 was not significantly perturbed by Rac CA expression (). The interaction between purified WAVE2 and purified IRSp53 was also decreased in the presence of GTPγS-loaded Cdc42 but not in the presence of Rac (). Therefore, Cdc42 appears to negatively regulate the binding of IRSp53 to WAVE2. We fractionated cells and examined the subcellular localization of proteins by conventional ultracentrifugation (). All of the proteins analyzed were found in all cell fractions, including the cytosol, membrane, and nucleus (unpublished data). WAVE2 was somewhat enriched in the membrane fraction; the membrane fraction contained 30% more WAVE2 than the cytosol fraction. Abi1 was much more enriched in the membrane fraction, as the membrane fraction contained approximately twice that of the cytosol fraction. Most IRSp53 was present in the membrane fraction, as it contained four times that of the cytosol fraction (). We then examined whether ruffle-inducing stimuli could translocate WAVE2 from the cytosol and nucleus to the membrane. Upon EGF treatment, the amount of WAVE2 in the membrane fraction was increased 17% (). When Rac CA was expressed, the amount of WAVE2 in the membrane fraction increased 26% (). However, when dominant-negative (DN) Rac (Rac DN) was expressed, the amount of WAVE2 in the membrane fraction decreased 16% (). The expression of Cdc42 CA also decreased the membrane localization of WAVE2 ( and S1 E). WAVE2 in the cytosol fraction was not significantly reduced in response to these stimuli, presumably because of the presence of nuclear WAVE2 translocated to the cytosol, as shown by antibody staining of the cells (unpublished data). Thus, ruffle-inducing stimuli induced the translocation of some WAVE2 to the membrane (). We then examined the contribution of IRSp53 to the membrane localization of WAVE2. When we knocked down IRSp53 by RNAi, the amount of WAVE2 at the membrane decreased by 30–40% (). WAVE2 was not involved in the membrane localization of IRSp53 because RNAi of WAVE2 did not have an effect on the amount of IRSp53 in the membrane (). To determine the relationship between protein localization and Arp2/3 complex activation, we purified WAVE2 protein complex with a FLAG tag. We purified WAVE2 complex from the cytosolic fraction; however, we were unable to resolubilize the pelleted membrane fraction. Therefore, we purified the WAVE2 complex from cell lysates, including the membrane fraction, with Triton X-100. Thus, WAVE2 complex from the Triton X-100 fraction contained cytosolic and membrane WAVE2 at a ratio of ∼1:1.3 (). Examination of the proteins in the WAVE2 complex in the Triton X-100 fraction by Western blotting showed the presence of Abi1 and IRSp53, as reported previously (; ; ; ). Abi1 also exists as several isoforms and splice variants, and several bands were identified. EGF stimulation induces the phosphorylation of WAVE2 (). However, significant alterations in the amounts of Abi1 and IRSp53 in WAVE2 immunoprecipitates were not observed in response to EGF stimulation. Specific binding of IRSp53 and Abi to WAVE2 was confirmed in experiments with vector-transfected cells (). Other WASP family proteins, including N-WASP and WAVE1, were not identified in the WAVE2 complex (). Proteins coeluted with WAVE2 did not differ significantly in response to EGF stimulation. Because WAVE2 in the membrane was increased only 17% upon EGF stimulation (), a lack of difference in the WAVE2 complex before and after stimulation may be reasonable. The expression of Rac CA or Rac DN also did not cause significant change in the WAVE2 complex (unpublished data). Mass spectrometry and Western blotting identified PIR121/Sra1, Nap1/p125NckAP1, and Abi1/Abi2/E3B1 in all WAVE2 complex preparations as reported previously (; ; unpublished data). We also identified IRSp53 comigrating with Abi1 and IgG in the Triton X-100 fraction only by SDS-PAGE independently of EGF stimulation (, bottom). From the band intensity in Coomassie brilliant blue–stained gels (unpublished data), the molar ratio of Abi, Nap1, or Sra1 to FLAG-WAVE2 was determined to be ∼0.5–0.6 (), suggesting the presence of monomeric WAVE2 upon FLAG-WAVE2 expression. Importantly, the amounts of PIR121/Sra1, Nap1, and Abi in the WAVE2 complex were consistently independent of stimuli or subcellular localization, indicating stable formation of the WAVE2 complex. Abi1 and IRSp53 bind to WAVE2 via the coiled-coil region of Abi1 and the SH3 domain of IRSp53 (; ). We measured the dissociation constant (Kd) between full-length WAVE2 and the coiled-coil region of Abi1 (aa 6–124) or between full-length WAVE2 and the SH3 domain of IRSp53 (aa 363–521) with a dual polarization interferometer, and the Kds were determined to be 0.6 and 5.7 μM, respectively (). We then measured the intracellular concentration of WAVE2 or IRSp53 by quantitative Western blotting with purified WAVE2 or IRSp53 as standards. WAVE2 concentration in whole cell lysates was 30–50 nM, whereas that of IRSp53 was 150–300 nM. Given the Kd of 5.7 μM for WAVE2 and IRSp53, only 2.5 nM WAVE2 was associated with IRSp53 in solution at WAVE2 and IRSp53 concentrations of 50 and 300 nM, respectively. Therefore, the Kd value between IRSp53 and WAVE2 suggests that these two proteins associate with the aid of other molecules such as Rac or PIP or in a restricted location such as the cell membrane. Even if WAVE2 and IRSp53 associate under limited situations, the 10-fold difference in Kd values appears to explain the lesser amount of IRSp53 in the aforementioned purified WAVE2 complex. Three washes in 10-fold volumes of buffer will result in a 1,000-fold difference in protein amount. Most WAVE2 molecules are reported to complex with Sra1/PIR121, Nap1, and Abi1. We fractionated lysates of control A431 cells or cells expressing FLAG-tagged WAVE2 in a 3–30% sucrose gradient. Most of the WAVE2 in control cells formed a protein complex, but a small population of monomeric WAVE2 was found (). Approximately half of the WAVE2 in cells stably expressing FLAG-WAVE2 was monomeric and not associated with Abi1 (). Therefore, the presence of monomeric WAVE2 in cells expressing FLAG-WAVE2 is caused by the ectopic expression of WAVE2. We also created a WAVE2 knockout MEF cell line stably expressing FLAG-WAVE2 in amounts similar to that of wild-type cells (). Most of the WAVE2 in WAVE2 knockout cells expressing FLAG-WAVE2 formed a complex () in a manner similar to wild-type cells (unpublished data). In all of the lysates we prepared, IRSp53 was not coeluted with WAVE2 (), indicating that IRSp53–WAVE2 association is very weak or transient. As reported previously, Abi1 was essential for WAVE2 complex formation because RNAi of Abi1 increased the amount of monomeric WAVE2 compared with that of the WAVE2 complex (). In contrast, RNAi of IRSp53 did not significantly affect WAVE2 complex assembly ( and Fig. S1 C). We next examined the stability of the WAVE2 complex by sequential sucrose gradient. The high molecular weight fraction of the FLAG-WAVE2 preparation obtained from a sucrose gradient was further subjected to another sucrose gradient (), showing that the WAVE2 complex was stable without any dissociation of the components in the second sucrose gradient. We then examined the dependence of protein complex formation on WAVE2 localization or EGF stimulation. The amount of WAVE2 in the smaller molecular weight fraction was not increased upon stimulation (; and Fig. S3, available at ). A small population of IRSp53 was found in the WAVE2 complex fraction and monomeric WAVE2 fraction, but it appeared to dissociate during fractionation (). Differences in the localization of WAVE2 (cytosolic or Triton X-100) did not result in differences in the amounts of the two populations of WAVE2 (; and Fig. S3). We added GTPγS-loaded Rac alone ( and Fig. S3) or a combination of GTPγS-loaded Rac, IRSp53, and PIP-containing liposomes ( and Fig. S3) to cell-purified WAVE2. However, no significant increase in WAVE2 in the smaller molecular weight fraction was observed. To examine the dependence of activities on WAVE2 complex formation, we analyzed the activity of each sucrose gradient fraction from Triton X-100 fractions of serum-starved cells in Arp2/3-mediated actin polymerization assays. As a control, actin polymerization was induced in sucrose gradients prepared in the absence of protein. Increased sucrose concentration decreased the rate of actin polymerization, but the maximum actin polymerization achieved was not altered (). When the rate of actin polymerization of each fraction was normalized to that induced by Arp2/3 alone, WAVE2 protein concentration, as determined by Western blotting, was associated with Arp2/3 activation (). Therefore, actin polymerization between monomeric WAVE2 and the WAVE2 complex did not differ significantly. Because we did not detect significant differences in Arp2/3 activation between the two WAVE2 populations, we examined the activity of the entire WAVE2 elution. We did not detect any significant alterations in actin polymerization in WAVE2 protein complexes alone (unpublished data). When we added purified Arp2/3, WAVE2 complex from various preparations showed Arp2/3 complex activation (). WAVE2 complex from the Triton X-100 fraction showed greater Arp2/3 complex activation than that from the cytosol fraction (). To assess differences in activity more quantitatively, we determined concentration-dependent activation curves for WAVE2 complexes from various preparations (). From these curves, we confirmed that WAVE2 complex from the Triton X-100 fraction had greater activity than that from the cytosol fraction. WAVE2 complex activity from the Triton X-100 fraction was similar to that of the VCA domain of WAVE2 (). EGF stimulation did not alter WAVE2 activity. Thus, cytosolic WAVE2 was not fully activated but membrane WAVE2 was. Because WAVE2 from the Triton X-100 fraction was fully active and WAVE2 in the protein complex had similar activity to that of WAVE2 not in the complex (), WAVE2 in the protein complex may be fully activated. We also measured the activity of the WAVE2 complex in MEFs. Almost all of the WAVE2 was complexed (). The activity of the WAVE2 complex prepared from the Triton X-100 fraction was also higher than that of the cytosolic WAVE2 complex, and no significant difference in the activities was induced by EGF stimulation (). To determine the absence or presence of IRSp53, we performed RNAi experiments to decrease IRSp53 expression. WAVE2 complex purified from the Triton X-100 fraction of cells with decreased IRSp53 expression by vector-based RNAi or by siRNA showed decreased Arp2/3 complex activation (; and not depicted). Restoration of IRSp53 in IRSp53 RNAi cells restored WAVE2 activity (unpublished data). The activity of the cytosolic WAVE2 complex was not altered (unpublished data), and the WAVE2 complex showed no detectable changes in silver-stained gels and in Western blot analysis in response to RNAi of IRSp53 (Fig. S1 C). We next examined whether exogenous IRSp53 activates the cytosolic WAVE2 complex. The addition of a 10-fold molar excess of purified IRSp53 to the cytosolic WAVE2 complex activated the Arp2/3 complex (), strongly indicating that IRSp53 activates WAVE2 at the membrane. To confirm the activation of WAVE2 by IRSp53, we prepared recombinant proteins (Fig. S4, available at ). We examined whether IRSp53 was able to activate recombinant WAVE2 alone or the reconstituted WAVE2 complex (Fig. S4). We also examined the effects of PIP-containing liposomes or control liposome composed of phosphatidylcholine (PC) and PI and that of lipidated Rac. Recombinant WAVE2 alone was activated by IRSp53 in the presence of GTPγS-loaded Rac and PIP-containing PC/PI liposomes (). However, activation was not observed with Cdc42 or with liposome lacking PIP, indicating that PIP, Rac, and IRSp53 are required for the activation of WAVE2. We also examined the activation of the reconstituted WAVE2 complex. Abi1, Nap1, and PIR121 did not show the activation of WAVE2 (). Lack of PIP, Rac, and/or IRSp53 decreased the activation of WAVE2. Cdc42 failed to activate the reconstituted WAVE2 complex in the presence of IRSp53 and PIP-containing liposomes (unpublished data). PIP-containing liposomes alone were not sufficient for activation of the WAVE2 complex. Rac did not activate the reconstituted WAVE2 in the presence of liposomes (). To confirm the PIP- and Rac-dependent activation of WAVE2 by IRSp53, various concentrations of IRSp53 were examined (). In the presence of PC/PI, IRSp53 did not activate WAVE2. However, in the presence of Rac and PIP/PC/PI, IRSp53 activated WAVE2 in a concentration-dependent manner. Thus, Rac requires IRSp53 to activate WAVE2 in the presence of liposomes. The A5 mutant of WAVE2, which is defective in PIP binding (), failed to be activated by IRSp53 in the presence of Rac and PIP-containing liposome (). Interestingly, IRSp53 alone activated WAVE2 in the absence of liposomes or the PIP-binding ability of WAVE2 (). Indeed, the activation of WAVE2 by IRSp53 was much stronger in the absence of liposomes (). Thus, liposomes may titrate out IRSp53 from WAVE2 by a nonspecific IRSp53–lipid association. PIP and Rac may then aid the association between IRSp53 and WAVE2 in the presence of liposomes. In the absence of liposomes, Abi1, Nap1, and PIR121 showed no significant WAVE2 activation (). The addition of Rac to WAVE2 alone or the reconstituted WAVE2 complex did not affect actin polymerization in the absence of liposomes ( and not depicted). PIP-dependent activation of WAVE2 by Rac and IRSp53 indicates that IRSp53 also binds to lipids. The crystal structure of the NH-terminal domain of IRSp53, termed the IMD or RCB domain, has been reported previously (). The overall surface of the IMD is positively charged, suggesting that the IMD binds to the electronegative inner leaflet of the cell membrane. We determined the binding of the RCB domain, the shorter fragment of the IMD, to lipids by ELISA (), in which various lipids were coated onto plates. The RCB domain bound to almost all of the lipid species, with increased binding corresponding to increased negative charges (). The RCB domain bound strongly to phosphatidylserine or weakly to PI, which are abundant lipids of the cell membrane (). The affinity of the RCB domains to various lipids indicated the nonselective binding of the RCB domain to the membrane. In the liposome cosedimentation assay, where bound proteins to the liposome were examined, no difference in affinity was observed between PIP-containing liposomes and PIP-lacking liposomes ( and not depicted). Without liposome, only a trace amount of RCB protein was found in the precipitates (). Therefore, although RCB domain bound to negatively charged lipids, there was no selective binding of the RCB domain to PIP or other phosphoinositides. IRSp53 appears to bind to the cell membrane via electrostatic interactions between the RCB domain and membrane. We next examined the association of Rac, IRSp53, and WAVE2 with PIP-containing liposomes. Recombinant IRSp53 bound to PIP/PC/PI liposomes () but also bound to PC/PI liposomes at this concentration. Rac purified from Sf9 cells also associated with liposomes, presumably through lipid modification (, A and B; ). WAVE2 alone also bound to PIP (), but in the presence of both activated (GTPγS loaded) Rac and IRSp53, the association of WAVE2 with PIP-containing liposomes increased fivefold (). Importantly, the increase in WAVE2 association in the presence of liposomes was not observed with GTPγS-loaded Cdc42, GDP-loaded Rac, or liposomes lacking PIP (). High molecular weight WAVE2 complex prepared by sucrose gradient from A431 cells stably expressing FLAG-WAVE2 also bound to PIP in the presence of IRSp53 and Rac (). Thus, PIP and Rac appear to greatly enhance the association between IRSp53 and WAVE2 or the WAVE2 complex on liposomes or cell membrane. The actin polymerization assay on PIP- or PI-coated beads confirmed the PIP-dependent association and activation of WAVE2 (Fig. S5, available at ). #text A431 cells were transfected with FLAG-tagged WAVE2 in pCMV-Tag 2 (Stratagene) or with vector alone. FLAG-tagged WAVE2 is functional because the transfection of FLAG-tagged WAVE2 into WAVE2 knockout cells eliminated WAVE2 deficits. After selection with G418, cells with stable FLAG-WAVE2 expression were grown. These cells did not show any significant change in growth or appearance compared with vector-transfected cells or parental cells. The amount of FLAG-WAVE2 expression was similar to the amount of endogenous WAVE2. 10 cells were plated onto 15-cm dishes and cultured for 5 d. After serum starvation overnight, some cultures were stimulated with EGF. To purify the WAVE2 complex from the cytosol (cytosol fraction), cells were harvested in buffer A containing 20 mM Tris-HCl, pH 7.5, 5 mM EDTA, 150 mM NaCl, 5 mM NaF, 1 mM NaVO, 1 mM PMSF, 10 μg/ml aprotinin, and 10 μg/ml leupeptin. To purify the WAVE2 complex, including that from the membrane fraction (Triton fraction), cells were harvested in buffer A supplemented with 1% Triton X-100 and 10% glycerol. Cells were sonicated and clarified by centrifugation at 20,000 for 20 min. The resulting supernatant was mixed with FLAG-agarose affinity gel (Sigma-Aldrich). After being mixed for 1 h, gels were washed with buffer A supplemented with 0.5% Triton X-100 and 10% glycerol and were washed with buffer XB containing 10 mM Hepes, pH 7.9, 100 mM KCl, 2 mM MgCl, 0.2 mM CaCl, and 5 mM EGTA. FLAG-WAVE2 complex was eluted with buffer XB plus 0.5 mg/ml FLAG peptide (Sigma-Aldrich). The eluted protein was analyzed or processed with sucrose gradient fractionation. 3–30% sucrose gradient was prepared in buffer XB supplemented with 10% glycerol, and the gradient was centrifuged at 200,000 for 15 h. After sucrose gradient, the WAVE2 complex at high molecular weight fraction (typically fraction No. 8–10) was used for further experiments after concentration by the Ultrafree filter (Millipore). Cells were transfected with FuGene 6 (Roche). 10 cells in 15-cm dishes were cultured for 24 h, and these cells were transfected for 24 h and cultured for 24 h. Cells were then starved overnight and used for assays. RNAi was performed as described previously with pSuper vector () or stealth RNAi (Invitrogen). The sequence of IRSp53 RNAi was GGAGCTGCAGTACATCGAC. Abi1 RNAi was performed as described previously (). Control RNAi was performed with pSuper vector with a sequence that caused no reduction of protein expression. Transfection efficiency was ∼80% when monitored by transfection with GFP-expressing plasmid. For cotransfection with WAVE2-GFP (), WAVE2-GFP and RNAi vector plasmids were mixed at 1:4 ratios to ensure the cotransfection of GFP-positive cells. Rac G12V and Cdc42 G12V were used as CA Rac and Cdc42, respectively, and were expressed in the pEF-BOS vector (). IRSp53 was tagged with Venus, a brighter variant of GFP (). Anti-IRSp53 and anti-Abi1 antibodies were raised by immunizing rabbits with full-length IRSp53 and Abi1 proteins purified from as GST fusion proteins. Anti-WAVE2 was raised as described previously (). Anti-GFP antibody was purchased from MBL International Corporation, and anti-FLAG M2 antibody was obtained from Sigma-Aldrich. Cell staining was performed as described previously, and actin filaments were stained with phalloidin (). Cells were then observed by confocal microscopy (Bio-Rad Laboratories). Anti-IRTKS antibody (M051-3) was purchased from MBL International Corporation and was found to be anti-IRSp53 antibody (Fig. S1 B). Cells were harvested in buffer A and sonicated. Lysates were clarified by centrifugation at 3,000 for 30 min to remove nuclei and debris. Supernatants were then clarified by ultracentrifugation at 400,000 for 45 min. The resulting supernatant was the cytosol fraction, and the pellet was the membrane fraction. Both were mixed with SDS-PAGE sample buffer to the same final volume, and the proteins were analyzed by Western blotting. All recombinant proteins were expressed in Sf9 cells with the Bac-to-Bac Baculovirus Expression System (Invitrogen). WAVE2 and Abi1 were tagged with GST at their COOH termini and expressed with pFastBac. PIR121 was tagged with GST at its COOH terminus and was coexpressed with Nap1 in the pFastBac Dual vector. WAVE2 tagged with GST and Abi1 were also coexpressed in the pFastBac Dual vector for better expression. Approximately half of WAVE2 that coexpressed with Abi1 was complexed with Abi1 after purification. Proteins were purified as described previously (). The interaction of WAVE2 with the coiled-coil region of Abi1 or the SH3 domain of IRSp53 using a dual polarization interferometer was investigated with AnaLight Bio 200 (Farfield) as described previously (). WAVE2 was cross-linked on sensortip coated with amine with BS3 cross-linker (Sigma-Aldrich). The Kd values were calculated from curve fitting. Actin polymerization assays were performed in XB buffer as described previously (). The Arp2/3 complex was used at a concentration of 50 nM. PIP, PC, and PI as well as PIP-coated beads were obtained from Echelon Biosciences. Liposome cosedimentation assays were performed as described previously (). For actin filament formation assays on PIP beads, 40 μg/ml WAVE2-GST coexpressed with Abi1, 100 μg/ml GST-IRSp53, and 30 μg/ml GTPγS- or GDP-loaded GST-Rac in 10 μl XB buffer were incubated with 2 μl PIP-coated beads. WAVE2 complex from A431 cells (0.5 μg/ml WAVE2) was also incubated with PIP-coated beads and Rac. After incubation at room temperature for 20 min, beads were washed with 100 μl XB buffer. Beads were then mixed with 50 μl XB buffer containing 50 nM Arp2/3 complex, 2 μM G-actin, 2 mg/ml BSA, and 2 unit/ml rhodamine-labeled phalloidin (Invitrogen) on ice. After incubation for 20 min on ice, 10 μl of bead-containing solution was placed between 18 × 18-mm BSA-coated glass and slide glass. After 10 min at room temperature, actin filaments were observed by phalloidin fluorescence. Fluorescence intensity was calculated by ImageJ software with the Oval profile plug-in (National Institutes of Health). All fluorescent images were taken through a microscope (Eclipse E600; Nikon) with a confocal microscopy system (Radiance 2000; Bio-Rad Laboratories) at room temperature. Fluorochromes used include AlexaFluor488, 546, and 647 and rhodamine (all purchased from Invitrogen). A 60× NA 1.40 oil immersion objective (Nikon) was used. Images were assembled with Adobe Photoshop. In each plate, photographs were cropped, and each fluorochrome was adjusted identically for brightness and contrast to represent the observed images. Time-lapse images were taken through a phase-contrast microscope (Axiovert S100; Carl Zeiss MicroImaging, Inc.) with a camera (CCD-782-Y/HS; Princeton Instruments). A 40× NA 1.30 FLUAR oil immersion objective (Carl Zeiss MicroImaging, Inc.) was used. The samples were electrophoresed in SDS-PAGE gels, transferred to polyvinylidene difluoride membrane, blocked with 5% nonfat dry milk in PBS and 0.1% Tween 20, incubated with primary antibodies, and incubated with alkaline-phosphatase–conjugated goat IgG secondary antibodies (Promega) followed by incubation with NBT/BCIP substrate (Roche). Resulting blots were scanned with a calibrated densitometer (GS-710; Bio-Rad Laboratories) and quantified with ImageJ software. Fig. S1 shows the WAVE2 complex and IRSp53 in A431 cells. Fig S2 shows the involvement of IRSp53 in ruffle formation in MEFs. Fig. S3 shows silver-stained gel images of fractionated WAVE2 in (F–K). Fig. S4 shows the reconstituted recombinant proteins used in this study. Fig. S5 shows the actin polymerization on PIP-coated beads by the WAVE2 complex or WAVE2 in the presence of IRSp53 and Rac. Videos 1 and 2 show cells treated with RNAi under stimulation. Online supplemental material is available at .

Cell migration is a spatiotemporally regulated process marked by the formation and disassembly of adhesions, which are complex supramolecular structures that connect the extracellular matrix to actin (). In the front of migrating cells, the continuous formation and disassembly of adhesions (adhesion turnover) is highly regulated and appears to be coupled to protrusion formation (). Although several regulators of adhesion turnover, including paxillin (), G protein–coupled receptor kinase–interacting protein 1 (GIT1; ), FAK (), Src (), and p21-activated kinase (PAK; ) are known, how these molecules act together to regulate adhesion turnover is not clear. Paxillin is a key regulator of adhesion turnover, as it interacts with several adhesion proteins () through its five NH-terminal LD domains, four COOH-terminal LIM domains, and multiple SH3 and SH2 binding domains (). The LD4 domain of paxillin binds FAK and GIT1 () and is implicated in adhesion turnover (; ). Paxillin targets GIT1 to the leading edge and adhesions (), and GIT1 overexpression sequesters paxillin from adhesions (), implicating this interaction in adhesion disassembly. Cells expressing an LD4-deletion paxillin mutant show perturbed migration and protrusion (). Also, GIT1 is a key regulator of protrusion (), raising the possibility that the paxillin–GIT1 interaction may regulate and thus link adhesion turnover and protrusion formation. GIT1, through its Spa2 homology domain (SHD), binds to the Rac exchange factor PAK-interactive exchange factor (PIX), which in turn binds the Rac effector PAK (), forming a trimolecular GIT1–PIX–PAK signaling complex (). GIT1 functions in part by targeting PIX and PAK to different subcellular zones (e.g., adhesions and the leading edge) in fibroblasts and epithelial cells (). This module is also implicated in neuronal synapse formation () and immunological synapse organization through local Rac and PAK activation (). PAK is also implicated in adhesion stability () through its kinase activity (), and the PIX–PAK complex is required for protrusion formation (). Phosphorylation is a likely mechanism by which paxillin–GIT1 binding is regulated, as it regulates the interaction of paxillin with other binding partners (; ). We recently identified a novel phosphorylation site at serine residue 273 in paxillin (S273-paxillin; ) that resides in its LD4 domain and thus can potentially regulate paxillin–GIT1 binding. We show that S273-paxillin phosphorylation is PAK-mediated and up-regulates adhesion turnover and protrusion by increasing paxillin–GIT1 binding and Rac activation. It also targets the components of the GIT1–PIX–PAK module near the leading edge to a population of small and highly dynamic adhesions. These adhesions exhibit very fast turnover and differ substantially in size and location from the adhesions studied earlier. Also, PAK activation is required for faster adhesion turnover and protrusion dynamics downstream of S273-paxillin phosphorylation through myosin. Collectively, we demonstrate a novel positive-feedback mechanism that regulates and couples adhesion and protrusion dynamics through the localization of a pax–GIT1–PIX–PAK complex. We recently used mass spectrometry to identify several utilized phosphorylation sites on paxillin (). Of these, S273 lies within the LD4 domain, and therefore its phosphorylation could regulate paxillin binding to FAK or GIT1. To determine whether S273-paxillin phosphorylation occurs in CHO-K1 cells, we assessed S273-paxillin phosphorylation in lysates prepared from CHO-K1 cells plated under migration-promoting conditions (see Materials and methods) by immunoblotting using a phospho–S273-paxillin–specific antibody that recognized phosphomimetic S273D-paxillin but not nonphosphorylatable S273A-paxillin (Fig. S1 a, available at ). Phospho–S273-paxillin antibody specificity was confirmed by phosphopeptide competition (Fig. S1 b). We observed a single band corresponding to the molecular mass of paxillin (∼68 kD) in lysates treated with CalyculinA (a Ser/Thr phosphatase inhibitor), with no detectable signal in untreated lysates (), suggesting that S273-paxillin is a labile phosphorylation site. During cell spreading, S273-paxillin phosphorylation was detected at low levels in suspended cells with an increase after 1 h of spreading until 3.5 h (Fig. S1, c and d). Paxillin is a substrate of PAK (), and the LD4 domain (N-LDELMAS*L-C) has a glutamic acid residue four residues upstream of S273; from peptide studies, this is a favored recognition determinant for PAK (). To determine whether S273-paxillin is phosphorylated by PAK, we performed a kinase assay with in vitro–synthesized FLAG-paxillin and either myc-tagged, kinase-dead (KD), or constitutively active (CA) PAK and assayed phospho–S273-paxillin levels by immunoblotting using the phospho–S273-paxillin antibody. Equal loading was confirmed by immunoblotting with anti-FLAG or anti-myc antibodies. Low levels of phospho–S273-paxillin were detected with KD-PAK, whereas a robust signal (eightfold increase) was observed with CA-PAK (). We also observed a similar eightfold increase in S273-paxillin phosphorylation when paxillin-GFP was immunoprecipitated from CHO-K1 cells coexpressing wild-type (WT)–paxillin-GFP and KD- or CA-myc-PAK (). Next, we assayed the binding of the S273-paxillin mutants with FAK and GIT1. Paxillin was immunoprecipitated from CHO-K1 cells coexpressing either WT-, S273D-, or S273A-paxillin-GFP and FLAG-GIT1 or myc-FAK. Similar levels of expression for paxillin-GFP and FLAG-GIT1 or myc-FAK were confirmed from immunoblots of the lysates (, d and e). GIT1 or FAK binding was assessed by immunoblotting. GIT1 binding to S273D-paxillin increased about threefold, whereas it decreased twofold with S273A-paxillin, compared with WT-paxillin (). In contrast, FAK binding changed modestly, if any, increasing by 25.0 ± 8.0% (P < 0.05) with S273A-paxillin and decreasing to 73.0 ± 3.0% (P < 0.01) with S273D-paxillin, compared with WT-paxillin (). To confirm the differential binding of phospho–S273-paxillin to GIT1, using an in vitro expression system, we synthesized FLAG-GIT1, untagged WT-paxillin, and myc-tagged KD- or CA-PAK. We then immunoprecipitated GIT1 and assessed the amount of phospho–S273-paxillin bound to GIT1 by immunoblotting. There was a sevenfold increase in the level of phospho–S273-paxillin bound to FLAG-GIT1 in the presence of CA-versus KD-PAK (). Similar results were obtained when we probed using an anti-paxillin antibody (unpublished data). This effect was specific because incubation with a phospho–S273-paxillin peptide abolished phospho–S273-paxillin–GIT1 binding, whereas the nonphospho–S273-paxillin peptide had no effect (). Together, our data demonstrate that S273-paxillin phosphorylation is directly mediated by PAK and regulates binding of paxillin to GIT1. The functional significance of S273-paxillin phosphorylation was determined by assaying its effects on cell migration. The migration rates for S273A-paxillin–expressing CHO-K1 cells showed a >40% decrease (20.0 ± 2.0 μm/h; = 30), whereas they increased by nearly 30% (45.0 ± 3.0 μm/h; = 30; P < 0.01) for S273D-paxillin, compared with WT-paxillin (35.0 ± 3.0 μm/h; = 30; P < 0.0001). shows the individual cell tracks of CHO-K1 cells expressing WT-, S273A-, or S273D-paxillin transposed to a common origin. When compared with WT-paxillin, S273A-paxillin–expressing CHO-K1 cells displayed shorter migration paths. In contrast, the migration paths of cells expressing S273D-paxillin were significantly longer. We also assayed the protrusiveness of S273D-paxillin–expressing cells that formed many protrusions (unpublished data), unlike S273A-paxillin–expressing cells. Protrusion rates, as assayed by kymography (; ), increased sevenfold (10.6 ± 1.3 μm/min) and reduced threefold (0.5 ± 0.1 μm/min) with S273D- and S273A-paxillin, respectively, when compared with WT-paxillin (1.5 ± 0.2 μm/min; ). The protrusion stability increased twofold (42.0 ± 5.0 min) and decreased slightly for S273A- and S273D-paxillin (17.0 ± 2.0 min), respectively, compared with WT-paxillin (22.0 ± 3.0 min; ). Therefore, S273-paxillin phosphorylation regulates the migration and protrusive activity of CHO-K1 cells. The LD4 domain of paxillin is implicated in the regulation of adhesion turnover (). for adhesion formation and disassembly in protrusive regions () of CHO-K1 cells expressing the S273-paxillin mutants. S273D-paxillin expression produced a large number of very small adhesions in the protrusions near the leading edge ( and Video 1, available at ). of <1 min (). In contrast, we did not observe these small, dynamic adhesions in the S273A-paxillin–expressing cells; instead, the majority of the adhesions were large and relatively stable (, , and Video 2). Cells expressing WT-paxillin showed an intermediate ratio of small to large adhesions (unpublished data). A similar effect on adhesion formation and disassembly was observed in paxillin-null ( ) mouse embryonic fibroblasts (MEFs) expressing S273D- and S273A-paxillin (unpublished data). To determine whether the rapid dynamics was a property of paxillin or of the adhesion as a whole, we imaged CHO-K1 cells coexpressing super-enhanced cyan fluorescent protein (seCFP)–S273D-paxillin and YFP-vinculin using time-lapse total internal reflection fluorescence (TIRF) microscopy (). for paxillin and vinculin disassembly decreased similarly (). for adhesion disassembly ( and ). Another adhesion marker, zyxin, colocalized with paxillin in the small adhesions in the protrusive areas of CHO-K1 cells (). Thus, S273-paxillin phosphorylation increases adhesion turnover in the protrusive regions of the cell. GIT1 is targeted to the leading edge through its interaction with paxillin (). MEFs using TIRF. MEFs, which exhibit large vinculin-containing adhesions (Fig. S2, available at ). In contrast, both GIT1 and vinculin localized in small adhesions near the cell periphery in WT MEFs (Fig. S2). Because paxillin–GIT1 binding increases upon S273-paxillin phosphorylation, we next asked whether the S273-paxillin mutants affected endogenous GIT1 localization. Both paxillin and GIT1 localized prominently near the leading edge in S273D-paxillin–expressing cells. However, in S273A-paxillin–expressing cells, which have only a few protrusions, GIT1 localized weakly in some large adhesions (). We saw a similar effect in CHO-K1 cells coexpressing YFP-GIT1 and seCFP-S273D-paxillin or -S273A-paxillin (unpublished data). To confirm the role of GIT1 in protrusion dynamics and adhesion turnover, we knocked down GIT1 expression in Rat2 fibroblasts using a GIT1 RNAi. Expression of a rat GIT1 RNAi dramatically reduced GIT1 expression compared with the control pSUPER vector alone (). GIT1 RNAi–expressing cells showed a fivefold decrease in the protrusion rate (0.03 ± 0.01 μm/min; , c and d) and a 2.5-fold increase in protrusion stability (; P < 0.0001) compared with the control (0.17 ± 0.02 μm/min; P < 0.0001). To determine whether this is due to loss of GIT1 expression, we rescued the rat GIT1 RNAi–expressing cells by coexpressing human GIT1, which is insensitive to the rat RNAi. This restored both the protrusion rate (0.21 ± 0.03μm/min; , c and d) and stability () to control levels. Also, in cells coexpressing GIT1 RNAi and WT-paxillin, the fraction of adhesions that turned over decreased approximately fourfold compared with control cells. for both adhesion formation ( = 3.5 ± 1.5 min) and disassembly ( = 4.5 ± 1.5 min) compared with control cells coexpressing the pSUPER vector and WT-paxillin ( <1 min; ). Thus, GIT1 directly regulates both protrusive activity and adhesion turnover. GIT1 targeting to the leading edge by S273D-paxillin prompted us to examine the subcellular localization of phospho–S273-paxillin, using the phospho–S273-paxillin–specific antibody. S273D-paxillin–expressing CHO-K1 cells revealed robust leading edge localization of phospho–S273-paxillin, whereas it was not readily detected in S273A-paxillin–expressing cells (). In CHO-K1 cells, endogenous phospho–S273- paxillin also localized in small puncta near the leading edge () that were not seen upon antibody preincubation with a competitive phosphopeptide (Fig. S3, available at ). Thus, S273-paxillin phosphorylation promotes both paxillin and GIT1 localization to the leading edge. To test whether PAK also functions downstream of S273-paxillin phosphorylation, we cotransfected KD-PAK and S273D-paxillin in CHO-K1 cells. KD-PAK strongly inhibited the S273D-paxillin phenotype (i.e., it reduced protrusive activity) and induced the formation of large adhesions, only a few of which disassembled over time (). of adhesion disassembly, comparable to that of the S273A-paxillin mutant (). CHO-K1 cells coexpressing KD-PAK and WT-paxillin also displayed reduced protrusiveness (Video 3, available at ) and the formation of large and stable adhesions (), only a few of which disassembled (8 ± 5%). for adhesion formation and disassembly for the adhesions that did turn over increased two- and threefold, respectively, compared with cells expressing WT-paxillin alone (). In contrast, CHO-K1 cells coexpressing CA-PAK and WT-paxillin were more protrusive (Video 4) and showed paxillin localization to numerous small and dynamic adhesions () near the leading edge. of <1 min for both adhesion formation and disassembly (). Thus, CA-PAK mimicked the S273D-paxillin phenotype. To show that activated PAK resides in the vicinity of dynamic adhesions, we determined its localization using a phosphospecific antibody that recognizes T423-phosphorylated active PAK (). Cells expressing S273D-paxillin showed robust phospho-PAK localization near the leading edge, whereas it did not show leading edge localization in cells expressing S273A-paxillin (). Thus, S273-paxillin phosphorylation promotes phospho-PAK localization near the leading edge. We next asked what regulated adhesion turnover downstream of PAK. Myosin II is regulated by PAK, either through direct phosphorylation of myosin light chain (MLC; ) or indirectly, though phosphorylation of MLC kinase (). To determine whether the turnover of the small paxillin-containing adhesions is dependent on myosin, we treated CHO-K1 cells coexpressing CA-PAK and WT-paxillin with 50 μM blebbistatin, a specific inhibitor of myosin II ATPase activity (). Immediately after exposure, the cells stopped protruding and the small adhesions stabilized and did not turn over; upon washout, the fast turnover rate recovered (), pointing to myosin as a key effector of this pathway. Our working hypothesis is that PAK is linked to paxillin indirectly via PIX, which in turn binds to GIT1 (). To test this hypothesis, we cotransfected CHO-K1 cells with S273D-paxillin and various mutants that disrupt the ternary GIT1–PIX–PAK module, namely, GIT1ΔSHD, PIXΔGBD, or PIXΔSH3. of adhesion disassembly compared with WT-GIT1 control (). However, cells coexpressing a GIT1 binding–deficient PIX mutant (PIXΔGBD) and S273D-paxillin still formed small and dynamic adhesions (Video 5, available at ), which are indistinguishable from WT-PIX control ( and ) and likely the result of PIX mislocalization, as reported by others (). In contrast, PIXΔSH3, a PAK binding–deficient PIX mutant (), abrogated the S273D phenotype, i.e., it led to decreased protrusiveness (Video 6) and formation of large and stable adhesions (). for adhesion disassembly compared with WT-PIX control ( and ). These results strongly implicate a requirement for GIT1–PIX–PAK interaction for fast adhesion dynamics. We examined PIX localization in CHO-K1 cells expressing the S273-paxillin mutants. PIX localized robustly to a region near the leading edge in CHO-K1 cells expressing S273D-paxillin. In contrast, PIX leading edge localization was not observed in S273A-paxillin–expressing cells (). These data indicate that PIX localizes to a region near the leading edge in response to S273-paxillin phosphorylation. PIX exhibits exchange factor activity for small GTPases, including Rac, which in turn can promote protrusive activity and adhesion formation (). To determine whether this activity is required for the effects of PIX, we used a guanine nucleotide exchange factor (GEF)–deficient mutant, PIX-LL (L238R and L239S; ). for adhesion disassembly when compared with S273D-paxillin alone (). The role of Rac was further confirmed by cotransfecting CHO-K1 cells with dominant-negative N17-Rac and S273D-paxillin. for adhesion disassembly ( and ). On the other hand, CHO-K1 cells coexpressing V12-Rac and WT-paxillin exhibited numerous small adhesions around the cell periphery ( and Video 8). However, the cells were not protrusive and the adhesions were not dynamic (, , and Video 8). We then transfected CHO-K1 cells with WT-paxillin and Tiam1, a potent Rac GEF. Tiam1 expression, like PIX, led to the formation of small adhesions () that turned over rapidly (). These data suggest a requirement of Rac GTPase cycling for fast adhesion dynamics. We next assayed for active Rac in cells expressing the S273-paxillin phosphomutants using a GST–p21 binding domain (PBD) pull-down assay. CHO-K1 cells coexpressing FLAG-WT-Rac and WT-, S273A-, S273D-paxillin-GFP or GFP vector alone were lysed 24 h after transfection, and Rac activity was assayed by its binding to the GST fusion of the PBD (). The positive control with WT-paxillin-GFP and V12-Rac exhibited maximum Rac-GTP binding to the GST-PBD bait. S273A-paxillin-GFP expression induced a marked decrease of bound Rac (approximately fourfold; = 3), whereas the GFP control and WT-paxillin-GFP showed comparable levels of bound Rac. Rac activation increased eightfold with S273D-paxillin when compared with S273A-paxillin and 1.5-fold when compared with WT-paxillin. Thus, S273-paxillin phosphorylation induces Rac activation. Using TIRF, we observed numerous small, highly dynamic adhesions near the leading edge of S273D-paxillin–expressing CHO-K1 cells ( and Video 9, available at ) and in other cell types, including NIH 3T3 fibroblasts and WT MEFs expressing S273D-paxillin (not depicted). of 30 ± 2 s for adhesion disassembly and 20 ± 1 s for assembly for S273D-paxillin (). To find out whether these adhesions were present under normal conditions, we examined cells expressing WT-paxillin using TIRF. An array of small and transient adhesions that are not readily apparent using wide-field configurations lined the region near the leading edge in the protrusive regions ( and Video 10). of 16 ± 2 s and 25 ± 2 s for adhesion formation and disassembly, respectively, suggesting that these adhesions are similar to those seen in S273D-paxillin–expressing cells (). Quantitative measurements of the adhesion size from their intensity profiles gave a diameter of 0.5 ± 0.1 μm, which did not vary with intensity, suggesting that they are subresolution. To determine the location of these adhesions with respect to the leading edge, we superimposed differential interference contrast (DIC) and fluorescence images of CHO-K1 cells expressing WT-paxillin. These adhesions were located 0.5–1.0 μm behind the leading edge (). Furthermore, cells coexpressing WT-paxillin-GFP and actin–monomeric RFP showed actin localization in a fluorescent band at the leading edge, whereas the small paxillin-containing adhesions localized at the boundary between the actin band and the remainder of the lamellipod (). In addition, interference reflection microscopy (IRM) showed that these adhesions are in very close proximity to the underlying glass coverslip, whereas the leading edge containing the actin band is above the surface (). Using TIRF, these small adhesions were also seen endogenously in CHO-K1 cells immunostained for paxillin and GIT1 (). In addition, these adhesions were observed in highly protrusive tumor-derived cells expressing WT-paxillin-GFP (e.g., B16 melanoma and MDA-MB-231 breast carcinoma cells), whereas a less protrusive MCF7 cell line showed larger and more stable adhesions (unpublished data). of adhesion formation and disassembly (). The presence of these small paxillin-containing adhesions and their dynamics in these cells suggests that these are a salient feature of highly protrusive cell types. Adhesion turnover at the front of a migrating cell appears to regulate migration by localizing and stabilizing the protrusion as the cell extends forward (adhesion–protrusion coupling). , , and cells exhibit impaired adhesion disassembly, protrusion, and migration (). We describe a novel PAK-mediated phosphorylation pathway that accelerates adhesion turnover and protrusion dynamics in migrating cells. Phosphorylation of paxillin on S273 by PAK promotes the localization of a signaling module containing the adaptor GIT1, the Rac GEF PIX, and the active form of the Rac effector PAK to a region near the leading edge. We conclude that PAK acts both upstream and downstream of S273-paxillin phosphorylation, in a positive-feedback loop, providing a mechanism for adhesion–protrusion coupling. It is tempting to speculate that the pax–GIT1–PIX–PAK module localizes Rac activity near the leading edge through the joint presence of PIX and PAK. This is consistent with previous studies that show Rac localization near the leading edge () and our own observation that Rac activation and cycling is required for rapid adhesion turnover. Genetic studies in also implicate a positive role for paxillin in the regulation of Rac activity (). Finally, recent evidence shows that the Rac-dependent spatial localization of protrusive activity is mediated by active PAK through the recruitment of PIX (; ). Our results demonstrate that active PAK is a key effector for fast adhesion turnover and protrusion dynamics after S273-paxillin phosphorylation. These observations are consistent with previous studies that have hinted at a role for PAK in adhesion stability () and shown active PAK localization near the leading edge (). We have extended these observations by clarifying the function and location of active PAK, demonstrating its direct role in adhesion turnover, and providing a pathway for regulating its localization. How does PAK regulate the rapid turnover of the highly dynamic adhesions? The inhibition of adhesion turnover by blebbistatin suggests that myosin is a major effector. PAK is known to affect myosin activity both by inhibiting MLC kinase () and through the direct phosphorylation of MLC (). Although the ATPase activity of myosin II generates contractile forces that are thought to mediate adhesion assembly (), there is also evidence that such contractility is involved in adhesion disassembly (). Thus, there are several possibilities for myosin-mediated regulation of adhesion turnover. The effect of PAK on protrusion also has multiple possibilities. A likely candidate is its effector LIM kinase, which regulates actin dynamics by inactivating actin depolymerizing factor/cofilin family members (). Modulating adhesion to the substratum is another possibility, as net protrusion is thought to result from the balance between actin treadmilling, retrograde actin flow, and the interaction of the actin filaments with adhesions (). Increased interaction with adhesions leads to more traction, less retrograde flow, and, hence, higher protrusion rates (). Although our results show a positive regulatory role for GIT1 at the leading edge, a recent study () in α integrin–expressing cells demonstrates an ARF–GTPase-activating protein domain–mediated inhibitory role for GIT1 at the sides and rear of migrating cells. This suggests that GIT1 serves complementary roles depending on the spatial cellular context. Though the events controlling adhesion signaling and migration via the α or α integrins differ substantially (), our studies do not exclude a role for the ARF–GTPase-activating protein domain of GIT1 in regulating protrusion. Our results show that S273-paxillin is a highly labile and regulated phosphorylation site. Interestingly, paxillin interacts directly with the serine/threonine phosphatase PP2A (), whose inhibition is observed in certain types of cancer and results in hyperphosphorylation of paxillin serine residues and dissolution of FAK–Src–paxillin complexes (; ). This suggests that S273-paxillin phosphorylation might also be under regulation by phosphatases opening a new facet of adhesion turnover regulation through paxillin dephosphorylation. Finally, the small adhesions that we observed have interesting properties that distinguish them from other adhesions. They are small (<0.5 μm), turnover rapidly (<1 min), contain GIT1 (as well as other components, such as FAK, vinculin, and zyxin), and reside in a region ∼1 μm behind the leading edge, which also contains phospho-PAK and PIX. They are present in the protrusive regions of normal cells and also highly motile tumor cells. Interestingly, rapidly locomoting cell types such as keratocytes (), neutrophils (), and macrophages () do not show highly organized adhesions. In contrast, most other adhesions are large, elongated, and centrally located; turnover with slower rates (several minutes); and do not have prominent concentrations of GIT1. Slower moving cells, e.g., fibroblasts, form these larger adhesions, whose presence corresponds with a decrease in the migration rate (). Therefore, we propose that these small, dynamic adhesions drive the migration of highly motile cells and therefore deserve intense study. CHO-K1 cells were cultured in low-glucose DME supplemented with 10% FBS, 4 mM -glutamine, 1 mM sodium pyruvate, 1% (vol/vol) nonessential amino acids, and penicillin/streptomycin and transfected with 0.25–1 μg DNA using Lipofectamine (Invitrogen). MEFs and Rat2 fibroblasts were cultured in high-glucose DME supplemented with 10% FBS and penicillin/streptomycin. Rat2 cells were transfected with 0.5–3.0 μg DNA using nucleofection. Cells were incubated 24–72 h before observation. Quickchange mutagenesis kit (Stratagene) was used to introduce the S273 mutations into paxillin-GFP (). 5′-GAGCTGATGGCGGCCCTCTCTGAC-3′ and 5′-GTCAGAGAGGGCCGCCATCAGCTC-3′ primers (forward and reverse) were used to generate S273A-paxillin. For S273D-paxillin, the primers used were 5′-GAGCTGATGGCGGACCTCTCTGAC-3′ and 5′-GTCAGAGAGGTCCGCCATCAGCTC-3′. Both mutations were confirmed using the sequencing primer 5′-CGTGTCAACGCCAGTCAGCAG-3′. seCFP-WT-paxillin was made by subcloning paxillin cDNA from paxillin-pcDNA3.1 Zeo () into the seCFP vector pKseCFP (a gift from A. Miyawaki, RIKEN, Saitama, Japan) using BamHI and EcoRI restriction sites. S273A and -D mutations were similarly introduced into seCFP-WT-paxillin using the Quickchange mutagenesis kit. The FLAG-paxillin, untagged WT-paxillin, FLAG-GIT1, GIT1ΔSHD, GIT1RNAi, PIXΔGBD, and PIXΔSH3 constructs were described previously (; , ; ). Myc-FAK (J.T. Parsons, University of Virginia, Charlottesville, VA), CA- and KD-myc-PAK1 (J. Chernoff, Fox Chase Cancer Center, Philadelphia, PA), Rac1 (A. Hall, University College London, London, UK), Tiam1 (J. Collard, The Netherlands Cancer Institute, Amsterdam, Netherlands), HA-ßPix (C. Turner, State University of New York Upstate Medical University, Syracuse, NY), HA-PIX-LL (L. Santy and J. Casanova, University of Virginia, Charlottesville, VA), and YFP-vinculin (S. Craig, The Johns Hopkins School of Medicine, Baltimore, MD) constructs were all gifts. The following primary antibodies were used: paxillin (BD Biosciences), c-myc 9E10 (Santa Cruz Biotechnology, Inc.), FLAG M2 (Stratagene), and GFP A-11122 (Invitrogen). The B71 zyxin (M.C. Beckerle, University of Utah, Salt Lake City, UT), phospho-PAK (J. Chernoff), and ßPIX antibodies (B. Xiao, The Johns Hopkins University, Baltimore, MD) were gifts. The GIT1 polyclonal antibody was previously described (). In vitro transcription–translation was performed using the TnT T7-coupled reticulocyte lysate system. 0.5 μg FLAG-paxillin and 0.75 μg of either CA- or KD-myc-PAK1 in T7-containing plasmids were transcribed and translated for 90 min at 30°C. After 90 min, kinase buffer containing 20 mM Hepes, 10 mM NaCl, 1 mM MgCl, 1 mM MnCl, and 20 μM ATP was used. The reaction was then allowed to continue for another 30 min at 30°C. For binding experiments, untagged WT-paxillin and FLAG-GIT1 were synthesized using the TnT system and incubated for 30 min with either CA- or KD-myc-PAK1 in kinase buffer. Phosphopeptide competition was performed by preincubating in vitro–synthesized mixtures of untagged WT-paxillin, FLAG-GIT1, and CA-myc-PAK1 with 500 molar excess of the phospho– or nonphospho–S273-paxillin peptide for 30 min. Immunoprecipitation was performed using anti-FLAG M2-conjugated agarose. Proteins were separated by 7.5% SDS-PAGE; transferred to Immobilon membranes; and probed with the phospho–S273-paxillin, anti-FLAG, anti-myc, or anti-paxillin antibodies. Cells were grown to 80–90% confluency, washed with ice-cold PBS, and lysed with ice-cold lysis buffer (25 mM Tris-HCl, pH 7.4, 100 mM NaCl, 0.5% NP-40, and protease inhibitors). The lysates were incubated on ice for 30 min and clarified by centrifugation (12,000 for 5 min). Equivalent amounts of the lysates were precleared with 30 μl mouse IgG agarose for 1.5 h at 4°C, followed by incubation with 2 μg of the anti-GFP polyclonal antibody for 1.5 h at 4°C. Complexes were incubated with protein A–agarose for 1 h and washed three times with ice-cold lysis buffer. The immunoprecipitates were analyzed by SDS-PAGE on 10% slabs, transferred to nitrocellulose, and detected by Western blot analysis. Protein binding levels were compared by densitometry of scanned Western blots using ImageJ software (NIH). Background-corrected densities were measured and normalized to GFP-paxillin densities run on the same gel. The GST-PBD fusion protein was purified with glutathione–Sepharose beads, and assays were performed as described previously (). CHO-K1 cells were cotransfected with FLAG-WT-Rac and GFP vector, WT-paxillin-GFP, S273A- paxillin-GFP, or S273D-paxillin-GFP. A positive control with CHO-K1 cells cotransfected with FLAG-V12-Rac and WT-paxillin-GFP was included. Lysates were collected 24 h after transfection and processed as described elsewhere (). Cells were plated on fibronectin-coated glass-bottomed 35-mm dishes () in CCM1 medium and fixed with 3% formaldehyde for 15 min. 0.15 M glycine was added for 10 min to stop the fixation followed by permeabilization with 0.2% (vol/vol) Triton X-100 for 5 min at room temperature. For immunostaining phospho–S273-paxillin, PIX, and phosphoactive PAK, the cells were fixed for 3–5 min with 3% formaldehyde, followed by chilled methanol for 15 min. After each step, the cells were washed three times with PBS, blocked with 2% BSA in PBS for 1 h, and incubated with primary antibodies for 1 h, followed by fluorescently conjugated secondary antibodies for 1 h at room temperature. The antibodies were diluted in PBS containing 2% BSA. Slips were mounted on slides with Vectashield mounting media (Vector Laboratories). For TIRF observation, coverslips were mounted using Slowfade antifade kit (Invitrogen). Protrusion parameters were quantified using kymography (). For CHO-K1 and Rat2 cells, images were captured at 30-s intervals for 60- and at 5-min intervals for 5 h, respectively. Kymographs were generated using ImageJ or Metamorph software along 1-pixel-wide regions oriented along the protrusion direction and perpendicular to the lamellipodial edge. Straight lines were drawn from the beginning to the end of single protrusion events in the kymographs; retraction events were ignored. Protrusion rates and protrusion stability were calculated from the slopes and x axis projection distance of these lines, respectively. A minimum of eight cells per treatment and at least three protrusions per cell from three independent experiments were analyzed. ImageJ or Metamorph software were used to measure the background-corrected fluorescent intensity of individual adhesions over time from cells expressing fluorescently tagged paxillin or vinculin (). Paxillin and vinculin incorporation into and departure from adhesions were linear on semilogarithmic plots of the background-corrected fluorescent intensity as a function of time. for formation and disassembly was determined from the slopes of these graphs. determination, measurements were obtained for 15−20 individual adhesions on four to six cells from three independent experiments. Fig. S1 shows that phospho–S273-paxillin antibody is specific and phospho– S273-paxillin levels are up-regulated during cell spreading. Fig. and WT MEFs. Fig. S3 shows endogenous phospho–S273-paxillin staining in CHO-K1 cells subject to peptide competition. Videos 1 and 2 show S273D- and S273A-paxillin-GFP dynamics in CHO-K1 cells. Videos 3 and 4 show WT-paxillin-GFP dynamics in CHO-K1 cells coexpressing KD- or CA-PAK and WT-paxillin. Videos 5–7 show S273D-paxillin-GFP dynamics in CHO-K1 cells coexpressing S273D-paxillin and PIXΔGBD, PIXΔSH3, or N17-Rac. Video 8 shows WT-paxillin dynamics in a CHO-K1 cell coexpressing V12-Rac and WT-paxillin. Videos 9 and 10 are TIRF videos of CHO-K1 cells expressing S273D- or WT-paxillin-GFP, respectively. Online supplemental material is available at .

Semaphorins comprise a large family of secreted and transmembrane molecules that play central roles in axon guidance in the developing nervous system (; ). The function of semaphorins is mediated by plexins, which are classified into four subfamilies: Plexin-A, -B, -C, and -D (). Semaphorins were originally identified as repulsive axonal guidance molecules, but they have recently been shown to regulate integrin-mediated cell migration in a variety of cells (). Sema3A exerts an essential permissive role in the execution of vasculature remodeling by inhibiting integrin-mediated adhesion of endothelial cells to the ECM (). Activation of Plexin-B1 negatively regulates integrin-based cell adhesion and migration of NIH-3T3 cells (). Plexin-C1 inhibits integrin-mediated adhesion and chemokine-induced migration of dendritic cells (). Thus, semaphorin/plexin signaling plays an important role in the migration of a variety of cells. However, the molecular mechanisms underlying the inhibition of integrin-mediated cell migration by semaphorins through plexins remain unclear. Rho family small GTPases are signal transduction molecules that remodel the actin cytoskeleton and play fundamental roles in numerous cellular processes (). The small GTPase Rnd1, a constitutively active GTPase (), is known to interact directly with the cytoplasmic domain of Plexin-B1 (). We recently revealed that Plexin-B1 functions as an R-Ras GTPase-activating protein (GAP) and directly and specifically down-regulates R-Ras activity in response to Sema4D, inducing repulsive response in hippocampal neurons, and that the expression of R-Ras GAP activity of Plexin-B1 requires Rnd1 association with the receptor (). Furthermore, expression of constitutively active R-Ras prevents growth cone collapse induced by Sema4D/Plexin-B1 as well as Sema3A/Plexin-A1, whereas R-Ras siRNA caused a growth cone collapse similar to those induced by semaphorins (). Integrins are a family of α/β heterodimeric cell surface receptors that bind to the ECM, such as collagens and fibronectins, and play a central part in regulating cell growth, survival, migration, and tumor metastasis (). Activation of integrins is essential for cell adhesion and cell migration, and several studies show that the Ras family of small GTPases regulates integrin activity (). Among the Ras family GTPases, activated R-Ras was shown to induce integrin activation and increase cell adhesion and matrix assembly, suggesting that R-Ras plays an important role in the regulation of integrin activity (; ). However, how R-Ras activity is regulated and how R-Ras activates integrins remain obscure. Significantly, Sema4D was the first extracellular stimulus shown to influence the activity of R-Ras. These facts collectively prompted us to speculate that plexins regulate integrin-mediated cell migration by their R-Ras GAP activity. In this study, we characterized the role of R-Ras downstream of Sema4D/Plexin-B1 in regulation of integrin activation and cell migration. The activation of R-Ras by ECM is required for ECM-mediated integrin activation and cell migration, and Sema4D/Plexin-B1 inhibits integrin activation and cell migration through R-Ras GAP activity. We also revealed that down-regulation of phosphatydylinositol 3-kinase (PI3-K) activity is responsible for Sema4D/Plexin-B1–induced suppression of β integrin activity and cell migration. We examined the effect of Sema4D on integrin-mediated migration of PC12 cells in a cell migration assay (). Transwell chambers were coated on the lower side with varying concentrations of collagen I. PC12 cells exhibited a collagen concentration–dependent promotion of cell migration, which was antagonized by Sema4D. The collagen-dependent PC12 cell migration is mediated by α and β integrin subunits, as functional blocking antibodies against α () and β () integrin subunits strongly impaired the migration. These results indicate that Sema4D antagonizes the collagen receptor, α/β integrin–dependent PC12 cell migration. R-Ras is implicated in integrin-mediated cell migration, and expression of a constitutively active form of R-Ras has been shown to stimulate cell migration (). We previously reported that Sema4D stimulation down-regulates NGF-stimulated R-Ras activity via the R-Ras GAP activity of Plexin-B1 to induce neurite retraction (). We next tested whether stimulation of PC12 cells with collagen and Sema4D affects R-Ras activity. PC12 cells were plated onto collagen-coated dishes and lysed, and the lysates were incubated with the GST-fused Ras binding domain of c-Raf-1 (GST-RBD) to pull down activated R-Ras (). As shown in , cells plated on collagen-coated dishes showed a collagen concentration–dependent increase in endogenous R-Ras activity, whereas those kept in suspension or plated onto the non–integrin-dependent substrate poly--lysine did not. Furthermore, the collagen-dependent activation of R-Ras was inhibited by a functional blocking antibody against β integrins, P5D2, and was enhanced by affinity-related activation of β integrins by the monoclonal antibody 8A2, which mechanically induces a high-affinity state of β integrins. These data suggest that β integrins are required for R-Ras activation upon ECM-mediated adhesion. Sema4D stimulation strongly inhibited the collagen-induced activation of R-Ras, and affinity-related activation of β integrins by the 8A2 antibody attenuated the inhibitory effect of Sema4D on ECM-mediated R-Ras activation. R-Ras is known to regulate β integrin activation (). To examine the effect of Sema4D on β integrin activity, we measured the activity of β integrins in cells with or without Sema4D stimulation by the immunoprecipitation assay with the monoclonal antibody against active conformations of β integrins, HUTS-4, which detects hybrid domain swing-out in β integrins, a process most commonly associated with ligand binding (). Sema4D antagonized the collagen-dependent activation of β integrins (). Inhibition of β integrin activity by Sema4D was also observed in the ELISA using the HUTS-4 antibody, which was performed under a detergent-free condition (). To further ascertain that Sema4D indeed affects the activity of β integrins, we performed flow cytometry analysis using the HUTS-4 antibody. As shown in , cells treated with Sema4D showed a decrease in the level of HUTS-4 binding (FITC staining). Mn treatment, which induces the activation of β integrins, resulting in the effective interaction with the ECM ligands and increased HUTS-4 binding (), completely overcame the Sema4D-induced decrease in HUTS-4 binding. These results suggest that decreased HUTS-4 binding induced by Sema4D is due to affinity modulation of β integrins. FAK is known to be autophosphorylated at tyrosine upon integrin activation (), and FAK phosphorylation downstream of β integrins is the important step for integrin-mediated cell migration (; ).As shown in , Sema4D inhibited the collagen-mediated FAK tyrosine phosphorylation. These results suggest that Sema4D inhibits ECM-mediated activation of R-Ras and β integrins. We also confirmed the involvement of the endogenous Plexin-B1 receptor in Sema4D-dependent inhibition of R-Ras activity and integrin functions. As shown in , both Sema4D-dependent inhibition of collagen-mediated activation of R-Ras and cell migration were blocked by the monoclonal antibody against Plexin-B1, which recognizes the extracellular ligand binding region of the receptor. These results suggest that Sema4D through Plexin-B1 inhibits ECM-mediated activation of R-Ras, functional activation of β integrins, and inhibition of cell migration. We tested whether Sema4D-mediated inhibition of cell migration is mediated by suppression of β integrin activity. PC12 cells preincubated with 5 μg/ml β integrin activating monoclonal antibody (8A2) were subjected to the transwell assay. We tested the migration at relatively low concentrations of collagen (∼3.0 μg/ml) because this antibody inhibits cell migration at high concentration of the ECM ligands by freezing β at a high-affinity state (). As shown in , affinity-related activation of β integrins by 8A2 stimulation overcame the inhibitory effect of Sema4D on collagen-mediated cell migration, whereas a control IgG2a antibody did not. These results suggest that the inhibition of β integrin activity is required for the inhibition of cell migration by Sema4D. We recently reported that Plexin-B1 encodes R-Ras GAP within its cytoplasmic tail and that Plexin-B1 associated with the Rho family GTPase Rnd1 functions as a specific GAP toward R-Ras (). We examined whether Sema4D/Plexin-B1–Rnd1–mediated R-Ras GAP activity suppresses adhesion-dependent R-Ras activation. COS-7 cells expressing R-Ras–wild type (WT) were plated onto fibronectin-coated dishes or nonadherent control dishes and lysed, and the lysates were incubated with GST-RBD to pull down activated R-Ras. The same cell lysates were also used for the immunoprecipitation assay using HUTS-4 for measurement of the activity of β integrins. In COS-7 cells, fibronectin stimulation activated both R-Ras and β integrins (). As shown in , expression of Plexin-B1–WT and Rnd1 inhibited the fibronectin-mediated R-Ras activation in the presence of Sema4D. However, this inhibitory effect was not observed in cells expressing Plexin-B1–GGA, a mutant lacking the ability to associate with Rnd1, or Plexin-B1–RA, a mutant lacking primary and secondary arginine residues required for the catalytic activity of GAP. These results suggest that Sema4D/Plexin-B1–Rnd1–mediated R-Ras GAP activity inhibits adhesion-dependent R-Ras activation. To examine the effect of Sema4D/Plexin-B1–Rnd1–mediated R-Ras GAP activity on β integrin activity, we measured the activity of β integrins in cells expressing Plexin-B1 and Rnd1 with or without Sema4D stimulation by the immunoprecipitation assay. As shown in , expression of Plexin-B1 and Rnd1 strongly inhibited the fibronectin-mediated β integrin activation in the presence of Sema4D, whereas inhibition of β integrin activation was not observed in cells expressing Plexin-B1–GGA or Plexin-B1–RA. The same results were also obtained by the ELISA using the HUTS-4 antibody performed under a detergent-free condition (). We also confirmed the results by flow cytometry analysis. COS-7 cells transiently cotransfected with GFP-Rnd1 and Plexin-B1 were treated with Sema4D, and GFP expression and HUTS-4 binding (phycoerythrin [PE] staining) were simultaneously analyzed by two-color flow cytometry. HUTS-4 binding (PE staining) was analyzed on a gated subset of cells positive for GFP expression to discriminate β integrin activity of transfected cells from that of untransfected cells. As shown in , a Sema4D-dependent decrease in HUTS-4 binding was observed in Plexin-B1–WT and Rnd1-expressing cells. However, cells coexpressing Rnd1 with Plexin-B1–RA or Plexin-B1–GGA, which lacks R-Ras GAP activity, did not show a Sema4D-dependent reduction in HUTS-4 binding. In addition, Sema4D/Plexin-B1–Rnd1 also inhibited the fibronectin-mediated FAK tyrosine phosphorylation, whereas inhibition of FAK phosphorylation was not observed in cells expressing Plexin-B1–GGA or Plexin-B1–RA (). These results suggest that Sema4D/Plexin-B1–Rnd1–mediated R-Ras GAP activity inhibits adhesion-dependent activation of R-Ras and thereby inhibits functional activation of β integrins. We next examined whether regulation of R-Ras activity plays key roles in the ECM-mediated activation of β integrins. As shown in , in untransfected cells, activity of β integrins was increased upon adhesion to fibronectin. This activation was completely blocked by the down-regulation of endogenous R-Ras activity by the expression of the myristoylated GAP domain of p98–R-RasGAP (Myr–R-RasGAP), which exhibits a specific GAP activity toward R-Ras (). R-Ras is implicated in integrin regulation, and the constitutively active form of R-Ras has been shown to increase the affinity of β integrins for fibronectin () and to stimulate cell migration (). Expression of R-Ras–QL actually induced remarkable activation of β integrins, and this was not further enhanced by fibronectin. We also tested whether R-Ras activity affects FAK tyrosine phosphorylation. As shown in , expression of Myr–R-RasGAP completely blocked the fibronectin-induced FAK phosphorylation, whereas R-Ras–QL markedly stimulated FAK phosphorylation independent of fibronectin, indicating that endogenous R-Ras activity is also required for ECM-mediated FAK phosphorylation. We further confirmed requirement of R-Ras in the ECM-mediated functional activation of β integrins. We reduced expression of R-Ras in COS-7 cells by R-Ras–specific siRNA expression vector and examined the effect on the activation of β integrins and phosphorylation of FAK. As shown in (C and D), expression of R-Ras siRNA effectively reduced endogenous R-Ras protein, and reduction in R-Ras protein blocked both the fibronectin-dependent activation of β integrins and phosphorylation of FAK. The ELISA using HUTS-4, under detergent-free conditions, also confirmed suppression of β integrin activation by inactivation of R-Ras by expression of Myr–R-RasGAP or knockdown of R-Ras by R-Ras RNA interference (). We also confirmed these results by two-color flow cytometry. COS-7 cells transiently transfected with Myr– R-RasGAP or an R-Ras siRNA together with GFP were stained with HUTS-4, and HUTS-4 binding (PE staining) was analyzed on GFP-positive cells. As shown in , the level of HUTS-4 binding was reduced in cells expressing Myr–R-RasGAP or R-Ras siRNA. These results demonstrate that activation of the endogenous R-Ras protein is essential for the ECM-mediated functional activation of β integrins. We next examined the effect of Sema4D/Plexin-B1 signaling on integrin-mediated cell migration. COS-7 cells expressing a control GFP alone exhibited a fibronectin concentration–dependent promotion of cell migration, and ectopic expression of GFP– R-Ras–WT enhanced this fibronectin-dependent cell migration (). Coexpression of Plexin-B1–WT and Rnd1 with R-Ras–WT blocked the R-Ras–induced promotion of cell migration toward fibronectin, in the presence of Sema4D at the lower well (). On the other hand, expression of Plexin-B1–RA, a mutant of Plexin-B1 that lacks R-Ras GAP activity, did not exhibit the Sema4D-dependent inhibition of cell migration toward fibronectin (). Association of Rnd1 with Plexin-B1 is essential for the expression of R-Ras GAP activity of Plexin-B1 (), and inhibition of cell migration was not observed in the cells without Rnd1 or in the cells expressing Plexin-B1–GGA, a mutant of Plexin-B1 unable to interact with Rnd1 (). The Plexin-B subfamily has been shown to activate RhoA via its COOH-terminal PDZ domain binding motif (; ; ). However, Plexin-B1–ΔC, a mutant of Plexin-B1 that lacks the PDZ domain binding motif but still has R-Ras GAP activity (), inhibited fibronectin-dependent cell migration in the presence of Sema4D (). Cell migration mediated by constitutively active R-Ras, R-Ras–QL, was not suppressed by the Sema4D/Plexin-B1–Rnd1 complex (). Expression levels of these constructs used in the assay were similar, as verified by immunoblot analysis (not depicted). Furthermore, R-Ras activity is essential for ECM-mediated cell migration, as both inactivation of R-Ras by expression of Myr–R-RasGAP or knockdown of R-Ras by R-Ras RNA interference almost completely suppressed the fibronectin-dependent cell migration (). We further confirmed that the R-Ras GAP activity exhibited by endogenous Plexin-B1 is required for Sema4D-mediated inhibition of ECM-mediated PC12 cell migration. We recently reported that the cytoplasmic region of Plexin-B1 by nature takes the intramolecularly tethered form and that disruption of the interaction between the NH-terminal region (N-Cyt) and the COOH-terminal region (C-Cyt) within the cytoplasmic domain () by Rnd1 binding to N-Cyt is essential for exhibiting the R-Ras GAP activity. C-Cyt associates with N-Cyt–GGA, which has no ability to interact with Rnd1, and Rnd1 cannot disrupt this interaction (). As shown in , overexpression of Plexin-B1–N-Cyt–GGA could effectively block the Sema4D/Plexin-B1–Rnd1 complex–mediated R-Ras GAP activity, suggesting that Plexin-B1– N-Cyt–GGA could be an effective tool to inhibit the R-Ras GAP activity of Plexin-B1 in a dominant-negative manner. Overexpression of Plexin-B1–N-Cyt–GGA in PC12 cells almost completely blocked the Sema4D-mediated inhibition of ECM-mediated cell migration (). We also examined the role of endogenous R-Ras protein in PC12 cell migration. Transfection of the R-Ras siRNA effectively reduced the expression of endogenous R-Ras protein in PC12 cells, whereas the control siRNA did not work (Fig. S1 A, available at ), and expression of R-Ras siRNA almost completely suppressed the collagen-dependent cell migration (Fig. S1, B and C), suggesting that R-Ras is a prime regulator for integrin-mediated cell migration in PC12 cells. These results demonstrate that activation of endogenous R-Ras protein is essential for the ECM-mediated cell migration and that regulation of R-Ras activity through Sema4D/Plexin-B1–mediated R-Ras GAP activity plays a key role in ECM-mediated cell migration. PI3-K is the predominant effector of R-Ras (; ), and R-Ras–mediated cell migration is sensitive to pharmacological PI3-K inhibitors (; ). Expression of R-Ras–QL induces the ECM-independent functional activation of β integrins and tyrosine phosphorylation of FAK () and causes COS-7 cell migration in the absence of ECM ligands (). The D64A mutation of R-Ras or the pharmacological PI3-K inhibitor LY294002 abrogated the cell migration induced by R-Ras–QL (). R-Ras–QL–64A, the effector loop mutant of R-Ras, impairs the ability of R-Ras to activate PI3-K (), and R-Ras–QL–mediated phosphorylation of the PI3-K effector Akt (PKB) was abolished by the D64A mutation (). We further examined the involvement of PI3-K in R-Ras–QL–induced activation of β integrins and subsequent FAK phosphorylation. As shown in , D64A mutation or LY294002 treatment markedly blocked both R-Ras–QL–induced activation of β integrins and phosphorylation of the downstream effector FAK. It has been reported that prominent PI3-K–dependent phosphorylation of Akt occurs in response to β integrin–mediated adhesion (). We examined the effect of Sema4D/Plexin-B1–mediated R-Ras GAP activity on PI3-K activity by measuring the phosphorylation of Akt. As shown in , expression of Plexin-B1–WT and Rnd1 inhibited the fibronectin-mediated Akt phosphorylation in the presence of Sema4D. However, this inhibition was not observed in cells expressing Plexin-B1–GGA or Plexin-B1–RA that had no ability to exhibit R-Ras GAP activity (). These results suggest that PI3-K activity is necessary for R-Ras–mediated activation of β integrins and that Sema4D/Plexin-B1–Rnd1 inactivates PI3-K through down-regulation of R-Ras activity. To clear out the role of PI3-K downstream of Sema4D/Plexin-B1, leading to suppression of β integrin activity, we transfected p110α-CAAX, a constitutively active form of PI3-K (), and tested the ability of Sema4D/Plexin-B1 to inhibit β integrin activation. In COS-7 cells, overexpression of p110α-CAAX by itself did not induce β integrin activation in the absence of fibronectin (; ). On the other hand, overexpression of the kinase-dead form of p110α blocked the fibronectin-dependent β integrin activation. These results suggest that PI3-K activity is necessary but that PI3-K activity by itself is not sufficient for inducing β integrin activation. We next examined whether the down-regulation of PI3-K activity downstream of Sema4D/Plexin-B1 is necessary for inhibition of β integrin activity. As shown in , overexpression of p110α-CAAX blocked Sema4D/Plexin-B1–dependent inactivation of β integrins. We also examined the ability of Sema4D to inhibit cell migration in cells expressing p110α-CAAX. As shown in , overexpression of p110α-CAAX in PC12 cells almost completely blocked the Sema4D-mediated inhibition of cell migration. These results suggest that down-regulation of PI3-K activity, downstream of Sema4D/Plexin-B1, is responsible for suppression of β integrin activity and inhibition of the ECM-mediated cell migration. Cell migration is a fundamental cellular process in many cell types, and semaphorins are known to act as a negative regulator for integrin-mediated cell migration. We show that the Sema4D receptor, Plexin-B1, down-regulates R-Ras activity and inhibits ECM-mediated integrin activation and cell migration through its R-Ras GAP activity. R-Ras is implicated in integrin regulation, and a constitutively active form of R-Ras has been shown to increase the affinity of integrins for fibronectin () and to stimulate cell migration (). We have examined a role of R-Ras in ECM-mediated integrin activation and cell migration and showed that R-Ras is markedly activated by the ECM and that this activation is required for activation of β integrins and subsequent cell migration, as inactivation of R-Ras activity by expression of the GAP domain of p98–R-RasGAP or knockdown of R-Ras by R-Ras–specific siRNA markedly reduces ECM-mediated integrin activation and cell migration. Our results also revealed that β integrins are required for R-Ras activation upon ECM-mediated adhesion. This suggests a positive feedback during cell-substrate adhesion, implicating R-Ras activation and the consequent further strengthening of integrin-mediated functions. Therefore, R-Ras is a central regulator for ECM-mediated integrin activation and cell migration, and the regulation of R-Ras activity is critical for integrin-mediated cell migration. Semaphorins are implicated in migration of a variety of cells. Stimulation of Plexin-B1 by Sema4D is reported to hamper integrin-based adhesion and cell migration in NIH-3T3 cells (). We have reported that Plexin-B1 encodes an R-Ras GAP in the cytoplasmic tail and that stimulation of the Plexin-B1–Rnd1 complex by Sema4D induces the R-Ras GAP activity and resultant repulsive response of neuronal growth cone (). We demonstrate here that Plexin-B1/Rnd1–mediated R-Ras GAP activity is also involved in Sema4D-induced inhibition of integrin activation and cell migration. Furthermore, the COOH-terminal PDZ domain binding motif of Plexin-B1 is dispensable for suppression of integrin activity and cell migration by Sema4D. In addition to Sema4D, class 3 semaphorins have been shown to control adhesion and migration of endothelial cells by inhibiting integrin function (), and Sema3A signaling–deficient mice have shown defective migration of neural crest cells (). Furthermore, Plexin-C1, a receptor of semaphorin A39R, was recently reported to inhibit integrin-mediated adhesion and chemokine-induced migration (). The R-Ras GAP–homologous domains are well conserved among plexin families, including Plexin-A and -C1. In addition, we recently reported that the down-regulation of R-Ras activity is also required for the Sema3A/Plexin-A–induced repulsive response in hippocampal neurons (). We speculate that the direct regulation of R-Ras activity by plexins is likely to be a mutual signaling pathway among plexin families and that this R-Ras GAP activity of plexin families may be a critical signaling system for semaphorin-regulated cell migration. Semaphorins were initially identified as repulsive factors for axon guidance, and many neurons use members of the integrin family of cell surface receptors for responses to neurite growth promoting factors, and integrin activation regulates neurite outgrowth (). Recently, expression of constitutively active R-Ras was shown to promote integrin-dependent neurite outgrowth of retinal neurons, suggesting that R-Ras activity plays an important role in integrin-dependent neurite outgrowth (). Therefore, it is proposed that the down-regulation of R-Ras activity by Plexin-B1 via R-Ras GAP activity suppresses R-Ras–mediated integrin activation and thereby induces growth cone collapse and inhibition of neurite outgrowth. With respect to signaling of other repulsive factors, the ephrin-B1 receptor EphB2, another family of the repulsive factor receptor, was also reported to suppress integrin-mediated functions by inactivating R-Ras (), suggesting that repulsive guidance cues inhibit integrin-mediated functions by inactivating R-Ras in general and that R-Ras acts as a common regulator of integrin activation and cell migration (). We also examined the downstream signaling of Sema4D/Plexin-B1–mediated R-Ras GAP activity leading to inactivation of β integrins and found that down-regulation of PI3-K activity is responsible for Sema4D/Plexin-B1–induced suppression of β integrin activity and cell migration. PI3-K activity is known to be required for R-Ras–mediated enhancement of cell migration (; ). PI3-K has emerged as the predominant effector for R-Ras, and R-Ras is a more potent activator of PI3-K than other Ras family members (; ). On the other hand, PI3-K activity has been shown to promote interaction between talin with the β integrin cytoplasmic tail, leading to the clustering and activation of integrins (; ; ). Integrin activation by mechanical stretch is also mediated by PI3-K and is followed by an increase in integrin binding to the extracellular matrix proteins (). Therefore, elevated PI3-K activity by activated R-Ras may trigger a sequence of events leading to clustering and activation of integrins, although overexpression of p110α-CAAX by itself is not sufficient for inducing β integrin activation (; ). We used the monoclonal antibody HUTS-4, which detects hybrid domain swing-out in β integrins, a process most commonly associated with ligand binding affinity (), to measure activity of β integrins and revealed that Sema4D/Plexin-B1–mediated R-Ras GAP activity suppresses affinity of β integrins through inactivation of PI3-K activity. Consistent with our results, a previous report demonstrated that an R-Ras–mediated increase in affinity of the β integrins is dependent on PI3-K activity by performing the ligand binding assay in mast cells (). On the other hand, have shown that PI3-K activity is not required for R-Ras–mediated integrin activation in CHO cells by using a ligand-mimetic antibody, PAC-1. Therefore, we speculate that this discrepancy may be due to the differences in ways to measure integrin activity or that R-Ras may regulate integrin activity via both PI3-K–dependent and –independent pathways, depending on the cell type. In conclusion, our results demonstrate that R-Ras activity is required for ECM-mediated integrin activation and cell migration and that the Sema4D/Plexin-B1–Rnd1 complex regulates integrin activation and cell migration through the R-Ras GAP activity. However, a variety of molecules such as ErbB-2 and Met have been known to be involved in plexin signaling, inducing diverse physiological functions (, ). It was recently shown that Plexin-B1 enhances chemotaxis of endothelial cells through the activation of multiple intracellular tyrosine kinase cascades independent of the R-Ras GAP activity (). Regulation of R-Ras activity, tyrosine kinases, and other signaling mechanisms may participate in diverse actions of plexins. Further work will be required to delineate the precise mechanism of R-Ras–mediated integrin activation and its regulation by plexins for cell migration during physiological and pathological processes, including neural cell migration, angiogenesis, and tumor metastasis. Plexin-B1 cDNA was provided by L. Tamagnone (Torino University, Torino, Italy). HA-tagged Rnd1; HA- and GFP-tagged human R-Ras and R-Ras–QL (Q87L); the GST-fused Ras binding domain of c-Raf-1 (amino acids 53–130); the NH-terminal HA-tagged myristoylated form of R-RasGAP; and Myc-tagged Plexin-B1, Plexin-B1–GGA (L1849G, V1850G, and P1851A), Plexin-B1–RA (R1677A, R1678A, and R1984A), Plexin-B1–ΔC (lacking the last seven COOH-terminal amino acids), and Plexin-B1– N-Cyt–GGA (amino acids 1511–1915) were described previously (,). The effector loop mutant of R-Ras, R-Ras–DA (D64A), was generated by a PCR-mediated mutagenesis. NH-terminal FLAG-tagged p110α was a gift from T. Katada (Tokyo University, Tokyo, Japan), and CAAX sequence was fused to the COOH terminus to create a constitutively active form as described previously (). The specific siRNA for R-Ras was designed to target 19 nucleotides at nucleotides 359 and 377 (5′-gcaagctcttcactcagat-3′), whereas the control siRNA was designed at nucleotides 426 and 444 (5′-caaggcagatctggagaca-3′), and both were expressed by using a siRNA expression vector (Ambion) as described previously (). The pharmacological PI3-K inhibitor LY294002 was purchased from Calbiochem. A soluble form of Sema4D fused to human IgG-Fc was a gift from H. Kikutani (Osaka University, Osaka, Japan). We used the following antibodies: mouse monoclonal antibodies against Myc and phosphotyrosine; a rabbit polyclonal antibody against p125-FAK (Upstate Biotechnology); mouse monoclonal antibodies against α-tubulin (Sigma-Aldrich), β integrins (BD Biosciences), and active β integrins, HUTS-4 (Chemicon); a rabbit polyclonal antibody against R-Ras (Santa Cruz Biotechnology, Inc.); a rat monoclonal antibody against HA (Roche); and HRP-conjugated secondary antibodies (DakoCytomation). For functional studies in the transwell assay, we used the following antibodies: the affinity-related β integrin–activating monoclonal antibody 8A2 (IgG2a); the functional blocking monoclonal antibody against the integrin α subunit, 3A3 (IgG1; Serotec); the functional blocking monoclonal antibody against the integrin β subunit, P5D2 (IgG1; Chemicon); and a mouse monoclonal antibody against the extracellular ligand binding region (raised against amino acids 771–1070 of human origin) of Plexin-B1 (IgG2b; Santa Cruz Biotechnology, Inc.). FITC- and PE-conjugated F(ab′)–specific secondary antibodies for flow cytometry were purchased from Jackson ImmunoResearch Laboratories. The PhosphoPlus Akt Antibody kit (Cell Signaling) was used for the analysis of the phosphorylation state of Akt. Proteins were separated by 12.5% SDS-PAGE and were electrophoretically transferred onto a polyvinylidene difluoride membrane (Millipore). The membrane was blocked with 3% low-fat milk in TBS and incubated with primary antibodies. The primary antibodies were detected with HRP-conjugated secondary antibodies and a chemiluminescence detection kit (Chemi-Lumi One; Nacalai Tesque). Images were captured using a LAS 1000 analyzer (Fuji) equipped with Image Gauge 4.0 software (Fuji). Cells on coverslips were fixed with 4% PFA in PBS for 15 min and washed with PBS five times. Cells were permeabilized with 0.2% Triton X-100 in PBS for 10 min and incubated with 10% FBS in PBS for 30 min to block nonspecific antibody binding. Cells were incubated with an anti–R-Ras antibody (1:200 dilution) for 1 h and then incubated with an Alexa Fluor 594–conjugated secondary antibody for 1 h. Cells were washed in PBS for 1 h and mounted in 90% glycerol containing 0.1% p-phenylenediamine dihydrochloride in PBS. Images were captured at RT using a microscope (Eclipse E800; Nikon) and a 40 × 0.75 objective (Nikon) equipped with a digital camera (DC350F; Leica). The images were arranged and labeled using Photoshop software (Adobe). COS-7 cells were cultured in DME containing 10% FBS, 4 mM glutamine, 100 U/ml penicillin, and 0.2 mg/ml streptomycin under humidified conditions in 95% air and 5% CO at 37°C. PC12 cells were maintained in RPMI 1640 with 10% horse serum (HS) and 5% FBS. Transient transfections were performed with Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. A soluble form of Sema4D was expressed as a fusion protein with the Fc fragment of human IgG. Stimulation with Sema4D was performed by incubation of the cells with Sema4D-Fc–containing medium at 37°C. 10 cells were detached with 1.5 mM EDTA in PBS, washed three times with serum-free medium, resuspended in DME containing 1% BSA, seeded on the upper side of 8-μm pore filters of Transwell chambers (Costar), which were coated on the lower side with varying concentrations of either fibronectin or collagen I (Sigma-Aldrich), and incubated for 7 h. Cells on the upper side of the filters were mechanically removed, and cells on the lower side were fixed with 4% PFA. The numbers of migrated cells through the filter were counted by the fluorescence of GFP or the staining with crystal violet (). At the same time, the cells were seeded onto 24-well plastic culture plates to count the total number of transfected cells (). Relative cell migration was then determined by the number of migrated cells normalized to the total number of transfected cells (). Unless described, the value from the GFP-transfected cells in the absence of coating was defined as 1. For functional studies using activating or inhibitory monoclonal antibodies, cells were pretreated with 5 μg/ml of antibodies or corresponding negative IgG controls for 5 min before seeding onto the transwells. Images were captured at RT in PBS using a microscope (Eclipse TE300-FN; Nikon) and a Plan Fluor 10 × 0.30 objective (Nikon) equipped with digital camera (DS-L1 and DS-5M; Nikon). The images were arranged and labeled using Photoshop 7.0 software. Measurement of β integrin activity by immunoprecipitation was performed as described previously (). 3 × 10 COS-7 cells were maintained in DME containing 1% FBS after transfection. 16 h after transfection, cells were detached with 1.5 mM EDTA in PBS, washed three times with serum-free medium, and resuspended in 10 ml of 1% BSA in DME with or without Sema4D-Fc. The cell suspension was plated onto 10-cm plates coated with or without 10 μg/ml fibronectin and incubated at 37°C for 15 min. The cells were lysed directly on dishes with ice-cold cell lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 10% glycerol, 1 mM sodium vanadate, 25 mM NaF, 10 μg/ml pepstatin, 1 mM PMSF, 10 μg/ml aprotinin, and 10 μg/ml leupeptin) containing 5 μg/ml HUTS-4, immunoprecipitated for 2 h, and subsequently incubated with protein G–Sepharose beads (GE Healthcare) for 1 h at 4°C. After the beads were washed twice with the ice-cold cell lysis buffer, the bound proteins were eluted in Laemmli sample buffer and analyzed by SDS-PAGE and immunoblotting with the monoclonal antibody against β integrins. To measure the activity of β integrins in PC12 cells, 10 cells were maintained in RPMI 1640 containing 1% HS for 12 h, detached with 1.5 mM EDTA in PBS, washed three times with serum-free medium, and resuspended in 10 ml of 1% BSA in RPMI 1640 with or without Sema4D-Fc. The cell suspension was plated onto 10-cm plates coated with or without 10 μg/ml collagen I and incubated at 37°C for 3 h. The cells were lysed directly on dishes with ice-cold cell lysis buffer. Measurement of the activity of β integrins by ELISAs under detergent-free condition was performed as described previously (). 10 cells transfected in 24-well plastic culture plates were detached with 1.5 mM EDTA in PBS, washed three times with serum-free medium, and resuspended in 1 ml DME containing 1% BSA, with or without Sema4D-Fc. One tenth of the resuspended cells (100 μl) were seeded onto the 96-well assay plates, which were coated with 10 μg/ml of either fibronectin or collagen I. Cell adhesion was allowed for 15 min at 37°C. Then, the cells were delicately washed once with PBS and the adherent cells were fixed with 4% PFA. After the fixative, the cells were thoroughly rinsed with PBS containing 0.1% BSA. To avoid nonspecific binding, the cells were incubated with PBS containing 5% BSA for 3 h at RT. Cells were then incubated overnight at 4°C with 2 μg/ml HUTS-4. After the incubation with primary antibody, the wells were rinsed and blocked with PBS containing 5% BSA for 3 h at RT before they were exposed to an HRP- conjugated secondary antibody. After the incubation, cells were rinsed again with PBS followed by distilled HO. The peroxidase color reaction was developed in the dark using -phenylenediamine according to the manufacturer's instructions (ELISA OPD kit; Nacalai Tesque), and the plate was read on a kinetic microtiter plate reader (GENios; Tecan) using the XFluor4 program (Tecan). The antibody concentration and incubation times were optimized to ensure testing in the linear range. Expression levels of the constructs used in the assay were also verified by immunoblot analysis. Analysis of cell surface expression of active β integrins by flow cytometry was performed as described previously (). 10 PC12 cells were seeded onto 6-cm noncoated plates in RPMI 1640 containing 10% HS and 5% FBS. 18 h after seeding, cells were treated with medium containing Sema4D-Fc or Sema4D-Fc plus 1 mM Mn for 3 h at 37°C. Cells were washed once with PBS and resuspended in blocking solution containing 5% dissociation buffer (Invitrogen) and 2% sheep serum in PBS. Cells were then incubated with 2.5 μg HUTS-4 or buffer alone for 1 h at 4°C, washed with the blocking solution, and labeled with FITC-conjugated secondary antibody for 30 min at 4°C. Cells were then washed and analyzed with an EPICS ELITE flow cytometer using the EXPO32 analysis program (Beckman Coulter). For the analysis of active β integrins in transiently transfected COS-7 cells, 10 cells were transfected with a GFP expression vector together with various other expression vectors. Cells were kept for 18 h in DME containing 10% FBS after transfection, stimulated for 5 min at 37°C with or without Sema4D-Fc, and were collected and incubated with HUTS-4 antibody or buffer alone as described previously in this section. Cells were labeled with a PE-conjugated secondary antibody, and expression of GFP and activity of β integrins (PE staining) were simultaneously analyzed by two-color flow cytometry. Analysis of the intensity of PE staining in a GFP-positive population was performed as described previously (). Approximately 10,000 cells were analyzed in each experiment, and the results shown are representative of two independent experiments. Measurement of R-Ras activity in cells was performed as described previously (). 7 × 10 COS-7 cells were maintained in DME containing 1% FBS after transfection. The cell suspension was prepared as described (see Measurement of the activity of β integrins by immunoprecipitation) and plated onto plastic dishes coated with or without 10 μg/ml fibronectin and incubated at 37°C for 15 min. The cells were lysed directly on dishes with ice-cold cell lysis buffer (25 mM Hepes-NaOH, pH 7.5, 150 mM NaCl, 1% NP-40, 0.25% sodium deoxycholate, 0.1% SDS, 10% glycerol, 10 mM MgCl, 1 mM EDTA, 1 mM DTT, 10 μg/ml aprotinin, and 10 μg/ml leupeptin) containing 75 μg of GST-fused Ras binding domain of c-Raf-1 (GST-RBD). To examine the effect of collagen I and Sema4D stimulation on R-Ras activity in PC12 cells, 10 cells were maintained in RPMI 1640 containing 1% HS for 12 h, detached with 1.5 mM EDTA in PBS, washed three times with serum-free medium, and resuspended in 10 ml of 1% BSA in RPMI 1640 with or without Sema4D-Fc. For samples indicated, cells were treated with 5 μg/ml of monoclonal β integrin blocking (P5D2) or activating (8A2) antibody before replating. Cells were either kept in suspension or plated onto 6-cm plates coated with (1 or 10 μg/ml) or without collagen I and incubated at 37°C for 15 min. The cells were lysed directly on dishes with ice-cold cell lysis buffer, and the lysates were used in a pull-down assay using GST-RBD. Detection of tyrosine phosphorylation of FAK was performed as described elsewhere (). The cells were lysed directly on dishes with ice-cold cell lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 0.25% sodium deoxycholate, 10% glycerol, 1 mM sodium vanadate, 25 mM NaF, 10 μg/ml pepstatin, 1 mM PMSF, 10 μg/ml aprotinin, and 10 μg/ml leupeptin) containing 4 μg/ml of the polyclonal antibody against FAK, immunoprecipitated for 2 h, and subsequently incubated with protein A–Sepharose beads (GE Healthcare) for 1 h at 4°C. COS-7 cells were maintained in DME with 0.5% FBS after transfection for 36 h. We added 20 μM LY294002 directly to the culture medium after transfection and changed it at every 12 h to reduce the basal levels of PI3-K activity. Cells were directly lysed on dishes with 1× Laemmli sample buffer and analyzed by SDS-PAGE and immunoblotting. Fig. S1 shows reduction in endogenous R-Ras protein by RNA interference in PC12 cells and requirement of endogenous R-Ras protein in collagen-mediated PC12 cell migration. Online supplemental material is available at .

The enzymes dipeptidyl aminopeptidase A/Ste13p and endopeptidase Kex2p, which process the α-factor mating pheromone in the yeast , undergo repeated cycles of vesicular transport between the TGN and endosomal system (; ; ). These enzymes possess large lumenal domains, a single transmembrane-spanning domain, and cytosolic domains of ∼100 amino acids. Within their trafficking itinerary, the best understood step is retrieval from the prevacuolar/endosomal compartment (PVC) back to the TGN, which is mediated by the retromer, an apparent vesicle coat complex (). Retromer recognition of Ste13p and the carboxypeptidase Y receptor Vps10p occurs via binding of the Vps35p retromer subunit with aromatic amino acid–based sorting signals such as FXFXD in Ste13p (, , ; ; ). However, the manner by which Ste13p and Kex2p reach the PVC is poorly understood. Some studies suggest that Ste13p and Kex2p may traverse via the early endosome (EE) en route to the PVC. A Ste13p-based reporter protein, A(F→A)–alkaline phosphatase (ALP), reaches the PVC slowly with a half-time of ∼60 min, whereas other proteins such as Vps10p and Cps1p reach the PVC within 5–15 min (; ). Deletion of the 2–11 region within the Ste13p cytosolic domain accelerates trafficking of A(F→A)-ALP into the PVC (; ). A signal analogous to 2–11 exists in Kex2p (). These data are consistent with a model in which A(F→A)-ALP reaches the PVC via the EE with the 2–11 signal conferring either EE to TGN retrieval or static retention within the TGN/EE. In support of an itinerary involving the EE, the loss of function of a yeast synaptojanin like protein Inp53p/Sjl3p, which is thought to play a role in TGN/EE traffic, accelerates the rate of trafficking of A(F→A)-ALP into the PVC but has no effect on Vps10p trafficking (). In addition, Kex2p has been colocalized with EE markers Tlg1p and chitin synthetase III (Chs3p) as well as Snc1p, a late secretory v-SNARE that recycles from the plasma membrane back to the TGN via an early endosomal compartment (; ; ; ). Finally, the loss of function of Soi3p, which appears to be required for efficient EE to PVC trafficking, delayed trafficking of a PVC retrieval-defective form of Kex2p to the PVC but did not affect Vps10p trafficking (). Clathrin-associated vesicular transport machinery clearly plays a role in trafficking between the TGN and endosomes of yeast. A loss of function in either clathrin heavy chain or Vps1p, a dynamin homologue that is thought to participate in the production of clathrin-coated vesicles, causes Ste13p and Kex2p to be mislocalized to the cell surface (; ; ). This suggests that there is a requirement for clathrin in both TGN to EE and TGN to PVC pathways. Furthermore, clathrin-coated compartments have been shown to contain Vps10p and Kex2p (), and clathrin is required for in vitro TGN to PVC trafficking of Kex2p (). Two types of clathrin-associated adaptors function within the yeast TGN/endosomal system. The and gene products (; ) are thought to function in the direct TGN to PVC pathway. Loss of GGA function causes the PVC-localized t-SNARE Pep12p to be mislocalized to the EE (). Likewise, the delivery of Vps10p to the PVC and Cps1p to the PVC/vacuole, cargoes that are thought to use the direct TGN to PVC pathway, are delayed in mutants (). No delay in A(F→A)-ALP transport into the PVC was observed in a strain lacking GGA function (); however, a moderate delay has been reported for Kex2p (). Thus, a role for GGA proteins in the trafficking of TGN resident proteins is still obscure. In contrast to the GGAs, the adaptor protein 1 (AP-1) complex appears to be involved in trafficking between the EE and TGN. Loss of AP-1 function caused the mislocalization of Chs3p to the cell surface under conditions in which it was sequestered to the EE (). A severe synthetic growth defect has been observed upon the simultaneous loss of function of the GGA proteins and AP-1 (). This synthetic growth defect suggests that both adaptors may mediate anterograde transport into the endosomal system, but this model has not been rigorously tested. In this study, we tested whether A(F→A)-ALP required the AP-1 complex for transport into the PVC when the GGA-mediated pathway and the plasma membrane route were blocked. We found that when AP-1 complex function was lost under these conditions, the transport of A(F→A)-ALP into the PVC was dramatically accelerated. In addition, we demonstrate that the AP-1 complex interacts with the 2–11 sorting signal within the Ste13p cytosolic domain. These results are most consistent with a model in which clathrin/AP-1 recognizes Ste13p in the EE and directs it into a retrograde pathway to the TGN. A-ALP is a model TGN membrane protein consisting of the NH-terminal cytosolic domain of Ste13p fused to the transmembrane and luminal domains of ALP (). When TGN/endosome retention is perturbed, A-ALP is transported to the vacuole, where its COOH-terminal propeptide is proteolytically removed. For example, mutation of the FXFXD retromer recognition motif prevents the retrieval of A-ALP from the PVC to the TGN (; ). This mutant, A(F→A)-ALP, is proteolytically processed with a half-time of ∼60 min, reflecting the kinetics of vacuolar delivery (). A-ALP has also been shown to reach the PVC in a class E mutant with similar kinetics (; ), indicating that the PVC to vacuole step is quite rapid and that the rate of A(F→A)-ALP processing largely reflects the rate of trafficking into the PVC. To test whether the A-ALP itinerary includes EEs, it was tagged with GFP and was expressed using the moderate strength promoter (). Like its untagged counterpart, pulse-chase immunoprecipitation analysis of GFP–A-ALP indicated that it was unprocessed after 60 min, whereas mutation of the FXFXD motif resulted in processing with a half-time of ∼60 min (unpublished data). Because the GFP tag did not interfere with the kinetics of transport into the PVC/vacuole, it is likely that the tag did not disrupt the normal trafficking patterns of A-ALP. Consistent with this, GFP–A-ALP exhibited a punctate staining pattern typical of yeast Golgi/endosomal proteins (). Cells expressing GFP–A-ALP were incubated at 0°C with the lipophilic dye FM4-64, allowing it to integrate into the plasma membrane but not be internalized (). The cells were then incubated at 30°C for 2 min to allow the dye to be transported from the plasma membrane to EEs. After 2 min, FM4-64 exhibited a punctate pattern typical of endosomes with little or no vacuolar or plasma membrane staining (). 56% of these FM4-64 structures colocalized with GFP–A-ALP (). In contrast, 11% of the FM4-64–positive structures contained the GFP-tagged PVC syntaxin Pep12p (), indicating that the vast majority are EEs rather than PVCs. After 8 min of internalization, FM4-64 exhibited some vacuolar staining in addition to punctate endosomal staining (unpublished data). Of the FM4-64 punctate structures observed after 8 min of internalization, 55% were positive for GFP–A-ALP, whereas 58% were positive for GFP-Pep12p. These data indicate that a pool of FM4-64 was transported from EEs to PVCs during the intervening 6 min. The significant FM4-64/GFP–A-ALP colocalization at both time points is consistent with A-ALP populating both EEs and PVCs. The minimal FM4-64/GFP-Pep12p colocalization at 2 min would be expected given that its localization is restricted to the PVC (; ). Vps8p and Soi3p are thought to be required for EE to PVC transport (; ), whereasVps8p appears to also function in PVC to vacuole trafficking (). A(F→A)-ALP processing was clearly slowed in a strain and was slowed to a lesser degree in a strain (). These results indicate that like Kex2p (), A(F→A)-ALP is transported along an EE to PVC route. We next tested whether trafficking via the PVC is absolutely necessary for A(F→A)-ALP to reach the vacuole or whether A(F→A)-ALP could reach the vacuole by an alternative pathway if trafficking to the PVC is blocked. Vacuolar processing of A(F→A)-ALP was analyzed in a Pep12p temperature-sensitive strain whose function is necessary for all vesicular trafficking into the PVC (; ). strains expressing A(F→A)-ALP were shifted to the nonpermissive temperature (36°C) for 10 min, pulsed for 10 min, and chased for the indicated times (). strain. These results demonstrate that A(F→A)-ALP must traffic to the vacuole via the PVC and, collectively, show that A(F→A)-ALP follows a TGN→EE→PVC pathway before reaching the vacuole. The mechanism by which TGN proteins reach the endosomal system appears to involve clathrin because Kex2p and Ste13p/A-ALP are mislocalized to the cell surface in clathrin mutants (; ; ). Therefore, it is possible that the GGA class of clathrin adaptors may function in this process. The rate of trafficking of A(F→A)-ALP to the PVC/vacuole was previously observed to be unchanged in a strain compared with wild type (). strain shifted to the nonpermissive temperature by pulse-chase immunoprecipitation. We found no decrease in the trafficking kinetics of A(F→A)-ALP in the , strain as compared with , or the wild-type strain (). Therefore, the GGAs are dispensable for anterograde trafficking of A(F→A)-ALP into the endosomal system. Moreover, this data suggest that there is a GGA-independent trafficking route that affords access to the PVC via a route that does not include the plasma membrane. The vacuolar protease Cps1 is initially synthesized as an inactive precursor that is proteolytically processed to yield the mature form upon reaching the vacuole (). Previously, it was reported that in a strain, Cps1p transport to the vacuole is delayed (). Cps1p contains a ubiquitin moiety that causes it to be sorted into multivesicular body vesicles upon reaching the PVC (; ). As GGA adaptors have been shown to recognize the ubiquitin moiety on cargo proteins at the TGN (; ), it seems likely that Cps1p enters clathrin/GGA-coated vesicles at the TGN and is directly delivered to the PVC. Using the same experimental regimen as for A(F→A)-ALP () except with different chase times, we also observed a marked delay in Cps1p processing in a strain (), albeit not as dramatic as reported by . strain, suggesting that in the absence of GGA function, a pool of Cps1p is mislocalized to the cell surface before being transported to the vacuole via the endocytic pathway. strain, Cps1p was slowly processed (61-min half-time), suggesting that in the absence of GGAs, Cps1p was capable of accessing the PVC by an intracellular route, presumably via the EE. mutation (); thus, like A(F→A)-ALP, Cps1p must transit via the PVC to then be transported to the vacuole. In summary, these results are consistent with a model in which Cps1p uses a GGA-dependent direct TGN to PVC pathway and that A(F→A)-ALP exits the TGN via a GGA-independent pathway leading to the EE. To address whether the clathrin adaptor AP-1 might function at the TGN for the transport of A(F→A)-ALP to the EE, random mutagenesis was used to generate a temperature-sensitive mutation. In the presence of the mutations, the allele exhibited near normal growth at 24°C but little or no growth at the nonpermissive temperature of 36°C (), which is consistent with the near synthetic lethality previously observed for mutations in these three genes (). We next asked whether the trafficking kinetics of A(F→A)-ALP was altered in this strain lacking both AP-1 and GGA function. In the event that TGN to EE trafficking of A(F→A)-ALP was blocked, including the mutations would prevent any A(F→A)-ALP from spilling into the GGA-mediated direct TGN to PVC pathway. strain exhibited markedly accelerated processing kinetics compared with wild type (20 vs. 56 min; ). strain (), the accelerated processing of A(F→A)-ALP was nearly as pronounced at 24 as at 36°C. Thus, the protein encoded by the allele is partially defective for trafficking at the permissive temperature. strain in which both the GGA-mediated direct TGN to PVC and plasma membrane routes are blocked, leaving only the TGN to EE pathway intact. These results imply that AP-1 is not required for A(F→A)-ALP transport to the EE and argue instead that AP-1 functions to slow transport into the PVC, presumably via EE to TGN retrieval. A(F→A)-ALP is obligated to reach the PVC before being transported to the vacuole (). strain by a route not normally used in wild-type cells. strain, regions of the TGN could be transported directly to the vacuole by an autophagic pathway. strain at 36°C compared with wild type (unpublished data). strain. The SA mutation blocks the delivery of A(F→A)-ALP to the PVC apparently by preventing its exit from the EE (). strain (), strongly suggesting that A(F→A)-ALP reaches the PVC/vacuole by its normal route in this strain. Furthermore, these results indicate that the SA trafficking block occurs after the trafficking step that is mediated by AP-1. Like the AP-1 adaptor complex, the role of the 2–11 region of Ste13p is to slow the transport of Ste13p/A-ALP into the PVC (; ; ). The 2–11 region could act as a static retention signal in the TGN or as a signal for EE to TGN retrieval. Another possibility is that it could be necessary for TGN to EE anterograde transport and that deletion of 2–11 might cause accelerated transport into the PVC because A(Δ2–11; F→A)-ALP is forced into alternative pathways (i.e., a direct TGN to PVC pathway or plasma membrane pathway). strain in which the alternative pathways were blocked (). strain. strain caused by deletion of the 2–11 region argues that this signal plays a role either in EE to TGN retrieval or in static TGN retention. If the 2–11 region and AP-1 act at the same step, the deletion of 2–11 and the loss of AP-1 function should not cause an additive effect on the acceleration of A(F→A)-ALP trafficking. strain (26-min half-time; ) that were similar to A(F→A)-ALP in (20 min; ) and A(Δ2–11; F→A)-ALP in (30 min; ). An in vitro binding assay was used to assess whether the heterotetrameric AP-1 complex associates with the Ste13p cytosolic domain. The Ste13p cytosolic domain (residues 1–118) was fused to the NH terminus of GST and the resulting fusion protein (Ste13-GST) purified from onto glutathione-agarose beads. The beads were incubated with protein extracts from a yeast strain containing an epitope-tagged allele () of the μ1 subunit of AP-1 (). Bead-associated proteins were analyzed by Western blotting for Apm1-HA and Apl2p (). Both AP-1 subunits were found to associate with Ste13-GST but not with GST alone (). Interestingly, we observed little or no association of Apm1-HA or Apl2p with Ste13(Δ2–11)-GST, indicating that the interaction is highly dependent on the 2–11 region. Fusions containing only residues 1–20 and 1–12 of Ste13p also bound to AP-1 with similar affinity to that of full-length Ste13-GST. Thus, residues 1–12 are necessary and sufficient for association with Apm1-HA and Apl2p. There appeared to be little, if any, difference in the binding of these AP-1 subunits to the wild-type, SA, and SD versions of Ste13-GST. Data from a previous study was consistent with the idea that phosphorylated S might antagonize the 2–11 signal (); however, this binding data coupled with the observation that the SA block is downstream of AP-1 would appear to argue against this. Finally, the data suggested that more Apm1-HA than Apl2p was bound to the Ste13-GST beads based on comparison with the amounts of these proteins detected in the input sample. This suggested that Apm1p was binding as a monomer in addition to binding within the context of intact AP-1. Thus, it could be the subunit that Ste13p interacts with directly. In animal cells, the μ or, less commonly, the β subunit of adaptor complexes recognize cargo proteins (; ). Given our results suggesting that yeast μ1, or Apm1p, may be the AP-1 subunit that recognizes Ste13p, we asked whether Apm1p expressed in rabbit reticulocyte lysates in the absence of the other yeast AP-1 subunits would bind to Ste13-GST. We found that Ste13-GST and Ste13(1–12)-GST associated with S-labeled Apm1p, whereas GST alone did not (, left). Mammalian μ subunits contain two independently folding domains: a domain in the NH-terminal one third of the protein that mediates association with the β subunit and a domain in the COOH-terminal two thirds of the protein that mediates binding to the YXXΦ class of cargo signals (where Φ stands for an amino acid with a bulky, hydrophobic side chain; ; ). The region of yeast Apm1p corresponding to the mammalian μ cargo-binding domain, Apm1-Δ2–158, associated with Ste13-GST and Ste13(1–12)-GST with a similar efficiency as full-length Apm1p (, right). Note that Apm1-Δ2–158 comigrated with Ste13-GST, affecting its mobility relative to the input and Ste13(1–12)-GST pull-down. As a more stringent test of whether Apm1p/Ste13p binding is direct, we fused full-length Apm1p and Apm1-Δ2–158 to maltose-binding protein (MBP), expressed the fusions in , and purified them using amylose resin affinity chromatography. Ste13-GST immobilized on beads but not GST alone, clearly associating with both MBP-Apm1p and MBP-Apm1-Δ2–158 while it sedimented only background levels of MBP (). Collectively, these data show that the 1–12 region of Ste13p associates with the COOH-terminal region of Apm1p directly (i.e., a bridging protein is not required). The broad implication is that AP-1 slows the transport of A(F→A)-ALP into the PVC by associating with the 1–12 region to recruit this cargo protein into clathrin/AP-1 vesicles. Cps1p clearly uses a different route than A(F→A)-ALP to reach the PVC because it is delayed by the loss of GGA function (). Cps1p does not normally appear to use the EE to PVC route, as a loss of Soi3p function has little or no effect on the rate of trafficking of Cps1p into the PVC (). Residues 1–12 of Ste13p are necessary and sufficient for binding to AP-1 in vitro (), and this region is clearly necessary to slow in vivo trafficking of A(F→A)-ALP into the PVC by acting at the TGN/EE. To test whether the AP-1–binding region of Ste13p is sufficient for in vivo function, we asked whether appending this region to the NH-terminal cytosolic domain of Cps1p would slow its trafficking in a strain. In the background, trafficking of Cps1p via the direct TGN to PVC pathway is prevented (), thus forcing these cargo proteins to access the PVC via the TGN→EE→PVC route or the plasma membrane→EE→PVC route. A construct containing residues 1–23 of Ste13p fused to Cps1p, Ste13-(1–23)-Cps1, was processed rapidly in wild-type cells at a rate similar to wild-type Cps1p, indicating that this fusion accesses the GGA-mediated pathway to the PVC and does not undergo any aberrant folding or transport delays in the early secretory pathway. However, in cells, Ste13(1–23)-Cps1 is processed significantly more slowly (54 ± 5-min half-time) than Cps1p in the strain (34 ± 2 min). This difference is comparable with the difference between A(F→A)-ALP (53 min) and A(Δ2–11; F→A)-ALP (30 min) observed in cells (). Thus, this Ste13p NH-terminal region slows the trafficking of Ste13-Cps1 most likely by mediating its retrieval from the EE. Importantly, a SA mutation in the Ste13(1–23)-Cps1 context markedly slowed trafficking in the strain (98 ± 11 min). Because the SA block appears to occur at the level of the EE (), this result indicates that the Ste13-Cps1 fusions do indeed traffic to the PVC via the EE as expected. A similar reduction in trafficking was observed when just residues 1–12 of Ste13p were fused to Cps1p (48 ± 3 min). Collectively, the data indicate that the first 12 residues of Ste13p are both necessary and sufficient to slow trafficking into the PVC. a j o r g o a l o f t h i s s t u d y w a s t o a d d r e s s t h e r e s p e c t i v e r o l e s o f t h e c l a t h r i n a d a p t o r s A P - 1 a n d G G A s i n t h e t r a f f i c k i n g o f A ( F → A ) - A L P , a m o d e l T G N p r o t e i n b a s e d o n S t e 1 3 p . O u r r e s u l t s s u g g e s t t h a t A P - 1 f u n c t i o n s i n r e t r o g r a d e E E t o T G N t r a n s p o r t o f A ( F → A ) - A L P . I n a d d i t i o n , w e d e m o n s t r a t e a p h y s i c a l i n t e r a c t i o n b e t w e e n a S t e 1 3 p c y t o s o l i c d o m a i n r e g i o n a n d A P - 1 a n d d e s c r i b e , f o r t h e f i r s t t i m e , a c a r g o - s o r t i n g s i g n a l r e c o g n i z e d b y t h e y e a s t A P - 1 a d a p t o r . The production of yeast media, the genetic manipulation of yeast strains, and all general molecular biology methods were performed as described previously () or as otherwise noted. Rabbit polyclonal antibodies against ALP have been previously described (; ). Rabbit anti-HA epitope and rabbit anti-MBP antibodies were obtained from Covance and New England Biolabs, Inc., respectively. Rabbit polyclonal antibodies against Cps1p were raised against a fusion protein consisting of GST fused to residues 46–577 of Cps1p, whereas rabbit anti-Apl2p antibodies were gifts from G. Payne (University of California, Los Angeles, Los Angeles, CA). Plasmids pSN55, pSN100, and pHJ63 have been previously described (; ). The GFP–A-ALP construct pCF17 is p416-CYC () containing PCR-derived sgGFP (Qbiogene) followed by a (GlyAla) linker and the coding region of () starting with codon 2. pCF4, CEN- was constructed by inserting a 4.97-kbp XbaI–XhoI fragment excised from pDP83-CPS1 () into the same sites in pRS316. Plasmid pCF2 was constructed by inserting a PCR fragment containing the full-length ORF into the EcoRI–SalI sites of pRS316. pSH46 was made by swapping a 0.5-kbp EagI–BglII fragment containing the 5′ region of with the Δ2–11, FA, and FA mutations for the corresponding fragment in pSN55. The Ste13-GST fusion constructs were made by introducing PCR or oligonucleotide duplex-derived inserts into the vector pETGEXCT (). Fusion of the Apm1p and Apm1-Δ2–158 coding sequences to the MBP sequence was performed by inserting PCR-derived inserts into the vector pMAL-c2 (New England Biolabs, Inc.). Ste13-Cps1 fusion protein constructs consisted of the PCR-derived promoter and relevant coding regions fused to the 5′ end of the coding region and 3′ untranslated regions cloned into pRS316 (). The fusion junction of the Ste13(1–23)-Cps1 and Ste13-(1–12)-Cps1 fusions were …KSSNGS… and …RKNGA…, respectively, with the numbered residue representing the last Ste13p residue and the underlined residues indicating the Cps1p sequence. To construct the allele, the ORF was amplified via PCR using an error-prone polymerase, Genemorph (Stratagene). The resulting population of PCR products was introduced into pRS313 via homologous recombination in yeast CFY6-2C/pCF2. His+ yeast transformants were then plated onto 5-FOA to lose pCF2 followed by screening for a lack of growth at 37°C and normal growth at 23°C. Finally, the mutagenized Apl2p-expressing plasmids rescued from temperature-sensitive yeast strains were retransformed back into CFY6-2C/pCF2 to test whether the growth phenotype was linked to the plasmid. All yeast strains are described in . SNY171-4D is a spore derived from a diploid made by crossing SNY165 and SNY94. SHY64 was constructed by mating type-switching UFY2. CFY6-2C is a spore derived from a diploid made by crossing SNY165/pCF2 and SHY64. CFY25-3B is a spore derived from a diploid made by crossing CFY6-2C/pCF6 with SNY94. CFY30, CFY31, CFY32, and CFY33 were all constructed using PCR-mediated gene replacement () of CPS1 with the NatR marker gene. The allele was integrated at the locus using plasmid pAPM1-HA∷URA3 (a gift from G. Payne), resulting in strain SNY190. The procedure for immunoprecipitation of wild-type and mutant A-ALP from [S]methionine/cysteine-labeled cells and Western blotting has been previously described (; ; ). Radioactively labeled proteins were quantified from gels using a phosphorimager system (FLA-2000; Fuji Film). For calculation of the half-time of A-ALP and Ste13-Cps1 processing, the log of the percentage of unprocessed precursor at each time point was plotted as a function of time, and the plots were analyzed by linear regression analysis. Immunoprecipitated Cps1p and derivatives were treated by endoglycosidase H before SDS-PAGE analysis according to a published protocol (). The precursor and mature forms of the immunoprecipitated A(F→A)-ALP, A(Δ2–11, F→A)-ALP, and Cps1p shown in – migrated on SDS-PAGE in a manner consistent with their predicted sizes. The various GST fusion proteins were expressed in BL21(DE3) (Novagen) and were affinity purified onto glutathione-agarose beads (Sigma-Aldrich). MBP, MBP-Apm1, and MBP-Apm1-Δ2–158 were expressed in BL21, affinity purified onto amylose resin (New England Biolabs, Inc.), and were eluted with buffer containing 10 mM maltose. The binding of Ste13-GST and its various mutant derivatives with AP-1 was assayed as previously described (). In brief, 1,000 OD units of SNY190 cells were spheroplasted, pelleted at 450 , and washed once in 40 ml of ice-cold 1.2 M sorbitol. Spheroplasts were resuspended in 10 ml of buffer C (25 mM Hepes-KOH, pH 7.2, 125 mM potassium acetate, 2.5 mM magnesium acetate, 1 mM DTT, 0.1% Triton X-100, 0.5 mM PMSF, 1 μg/ml leupeptin, and 1 μg/ml pepstatin A), Dounce homogenized, and centrifuged at 20,000 for 15 min. The supernatant was then transferred to a fresh tube and was incubated with a 0.75-ml bed volume of glutathione-agarose beads for 1 h at 4°C. The beads were pelleted at 750 a 100-μl input sample was saved for gel analysis, the remaining supernatant was split into seven equal portions, and each portion was incubated with a 100-μl bed volume of glutathione-agarose coated with Ste13-GST (or derivatives) for 3 h at 4°C. Samples were washed five times in buffer C followed by elution with SDS-PAGE sample buffer and heating at 100°C for 5 min. Eluted proteins were analyzed by SDS-PAGE and immunoblotting. Pull-down of in vitro translated wild-type and mutant Apm1p was performed by first translating Apm1p and Apm1-Δ2–158 in rabbit reticulocyte lysates (Ambion) in the presence of [S]methionine according to the manufacturer's instructions. Glutathione-agarose prebound to GST and GST fusions were incubated with aliquots of each translation reaction diluted with buffer C for 90 min at 4°C in a total volume of 400 μl. The beads were then washed five times with buffer C and were eluted with SDS-PAGE sample buffer at 100°C. Eluted proteins were analyzed by SDS-PAGE and autoradiography. Cells harvested from log-phase cultures were incubated for 30 min in YPD media containing 0.02 mg/ml FM4-64 (Invitrogen). They were then washed twice with YP media (lacking glucose) and once with SD-ura media before being resuspended in SD-ura media. Up to this point, all steps were performed at 0–4°C. The cells were then incubated at 30°C for 2 or 8 min, immediately placed on ice, and metabolic activity was stopped with 10 mM NaN and 10 mM NaF. The cells were mounted on 2% agarose pads on microscope slides containing 10 mM NaN and 10 mM NaF. The cells were immediately imaged at room temperature for FM4-64 and GFP staining using an epifluorescence microscope (DM5000B) equipped with a 100× NA 1.4 HCX plan-Apo lens, digital camera (DFC350X), and FW4000 software (all from Leica). Images were overlaid using the FW4000 software, adjusted slightly for brightness and contrast, and formatted using Adobe Photoshop 7.0.

(also referred to as meningococcus) is a Gram-negative bacterium that is an obligate commensal of the human nasopharyngeal mucosa. Meningococci can cause fulminant, rapidly fatal sepsis and can cross the blood–meningeal barrier, causing meningitis. We have recently shown that virulent encapsulated bacteria first adhere to endothelial target cells through their type IV pili and then proliferate, locally forming a colony at their site of attachment on the cell surface. Adhesion then promotes the local formation of membrane protrusions that surround bacteria favoring bacterial internalization within intracellular vacuoles and their transcytosis (). The formation of membrane protrusions stems from the organization of specific molecular complexes involving the molecular linkers ezrin and moesin (known as ERM [ezrin-radixin-moesin] proteins), along with the clustering of several membrane-integral proteins, including CD44, intracellular adhesion molecule (ICAM) 1, and cortical actin polymerization (; ; ). Recruitment of blood leukocytes to the site of infection involves a sequential, multistep process, from the tethering of leukocytes, followed by their arrest on the surface of activated endothelium to transendothelial migration. Leukocyte arrest, or firm adhesion, is mediated by the interaction of endothelial vascular cell adhesion molecule (VCAM) 1 with the integrin αβ (VLA-4) and of ICAM-1 and -2 with the integrins αβ (LFA-1) and αβ (Mac-1). Shortly after arrest, most leukocytes spread and begin to migrate laterally over the apical surface of the endothelium to reach the nearest intercellular junction, a step referred to as locomotion (). Leukocytes then migrate through the endothelial cell junctions (diapedesis or paracellular migration) or transmigrate directly through individual endothelial cells (transcellular migration; ; ; ). Active roles of the endothelium in facilitating leukocyte extravasation have been suggested by a variety of studies (; ). We and others have shown that endothelial adhesion molecules are involved in transducing leukocyte adhesion-mediated signaling responses to endothelium, leading to actin cytoskeletal reorganization (; ). Moreover, recent studies have demonstrated that leukocyte adhesion promotes the remodeling of the apical endothelial plasma membrane into projections that surround adherent leukocytes (; ; ). These structures, referred to as endothelial docking structures or transmigratory cups, are essential to promote firm adhesion and extravasation of leukocytes through paracellular as well as transcellular routes. These docking structures result from the dynamic redistribution of VCAM-1 and ICAM-1 at the leukocyte–endothelial contact area, accompanied by the recruitment of activated ERM proteins and by cortical actin polymerization. Extravasation is therefore an active, sequential process that requires drastic morphological changes involving the clustering of adhesion receptors on both leukocytes and endothelial cells. Because adhesion, like leukocyte adhesion, induces important cytoskeletal modifications at the endothelial surface, we analyzed the consequences of bacterial infection on leukocyte extravasation. Our results provide evidence that , by recruiting the ERM proteins ezrin and moesin, prevents the formation of the endothelial docking structures that are crucial in providing directional guidance for leukocyte emigration. To investigate whether leukocyte diapedesis process was altered by infection of human endothelial cells with , we analyzed the interaction of freshly isolated human monocytes or neutrophils with infected monolayers of a human bone marrow endothelial cell line (HBMEC). Endothelial cell monolayers, either untreated or preactivated with 100 U/ml TNF-α for 24 h, were left uninfected or were infected with a piliated encapsulated strain of (2C43 strain) for 1 h, during which small bacterial colonies developed on the cell surface of ∼30–40% of the total monolayer. Leukocyte adhesion to infected endothelial cells was assessed by measuring their attachment under flow conditions (). As expected, when a shear stress was applied (20 ml/h; 0.88 dyn/cm), both monocytes and neutrophils adhered poorly to unactivated HBMECs, whereas they attached to endothelial cell monolayers preactivated by TNF-α treatment. Interestingly, infection of the endothelial cells by promoted adhesion of leukocytes to unactivated monolayers but did not further increase leukocyte adhesion to TNF-α–activated monolayers. These results indicate that infection of endothelial cells by promotes strong adhesion by leukocytes, similar to the firm adhesion induced upon endothelial activation by inflammatory cytokines. Leukocyte diapedesis through monolayers of TNF-α–activated endothelial cells infected by was then assessed. In the absence of the chemoattractant, stromal cell–derived factor-1 (SDF-1α/CXCL12), a minimal migration of monocytes or neutrophils across noninfected endothelial monolayers was observed (). In the presence of 25 ng/ml SDF-1α, the transmigration of both monocytes and neutrophils through noninfected endothelial monolayers was enhanced. Unexpectedly, this SDF-1α–mediated transmigration was decreased by 60–90% when endothelial monolayers were infected by on their apical surface before leukocyte–endothelial cell interaction. When adhered to the endothelial cell surface, cell-associated bacteria promoted the firm adhesion of leukocytes but inhibited their SDF-1α–induced chemotaxis across a monolayer of endothelial cells. To investigate the molecular events responsible for the blockade of leukocyte transendothelial migration, real-time microscopy experiments were performed to analyze neutrophil behavior on the apical surface of infected endothelial cells (). Freshly isolated neutrophils were injected under flow condition (20 ml/h; 0.88 dyn/cm) onto TNF-α–activated HBMECs, which were either noninfected or previously infected for 1 h by . During a 30-min period, we observed that 30–50% of the neutrophils adhering to noninfected cells rapidly moved (within 2 min) from the site of firm adhesion to intercellular junctions and underwent diapedesis ( and Video 1, available at ). After neutrophils had crossed through the endothelial cell monolayer, they became phase dark and continued to move underneath the monolayer. In contrast, after initial adhesion on infected cells, ∼90% of the neutrophils rapidly moved along the surface of the endothelial monolayers toward bacterial colonies, sometimes covering a distance as long as 90 μm, with a mean speed of 0.6 μm/s ( and Video 2). However, once neutrophils reached a bacterial colony, their locomotion stopped and they remained trapped for the rest of the observation period (; and Videos 3 and 4). Interestingly, we noticed that all the trapped neutrophils were initially very active and attempted to migrate in many different directions ( and Videos 3 and 4). After 15 min, these neutrophils rounded up and remained immobile, and some of them (∼20%) ultimately detached from the infected monolayers. These results were further confirmed by scanning electron microscopic analysis of freshly isolated monocytes adhering for different periods of time to TNF-α–activated endothelial cells infected by or to uninfected endothelial cells as control (Fig. S1, available at ). Although monocytes rapidly migrated underneath the endothelial monolayer after initial attachment to noninfected monolayers, few monocytes were observed underneath infected endothelial monolayers. Monocytes rather accumulated in the vicinity of bacterial colonies. No morphological changes characteristic of leukocyte phagocytic activity were observed, consistent with the reported observations that the polysaccharide capsule of virulent strains protects the bacteria from phagocytosis by neutrophils or macrophages (). These results demonstrate that infection of endothelial cells by induces the migration of leukocytes at the endothelial cell surface toward bacterial colonies, likely because of the release of chemotactic factors by the bacteria. Unexpectedly, the locomotion of leukocytes halted once they reached the vicinity of a bacterial colony, preventing subsequent leukocyte migration to the endothelial cell junctions. To unravel the molecular mechanisms mediating the blockade of leukocyte locomotion and transmigration, we examined the distribution of endothelial adhesion molecules at the surface of infected monolayers (). As we previously showed (), the ERM binding proteins ICAM-1 and CD44 were massively recruited to the bacterial adhesion sites, together with the ERM proteins ezrin () and moesin (not depicted). Moreover, we observed that E-selectin, ICAM-2, and VCAM-1, which also contain an ERM binding motif in their cytoplasmic domains, were also recruited underneath bacterial colonies. 3D reconstructions indicate that all the recruited proteins concentrated in cellular projections surrounding bacteria, as illustrated for ICAM-1 (). In contrast, the junctional adhesion molecule JAM-A was not recruited at the bacteria adhesion site (not depicted), even though it is known to also play a role in leukocyte adhesion to and migration through endothelial monolayers (). These observations indicate that induces the localized recruitment of most of the endothelial adhesion molecules known to be involved in leukocyte adhesion. The cortical cytoskeleton is known to regulate the membrane localization of several adhesion molecules through one or more ERM proteins. These proteins, which are kept inactive in the cytoplasm through an intramolecular interaction, are activated by phosphatidylinositol 4,5-bisphosphate (PIP) binding to their NH-terminal domain and via phosphorylation of their COOH-terminal actin binding site by Ser/Thr kinases, such as Rho kinase (; ). Accordingly, using a PIP binding probe, the GFP-tagged PH domain of PLCδ (GFP-PLCδ-PH), we observed that infection of endothelial cells by induced a robust production of PIP at bacterial entry sites (). Moreover, an ezrin variant deleted of its PIP binding site (ezrin PIP ) was not recruited but remained diffusely distributed within the cytoplasm (). Those observations clearly confirmed that ezrin recruitment at the bacteria adhesion site was dependent on PIP binding. In contrast, treatment by the Rho kinase inhibitor Y-27632, which completely prevented cortical actin polymerization induced by as previously described (), did not affect ezrin recruitment, nor did it affect the recruitment of CD44, ICAM-1, or VCAM-1 (). We then expressed the NH-terminal domain of ezrin (FERM domain), which contains PIP and adhesion molecule binding sites but not the F-actin binding motif, therefore promoting a dominant-negative effect on the actin-dependent function of ezrin (). We previously showed that this ezrin domain prevents both the recruitment of endogenous ezrin and the polymerization of cortical actin induced by (). We present evidence that expression of a GFP-FERM construct did not affect the recruitment of CD44 (), ICAM-1, or VCAM-1 (not depicted). Collectively, these results demonstrate that recruitment of ICAM-1, ICAM-2, VCAM-1, or CD44 by does not require an active cytoskeletal-driven mechanism. Their recruitment most likely depends on their interaction with ezrin and moesin, which are both massively recruited at the site of bacterial adhesion by a PIP-dependent mechanism. These events provide an efficient sequestration of the endothelial adhesion molecules beneath bacterial colonies, raising the question of whether this process affects the leukocyte diapedesis process. Some recent studies have highlighted a crucial role of the redistribution of ICAM-1, VCAM-1, ezrin, moesin, and F-actin at leukocyte–endothelial contacts for the formation of endothelial docking structures that prevent leukocyte detachment by shear stress and promote leukocyte transmigration (). In agreement with these studies, we observed that both monocyte and neutrophil adhesion to noninfected HBMEC promoted the clustering of ICAM-1, VCAM-1, and ezrin as well as cortical actin polymerization around attached leukocytes. These clusters of adhesion molecules were assembled into endothelial docking structures, as assessed by 3D reconstructions of the horizontal sections (, top; and Video 5, available at ). Staining with antibodies directed against other endothelial adhesion molecules also revealed a considerable enrichment in CD44 and ICAM-2. Moreover, in line with previous studies suggesting that PIP was required for the generation of these docking structures (), we noticed a strong accumulation of the PIP binding probe GFP-PLCδ-PH within these structures (, top). Strikingly, the clustering of ICAM-1, VCAM-1, CD44, and ezrin was no longer observed around leukocytes that were in contact with a bacterial colony, whereas these proteins were highly enriched in the cellular projections surrounding bacteria (, bottom; and Video 6, available at ). Interestingly, PIP formation was observed around leukocytes in contact with a bacterial colony, as shown by the accumulation of GFP-PLCδ-PH (, bottom). This result strongly suggests that the lack of ezrin recruitment at leukocyte–endothelial contacts was not due to any bacterial inhibitory signal that would alter leukocyte-induced intracellular signaling in endothelial cells. These observations were further confirmed by a quantitative analysis of ezrin accumulation at leukocyte–endothelial contacts (). Interestingly, we noticed that endothelial docking structures still partially formed around leukocytes adhering at the vicinity of small bacterial colonies (, c) but not around leukocytes adhering at the vicinity of large colonies (, d), suggesting that the level of inhibition of these structures might be correlated with the size of bacterial colonies. By performing neutrophil transmigration experiments at different time points after bacterial adhesion to endothelial cells, corresponding to different sizes of bacterial colonies, we indeed clearly established that only large colonies (>100 μm) totally prevented neutrophil diapedesis (). Importantly, leukocytes interacting with noninfected endothelial cells adjacent to infected cells induced the formation of endothelial docking structures (, b) as efficiently as leukocytes adhering on noninfected monolayers (, a). Moreover, no inhibition of neutrophil transmigration through a fibronectin matrix was observed even when large bacterial colonies were grown on this matrix (, dotted line). These control experiments indicated that bacteria did not directly affect the intrinsic migratory response of neutrophils. Although we cannot formally exclude the possibility that partially interferes with leukocyte diapedesis by a bacterial chemotactic signal, our data demonstrate that the inhibition of leukocyte diapedesis is due to the absence of endothelial docking structures around the adherent neutrophils in contact with colonies. The level of inhibition is correlated with the size of bacterial colonies, strongly suggesting that titrates away some of the key endothelial mediators of leukocyte adhesion and guidance normally present in the endothelial docking structures. To test this hypothesis, we assessed by immunofluorescence analysis and 3D reconstructions whether overexpression in endothelial cells of any known component of the docking structures (ICAM-1, VCAM-1, ezrin, or moesin) would rescue their formation at the surface of infected cells. No rescue was observed after overexpression of GFP-VCAM-1 (, top) or GFP-ICAM-1 (not depicted). However, when cotransfected with VSVG-ezrin, GFP–VCAM-1 (, bottom) or GFP–ICAM-1 (not depicted) now appeared within docking structures on infected cells. In addition, overexpression of GFP-ezrin or -moesin alone rescued the formation of docking structures containing the endogenous ERM-binding proteins VCAM-1 and ICAM-1 (not depicted). These data strongly suggested that ezrin and/or moesin are required for docking structure formation and are titrated away by , thus preventing the formation of these docking structures at the surface of infected endothelial cells. We then assessed whether overexpression of the same proteins would rescue leukocyte diapedesis after infection of endothelial cells. Interestingly, although transfection of endothelial cells with GFP alone, GFP–ICAM-1, or GFP–VCAM-1 had no effect, overexpression of GFP-ezrin or -moesin efficiently rescued neutrophil diapedesis (). Coexpression of GFP-ezrin with GFP–ICAM-1 or GFP-moesin with GFP–VCAM-1 did not further enhance diapedesis (). In addition, transfection with increasing amounts of ezrin cDNA () or moesin cDNA (not depicted) clearly indicated a dose–response effect. To further demonstrate that ERM proteins are required for the formation of docking structures and leukocyte diapedesis, we evaluated the consequences in endothelial cells of overexpression of the dominant-negative FERM domain of ezrin. Endothelial cells were cotransfected with GFP-actin and VSVG-ezrin (, top) or VSVG-FERM (bottom) before adhesion of neutrophils. We observed that the expression of the FERM domain totally prevented the formation of the docking structures. Moreover, the expression of either GFP-FERM () or VSVG-FERM (not depicted) decreased neutrophil diapedesis by 50–60%, whereas the expression of GFP alone or GFP-ezrin () had no effect. Altogether, our results (see for summary) provide clear evidence that ERM proteins are absolutely required for the formation of endothelial docking structures and that prevents the formation of these structures by promoting the massive recruitment of both ezrin and moesin at bacterial adhesion sites. This process leads to the sequestration of key endothelial mediators of leukocyte adhesion and guidance, such as CD44, ICAM-1, and VCAM-1, and ultimately results in a strong inhibition of leukocyte diapedesis. Bacteria that have coevolved with their hosts use different strategies to overcome protective host defenses, such as phagocytosis or humoral defense mechanisms, by interfering with cytokine secretion or antigen presentation or by directly inhibiting T and B cell effector functions (; ). We provide evidence that infection of human endothelial cells interferes with leukocyte transendothelial migration by preventing the formation of the endothelial docking structures required for proper leukocyte diapedesis. This inhibitory mechanism is likely selective of invasion, as the infection of endothelial cells by bacterial pathogens, such as or , generally leads to the up-regulation of endothelial adhesion molecules, triggering leukocyte adhesion and transmigration (; ). Some bacterial pathogens may, however, inhibit transendothelial migration by distinct mechanisms, such as , which induces the degradation of Interleukin-8, a key chemokine involved in the transendothelial migration of neutrophils (). The most striking result of the present study is the observation that colonies on the endothelial surface actively recruit several adhesion molecules (ICAM-1, ICAM-2, VCAM-1, and CD44), preventing them from assembling within endothelial docking structures that were recently shown to be necessary for proper leukocyte transendothelial migration (; ). As a consequence, leukocytes in the vicinity of a bacterial colony remained trapped, spinning in circles and then becoming immobilized, unable to reach endothelial junctions, whereas leukocytes adhering to noninfected endothelial cells rapidly moved to the nearest junction for diapedesis. These results are highly consistent with a previous study reporting that blocking antibodies against ICAM-1 and -2 prevent monocyte locomotion on the endothelial cell surface, causing them to repeatedly “pirouette” and cross junctions without undergoing diapedesis (). Our results reveal that bacteria-induced inhibition of the formation of endothelial docking structures is strictly dependent on the recruitment of ezrin and moesin at the site of bacterial adhesion. These ERM proteins can simultaneously bind F-actin and the cytoplasmic domains of multiple adhesion molecules through their COOH- and NH-terminal domains, respectively. We show that expression of the truncated NH-terminal domain of ezrin, which displays a dominant-negative activity on actin-dependent ezrin function (), prevents the formation of the docking structures and impairs leukocyte diapedesis. Our results further reveal that by inducing the massive recruitment of ezrin and moesin underneath bacterial colonies, prevents their accumulation at leukocyte–endothelial cell contact area and the formation of docking structures. Overexpression of ezrin or moesin in cells infected with was sufficient to rescue both the formation of functional docking structures and efficient leukocyte diapedesis through infected monolayers. Altogether, these results firmly establish the pivotal and redundant role of ezrin and moesin in the formation of endothelial docking structures and leukocyte diapedesis. After infection by , ERM proteins are titrated away from the leukocyte–endothelial contact area, preventing the endothelial adhesion molecules from assembling in docking structures. It is possible, however, that a fraction of ICAM-1 and VCAM-1 molecules remains at the apical surface of infected cells and may participate in leukocyte attachment without leading to a functional interaction with the actin cytoskeleton. In addition, we cannot exclude the possibility that other leukocyte–endothelial interactions are involved in leukocyte attachment to infected cells, such as JAM-A interaction with LFA-1 (). It is indeed interesting to note that JAM-A distribution was not altered by the bacterial infection, as JAM-A is not involved in the formation of the endothelial docking structures and seems to play a minor role in leukocyte transmigration (). Interestingly, the same set of endothelial proteins is present in the membrane protrusions induced by and in the docking structures promoted by leukocyte adhesion. Ezrin translocation in these two structures is dependent on PIP binding, in agreement with previous studies showing that PIP binding is essential for the membrane localization of ERM proteins (; ). However, our data shed light on some disparities in the molecular events controlling the maintenance of these two different structures. Indeed, it was previously showed that the recruitment of VCAM-1 and ICAM-1 during leukocyte adhesion requires the polymerization of cortical actin for anchorage of the adhesion molecules at the leukocyte–endothelial contact area (; ). Our results further confirm that ezrin anchorage to the actin cytoskeleton is required for the formation of the endothelial docking structures. In contrast, such anchorage was not required for ezrin clustering at the bacterial interaction sites. Moreover, inhibition of the induced cortical actin polymerization by the Rho kinase inhibitor Y-27632 did not affect ezrin recruitment; neither did the clustering of ERM binding endothelial adhesion molecules induced by require cytoskeletal anchoring. These observations strongly suggest an essential role of ezrin and moesin for the recruitment of multiple adhesion molecules at bacterial interaction sites by an actin cytoskeleton–independent mechanism. Interestingly, a similar mechanism of protein complex assembly was previously documented regarding the formation of the immunological synapse between T cells and antigen-presenting cells (). Systemic infection by is a highly complex process that involves multiple interactions with host cells and host immune responses. Once meningococci have reached the bloodstream, they can cause fulminant sepsis and/or invade the meninges by crossing the blood–brain barrier. Once in the subarachnoid space, lacking the main humoral and cellular host defense mechanisms, bacterial proliferation can proceed in an uncontrolled manner. It is widely held that endotoxin release elicits local secretion of proinflammatory cytokines (such as TNF-α or IL-1β), which in turn increase blood–brain barrier permeability and promote neutrophil influx. The subsequent release of neutrophil products contributes to the development of clinically overt meningitis (). Because neutrophils are normally rapidly recruited to the sites of bacterial invasion and thus constitute the first line of defense against bacterial pathogens, it is tempting to speculate that the mechanism described here could confer a selective advantage to meningococci invading the brain. In conclusion, our results offer new, unexpected insights into how a bacterial pathogen may hamper the triggering of a host inflammatory response by blocking the transendothelial migration of leukocytes. Moreover, the study of this host–pathogen interaction has provided a unique window into the molecular mechanisms controlling leukocyte adhesion and transendothelial migration, thus expanding our knowledge of this fundamental cellular process. mAbs to ICAM-1 (11C81) and VCAM-1 (4B2) were purchased from R&D Systems, mAb to CD44 (J173) was purchased from Immunotech, mAb to E-selectin was obtained from BD Biosciences, mAb to ICAM-2 (MAB2147) was obtained from Chemicon, and mAb to LFA-1 (sc-7306) was purchased from Santa Cruz Biotechnology, Inc. Polyclonal antibodies against ezrin and moesin were kindly provided by C. Roy and P. Mangeat (Centre National de la Recherche Scientifique, Montpellier, France). Rhodamine-phalloidin was purchased from Sigma-Aldrich. The Rho kinase inhibitor Y27632 was purchased from Calbiochem. Fibronectin and human recombinant TNF-α and SDF-1α were purchased from AbCys. Leukocytes were isolated from the peripheral blood of healthy volunteers by density gradient sedimentation in Ficoll (Pan Biotech), washed in PBS plus 0.1% NaCl, and resuspended at the density of 10/ml in RPMI medium containing 10% SVF plus 20 ng/ml GM-CSF (R&D Systems). HBMECs (provided by B. Weksler, Cornell University, Ithaca, NY) were cultured and infected with 2C43 (formerly clone 12), a piliated capsulated Opa variant of the serogroup C meningococcal strain 8013, as previously described (; ). Endothelial cells were seeded at 10 cells/cm on 1-well glass chamber slides (LabTek) precoated with 2% gelatin (DIFCO) and were cultured for 48 h. When indicated, cells were activated with 100 U/ml TNF-α for 12 h and then infected with for 1 h. Live bacteria were stained for 15 min with polyclonal antibody to the bacterial capsule coupled to Cy2. The glass slide was then assembled in a parallel plate laminar flow chamber (Immunetics). A thermostated flow chamber was mounted on the stage of an inverted microscope (Eclipse TE300; Nikon) equipped with a 20× objective (S fluor [Nikon]; NA of the objective lens, 0.75), and 8 × 10 leukocytes were injected using a flow of 20 ml/h (0.88 dyn/cm) in RPMI medium at 37°C. Nomarski differential interference contrast images and fluorescence were captured every 2 s through a cooled charge-coupled device camera (CoolSNAPfx; Roper Scientific) controlled by MetaMorph software (Universal Imaging). Images were analyzed with transparencies of the labeled bacteria superimposed over QuickTime videos (Apple Computers) of the leukocytes on the surface plane of the endothelial monolayer. Videos were accelerated two times and compressed with Premiere 6.5 (Adobe). The number of adherent or transmigrated cells was quantified by direct visualization of 20 different fields (40× phase-contrast objective) of three independent experiments. Vectors encoding GFP–ICAM-1, GFP–VCAM-1, and GFP-moesin were provided by F. Sanchez-Madrid (Hospital Universitario de La Princesa, Madrid, Spain). Vectors encoding the VSVG-FERM, the VSVG-ezrin, the GFP-ezrin, and the GFP-tagged PIP ezrin mutant were provided by M. Arpin (Institut Curie, Paris, France). Vector encoding the GFP-FERM and the GFP-PLCδ-PH were provided by J. Delon and G. Bismuth (Institut Cochin, Paris, France), respectively. HBMECs were transfected using the nucleofector system developed by Amaxa, Inc., as previously described (). HBMECs were seeded on 3-μm-pore Transwell filters (Costar) precoated with 2% gelatin at 10 cells/cm, and cultured for 48 h. Cells were then treated for 24 h with 100 U/ml TNF-α. Transfected HBMECs were seeded on Transwell filters at 5 × 10 cells/cm for 6 h and then treated for 24 h with 100 U/ml TNF-α. When indicated, Transwell filters were precoated with 10 μg/ml fibronectin for 1 h. Filters covered with fibronectin or HBMECs were overlaid for 15 min with bacterial inoculum in starvation medium (DME supplemented with 0.1% BSA). Filters were then washed three times with starvation medium to remove nonadherent bacteria, and infection was allowed to proceed for the indicated period of time. Filters were then washed twice, and 1.5 × 10 leukocytes were added in the upper chambers. When indicated, 25 ng/ml of the chemoattractant SDF-1α was added in the lower chambers. After 3 h, transmigrated leukocytes were recovered from the lower chambers and the number of transmigrated cells was quantified by flow cytometry. To ensure that endothelial permeability was not affected during the migration assay, cells were washed three times to remove the leukocytes and the culture medium was replaced by medium containing Evans blue. The passive diffusion of Evans blue through the monolayer over a period of 10 min was determined by densitometry analysis, and the coefficient of permeability was determined. HBMECs were grown on Permanox coverslips to confluence. After infection and leukocyte adhesion, cells were fixed and labeled as previously described (; ). Series of optical sections were obtained with a confocal laser-scanning microscope (TCS-SP2; Leica), using a 63× oil-immersion objective. 3D reconstructions were obtained using Amira software (Mercury Computer Systems) and quantification histograms with ImageJ software (NIH). Cells were fixed with 2.5% glutaraldehyde in 0.066 M cacodylate, pH 7.5, for 1 h at room temperature and then rinsed twice with 0.1 M cacodylate, pH 7.5. Dehydration was performed by incubating samples with increasing concentrations of ethanol, up to 100%. Samples were submitted to critical-point drying in CO, spatter-coated to produce 15-nm gold particle coating, and examined using a microscope (JSM 6300F; JEOL) at an accelerating voltage of 5 kV. The number of adherent, migrating, transmigrated, or rounded cells or cells in contact with a bacterial colony was quantified by direct visualization of ∼20 different fields, corresponding to a total of 80–100 leukocytes. Fig. S1 shows scanning electron microscopy analysis of monocytes adhering for different periods of time on endothelial cells infected by . Video 1 shows live imaging of neutrophil diapedesis through endothelial cell monolayer. Video 2 shows live imaging of a neutrophil migrating at the surface of an endothelial cell monolayer infected by . Video 3 shows live imaging of a neutrophil trapped in the vicinity of a bacterial colony. Video 4 shows live imaging of a trapped neutrophil attempting to migrate in many different directions. Video 5 shows ezrin accumulation in a 3D docking structure formed around an adherent neutrophil. Video 6 shows ezrin accumulation in the cellular projections induced by . Online supplemental material is available at .

Mitochondria are essential organelles whose structure and function adapt to different cellular conditions through continuous fusion and fission events (). Mitochondrial dynamics exert essential physiological and developmental roles, regulate apoptotic processes, and affect energy production within mitochondria (). Two neuropathies, Charcot-Marie-Tooth type 2A and autosomal dominant optic atrophy, are caused by mutations in essential fusion components, namely, mitofusin 2 or OPA1 (). Although several components involved in fusion have been identified, many of them in yeast, molecular mechanisms underlying mitochondrial fusion events are poorly understood. A central role has been assigned to yeast Fzo1 in the outer membrane of mitochondria (or to its mammalian homologues mitofusin 1 and 2) as part of a fusion complex that also contains Ugo1 and Mgm1 in yeast (). Less clear is the role of the F-box protein Mdm30, whose loss leads to the accumulation of aggregated and fragmented mitochondria (). Among the 21 annotated proteins of with an F-box motif (), some assemble in Skp1–Cdc53–F-box (SCF) E3 ubiquitin ligase complexes, which mediate proteasomal proteolysis of specific substrates (). Indeed, Mdm30 has been linked to the turnover of the transcription factor Gal4 in the nucleus (). Mdm30 was not only localized to the cytosolic fraction but also found in association with mitochondria (). Therefore, in analogy to the degradation of resident ER proteins, Mdm30 may affect mitochondrial dynamics by coupling the mitochondrial fusion machinery to the ubiquitin–proteasome system (UPS) in the cytosol. Consistently, accumulation of Fzo1 has been observed in cells lacking Mdm30 (). However, it remained unclear whether this indeed reflects Mdm30-dependent proteolysis of Fzo1 and whether the UPS is involved. Notably, 26S proteasomes have been linked to the degradation of Fzo1 in α-factor–arrested yeast cells (). However, degradation does not depend on Mdm30 under these conditions (). Moreover, the involvement of 26S proteasomes remained controversial, as only proteasome inhibitors have been used, which are known to be effective only in yeast cells with increased membrane permeability or lacking drug-efflux pumps (). We have analyzed the role of Mdm30 for the regulation of mitochondrial dynamics and demonstrate for the first time Mdm30-dependent proteolysis of Fzo1 in vegetatively growing yeast cells. Mdm30 is part of a novel proteolytic pathway that does not involve SCF complexes and 26S proteasomes and thus is strikingly different to the proteasomal degradation of Fzo1 in α-factor–arrested yeast cells. The accumulation of Fzo1 in cells suggests that the F-box protein Mdm30 is involved in the degradation of Fzo1 (; ). We therefore assessed the stability of Fzo1 in wild-type and cells after inhibition of protein synthesis with cycloheximide. This analysis revealed that Fzo1 is constitutively degraded in wild-type cells, whereas it remained stable in the absence of Mdm30 (). Similar experiments were performed in yeast cells in the exponential or post–diauxic-shift phase, but a significant dependence of Fzo1 stability on the growth phase was not observed (). To distinguish between complete Fzo1 turnover and processing, we followed Fzo1 degradation using antibodies directed against either the NH-terminal GTPase domain or a peptide located at the COOH-terminal segment of Fzo1. Moreover, the stability of an Fzo1 variant carrying an NH-terminal HA tag was examined. As no proteolytic fragments of Fzo1 were detected, we conclude that Fzo1 is completely degraded in an Mdm30-dependent manner. Mdm30 might affect the localization and/or complex assembly of Fzo1 and thereby affect Fzo1 stability and mitochondrial fusion. Therefore, mitochondrial and postmitochondrial supernatant fractions of wild-type and cells were analyzed for the presence of Fzo1 (). Regardless of the presence of Mdm30, Fzo1 was recovered from the mitochondrial fraction, which also contained the mitochondrial outer membrane protein Tom40 but was devoid of the soluble cytosolic protein Tpi1 (). Fzo1 assembles into an 850-kD complex in wild-type cells () and was found to be present in a complex of the same size in cells (). Thus, Mdm30 does not affect the assembly of Fzo1 in the outer membrane of mitochondria. Fzo1 was recently reported to be degraded upon mating factor–induced cell cycle arrest (; ). Intriguingly, under these conditions, proteolysis occurs independent of Mdm30 (; ) or any other nonessential yeast gene encoding an F-box protein (unpublished data). We therefore conclude that Fzo1 is degraded along two physiologically and mechanistically distinct pathways, one prevalent in vegetatively growing cells and the other in α-factor–arrested cells. To assess the requirement of the Mdm30 F-box motif for Fzo1 degradation, four conserved amino acid residues within the motif were replaced, namely, L19P, P20A, E22Q, and I23A (). In a second construct, the entire F-box was deleted (amino acids 1–58; , ΔF-box). Wild-type or mutant forms of Mdm30 were found in equal association with mitochondria upon subcellular fractionation (unpublished data). After expression in wild-type and strains, the steady-state level of Fzo1 was monitored by immunoblotting. In wild-type cells, the steady-state concentration of Fzo1 was unaffected by the presence of the mutant Mdm30 forms (). In cells, however, only wild-type Mdm30 but not Mdm30 variants with a defective F-box were able to restore degradation and low Fzo1 steady-state concentrations (). These results demonstrate that proteolysis of Fzo1 depends on the integrity of the Mdm30 F-box. To examine whether Mdm30 interacts directly with Fzo1, we performed coimmunoprecipitation experiments using Fzo1-specific antibodies in cells expressing Mdm30 or F-box variants thereof. Mdm30 was precipitated with Fzo1-specific antibodies, demonstrating complex formation of both proteins in wild-type cells (). Thus, in agreement with studies on other F-box proteins (), Fzo1 recognition by Mdm30 does not depend on its F-box (). It should be noted that after solubization of mitochondrial membranes with digitonin, Mdm30 or mutant variants did not coelute with Fzo1 upon sizing chromatography, indicating a rather dynamic interaction (unpublished data). Mdm30 was found associated with both Skp1 and Cdc53 in high throughput protein–protein interaction studies () and in vitro (). Its requirement for Fzo1 proteolysis therefore suggests that degradation occurs through ubiquitylation by SCF complexes (). Protein shut-off experiments were performed in cells deficient for core components of SCF complexes (, , and ) or for the E2 enzyme Cdc34 (). Fzo1 was degraded in these SCF-deficient cells with kinetics similar to those of wild-type cells (). This was in contrast to Grr1, a known substrate of the SCF complex (), which stably accumulated in these mutants (). We also analyzed the stability of Fzo1 in cells that allow tetracycline-sensitive expression of the essential genes and () but did not observe a stabilization of Fzo1 upon repression of or (unpublished data). Thus, Mdm30-dependent proteolysis of Fzo1 does not require SCF ubiquitin ligase complexes. These findings are reminiscent of the F-box proteins Ctf13 and Rcy1, which do not assemble in SCF complexes (; ), and substantiate that F-box proteins may exert cellular functions other than ubiquitin-dependent protein degradation. It should be emphasized, however, that Mdm30 could mediate the degradation of other substrate proteins by interaction with SCF complexes. This hypothesis is supported by the recently observed Mdm30-dependent ubiquitylation of Gal4 in the nucleus (). To directly examine a possible ubiquitylation of Fzo1 by other E3 ubiquitin ligases, cellular ubiquitin pools were limited either by using a mutant E1 enzyme () or through the overexpression of mutant ubiquitin forms, unable to form ubiquitin chains (K29R, K48R, K63R, or RRR). Neither approach caused a stabilization of Fzo1 (). We conclude that Mdm30-dependent constitutive Fzo1 degradation is independent of SCF and ubiquitin. We also examined the role of ubiquitylation for Fzo1 proteolysis upon mating factor–induced cell cycle arrest in cells and found it to be impaired (). Thus, in striking contrast to its constitutive and Mdm30-dependent degradation, the proteolytic breakdown of Fzo1 in arrested cells is ubiquitin dependent (). We next examined the stability of Fzo1 in yeast cells with deficient proteasomal activity. Proteolysis of Fzo1 was not impaired in or cells carrying mutations in a proteolytic subunit of the 20S core particle or in an ATPase subunit of the 19S regulatory complex, respectively (). Similarly, deletion of the proteasomal assembly factor Ump1 did not interfere with the proteolytic breakdown of Fzo1 in vegetatively growing cells (). This is in contrast to Fzo1 proteolysis in cells arrested in the presence of mating factor (). In agreement with previous inhibitor studies (), Fzo1 remained stable after treating or cells with α-factor (). We conclude that the constitutive Mdm30-dependent turnover of Fzo1 occurs independently of the 26S proteasome, whereas degradation of Fzo1 upon α-factor arrest involves ubiquitylation and 26S proteasomes. To further characterize the Mdm30-mediated Fzo1 proteolysis, we examined a potential ATP dependence of this proteolytic pathway. Two chemical inhibitors of ATP production in mitochondria were used: NaN, an inhibitor of the cytochrome oxidase, and carbonylcyanide -chlorophenylhydrazone (CCCP), an uncoupler dissipating the membrane potential across the mitochondrial inner membrane. Both treatments completely blocked degradation of Fzo1, which appears to be ATP dependent (). We therefore analyzed the stability of Fzo1 in mitochondria lacking the ATP-dependent -AAA protease subunit Yme1, which is active in the intermembrane space of mitochondria (), but did not observe any effect on Fzo1 degradation (). Moreover, we examined the involvement of vacuolar degradation, which is responsible for the general removal of mitochondrial proteins under rapamycin-induced autophagic conditions (). However, Fzo1 was not stabilized in cells lacking active vacuolar peptidases (). We conclude that Fzo1 is constitutively degraded by a yet unknown proteolytic system whose activity is apparently ATP dependent. Deletion of impairs mitochondrial fusion and results in the formation of fragmented mitochondria accompanied by the accumulation of Fzo1 (; ). To examine whether Fzo1 accumulation in cells is a general consequence of impaired mitochondrial fusion, the steady-state concentrations of Fzo1 were determined in , , and mutant cells. Fzo1 did not accumulate in either of these mutant cells, demonstrating that there is no direct link between the cellular level of Fzo1 and ongoing mitochondrial fusion (). To further substantiate the specific effect of Mdm30 on Fzo1 stability and mitochondrial fusion, we analyzed the effect on the steady-state level of Fzo1 of the 131 nonessential deletion strains and in the 16 tet-off essential genes known to have fragmented mitochondria (; ). However, apart from cells, we did not observe altered Fzo1 levels (unpublished data). It therefore seems likely that Mdm30 affects mitochondrial dynamics by specifically regulating the stability of Fzo1. To substantiate the specific effect of Mdm30 on Fzo1, we assessed mitochondrial morphology upon overexpression of Fzo1 in wild-type cells. Strikingly, mitochondria in cells overexpressing Fzo1 in the presence of an wild-type allele resembled the fragmented mitochondria seen in cells (). In agreement with this observation, proteolysis of Fzo1 was only observed when Fzo1 was expressed from a centromeric low-copy plasmid but not when expressed from a multicopy plasmid (). These findings were further supported by modulating Fzo1 expression in wild-type and cells using the inducible promoter. Growing yeast cells in the presence of glucose, raffinose, or galactose results in repression, nonrepression, or induction of Fzo1 synthesis. Importantly, tubular mitochondria were observed in cells grown under nonrepressed conditions in the presence of raffinose, i.e., at intermediate expression levels of Fzo1 ( and ). This confirms that mitochondrial morphology can be modulated by changing Fzo1 expression levels. Thus, mitochondrial fusion requires a tight control of the steady-state concentration of Fzo1. We conclude that lowering Fzo1 protein levels allows bypassing of Mdm30. Therefore, Mdm30 promotes mitochondrial fusion exclusively by controlling the steady-state level of Fzo1. Overexpressing Fzo1 in wild-type cells prevents its degradation, suggesting that Mdm30 could be the limiting factor. Therefore, Mdm30 was overexpressed in cells containing a galactose-inducible allele. Surprisingly, we found that the simultaneous overexpression of Fzo1 and Mdm30 is deleterious for cellular growth and leads to abnormal cellular and mitochondrial morphologies (). It is therefore conceivable that the Fzo1–Mdm30 complex sequesters an as-yet-unknown factor that is essential for growth. Our experiments demonstrate a crucial role of Fzo1 proteolysis for the maintenance of mitochondrial morphology. The stability of Fzo1 is controlled by a novel, apparently ATP-dependent proteolytic system. This system involves the F-box protein Mdm30 but not ubiquitylation and SCF complexes, opening new perspectives for cellular activities of F-box proteins. Our results also demonstrate that proteins in the outer membrane of mitochondria can be degraded in at least two ways, only one of which is dependent on the UPS. At the same time, they suggest a mechanism that regulates organellar morphology by the use of two different proteolytic systems during different physiological states of a eukaryotic cell. Yeast strains were grown according to standard procedures on complete or synthetic media supplemented with 2% (wt/vol) glucose or, when indicated, 2% (wt/vol) galactose or raffinose and 0.02% (wt/vol) glucose. 100 μg/ml cycloheximide, 10 μM α-factor (or 0.1 μM in the case of cells), 2 mM NaN, or 1 mM CCCP was added when indicated. The cell cycle arrest in the presence of α-factor was verified microscopically. Yeast strains used were derivatives of W303 and BY4741. The strains , , , , or and the isogenic wild-type strain BY4741 were obtained from Euroscarf. For expression of Fzo1 in wild-type and cells under the control of the promoter, previously described strains were used (). and were deleted by PCR-based homologous recombination in W303. To increase the efficiency of the α-factor arrest, was deleted by PCR-based homologous recombination in wild-type, , and cells. , , and mutant variants (F-box*, with the L19P, P20A, E22Q, and I23A mutations; ΔF-box, without the residues 1–58) encoding an NH-terminal FLAG tag were cloned into the multicopy vector pJDCEX2 (2μ; and promoter). Fzo1 harboring an NH-terminal HA tag was expressed under the control of the endogenous promoter using the centromeric plasmid pRS315 or the multicopy plasmid YEplac181. For visualizing mitochondria, the plasmid pVT100U-mtGFP (mitochondria-targeted GFP) was used. To monitor constitutive proteolysis of Fzo1, cycloheximide was added to logarithmically growing yeast cultures at a final concentration of 100 μg/ml. Thermosensitive strains were incubated for 1 h at 37°C before adding cycloheximide. To examine Fzo1 degradation upon α-factor–induced cell cycle arrest, early log phase cells grown on YPD were treated with 20 mM sodium citrate and 10 μM α-factor (or 0.1 μM in strains). At the indicated time points, cells corresponding to 3 OD units were collected and lysed at alkaline pH (), and protein extracts were analyzed by SDS-PAGE and immunoblotting. Mean values of at least three independent experiments are shown. Wild-type cells expressing Mdm30 and variants thereof from a copper-responsive promoter were grown on synthetic media without an additional copper supplement. Crude mitochondria obtained from 60 OD units of these cells were solubilized for 15 min in 1% (wt/vol) digitonin in immunoprecipitation buffer (150 mM KAc, 4 mM MgAc, 30 mM Tris/HCl, pH 7.4, 1 mM PMSF, and 1 mM ATP). After a clarifying spin for 15 min at 45,000 , supernatants were incubated for 1 h at 4°C under constant mixing with protein A–Sepharose and polyclonal antibodies against the GTPase domain of Fzo1. After two washing steps in immunoprecipitation buffer and one washing step in 10 mM Tris/HCl, pH 7.4, bound material was eluted with SDS sample buffer and used for Western blot analysis. Antisera against the COOH terminus of Fzo1 (), the GTPase domain of Fzo1 (provided by J. Nunnari, University of California, Davis, Davis, CA), the HA epitope (3F10; Roche), the FLAG epitope (M2-F3165; Sigma-Aldrich), or the myc epitope (9B11; Cell Signaling Technology) were used. Overnight cultures were analyzed by epifluorescence microscopy on a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) using a 100× oil-immersion objective. GFP fluorescence was visualized in living cells. Images were acquired with a camera (Quantix; Photometrix) and processed with MetaMorph 4.5 software (Universal Imaging Corp.).

Mitochondrial morphology is essential for the functions of the organelle. Mitochondria fuse and divide in highly regulated manners, and these two activities play a central role in morphogenesis of mitochondria (; ). Many components involved in mitochondrial fusion and division are highly conserved in a variety of organisms ranging from yeast to humans (; ). Among them, Mgm1p is a conserved dynamin-related GTPase essential for fusion, morphology, inheritance, and the genome maintenance of mitochondria in yeast (; ; ; ). Mutations in the human homologue of Mgm1p, OPA1, are linked to autosomal dominant optic atrophy (; ). There are two species of Mgm1p, an 84-kD form (s-Mgm1p) and a 97-kD form (l-Mgm1p) (). These two isoforms exert partially overlapping but distinct functions. Both forms are required for mitochondrial morphology and mitochondrial DNA (mtDNA) maintenance, whereas only l-Mgm1p is required for mitochondrial fusion (; ; ). In addition, these two isoforms have distinct localization and topologies (; ). l-Mgm1p is inserted in the inner membrane with a single transmembrane domain. In contrast, s-Mgm1p lacks the TM domain and is peripherally associated with both the outer and inner membranes (OMs/IMs) in the intermembrane space (IMS). Generation of the two forms of Mgm1p is mediated by a unique sorting process called alternative topogenesis (). Mgm1p is synthesized in the cytosol with a presequence followed by two adjacent hydrophobic segments (TM). Mgm1p is imported into mitochondria via the translocase of the OM, and its first TM segment (TM1) is then inserted into the IM-localized TIM23 translocon. After cleavage of the presequence by the matrix-localized processing protease, one of two choices is made. In one case (, blue arrow), Mgm1p diffuses out of the TIM23 channel into the IM, forming l-Mgm1p. Alternatively (, yellow arrows), Mgm1p is further translocated into TIM23 by the action of the matrix mtHsp70 motor and the PAM complex (). When the second TM (TM2) reaches the IM, Mgm1p is cleaved by the rhomboid protease Pcp1p (; ; ), forming s-Mgm1p. Supporting this branched sorting pathway, pulse-chase studies show no precursor–product relationship between l-Mgm1p and s-Mgm1p (; ). Instead, both forms of Mgm1p appear simultaneously. Here, we identify Ups1p, a novel IMS protein which regulates the sorting of Mgm1p in the topogenesis pathway and mitochondrial morphology. To identify components required for the novel topogenesis pathway for Mgm1p, we screened yeast mutants for those with altered amounts of l- and s-Mgm1p. In particular, we examined the steady-state levels of Mgm1p in 23 yeast knockouts recently shown to have mitochondrial morphology defects (; ). In addition to cells lacking the protease Pcp1p, we found that knockouts of showed dramatically reduced amounts of s-Mgm1p (). We named the gene for “unprocessed”. cells contain small but detectable amounts of s-Mgm1p. Immunoblotting of isolated mitochondria showed that l-Mgm1p is located in the mitochondria of cells (). In addition, all mitochondrial proteins tested (except for the IMS protein Cyb2p) were present in similar amounts in wild-type (WT) and mitochondria. We also found that the integrity of the OM is not compromised in mitochondria in protease studies (Fig. S1 A, available at ). Interestingly, the level of s-Mgm1p in cells varied with carbon sources. When cells were grown in the fermentable carbon source glucose (YPD), 53% of Mgm1p was converted to s-Mgm1p in WT cells. In contrast, only 1% of Mgm1p was present as s-Mgm1p in cells in YPD (). However, similar amounts of s-Mgm1p were found in WT (46%) and cells (47%) when grown in the nonfermentable carbon sources glycerol and ethanol (YPGE). The carbon source dependence of Ups1p for Mgm1p processing suggests that yeast cells carry a redundant Ups1p-like activity that is repressed when cells are grown on glucose. In addition to Mgm1p processing, Ups1p is also required for normal mitochondrial shape () and cell growth () in a carbon source–dependent manner. Mitochondria in cells in YPD, but not in YPGE, showed strikingly altered morphology, including fragments, short tubules, and aggregates of fragments (). These altered morphologies are similar to those seen in and mutants (; ; ,). Quantitation of mitochondrial morphology in cells grown in YPD shows that only ∼25% of cells contained normal tubular mitochondria, unlike WT cells (98%) (). In contrast, in YPGE, both WT (98%) and cells (97%) contained normal mitochondrial tubules. Other organelles including the actin cytoskeleton, the ER, and vacuoles showed WT morphology in cells (Fig. S2, available at ). Although and cells both contain lower amounts of s-Mgm1p, there are clearly differences between the two strains. For example, although cells lack mtDNA and are inviable on YPGE, cells contain normal mtDNA nucleoids () and are able to grow on nonfermntable medium (). Because s-Mgmp1p is required for mtDNA maintenance (; ; ), it is likely that the small but significant amounts of s-Mgm1p in cells are sufficient for mtDNA maintenance. Pcp1p has been shown to cleave another IMS protein, Ccp1p (), and, cells accumulated the intermediate form of Ccp1p (i-Ccp1p; ). In contrast, cells contained normal amounts of the mature form of Ccp1p (m-Ccp1p; ), suggesting that Ups1p specifically regulates Mgm1p processing and is not simply required for the activity of Pcp1. encodes a 20-kD protein (176 amino acids). Sequence analysis predicts that Ups1p contains an MSF1 domain but lacks a typical presequence and TM domain (Fig. S2). The MSF1 domain consists of ∼150 residues and is named after the yeast Msf1' protein. Although Msf1'p might be involved in intra-mitochondrial protein sorting (), the exact function of Msf1'p is unknown. We found that Ups1p is homologous to multiple proteins in many organisms, from yeast to humans ( and Fig. S2). In the yeast genome, Ups1p is similar to two other proteins: Msf1'p (30% identical) and Ydr185cp (30% identical). However, unlike cells, and cells contained only slightly reduced amounts of s-Mgm1p (not depicted). Humans have four proteins related to Ups1p, with PREL1 showing the highest homology (31% identical; and Fig. S2). The function of PRELI is unknown, but it is highly expressed in liver, lymph node, and leukocytes () and has been localized to mitochondria (). We found that PRELI can functionally replace Ups1p in cells. When PRELI is expressed from the constitutive promoter, the levels of s-Mgm1p in cells were indistinguishable from WT cells (). PRELI also rescued the mitochondrial shape defect () and growth defect () in cells. Confirming the mitochondrial location of PREL1, we found that a PREL1-GFP fusion protein colocalized with mitochondria in HeLa cells (). Our observations indicate that the function of Ups1p is evolutionarily conserved among eukaryotes. Ups1p is peripherally associated with the outside of the mitochondrial IM. Subcellular fractionation () and fluorescence microscopy () show that Ups1p is associated with mitochondria. Interestingly, truncation studies showed that the first 80 residues of Ups1p, but not the first 40 residues, are sufficient for mitochondrial localization (). Protease studies demonstrated that Ups1p is located in the IMS. We found that Ups1p-myc was completely digested by trypsin treatment only after the OM was disrupted by osmotic shock, similar to the IMS-facing protein Tim23p (). In contrast, the surface receptor Tom70p was digested by trypsin in intact mitochondria. We noticed that after osmotic shock, ∼50% of Ups1p-myc was associated with the mitoplast membrane (). To further characterize the membrane-bound fraction of Ups1p, mitoplasts were treated with 1.5 M sodium chloride or 0.1 M sodium carbonate (). We found that Ups1p-myc, but not the integral protein Tim23p, was completely extracted from the mitoplasts with sodium carbonate, suggesting that the association of Ups1p with membranes is peripheral. When membrane vesicles from Ups1p-myc mitoplasts were separated on sucrose gradients, Ups1p-myc cofractionated with the IM protein Fβ, but not with the OM protein porin (). Thus, Ups1p is a mitochondrial IMS protein, a half of which is peripherally associated with the IM. We also found that Ups1p-myc is present in a protein complex of ∼170 kD using blue-native gel electrophoresis of digitonin-solubilized mitochondria (). Importantly, the Ups1p complex does not contain Fzo1p, Ugo1p, Mgm1p, or Pcp1p because the size of the complex did not change in , , , or mitochondria (). We also noticed that the level of the Ups1p complex is slightly reduced in mitochondria. This probably reflects reduced levels of Ups1p in mitochondria (not depicted), suggesting that Mgm1p may be important for the stability of Ups1p. Furthermore, we also found no interaction of Ups1p with Fzo1p, Ugo1p, or Mgm1p in immunoprecipitation studies (not depicted). Suggesting that Ups1p functions in the sorting of Mgm1p, we find that alterations in the first hydrophobic segment of Mgm1p bypass the requirement of Ups1p for the formation of s-Mgm1p. In earlier studies, TM1 of Mgm1p was found to be critical for proper Mgm1p topogenesis (). When the hydrophobicity of TM1 was decreased by replacing glycine with aspartic acid (Mgm1p), TM1 was no longer efficiently arrested in the TIM23 translocon, and very little l-Mgm1p was formed. Instead, most of Mgm1 is now pulled further through the TIM23 channel, allowing access of TM2 to the Pcp1 protease; most of Mgm1 was thereby converted to s-Mgm1p (, lanes 1 and 2). We found that the Mgm1 alteration alleviates the need for Ups1p in the production of s-Mgm1p. As shown in (lane 3), cells expressing WT Mgm1p from a plasmid contained mostly l-Mgm1p, and only a small amount of s-Mgm1p was seen. cells, most of Mgm1p was converted to s-Mgm1p (, lane 4). cells expressing Mgm1p (, lanes 2 and 4). Therefore, Ups1p is not simply required for the activity of Pcp1p, or for the stability of s-Mgm1p. Our results also suggest that additional substrates of Ups1p activity exist besides Mgm1p. In cells expressing Mgm1p, normal ratios of l-Mgm1p and s-Mgm1p are seen () and Fzo1p and Ugo1p are normally expressed (), yet these cells contain fragmented mitochondria () and still show growth defects (). We found that Ups1p is not directly required for protein import into mitochondria using an in vitro import assay (Fig. S1 C). We integrated the 1 promoter in front of the open reading frame and isolated mitochondria from cells grown in YPD for 13 h. Immunoblotting shows that Ups1p is undetectable in the Ups1-depleted mitochondria (Fig. S1 B). In addition, the amount of s-Mgm1p is also highly reduced. Using the mitochondria, we examined protein import for Mgm1(1–228)-DHFR (), the matrix-targeted protein Su9-DHFR, and the IM proteins Tim22p and Tim18p. Mgm1(1–228)-DHFR consists of DHFR fused to the first 228 residues of Mgm1p, which contains two TM segments and can be cleaved by Pcp1p (). We found that the Ups1p-depleted mitochondria show no detectable import defect for all proteins we tested. Therefore, these data demonstrate that Ups1p is not required for mitochondrial protein import, and suggest that the processing defect for Mgm1p does not result from protein import defects. In this assay, we found similar amounts of the processed form of Mgm1(1–228)-DHFR in WT and the Ups1p-depleted mitochondria (Fig. S1 C). However, in both mitochondria the efficiency of the cleavage is considerably low (∼10%; Fig. S1 C) compared with Mgm1p processing seen in vivo (∼50%; ). This is consistent with previous observations that only a minor fraction of Mgm1p(1–228)- DHFR is processed in vitro (). The previous and our current studies suggest that the sorting of Mgm1p may not be a rate-limiting step in in vitro assays. It is possible that the protease activity of Pcp1p is highly reduced in these in vitro assays and becomes limiting. Supporting this idea, the mutation G→D in Mgm1p(1–228)-DHFR only slightly increases the efficiency of the processing (approximately twofold) in in vitro import assays, whereas in vivo the same mutation converts almost all Mgm1p to s-Mgm1p (; ). Our study identifies a novel IMS protein, Ups1p, which regulates the alternative topogenesis of Mgm1p. If, as our results suggest, Ups1p plays a role in Mgm1p sorting, then how does it achieve this function? In particular, how does it increase the yield of s-Mgm1p during import? In the normal sorting of Mgm1p, about half of Mgm1p arrest with TM1 in the TIM23 complex and then diffuse laterally out of the translocon into the lipid bilayer of the IM as l-Mgm1p. In one scenario, Ups1p may control the gating activity of TIM23, thus preventing all of Mgm1 from exiting the import channel, allowing some of Mgm1p to be further translocated by TIM23 and processed by Pcp1p to form s-Mgm1p. In another scenario, Ups1p may recruit or regulate the mt-Hsp70–containing PAM complex. By increasing the activity of PAM, the stop-transfer function of TM1 of Mgm1p could be countered, and more molecules could be pulled through the TIM23 complex. Arguing against these two possibilities, we find that Ups1p does not stably associate with the import machinery. Ups1p-myc is found in a complex of ∼170 kD, clearly distinct from the TIM23 complex (90 kD) (), and Ups1p does not coimmunopreciptate with known IM import components including Tim23p, Tim50p, Tim17p, Tim21p, Pam18p, Hsp70p, and Tim44p (unpublished data). In a third scenario, Ups1p may regulate the intramitochondrial level of ATP. Consistent with this idea, inactivation of Atp6p, a subunit of FF-ATP synthase, altered the ratio of l-Mgm1p to s-Mgm1p (). However, the level of s-Mgm1p in the mutants is only slightly decreased (Halern et al., 2004). Therefore, the dramatic decrease in the s-Mgm1p levels in cells cannot be explained simply by reduction in the ATP level. In addition, cells without mtDNA, thereby defective in respiration, normally produce s-Mgm1p (, lane 1). In a fourth model, Ups1p may bind directly to Mgm1 to facilitate sorting. Acting like a chaperone, Ups1p may stabilize a conformation of Mgm1p that masks the stop-transfer function of TM1. Those Mgm1 proteins bound by Ups1p would be pulled further toward the matrix and converted to s-Mgm1p. Although we found no stable interaction between Ups1p and Mgm1p, Ups1p might transiently interact with Mgm1p during the topogenesis. Additional studies are clearly needed to clarify the role of Ups1p in Mgm1p sorting. Yeast cells actively remodel their mitochondrial shape and number in response to different growth conditions and carbon sources (). This plasticity is performed, at least in part, by regulated fusion and division (; ). Although the precise mechanisms of l-Mgm1p and s-Mgm1p are not known, it is clear that both forms of Mgm1p are required for normal mitochondrial dynamics (; ; ). By affecting the ratios of the two forms of Mgm1p, Ups1p is in an ideal position to control yeast mitochondrial shape and number. Because human PREL1 can substitute for Ups1p in Mgm1p sorting, it is likely that PREL1 normally acts on OPA1, the human homologue of Mgm1p. However, it has not been directly tested if the human rhomboid protease cleaves OPA1. Alternative topogenesis of the OPA1 isoforms may help explain the incredible diversity of mitochondrial shape, number, and distribution observed in different mammalian cell types (; ). Yeast strains used in this study are listed in Table S1. Complete disruption of the gene was constructed by PCR-mediated gene replacement as described previously () using the gene from the pRS400 plasmid () with primers 1729 and 1730, into diploid strain FY833/844 (). Heterozygous diploids were sporulated and dissected to obtain strains. strain was constructed by crossing a strain and α strain. -myc- strain, which expresses Ups1p-myc, was constructed by homologous recombination in FY833/844 using the myc-TRP1 cassette from pFA6a-13myc-TRP1 () with primers 1731 and 1732. Heterozygous diploids were sporulated and dissected to obtain strain. stain, which expresses Ups1p from the promoter, was constructed by homologous recombination using the kanMX6-GAL1 cassette from pFA6a-kanMX6-PGAL1 () with primers 1749 and 1750. Yeast cells were grown in media including YPD (YP medium containing 2% glucose), YPGE (YP medium containing 2% glycerol and 3.2% ethanol), and YPGalSuc (YP medium containing 2% galactose and 2% sucrose). Standard genetic techniques were used (). pRS314-PRELI, a plasmid expressing human PRELI from the promoter, was constructed as follows. The promoter was PCR amplified from yeast genomic DNA using primers 1995 and 1996, and digested with KpnI and XhoI. Human PRELI was PCR amplified from pOTB7-PRELI (IMAGE ID 3505068; Open Biosystems) using primers 1753 and 1999, and digested with XhoI and SacII. A DNA fragment carrying the terminator was obtained by digesting pAA2 (), using SacII and SacI. The DNA fragments were subcloned into KpnI/SacII-digested pRS314. pRS313-Su9-GFP, a - plasmid expressing GFP fused to the presequence of subunit 9 of the F-ATPase of from the promoter was constructed as follows. The Su9 presequence (residues 1–69) was PCR amplified from pGEM4-Su9-DHFR using oligos 1871 and 1872. The PCR fragment was digested with EcoRI and XbaI, and cloned into EcoRI/XbaI-digested pRS313-ADH1-COX4-GFP () to replace the Cox4p presequence by the Su9 presequence. To form pRS314-UPS1, a plasmid expressing Ups1p, was PCR amplified from yeast genomic DNA using primers 1727 and 1728, digested with XhoI and NotI, and cloned into pRS314. To form pRS315-UPS1-GFP, a plasmid expressing Ups1p fused to GFP, was PCR amplified from pRS314-UPS1 using primers 1727 and 1994, digested with XhoI and NotI, and cloned into pAA1, which encodes GFP. To form truncated versions of Ups1p, primers 2060 for Ups1p(1–80) and 2061 for Ups1(1–40)p were used instead of 1994. To form pRS314-MGM1, glycine at residue 100 of pRS314-Mgm1p () was replaced by aspartic acid using site-directed mutagenesis according to manufacturer's instructions (QuickChange; Stratagene). pSP6-MGM1(1–228)-DFHR was constructed as described previously (). Cell lysates were prepared as described previously (). In brief, cells (1 OD unit) at log phase were collected by centrifugation and washed in buffer A (2 mM EDTA, 2 mM PMSF, 100 μM TPCK, and protease inhibitor cocktail; Sigma-Aldrich). Cells were resuspended in 500 μl of buffer A, and 100 μl of lysis buffer (1.85 M NaOH and 106 mM β-mercaptoethanol) was added. After incubation for 10 min on ice, 316 μl of 100% TCA was added. Samples were incubated for 15 min on ice, washed twice with acetone, resuspended in 50 μl SDS-PAGE sample buffer, and boiled for 5 min. Proteins (8.3 μl per lane) were analyzed by SDS-PAGE and immunoblotting. Antibodies raised against the COOH-terminal 10 amino acids of Mgm1p () were affinity purified against the same peptide using SulfoLink Coupling Gel (Pierce Chemical Co.) according to the manufacturer's instructions. Antiserum to Ups1p was produced using the peptide corresponding to residues 162–175 (FVIQKLEEARNPQF) and affinity purified against the same peptide as described above. WT cells were grown in minimal medium lacking methionine and cysteine to log phase. Cells were harvested, resuspended in 40 mM KPi, pH 6.0, and 1% glucose, and labeled for 5 min at 30°C in the presence of 0.1 mCi/ml [S]Translabel (MP Biomedicals). The chase was initiated by adding unlabeled methionine and cysteine to 20 μM. Cells were collected and processed for immunoprecipitation as described previously (). 600 μg of mitochondria were resuspended on ice in 96 μl of 50 mM NaCl, 5 mM 6-aminocaproic acid, 100 mM Bis Tris, pH 7.0, 50 μg/ml α2 macroglobulin, and protease inhibitor cocktail (Sigma-Aldrich). Then, 24 μl of 10% digitonin was added, and mitochondria were solubilized for 15 min. After centrifugation at 12,500 for 10 min, 165 μg of mitochondrial proteins was separated on 6–16% acrylamide gradient gels (). HeLa cells were grown in DME (Invitrogen) with 10% fetal calf serum (Atlanta Biologicals). The open reading frame of PRELI was PCR amplified from pOTB7-PRELI using oligos 1753 and 1760, digested with XhoI and EcoRI, and subcloned into XhoI/EcoRI-digested pEGFP-N1. (CLONTECH Laboratories, Inc.). 4 × 10 HeLa cells were transfected with 1 μg pPRELI-GFP using Lipofectamine (Life Technologies, Inc.) following the manufacturer's directions. Yeast cells were observed using a microscope (Axioskop; Carl Zeiss MicroImaging, Inc.) with a 100× Plan-Neofluar objective. Fluorescence and differential interference contrast images were captured with a CCD camera (Orca ER; Hamamatsu) using OpenLab software version 3.0.8 (Improvision, Inc.). HeLa cells were viewed using a microscope (Axiovert; Carl Zeiss MicroImaging, Inc.) with a 40× Achrostigmat objective. Images were captured with a Photometrics CoolSNAP camera (Roper Scientific) using IPLab software (Scanalytics). Mitochondria were isolated from WT and strains and examined for protein import as described previously (). Immunoprecipitation was performed as described previously (), except that 1% digition was used instead of Triton X-100. Fig. S1 shows that Ups1p is not required for the integrity of the OM, for protein import into mitochondria, and for the morphology of actin cytoskeleton, the ER and vacuoles. Fig. S2 shows that Ups1p is evolutionarily conserved among eukaryotes. Table S1 shows yeast strains and PCR primers used in this study. Online supplemental material is available at .

The NF-κB family of transcription factors controls diverse mammalian signaling responses that mediate cell survival, inflammation, and immune response (; ; ). Functional NF-κB exists in a dimeric form that is composed of combinations of five proteins containing a Rel homology region, i.e., cRel, RelA, RelB, p50, and p52 (; ). In resting cells, the NF-κB dimer is bound to inhibitor IκB proteins, i.e., IκBα, -β, and -ɛ, which inhibit NF-κB DNA-binding activity and prevent its nuclear accumulation. Activation of the NF-κB signaling pathway relies upon signal-dependent phosphorylation and degradation of the IκB proteins that result in subsequent nuclear translocation of the NF-κB dimer (). Termination of NF-κB activity after cellular stimulation is critical, as deregulated inflammatory gene expression can be detrimental to the health of the organism, and several attenuation mechanisms have been described (; ). Importantly, IκBα, which is a target gene of NF-κB, is induced by numerous NF-κB–inducing stimuli, resulting in the termination of NF-κB DNA-binding activity and nuclear localization (; ). Temporal control of NF-κB activity has been shown to mediate stimulus-specific gene expression programs in response to different inflammatory stimuli (), and understanding the dynamic regulation of NF-κB by IκB proteins is of critical importance. Negative feedback that is mediated by IκBα was shown to confer the propensity for oscillatory NF-κB nuclear activity, both when examined biochemically in gene knockout cells containing only the IκBα isoform (; ) and when examined by microscopy using transiently transfected IκBα and RelA proteins fused to fluorescent moieties (; ). The oscillations are not apparent in cells containing all three IκB proteins at normal expression levels (, A and C; ; ). These observations suggested that IκBβ and/or -ɛ proteins play a role in dampening IκBα-mediated oscillations and determining the dynamics of NF-κB activity. The mechanism that confers dampening of oscillations in NF-κB activity was proposed to involve the nuclear accumulation of newly synthesized IκBβ that binds nuclear and promoter-bound NF-κB and shields it from IκBα-mediated nuclear export (; ). This mechanism was included in our computational model that recapitulates NF-κB activation in response to TNF stimulation (), but our later studies were unable to observe the IκBβ effect in murine embryonic fibroblasts (MEFs), which are the cells for which the model was constructed (unpublished data). Removal of the mathematical term for this mechanism from the model resulted in highly oscillatory NF-κB responses (). Although genetic evidence points to important roles for IκBβ and -ɛ in regulating the dynamics of NF-κB signaling, the mechanisms by which they function remained unclear. In this study, we investigated the dynamic behavior of all three canonical IκB isoforms, especially by contrasting IκBɛ and -β functions with that of IκBα. Our studies revealed that IκBɛ expression is, in fact, highly NF-κB inducible, and that it mediates functional negative feedback on NF-κB activity; however, it does so in antiphase to that of IκBα. Two antiphase negative feedbacks emerge as an important regulatory module that may be present for the dynamic control of signaling in other pathways as well. We constructed probes for RNase protection assays (RPAs) that allowed for the simultaneous quantitative monitoring of all three IκB mRNAs to characterize the regulation of IκBɛ and -β gene expression in response to stimulation. Analysis of the mRNA levels in MEFs that were stimulated with TNF confirmed that IκBα was strongly induced (). IκBβ showed only weak induction, which suggests that it is not a strong NF-κB–responsive gene in this cell type. Remarkably, IκBɛ transcription was highly induced by TNF stimulation, and quantitation of these results showed that IκBɛ was induced to a higher degree than IκBα (). Although the exact induction folds varied in replicate assays using separate MEF cell stocks (unpublished data), IκBɛ fold induction was consistently higher than that for IκBα. Furthermore, IκBɛ expression was also induced in response to LPS in MEFs, the macrophage cell line RAW264.7, and the B cell line 70Z (unpublished data), indicating that the dynamic control of its synthesis is not specific to TNF or to fibroblasts. Analysis of IκB mRNA levels in MEF cells deficient in NF-κB showed no induction of any of the three IκBs (). The temporal profile of IκBα transcript induction during chronic TNF stimulation is well described (). It shows rapid activation as early as 15 min, a peak within 1 h, and a slow attenuation over many hours. We observed a similar activation profile for IκBɛ induction, but were surprised to note a distinct 45-min onset delay (). This suggests that the IκBɛ promoter may involve a delay mechanism, such as a requirement for the activation of an NF-κB–responsive transcription factor (a feed-forward regulation). However, the inhibition of protein synthesis by cycloheximide did not attenuate transcriptional activation of either IκBα or -ɛ ( and Fig. S1, available at ) and, thus, does not support the notion of feed-forward regulation. It has been previously shown that transient TNF stimulation leads to transient nuclear NF-κB activity lasting only ∼60 min (). To determine whether short stimulation would efficiently induce IκBɛ expression, an RPA was performed on MEF cells that were stimulated for 15 min with TNF. The results show that IκBɛ transcription is still activated with an onset delay and with an induction profile similar to that of chronically stimulated cells ( and Fig. S1). Because NF-κB activity is diminished when IκBɛ mRNA levels are still rising, and because cycloheximide treatment precludes a feed-forward mechanism, we suggest that the time delay in the activation of IκBɛ transcription occurs after NF-κB recruitment to the promoter. Further study is required to elucidate the mechanism of this delay. NF-κB–responsive syntheses of IκBɛ and -β were added to the mathematical model to explore the functional consequences imparted by these negative feedback regulators upon NF-κB activity. To determine the kinetic parameter values for the inducible transcription of each isoform, the temporal responses of IκB transcription in the model were constructed such that the mRNA induction profiles calculated by the model correlated with our RPA data (). This required the fitting of parameters defining transcription and translation rates and mRNA stability. The new model also includes revised IκB protein degradation parameters from earlier studies (unpublished data). The revised model recapitulates IκBα protein degradation immediately after IκB kinase (IKK) activation and rapid synthesis in response to NF-κB nuclear localization (). Chronic stimulation results in repeated IκBα protein degradation and synthesis (). In addition, the model shows delayed induction of IκBɛ and -β protein syntheses. The low inducibility of IκBβ transcription results in very low IκBβ protein synthesis, whereas the high inducibility of IκBɛ transcription results in notable accumulation of IκBɛ protein (). Earlier studies revealed oscillatory NF-κB activity in cells lacking IκBβ and -ɛ () and in cells in which NF-κB–inducible IκBα was overexpressed (), whereas in wild-type cells late NF-κB activity (beyond 2 h) was remarkably steady (). However, the underlying dampening mechanism that results in steadied late activity remained obscure. Because induced synthesis of IκBɛ is delayed, we reasoned that IκBɛ may mediate an antiphase negative feedback that provides effective dampening of IκBα-mediated oscillations. Indeed, our simulations of signaling modules lacking IκBɛ revealed oscillations in nuclear NF-κB that persist with a higher amplitude than those that contain IκBɛ and represent wild-type cells (). In contrast, simulations of cells lacking IκBβ do not show such aberrant oscillations, whereas systems lacking both IκBɛ and -β do. We set out to examine these predictions experimentally, using nuclear extracts prepared from TNF-treated MEFs harboring genetic deficiencies for IκBɛ and/or -β. We measured NF-κB DNA-binding activity by electrophoretic mobility shift assay (EMSA; ) and nuclear localization of the NF-κB protein RelA by Western blot (Fig. S2, available at ). In both assays, all four cell types exhibited fast induction of nuclear NF-κB in response to TNF stimulation by 15 and 30 min, a transient trough at 60–75 min, and subsequent recovery at 90–120 min. However, a second trough at 135–150 min was most pronounced in κɛ and κβɛ cells, as was a third trough at ∼225 min. These studies suggest that negative feedback, provided by IκBɛ in antiphase to that of IκBα, is the primary mechanism that dampens the propensity for oscillations in NF-κB activity. The regulatory motif consisting of two antiphase negative feedback systems may be present in other signaling pathways to control the dynamics of signal transduction, and variations in the relative strength of the two systems may provide for altered response dynamics to the same stimulus. A similar model has recently been proposed for two NF-κB–inducing signaling pathways emanating from the TLR4 receptor (). In contrast to the two interacting negative feedback mechanisms, this model depicts the coupling of two positive oscillatory signals in an antiphase relationship that produces stable NF-κB activity in response to LPS stimulation in wild-type cells. As the temporal control of NF-κB activity determines NF-κB–responsive gene expression (; ; ), the interaction of antiphase regulation by IκBα and -ɛ may contribute to the regulation of stimulus-specific and cell type–specific gene expression programs by modulating the dynamics of this transcription factor. We used the computational model to identify conditions in which NF-κB–responsive IκBɛ expression would mediate negative feedback on stimulus-induced NF-κB activity and found that the most significant role for IκBɛ was in systems with reduced IκBα. To model such conditions, we used computational simulations to study the temporal profile of nuclear NF-κB in response to 15-min stimulation in systems lacking IκBα, -ɛ, or both (). Systems containing all three IκBs show rapid nuclear localization of NF-κB followed by removal from the nucleus within 1 h, as previously shown (). However, in systems lacking IκBα, we predicted effective down-regulation of NF-κB activity in the third hour and beyond. In this context, IκBɛ deficiency results in prolonged NF-κB activity, whereas in systems containing high IκBα expression it does not have an effect. We used IκBα-deficient MEFs as a model for cell types that have reduced IκBα expression. These MEFs showed NF-κB activity to last ∼3 h in response to 15-min transient TNF stimulation, after which it was dramatically attenuated (). In contrast, cells that were deficient in both IκBα and -ɛ showed a pronounced delay in attenuation, with NF-κB still present in the nucleus even at 6 h. Wild-type and IκBɛ-deficient cells are nearly indistinguishable, and both have strong NF-κB accumulation at 30 min and attenuation within 1 h. Collectively, these data strongly suggest that IκBɛ is responsible for the removal of NF-κB from the nucleus at late time points, allowing for dynamic functional interplay with the faster-acting feedback of IκBα. Temporal control of NF-κB localization by IκBα was shown to control NF-κB–responsive gene expression not only quantitatively () but also qualitatively (). To study the effects of IκBɛ-negative feedback on NF-κB–dependent gene expression, the transcription of five NF-κB–responsive genes was monitored by RPA after transient TNF stimulation in wild-type, κα, and κα κɛ cells. The genes encoding TNF, G-CSF, and LIF are inducibly expressed in fibroblasts upon TNF stimulation, but mRNA levels return to baseline within 3 h in wild-type cells. In IκBα-deficient cells, these genes are attenuated within 4 h (). In this context, the loss of IκBɛ-negative feedback results in a further delay in attenuation and quantitative deregulation of TNF, G-CSF, and LIF expression. Interestingly, the loss of both IκBα- and -ɛ-negative feedback has a dramatic qualitative effect for GM-CSF and MIP-2. Although these genes are not induced in wild-type or IκBα-deficient cells, both are strongly responsive to NF-κB activation when both IκBα- and -ɛ-negative feedbacks are absent. The data presented here demonstrate that IκBɛ-dependent negative feedback regulates the termination of NF-κB–responsive gene expression in both a quantitative (in the cases of TNF, G-CSF, and LIF) and qualitative (in the cases of GM-CSF and MIP-2) manner. The functional interplay between the antiphase IκBα- and -ɛ-negative feedback responses may explain differences in NF-κB–dependent gene expression profiles seen in various cell types. In MEFs, IκBɛ-mediated negative feedback appears to be secondary to that provided by IκBα in response to transient inflammatory stimuli, and it is therefore assumed that IκBα controls the bulk of the NF-κB–responsive gene expression (). However, the ratio of the abundance of IκBɛ in relation to IκBα is cell type–specific (; ; ; ), suggesting that IκBɛ may play a predominant role in NF-κB–responsive gene expression in particular cell types. Indeed, in vivo studies have shown that a deficiency of functional IκBɛ has physiological consequences (; ; ; ) and, thus, emphasize the notion that no IκB isoform functions on its own. To understand regulation of NF-κB activity in different cell types and in response to diverse stimuli, the interplay of all IκB isoforms within the IKK–IκB–NF-κB signaling module must be considered. Our studies aimed to quantitatively characterize the temporal expression profiles of the three IκB isoforms and to examine their functional consequences on NF-κB regulation. Earlier studies showed inducible IκBɛ expression (; ). We have demonstrated that induction of IκBɛ is NF-κB dependent and functions to attenuate NF-κB activity and terminate NF-κB–responsive gene expression. Based on these three criteria we conclude that IκBɛ mediates bona fide functional negative feedback regulation on NF-κB activity. Importantly, our studies reveal that inducible expression of IκBɛ is delayed by 45 min with respect to that of IκBα, thus, creating a two–negative feedback regulatory module that critically controls the dynamics of NF-κB activity. We suggest that the relative strength of the two feedback mechanisms and their temporal relationship to each other may account for cell type–specific dynamic regulation of NF-κB activity. The immortalized MEF cells used were previously described (). Cells were grown to confluency in DME containing 10% bovine calf serum and starved for 24 h in media containing 0.5% bovine calf serum. Stimulations were performed with 10 ng/ml TNF (Roche). Cells that were transiently stimulated with TNF were washed twice with 1× PBS after stimulation and returned to untreated media. EMSAs were performed as previously described (). Western blots using whole-cell extracts were performed as previously described (). IκBα and -β antibodies were obtained from Santa Cruz Biotechnology, Inc. (SC-371 and SC-945, respectively). An antigen-purified polyclonal mouse antiserum raised against recombinant full-length mouse protein was used for IκBɛ. Total cellular RNA was isolated from confluent and serum-starved cells with Trizol reagent (Invitrogen). Transcript levels were monitored with α-[P]UTP–labeled probes using a RiboQuant kit (BD Biosciences) according to the manufacturer's instructions. Data was obtained using a storage phosphor screen (GE Healthcare) and a variable mode imager (Typhoon 9400; GE Healthcare). Data was quantitated using ImageQuant software version 5.2 (GE Healthcare) by normalization to L32 and/or glyceraldehyde- 3-phosphate dehydrogenase after local background subtraction. IκB probes were designed to select for mature mRNA species by spanning exon–exon junctions. The following primer pairs were used to amplify fragments from reverse-transcribed RNA: 5′-TCGCTCTTGTTGAAATGTGG-3′ and 5′TGGAGATTTTCCAGGGTCAG-3′ (IκBα); 5′-GCCCTTAGTCTTTGGCTACG-3′ and 5′-TCTCAGCCACCAACACTCCT-3′ (IκBβ); and 5′-GGCAGACAGCTTTCTCATCC-3′ and 5′-TGAGGTCGCAGTCTTCAATG-3′ (IκBɛ). G-CSF, LIF, MIP-2, TNF, L32, and glyceraldehyde-3-phosphate dehydrogenase probes were obtained from RiboQuant sets (BD Biosciences). We previously constructed a computational model to describe NF-κB activation events in response to IKK activation by TNF (). This model comprises a singular NF-κB species, three IκB isoforms (IκBα, -β, and -ɛ), and IKK. Synthesis and degradation of the IκBs and cellular localization and interactions for all components were calculated using a system of ordinary differential equations. The model used in this study includes NF-κB–induced IκBɛ and -β transcription and was written in MatLab V7.0 (MathWorks) using previously described methods (). MatLab simulation files are available upon request. Fig. S1 supports , with quantitation of IκBα and -ɛ gene induction profiles. Fig. S2 supports , with super-shift and oligonucleotide competition EMSAs, as well as nuclear westerns for RelA. Table S1 contains the computational model parameters and reactions.

Insulin resistance, a condition in which the cells become resistant to the effects of insulin, is a major risk factor for type 2 diabetes as well as hypertension, dyslipidemia, and atherosclerosis (). Despite several investigations, the molecular mechanism underlying insulin resistance has not been adequately clarified. TNF-α is an adipocytokine and induces insulin resistance (). A TNF-α signal results in the phosphorylation of Ser of insulin receptor (IR) substrate 1 (IRS-1), in turn attenuating the metabolic insulin signal (). Many serine kinases such as JNK, glycogen synthase kinase 3, and mammalian target of rapamycin have been reported to phosphorylate serine residues of IRS-1 (). However, the serine kinase that precisely regulates metabolic insulin action is unclear. After the first report of type 2 diabetes being successfully treated with high-dose salicylate in 1901 (), numerous attempts have been made to identify the target molecules of salicylate. In 1998, salicylate was reported to be a strong inhibitor of the kinase activity of IκB kinase (IKK) β (). Since then, studies have focused on the IKK complex as a critical molecule for the development of insulin resistance (). The IKK complex consists of two catalytic subunits, IKK-α and IKK-β, and one scaffold subunit designated nuclear factor κB essential modulator (NEMO)/IKK-γ (; ; ). Among these subunits, IKK-β is a key insulin resistance molecule, as demonstrated by a study using the IKK-β knockout mouse (). A recent study showed the IKK complex to phosphorylate IRS-1 at Ser, which is associated with TNF-α stimulation and diminished insulin signaling (). However, whether IKK-β itself physically binds to IRS-1 is uncertain. Furthermore, the role of NEMO is also unclear. Myo1c is a motor protein that is classified as an unconventional myosin I. This class of myosins is widely distributed, having been identified in organisms from yeast to human. In adipocytes, Myo1c reportedly facilitates the recycling of vesicles containing glucose transporter 4 (). However, little is known about the molecular mechanisms regulating motor Myo1c–cargo interactions. We investigated the formation of the functional complex of signaling molecules containing IKKs and IRS-1 in response to insulin. We found that NEMO functions as a motor receptor, whereas Myo1c and the actin cytoskeleton facilitate translocation of the IKK complex to membrane ruffles or to the vicinity of IRS-1. This interaction between IKKs and IRS-1 is essential for TNF-α–induced phosphorylation of IRS-1 at Ser, which results in the inhibition of glucose uptake. Our present results suggest a novel mechanism whereby Myo1c–NEMO-mediated signaling complex formation plays a role in TNF-α–induced insulin resistance. Researchers have reported that IKK-β is crucial for TNF-α–induced IRS-1 serine phosphorylation (; ). However, the role of the NEMO/IKK-γ subunit is poorly understood. We first examined the intracellular localization of NEMO in differentiated 3T3-L1 adipocytes using anti-NEMO antibody. As shown in , NEMO results in a fine punctate or granular appearance throughout the cytoplasm under basal and TNF-α–treated conditions. In contrast, the addition of insulin to culture adipocytes yields the rapid translocation of NEMO to the cell periphery, especially in membrane ruffles visualized by staining with AlexaFluor596-phalloidin. This translocation is similar to that seen in other cell types (). Interestingly, treatment with the actin depolymerizer latrunculin B inhibited NEMO translocation, whereas the microtubule disrupter nocodazole did not (). These data suggest that insulin stimulates the accumulation of NEMO at membrane ruffles through the cortical actin network. Because NEMO has neither an actin-binding motif nor a membrane-targeting domain, we attempted to identify NEMO-binding proteins using mass spectrometry. 3T3-L1 adipocytes were infected with an adenovirus vector containing myc-tagged full-length NEMO and were treated with 20 ng/ml TNF-α or 100 nM insulin for 20 min. The cell lysates were immunoprecipitated with anti-myc antibody. The precipitates were resolved by SDS-PAGE and visualized with silver staining. With in-gel digestion followed by peptide mass fingerprinting, we identified two candidate proteins, Myo1c and actin, showing increased binding to NEMO in the presence of insulin (, arrowheads). A series of experiments were performed to confirm the interaction between NEMO and Myo1c. We first examined endogenous protein–protein interactions by immunoprecipitation using polyclonal anti-NEMO– and polyclonal anti-Myo1c–specific antibodies. As shown in , insulin treatment increased NEMO–Myo1c binding. We next confirmed this interaction using recombinant proteins in vitro. GST-tagged NEMO and His-tagged Myo1c were purified, mixed, and pulled down with each other. As shown in Fig. S1 (available at ), the interaction was easily detected by immunoblotting. These results suggest that NEMO and Myo1c interact directly in an insulin-dependent manner. Based on the results presented in and , we hypothesized that the IKK complex containing NEMO is transported from the cytosol to membrane ruffles by Myo1c. To examine this possibility, we conducted experiments using the dominant inhibitory cargo domain of Myo1c. Overexpression of this cargo domain (residues 767–1,028) has been shown to result in the dominant inhibition of cargo binding (). 3xFlag-tagged full-length Myo1c (wild type [WT]) or the dominant inhibitory cargo domain of Myo1c was cotransfected into culture adipocytes with enhanced GFP (eGFP)–tagged NEMO. In cells coexpressing Myo1c WT and eGFP-NEMO, NEMO showed marked translocation to membrane ruffles with insulin stimulation. Interestingly, Myo1c WT also accumulated in the membrane and enhanced Myo1c expression, resulting in the extensive formation of membrane ruffles (). In contrast, cells expressing Myo1c cargo domain and NEMO showed the marked inhibition of insulin-stimulated NEMO translocation. Similar inhibition of NEMO translocation was observed in Myo1c knockdown cells using adenovirus encoding short hairpin RNA (shRNA [Myo1c]; ). We observed the association between NEMO and Myo1c biochemically (). We also found that membrane targeting of NEMO requires a motor protein, Myo1c (). Collectively, the data indicate the scaffold protein NEMO to be transported to membrane ruffles by the motor protein Myo1c. These observations are consistent with our aforementioned hypothesis. A recent study showed that IKKs interact with IRS-1 and interfere with insulin signaling (). To confirm this interaction in culture adipocytes, we first examined the localization of endogenous NEMO and IRS-1 () or Xpress-tagged NEMO and eGFP-tagged IRS-1 (). In the basal state, IRS-1 was present in the cytoplasm, whereas with insulin stimulation, IRS-1 and NEMO colocalized to discrete foci in the cytoplasm as well as membrane ruffles. These observations on the intracellular localization of IRS-1 were very similar to those described in a study by . They reported a mechanism of IRS-1 signal down-regulation involving the formation of a sequestration complex containing IRS-1 in CHO-K1 cells. Interestingly, the large intracellular IRS-1 complexes (foci) appeared to be the negative regulatory machinery. We next examined the direct interaction between endogenous IRS-1 and NEMO by immunoprecipitation (). Interestingly, the overexpression of Myo1c markedly increased this interaction. In contrast, cells overexpressing the dominant inhibitory cargo domain of Myo1c showed a diminished interaction. One interpretation of these findings is that Myo1c mediates the interaction by delivering NEMO to IRS-1. This explanation is supported by another set of experiments shown in Fig. S2 (A and B; available at ). These experiments focused on Myo1c–IKK-β and IRS-1–IKK-β associations. When NEMO was knocked down, the Myo1c–IKK-β interaction was disturbed. Similarly, the overexpression of ΔN-NEMO (detailed in the next section) diminished the IRS-1–IKK-β interaction. These results, combined with the data shown in , are consistent with our hypothesis that Myo1c transports the IKK complex via binding to NEMO. Another interesting observation illustrated in was that insulin enhanced the association between IRS-1 and IKK-β. These data raise the possibility that insulin may assemble clusters of signaling molecules to facilitate the interaction between IKKs and IRS-1. To assess this possible new role of insulin, we performed two additional experiments. First, we observed IRS-1 Ser phosphorylation induced by TNF-α after treatment with various concentrations of insulin (). Although TNF-α–induced serine phosphorylation of IRS-1 was detected within 20 min even in the absence of insulin, a low concentration of insulin markedly enhanced TNF-α–induced Ser phosphorylation. These data are consistent with the results presented in . Next, we also observed Ser phosphorylation of IRS-1 in adipocytes expressing WT Myo1c and dominant inhibitory Myo1c. Overexpression of dominant inhibitory Myo1c diminished Ser phosphorylation of IRS-1 (Fig. S3 A, available at ). These results show that Myo1c promotes the interaction between IRS-1 and NEMO and mediates Ser phosphorylation of IRS-1. Furthermore, a low dose of insulin and an intact actin cytoskeleton may be necessary for clustering molecules related to TNF-α to down-regulate IRS-1. Because IRS-1 protein is a key mediator of insulin signaling, we next focused on the roles of Myo1c and NEMO in insulin signaling and glucose transport. We prepared WT NEMO and NH-terminal–deleted (ΔN; residues 101–412) NEMO constructs and performed a series of experiments. The NH terminal of NEMO is the IKK-β–binding site, and deletion of this site was shown to impair the binding of NEMO with IKK-β (). First, we assessed the effects of NEMO expression on upstream insulin signal cascades in 3T3-L1 adipocytes. Overexpression of WT NEMO induced the phosphorylation of IRS-1 Ser while decreasing Akt phosphorylation in the absence of TNF-α. Overexpression of ΔN-NEMO prevented the phosphorylation of IRS-1 Ser and inhibited Akt phosphorylation. We detected no changes in the tyrosine phosphorylation of the IR β chain (). Second, we measured 2-deoxyglucose uptake in 3T3-L1 adipocytes expressing NEMO constructs. Overexpression of WT NEMO abolished insulin-stimulated glucose uptake in the absence of TNF-α. In contrast, the overexpression of ΔN-NEMO completely blocked the inhibitory effects of TNF-α (). Third, we introduced shRNA into culture adipocytes to induce specific degradation of NEMO mRNA. NEMO protein expression was decreased to 20–30% of the control level (unpublished data). As expected, the deletion of NEMO almost completely blocked the inhibition of insulin-stimulated glucose uptake by TNF-α (). Under these conditions, IR-β tyrosine, IRS-1 Ser, and Akt Ser phosphorylations were examined in 3T3-L1 adipocytes. NEMO silencing by shRNA also prevented TNF-α–mediated IRS-1 Ser phosphorylation and inhibition of Akt phosphorylation without affecting IR tyrosine phosphorylation (). These data, combined with the data presented in and , suggest that motor protein Myo1c and its receptor protein NEMO mediate TNF-α–induced down-regulation of IRS-1 and glucose uptake. However, it was previously shown that the overexpression of ΔN-NEMO results in the loss of IKK kinase activity (). We confirmed the inhibition of IKK kinase activity in 3T3-L1 adipocytes expressing ΔN-NEMO as well as NEMO knockdown cells (Fig. S3, B and C). These observations indicated that NEMO plays a role in assembling the IKK complex and that this step is critical for IKK kinase activity. To avoid this bias and clarify the roles of Myo1c and NEMO in glucose uptake, we conducted another experiment using Myo1c cargo domain constructs. As shown in , overexpression of the cargo domain inhibited the TNF-α–induced suppression of glucose uptake. Together, these results provide evidence that NEMO may function as a receptor molecule for Myo1c and that Myo1c promotes the TNF-α–induced suppression of metabolic insulin action. In agreement with a previous study (), we confirmed that Myo1c cargo domain expression itself decreased insulin-stimulated glucose uptake by 30%. This may be a result of the inhibitory role of Myo1c on GLUT4 recycling (). Based on the aforementioned data, we propose a simple model whereby Myo1c and its receptor NEMO cooperatively facilitate IKK–IRS-1 complex formation, as illustrated in . NEMO is a scaffold protein of the IKK complex. Recent studies suggest that some scaffold proteins serve as links between molecular motors and intracellular vesicles, thereby functioning as cargo proteins (; ). In contrast, our data suggest that the scaffold protein NEMO links motor and signaling molecules as cargos. It is noteworthy that Myo1c organizes the signaling complex and serves as a platform for the two distinct signals to interact (i.e., the insulin signal and the TNF-α signal mediating insulin resistance). Finally, our results allow us to draw three conclusions. First, the motor protein Myo1c appears to participate directly in the mechanism of IRS-1–IKK complex formation in culture adipocytes. It is possible that NEMO is a molecular receptor linking motor (Myo1c) and cargo (IKK-α and -β). Second, NEMO and Myo1c may be involved in the TNF-α–induced Ser phosphorylation of IRS-1, resulting in the attenuation of insulin signaling and glucose transport. Third, Myo1c and the actin cytoskeleton may facilitate formation of the signaling molecule complex that participates in the TNF-α–induced down-regulation of IRS-1. In summary, our data suggest that Myo1c and NEMO are responsible for the mechanism of TNF-α–induced insulin resistance. Mouse full-length NEMO and IRS-1 were cloned by RT-PCR amplification with total mRNA from 3T3-L1 adipocytes. WT and an NH-terminal deletion mutant of NEMO containing amino acid residues 101–412 were subcloned into pEGFP-C2, pcDNA3.1His, pET-16b, and/or pGEX-6p-1 vectors. Mouse Myo1c cDNA was purchased from DNAFORM and subcloned into p3xFLAG-CMV7.1, pET-16b, and pGEX-6p-1 vectors. IKK-α and IKK-β cDNA were gifts from H. Nakano (Juntendo University School of Medicine, Tokyo, Japan). The following antibodies were used: anti-NEMO, antiphospho-Ser IRS-1, and antiphospho-Akt antibodies (Cell Signaling); anti-NEMO and anti–IRS-1 antibodies (Upstate Biotechnology); anti-Flag antibody (Sigma-Aldrich); anti-Xpress antibody (Invitrogen); and Cy3-conjugated anti–mouse IgG (Jackson ImmunoResearch Laboratories). Rabbit polyclonal anti-Myo1c antibody was generated against the peptide sequence DKSELSDKKRPE. All other antibodies were purchased from Santa Cruz Biotechnology, Inc. AlexaFluor596-phalloidin was obtained from Invitrogen. Mouse TNF-α was purchased from PeproTech. 3T3-L1 fibroblasts were grown in DME with 10% FBS at 37°C. The cells (3–4-d after confluence) differentiated into adipocytes with incubation in the same DME containing 0.5 mM isobutylmethylxanthine, 0.25 μM dexamethasone, and 4 μg/ml insulin for 3 d and were then grown in DME with 10% FBS for an additional 3–6 d. Differentiated 3T3-L1 adipocytes were transfected by electroporation. The cells were then replated onto coverslips and allowed to recover for 48 h followed by stimulation with 100 nM insulin or 20 ng/ml TNF-α for 15 min at 37°C. Then, 5 μM latrunculin B or 30 μM nocodazole were added 60 min before treatment with insulin. Minimum concentrations of these agents required for disrupting the cytoskeleton in culture adipocytes were determined previously (). Cells were fixed with 3.7% formaldehyde in PBS, permeabilized with buffer A (0.5% Triton X-100 and 1% FBS in PBS) for 15 min, and incubated for 2 h with primary antibodies at room temperature. The cells were washed and incubated with an appropriate secondary antibody or AlexaFluor596-phalloidin for 30 min. The coverslips were washed thoroughly and mounted on glass slides. Immunostained cells were observed at room temperature with a laser-scanning confocal microscope (LSM5 PASCAL; Carl Zeiss MicroImaging, Inc.) and its two-channel scanning module equipped with an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.). The inverted microscope used the 63× NA 1.4 oil objective lens run by LSM5 processing software (Carl Zeiss MicroImaging, Inc.) and Adobe Photoshop CS2. Adenovirus producing mouse WT NEMO, deletion mutant NEMO (residues 101–412), mouse WT Myo1c, and dominant inhibitory Myo1c (residues 767–1,028) were prepared using an AdEasy Adenoviral Vetor System (Stratagene). shRNA was designed to have a 5′-AAGGATTCGAGCAGTTAGTGAGC-3′ sequence. Synthetic complementary single-stranded oligonucleotide DNA was annealed, and the double-stranded DNA of the target sequence was created. This annealed DNA was inserted into a pcPUR+U6i cassette (), and the insert was transferred to an AdEasy Adenoviral Vector System. This shRNA system decreased NEMO protein expression to 20–30% of the control level. Myc-NEMO was expressed in 3T3-L1 adipocytes. 2 d thereafter, cells were serum starved for 2 h and stimulated with 100 nM insulin or 20 ng/ml TNF-α for 20 min. Cell lysates were prepared and immunoprecipitated with anti-myc antibody. Samples were resolved by SDS-PAGE, and proteins were visualized by silver staining. The bands were excised and subjected to in-gel digestion according to the method described by with minor modifications. Mass spectra were acquired using a time of flight mass spectrometer (Voyager DE Pro; Applied Biosystems). The search engine for the peptide mass fingerprint was the web-based Mascot (Matrix Science). GST-fused NEMO and Myo1c were expressed using a pGEX-6p-1 vector in BL21 cells. His-tagged NEMO and Myo1c were expressed using a pET-16b vector in BL21 (DE3) cells. GST-fused protein and His-tagged protein were mixed in PBS and pulled down with glutathione–Sepharose beads (GE Healthcare). Protein interactions were detected by Western blotting using anti-His antibody and anti-GST antibody. Target sequences used in shRNA were the same as those described previously (). Synthetic complementary single-stranded oligonucleotide DNAs were annealed to make double-stranded DNAs of the target sequences. These annealed DNA were inserted into a pcPUR+U6i cassette vector, and the plasmids were electroporated into differentiated 3T3-L1 adipocytes. Cells were lysed in lysis buffer (20 mM Hepes, pH 7.2, 100 mM NaCl, 1 mM EDTA, 25 mM NaF, 1 mM sodium vanadate, 1 mM benzamidine, 5 μg/ml leupeptin, 5 μg/ml aprotinin, 1 mM PMSF, and 1 mM DTT), and the protein concentration was measured with bicinchoninic acid protein assay reagent (Pierce Chemical Co.). For immunoprecipitation, the cell lysate was preincubated with protein G–Sepharose beads at 4°C for 30 min to remove nonspecific bound protein. Then, samples were incubated with primary antibody at 4°C for 2 h followed by incubation with protein G–Sepharose beads. Lysates and immunoprecipitates were resolved by SDS-PAGE and transferred to a polyvinylidene difluoride membrane (GE Healthcare). The membrane was preblotted in milk buffer for 1 h and immunoblotted with primary antibody for 2 h. HRP-conjugated secondary antibodies (Jackson ImmunoResearch Laboratories) were used, and proteins were visualized using an enhanced chemiluminescence substrate kit (GE Healthcare). Differentiated adipocytes were prepared in 24-well plates. Cells were infected with the recombinant adenoviruses. 2 d thereafter, the cells were serum starved for 2 h at 37°C in Krebs-Ringer phosphate buffer (130 mM NaCl, 5 mM KCl, 1.3 mM CaCl, 1.3 mM MgSO, and 10 mM NaHPO, pH 7.4) and were treated with or without 20 ng/ml TNF-α for 4 h. Next, the cells were stimulated with or without 100 nM insulin for 5 min, and 2-deoxyglucose uptake was determined by 2-deoxy-D-[2,6 H] glucose incorporation. Nonspecific deoxyglucose uptake was measured in the presence of 20 μM cytochalasin B and subtracted from each determination to obtain specific uptake. The deoxyglucose uptake was corrected using the protein amount. 3T3-L1 adipocytes were infected with the recombinant adenoviruses as indicated. 2 d after the infection, cells were stimulated with or without 20 ng/ml TNF-α for 5 min, and cell lysates were prepared. After adjusting the protein concentration, immunoprecipitation using anti–IKK-β antibody was performed. Precipitates were mixed with IKK substrate peptide (KKKKERLLDDRHDSGLDSMKDEE; Upstate Biotechnology) and γ-[P] ATP. After a 10-min incubation at 30°C, samples were transferred to P81 paper (Whatman) and washed with 0.75% phosphoric acid and acetone. Radioactivity was counted using a scintillation counter. Multiple comparisons among groups were performed using the one-factor analysis of variance test (post-hoc test; Turkey-Kramer). Results are presented as means ± SD. Values of P < 0.05 were considered statistically significant. Fig. S1 shows the results of pull-down experiments to demonstrate the direct interaction of NEMO with Myo1c. Fig. S2 shows the effects of NEMO knockdown on Myo1c–IKK-β interaction (A) and the effects of NEMO expression on IRS-1–IKK-β interaction (B). Fig. S3 shows the effects of Myo1c expression on the TNF-α–induced phosphorylation of IRS-1 Ser (A) and the effects of WT NEMO and ΔN-NEMO expression or NEMO knockdown on IKK-β kinase activity (B and C). Online supplemental material is available at .

In eukaryotes, precise duplication of the genome during S phase is achieved through the initiation of replication at numerous origins distributed throughout the DNA. In late mitosis and early G1, origins are licensed for replication by loading complexes of the minichromosome maintenance proteins 2–7 (Mcm2–7), thus, forming a “prereplicative complex” (; ). Licensing involves the coordinated action of the origin recognition complex (ORC) Cdc6 and Cdt1 proteins (), which probably act to clamp Mcm2–7 around the DNA. Mcm2–7 are displaced from replication origins as they initiate, and they have been suggested to provide a helicase activity to unwind DNA ahead of the replication fork (). To prevent replication origins from firing more than once in a single S phase, the ability to license new origins is prevented from late G1 through to midmitosis (; ). This means that whatever the problems encountered by replication forks during S phase, only previously licensed sites can be used. If two converging replication forks stall irreversibly during S phase, the DNA between them will probably remain unreplicated, potentially leading to cell death or chromosome rearrangement. Therefore, it is crucial for cells to organize their replication machinery so as to minimize the risk of this happening. Previous work in a range of eukaryotes, including , humans, and , has demonstrated that Mcm2–7 complexes are loaded onto DNA in an ∼20-fold excess over the number of DNA-bound ORC molecules and over the number of replication origins (; ; ; ; ; ). However, normal replication rates are maintained when the number of Mcm2–7 molecules is reduced to 1–2 per origin (; ; ; ; ). This observation, termed the “MCM paradox” (), has led to speculation on the possible function of these extra complexes, particularly as MCM proteins do not exclusively colocalize with sites of DNA synthesis in S phase (; ; ). There is evidence that Mcm2–7 complexes are found on DNA at significant distances from where ORC is bound (; ; ; ). It has been suggested that these distant multiple MCM complexes could cooperatively pump double-stranded DNA and, thus, unwind it (). More recently, a role for excess MCM proteins in checkpoint activation has been proposed, based on the observation that cells partially depleted of Mcm7 display a reduction in replication checkpoint signaling (; ). A third possibility that has been suggested is that excess Mcm2–7 can provide replication origins for use under certain contingencies, such as incomplete DNA replication (; ; ; ). These different proposals are not mutually exclusive. In this study, we demonstrate a role for excess Mcm2–7 complexes in licensing “dormant” replication origins. These dormant origins are effectively suppressed during unperturbed DNA replication, but can support initiation when replication forks are stalled in response to a range of replicative stresses, including replication inhibition by aphidicolin, mitomycin C, etoposide, or actinomycin D. In support of this model, we demonstrate that in , partial knockdown of MCMs induces hypersensitivity to otherwise nontoxic levels of hydroxyurea (HU). When demembranated sperm nuclei are incubated in egg extracts, an average of 10–20 Mcm2–7 complexes are loaded onto each replication origin before entry into S phase (; ; ); we call this DNA “maximally licensed.” If the Cdt1 inhibitor geminin is added to the extract shortly after the sperm DNA, the number of DNA-bound MCM complexes can be limited to the minimum required to support approximately normal replication kinetics (); we call this DNA “minimally licensed.” To examine the role of excess Mcm2–7 complexes, we first investigated whether minimum licensing causes an alteration in the spacing between adjacent replication origins. Nascent DNA was labeled with biotin-16-dUTP during early S phase, after which the DNA was isolated, spread on microscope slides, and analyzed by fluorescence microscopy. Short fluorescent tracks were seen, which were caused by bidirectional replication from origins at the center of each track, and the spacing between replication origins was determined by measuring the distance between the centers of adjacent tracks (; ). shows that there was no significant difference in the average interorigin distance between the two samples (15.8 kb for maximum licensing; 17.1 kb for minimum licensing). Under normal circumstances, clusters of 3–7 adjacent origins initiate at similar times, with different clusters being activated at different stages of S phase (). We observed no major difference between minimally and maximally licensed DNA in the clustering of replication origins, with an average of 6.1 origins per cluster for maximally licensed DNA and 4.8 for minimally licensed DNA. Therefore, we conclude that minimum licensing does not significantly change replication origin usage under normal conditions. Next, we examined whether excess Mcm2–7 complexes become important if replication forks are put under stress by supplementing extracts with low concentrations of the DNA polymerase inhibitor aphidicolin. (crosses and closed circles) shows that the replication of minimally and maximally licensed DNA was inhibited to similar extents by 10 μM aphidicolin. We have previously shown that, at this concentration, aphidicolin slows replication forks by approximately threefold, but that the major effect on replication is to induce an ATR-dependent checkpoint that suppresses further initiation events (). Therefore, we also supplemented extracts with caffeine, which is an ATR kinase inhibitor (, crossed boxes). Consistent with a previous study (), the addition of caffeine completely rescued the aphidicolin-induced inhibition of replication with maximally licensed DNA. In contrast, caffeine only partially rescued replication when minimally licensed DNA was replicated in the presence of aphidicolin. This suggests that excess Mcm2–7 complexes are required in some way to allow rescue of DNA replication after the inhibition of replication fork progression. Replication forks in normally progress at ∼10 bp/s (; ; ; ), but are slowed approximately threefold by 10 μM aphidicolin (). Fork rate is not significantly affected by the presence of caffeine (). To replicate at normal rates in the presence of aphidicolin and caffeine, maximally licensed DNA must use more replication forks than normal. Minimally licensed DNA, in contrast, does not appear capable of using more replication forks. Therefore, we assessed the number of active forks by measuring the quantity of Cdc45 loaded onto chromatin. Cdc45 loads onto replication forks (probably binding Mcm2–7) just before initiation and moves with the forks as they elongate (; , ; ; ). shows that both maximally and minimally licensed chromatin contained similar quantities of Cdc45 during an undisturbed S phase, as expected from their similar replication rates. Consistent with a previous study (), the amount of Cdc45 on maximally licensed chromatin increased ∼20-fold when extract was treated with a combination of aphidicolin and caffeine. In contrast, minimally licensed chromatin showed only a two- to threefold increase under similar conditions. A slight increase of chromatin-bound Cdc45 was also seen when both maximally and minimally licensed chromatin were treated with caffeine, which is consistent with the acceleration of the origin firing that caffeine causes (; ). Next, we performed quantitative immunoblotting to estimate the amount of Cdc45 loaded onto chromatin under these conditions (). A fixed quantity of sperm nuclei was incubated in increasing quantities of egg extract supplemented with aphidicolin and caffeine. Extracts were optionally supplemented with geminin shortly after sperm addition to create minimally licensed chromatin. In the absence of added geminin (maximum licensing), the quantity of chromatin-bound Cdc45 increased with increasing extract volume (, shaded bars), caused, in part, by increased quantities of Mcm2–7 being loaded onto the DNA (). On the minimally licensed chromatin, however, Cdc45 remained at a fairly constant level of 3–5 ng (, open bars). This level of Cdc45 on minimally licensed chromatin corresponds to 1–2 molecules of Cdc45 per 10-kb DNA, or 1–2 molecules of Cdc45 per molecule of ORC (). If there is a single molecule of Cdc45 at each replication fork, these results suggest that under normal conditions a single molecule of ORC allows, on average, a single pair of replication forks to initiate; the number of forks can be increased under certain circumstances, but only if there are excess Mcm2–7 molecules present on the DNA. These results suggest that maximally licensed DNA should contain an increased density of nascent strands under conditions where there are an increased number of forks. Because recent studies have shown that caffeine can accelerate origin firing (; ), we repeated the fluorescent fiber-labeling experiments using extracts supplemented with caffeine. (left) shows that when maximally licensed chromatin was replicated in extract containing caffeine the average origin spacing declined dramatically (compare with ), with the observed spacing between nascent strands falling to near the resolution limit of this technique (1–2 kb). When the fiber-labeling experiments were repeated with minimally licensed chromatin, the reduction in spacing was much less pronounced than with maximally licensed DNA (, right). These data suggest that the extra Mcm2–7 molecules on maximally licensed DNA license dormant replication origins that can be used in the presence of caffeine. If this is correct, then nascent strands from these additional origins will very rapidly fuse into high molecular weight products that are observable on alkaline agarose gels (, left). On minimally licensed DNA, interorigin distances are larger, and nascent strand fusion should occur later (, right). To test this, maximally and minimally licensed DNA samples were pulsed with α-[P]dATP at the beginning of S phase, and then chased with unlabeled dATP for different periods. shows that maximally and minimally licensed DNA gave very similar profiles in the absence of any drug treatment. Although the long smear of nascent strands makes an accurate rate determination impossible, the modal size of nascent strands increased at a rate of ∼25 nt/s. Within expected error, this is consistent with each strand being elongated by two forks, each moving at 10 nt/s, as previously reported (; ; ; ; ). In contrast, the behavior of maximally and minimally licensed DNA in the presence of aphidicolin and caffeine () showed a dramatic difference that was consistent with the model shown in . This is highlighted in , where the relative abundance of nascent strands in each sample lane of the corresponding alkaline gel has been plotted, with the earliest sample lane plotted nearest the top. Both maximally and minimally licensed nascent strands started off with a more uniform size than in the untreated sample because aphidicolin had limited the rate of elongation. During the first few minutes, both maximally and minimally licensed nascent strands increased in size at ∼6 nt/s (∼3 nt/s per fork). For the minimally licensed sample (), most strands maintained this rate over the time course. In later samples, however, some strands doubled in size, which is consistent with fusion between adjacent strands (, arrows). The larger strands first became visible at ∼10 kb, which is consistent with them being fusions between the smallest replicons of ∼5 kb seen in . A very different pattern was seen with maximally licensed DNA (). A smear of higher molecular weight strands built up much more rapidly than it did for the minimally licensed sample, so that by the middle of the time series the majority of the strands had shifted to the higher molecular weight form. By the end of the time series, the majority of the strands were above the exclusion limit of the gel. Unlike the case for the minimally licensed DNA, there was no obvious lower size limit at which the strand fusion events occurred. These results suggest that in the presence of aphidicolin and caffeine, maximally licensed DNA displays a dramatic reduction in replicon size, an effect that is not seen with the minimally licensed DNA. Note that, consistent with a previous work (), there was no sign of caffeine reducing fork stability, as we did not observe buildup of low molecular weight products. When extracts were supplemented with caffeine alone, there was no obvious difference between maximally and minimally licensed DNA, with both samples yielding fully replicated product (). Our fiber-labeling experiments with caffeine () predict that caffeine alone can increase the number of active forks when the DNA is maximally licensed. However, without the fork-slowing effect of aphidicolin, there is rapid fusion of the nascent strands arising from additional origins that would be dormant in the absence of caffeine. Once these initial fusion events have taken place in the maximally licensed sample, the maximally and minimally licensed samples behave similarly. This effect also explains why we do not see very high levels of Cdc45 loading in maximally licensed DNA treated with caffeine alone (). Our results so far suggest that excess Mcm2–7 complexes license dormant origins, which are not used under normal conditions, but which can fire to rescue DNA synthesis in response to replication fork inhibition. It is known that Cdks are required for the initiation of replication forks in the system, but not for fork elongation (; ). It is also known that Cdks act on individual origins very shortly before the origin fires (; ). Therefore, we predicted that if Cdks were inhibited at the start of S phase, when only the very first origins had fired, dormant origin firing would be prevented and no rescue of DNA replication would take place. To inhibit Cdks, we used roscovitine, which is a purine analogue that inhibits initiation in the system (; ). A model of the predicted behavior is shown in . The model on the left shows initiation occurring at dormant origins when replication occurs in the presence of aphidicolin and caffeine. The model on the right shows that if roscovitine is added shortly after the first origins have fired, initiation cannot occur at the dormant origins, so the replication profile resembles that of minimally licensed DNA treated with aphidicolin and caffeine. show experimental results that support this model. (left) shows the effect on DNA replication when different combinations of caffeine, aphidicolin, and roscovitine were added to egg extract at the start of S phase. Consistent with previous work (), the addition of roscovitine in early S phase allowed existing forks to elongate, but the replication rate subsequently trailed off as these clustered nascent strands fused with one another. As expected for maximally licensed DNA, aphidicolin inhibited replication and this inhibition could be completely rescued by the addition of caffeine. However, when roscovitine was added together with aphidicolin and caffeine, there was no rescue of replication. A related experiment is shown in (right), the difference being that the inhibitors were added later in S phase. At this time, addition of roscovitine alone had no significant effect on replication kinetics, suggesting that virtually all the normal replication origins had already fired by the time the inhibitors were added. As before, inhibition of replication by aphidicolin could be rescued by coaddition of caffeine, and this rescue was substantially dependent on further Cdk activity. Some roscovitine-resistant rescue by caffeine was observed, possibly reflecting additional recovery pathways, such as the restart of previously stalled forks. The significance of this version of the experiment is that a requirement for Cdks is seen at a late stage in S phase when most origins have already initiated, suggesting that complete replication is dependent on the use of origins that would never normally have fired (dormant origins). To confirm our interpretation of these results, we performed alkaline gel analysis of the nascent strands. Nascent DNA was labeled with α-[P]dATP at the beginning of S phase, in the presence of aphidicolin and caffeine. The P was then chased with unlabeled dATP, while, at the same time, extract was supplemented minus () or plus roscovitine (). At different times thereafter, labeled DNA was analyzed on alkaline gels. The sample treated with aphidicolin and caffeine () closely resembled the corresponding sample shown in , with nascent DNA increasing rapidly in size so that most migrated at the exclusion limit of the gel at the end of the time series. Addition of roscovitine () almost completely inhibited the increase in modal strand length, so that most strands increased in size at a constant rate of ∼6 nt/s (∼3 nt/s per fork). At later time points, a small number of strands migrated in a higher molecular weight smear at approximately twice the size of the majority, which is consistent with the fusion of replicons 5–10 kb in size (, arrows). This profile resembles that of minimally licensed DNA treated with aphidicolin plus caffeine (). As predicted, by inhibiting Cdk-dependent initiation of dormant origins in maximally licensed DNA, the pattern of nascent strand labeling is made to resemble that of minimally licensed DNA, in which dormant origins do not exist. We next investigated whether the activation of dormant origins in maximally licensed DNA is useful in recovering from any replication-induced stresses other than aphidicolin (). Both mitomycin C (a cross-linking agent) and etoposide (a topoisomerase II inhibitor) induced a block to replication that could only be efficiently rescued by caffeine when excess Mcm2–7 were present on the DNA. This is similar to the results with aphidicolin () and suggests a general requirement for dormant origins in rescue of DNA synthesis after replication block. Actinomycin D, which acts both as a DNA intercalator and an inhibitor of the primase component of DNA polymerase α showed an even more dramatic effect. Replication was only slightly inhibited when maximally licensed DNA was treated with 4 ng/μl actinomycin D, but the inhibition was much more severe on minimally licensed DNA. Although the small amount of inhibition of maximally licensed DNA was fully reversed by caffeine, virtually no rescue with caffeine was observed with minimally licensed DNA. This result is explained by the inability of actinomycin D to activate a strong ATR-dependent checkpoint response, despite strong inhibition of replication (). Dormant origins in such cases can fire and promote significant recovery from replication inhibition without the need for checkpoint alleviation by caffeine. Consistent with this interpretation, treatment of egg extracts with actinomycin D greatly enhances Cdc45 loading (). The large amounts of replicative stress that we need to use in the cell-free system activate global checkpoint pathways that prevent replication from completing. To test the physiological significance of our findings, we turned to an in vivo system, where the effect of much lower levels of replicative stress can be reliably assessed. We took advantage of the ability to lower specific gene expression in by providing siRNA-expressing bacteria as a food source. We first mixed different ratios of bacteria expressing siRNA and bacteria expressing empty vector to determine the maximum quantity of siRNA that had no observable effect on the life cycle (Table S1, available at ). This creates a state that is functionally equivalent to the minimally licensed state in . We also determined a low concentration of HU (9.5 mM), which had no observable effect on the life cycle (Table S1). HU lowers dNTP pools by inhibiting ribonucleotide reductase and thereby inhibits DNA synthesis. shows that treatment with a combination of the siRNA and 9.5 mM of HU completely abrogated proliferation. There was a dramatic reduction in the number of adult worms in the first generation that had fed on the siRNA. The few adult worms in this generation appeared to be sterile (not depicted), and produced virtually no eggs or larvae (). Similar results were obtained using knockdown of and (Tables S2 and S3). These experiments show that a reduction of Mcm2–7 proteins causes a dramatic hypersensitivity to HU, while showing no obvious defect in the absence of replicative stress. Although there are many possible explanations for what is happening in these experiments, the results are fully consistent with the results obtained in and the idea that excess Mcm2–7 license dormant origins that are not used during undisturbed S phases. The results presented in this work provide a solution to the MCM paradox by showing that excess Mcm2–7 license many dormant replication origins that are not normally used because of suppression by a caffeine-sensitive pathway. In , there are ∼10 dormant origins for each active origin, and when dormant origin activation is permitted by treatment with caffeine, the average origin spacing drops 5- to 10-fold. The caffeine sensitivity suggests the involvement of an ATR-dependent checkpoint in suppressing the activation of dormant origins. This conclusion is consistent with results in mammalian cells, which show that suppression of ATM–ATR or Chk1 checkpoint pathways leads to a dramatic reduction in the average spacing between adjacent replication origins (unpublished data). We also show that these dormant origins can be used to allow complete replication under conditions of replicative stress, and we propose that this is at least part of the reason for their existence. Detailed analysis of this effect has been performed in egg extracts, but we have also tested a key consequence of this model by lowering MCM levels in to operationally mimic the minimally licensed state. Reduction of MCM levels causes a dramatic hypersensitivity to HU, while showing no obvious defect in the absence of replicative stress. There is also published evidence that this effect may operate in mammals. DNA fiber autoradiography studies have shown an increase in the density of replication origins after replication has been inhibited by FdUrd, which is a thymidylate synthase inhibitor (; ), or by treatment with ultraviolet light (, , ; ). Similarly, reduced nucleotide availability has been shown to increase the frequency of initiation at origins that are not normally used (). Therefore, we believe that dormant origins are present in many metazoans, where they can potentially provide a role in responding to replicative stresses. egg extracts were prepared as previously described () and supplemented with 250 μg/ml cycloheximide, 25 mM phosphocreatine, 15 μg/ml creatine phosphokinase, and 300 μM CaCl before use. Demembranated sperm nuclei () were added to extract to a final concentration of 10–15 ng DNA/μl and incubated at 23°C. For replication assays, extracts were supplemented with α-[P]dATP, and DNA synthesis was measured by trichloroacetic acid precipitation, as previously described (). To obtain minimally licensed DNA samples, recombinant geminin lacking the destruction box (; ) was added to a final concentration of 10 ng/μl. The appropriate geminin addition time was determined for each extract to produce minimal licensing, as previously described (). Chromatin was isolated and subjected to Western blot analysis, as previously described (). Aphidicolin (Sigma-Aldrich) was dissolved in DMSO at 18 mM and used at a final concentration of 10 μM in extract. Caffeine (Sigma-Aldrich) was dissolved in H0 at 100 mM and was used at a final concentration of 5 mM. Roscovitine (Calbiochem) was dissolved in DMSO at 400 mM and used at a final concentration of 500 μM. Mitomycin C was dissolved in H0 at 5 mM and used at a final concentration of 500 μM. Etoposide was dissolved in DMSO at 25 mM and used at a final concentration of 200 μM. Actinomycin D was dissolved in DMSO at 1 μg/μl and used at a final concentration of 4 ng/μl. Anti-Cdc45 antibody () was a gift from H. Takisawa (Osaka University, Osaka, Japan). Geminin was a gift from A. Ferenbach (University of Dundee, Dundee, Scotland). sperm DNA was incubated in extract supplemented with 50 μM biotin-16-dUTP (Roche) under the desired conditions. Reactions (20 μl) were stopped by resuspension in 350 μl of ice-cold NIB (50 mM Hepes-KOH, pH 7.6, 50 mM KCl, 5 mM MgCl, 2 mM DTT, 0.5 mM spermidine-3HCl, 0.15 mM spermine-4HCl, 100 μM PMSF, and 1 μg/ml each of aprotinin, leupeptin, and pepstatin) containing 30 μM aphidicolin. The resuspended extract was underlayered with 100 μl NIB containing 30 μM aphidicolin and 10% (wt/vol) sucrose and spun at 2,000 in a swinging bucket rotor for 5 min at 4°C. The supernatant was removed to leave only the sucrose cushion, and the top of the cushion was then washed twice with 100 μl NIB plus 30 μM aphidicolin. The top part of the cushion was carefully removed to leave 40 μl, which was resuspended in 400 μl ice-cold PBS plus 0.1% (vol/vol) Triton X-100. This was underlayered with PBS plus 10% sucrose and spun like the resuspended extract. The supernatant was removed and the top of the cushion was washed twice with 100 μl PBS. The cushion was carefully removed to leave the pelleted nuclei, which were gently resuspended in 50 μl PBS before freezing on dry ice. DNA was spread on glass slides according to the previously described standard conditions (). Slides were rehydrated with HPLC pure water and incubated for 2 h in blocking solution containing PBS, 1% (wt/vol) BSA, and 0.05% (vol/vol) TWEEN 20. Slides were incubated with anti-biotin (B7653; Sigma-Aldrich) at 20 μg/ml in this buffer for 2 h, washed extensively (10 changes of buffer over 2 h), and labeled with 1 μg/ml cy3-conjugated anti–mouse antibody (715-165-150; Jackson ImmunoResearch Laboratories) in the same buffer for 2 h. Samples were washed extensively (10 changes of buffer over 2 h), and then washed twice in PBS and DNA stained with YOYO (Y-3601 diluted at 1:10,000 from a 1-mM stock; Invitrogen) for 10 min. Samples were rinsed five times in PBS and mounted in Vectashield. Images were captured using a microscope (LSM510-META; Carl Zeiss MicroImaging, Inc.) and measurements were made using the associated LSM software. Random fields were selected using YOYO staining to ensure that only single DNA fibers, and not fiber bundles, were scored. Fields rich in unambiguous single fibers were recorded using a 100×, NA 1.4, lens for optimal resolution and sensitivity. For each sample a minimum of 200 measurements was made. Pulse-chase experiments were performed by incubating sperm DNA in interphase egg extract, with inhibitors added as was appropriate. At the beginning of S phase (the exact timing being determined independently for each extract), the extract was supplemented with 0.2 mCi/ml α-[P]dATP (from a stock of 10 mCi/ml and 3,000 Ci/mmol; GE Healthcare). After a 2-min pulse (5 min for samples containing aphidicolin), the radioactive nucleotide was chased by addition of 2.5 mM unlabeled Mg-dATP. Aliquots (10 μl) were subsequently removed from the sample at the desired time points, reactions were stopped, and the DNA was processed for electrophoresis on 0.7% agarose gels, as previously described (). HindIII-digested λ-phage DNA (New England Biolabs) was end-labeled with α-[P]dATP and loaded for comparison (12.5 ng per lane). Alkaline agarose gels were exposed to autoradiography film (Kodak), which was scanned using a scanner (CanoScan 9900F; Canon) and Photoshop software (Adobe). Lane densities were analyzed using GelEval software (FrogDance Software). RNAi feeding was performed as previously described, using (R10E4.4), (ZK632.1), and (F32D1.10) from the genome-wide RNAi library (; ). RNAi-expressing bacteria were titrated with bacteria expressing RNAi against GFP. Three P0 worms were placed on RNAi plates containing 100 μg/ml ampicillin, 5 mM IPTG, and the indicated concentration of HU at the L4 larval stage, and the resulting adult worms were taken off those plates 40 h later. After 4 d, the number of adult F1 worms was scored, and sterile worms were determined by the absence of eggs in their body. The number and viability of F2 animals was scored 2–3 d later. At this stage, under wild-type conditions, plates were already starved because of excessive F2 larvae. Plates were viewed with a microscope (Stemi SV11; Carl Zeiss MicroImaging, Inc.) using a 20× objective. Representative images were captured using Openlab 3.1.2 software (Improvision) and exported as tiff files. Tables S1–S3 show the number of adult F1 worms observed at different HU concentrations and different amounts of siRNA directed against (Table S1), (Table S2), and (Table S3). Online supplemental material is available at .

The defining feature of eukaryotic cells is the separation of the chromosomes from the cytoplasm, which is achieved by the nuclear envelope (NE), a highly selectively permeable double membrane system. The enclosure of the genetic information within the nucleus generates a unique environment that separates DNA replication, transcription, and RNA processing from protein synthesis (for reviews see ; ). Communication between the nucleus and the cytoplasm occurs through nuclear pore complexes (NPCs), specialized protein-filled “holes” that span the NE. The NE of eukaryotes consists of a double lipid bilayer named the inner and outer nuclear membranes. The outer nuclear membrane is structurally and functionally continuous with the ER, whereas the inner nuclear membrane harbors a unique set of proteins that helps connect the membrane to the underlying lamina and provide specialized chromatin regulatory domains (; for reviews see ; ; ; ; ). The inner nuclear membrane is anchored to the nuclear lamina, a polymeric meshwork composed of specialized intermediate filament proteins named lamins. The major lamin type present in eggs, lamin B3, is associated with the inner nuclear membrane via posttranslational isoprenyl groups (for review see ) as well as integral inner nuclear membrane and lamina-associated polypeptides (LAPs; for reviews see ; ; ). The nuclear lamina provides structural support to the nucleus (), and its expansion is required for continued nuclear growth (; ). During cell division in metazoa, NE components are disassembled and dispersed throughout the dividing cell (; for review see ). After NE breakdown, the mitotic spindle assembles and then segregates the duplicated chromosomes. At the end of mitosis, the mitotic spindle disassembles and a new NE reassembles around the segregated chromosomes. Postmitotic nuclear reassembly requires at least two distinguishable steps: enclosure of the chromatin within a functional NE and growth and expansion of this NE. Initial steps during enclosure of the chromatin include the targeting of nuclear vesicles to the chromosomes, the fusion of vesicles to reform the nuclear membranes, the assembly of NPCs, and the formation of the nuclear lamina. Continued NE growth requires further vesicle fusion, nuclear protein import, and lamina expansion (for reviews see ; ). Although it is clear that phosphorylation and dephosphorylation of key structural elements are required for the large-scale molecular rearrangements that take place during NE breakdown and reformation, the molecular mechanisms of these processes remain largely unknown. One class of proteins implicated in NE reformation is LAPs. LAP2β (also known as thymopoietin) is a ubiquitously expressed and highly conserved (; , ; ; ; ) inner nuclear membrane protein that binds to lamin B and chromosomes (; , ). The six isoforms of mammalian LAP2 (designated α, β, γ, δ, ɛ, and ζ) are generated by alternative splicing of the same transcript (; ; ). With the exception of mammalian LAP2α and LAP2ζ, all LAP2 isoforms have a common transmembrane domain at their COOH terminus and are type II integral membrane proteins of the inner nuclear membrane (for review see ). LAP2γ, δ, and ɛ are closely related to LAP2β, but each lacks one or more of several short regions of the nucleoplasmically oriented NH-terminal domain, which is involved in both lamin B and chromatin binding. In mammals, LAP2α, β, and γ are the major LAP2 isoforms (; ; ). Several LAP2 isoforms have also been found in tissue culture cells (), oocytes (; ), or early embryos (), but their molecular structure has not been elucidated. The three LAP2 cDNAs isolated from oocyte mRNAs (coding for proteins with predicted molecular masses of 62.84 kD [clone 2], 46.4 kD [clone 3], and 58.7 kD [clone 4]) most closely resemble human LAP2β in primary sequence and also appear to be related by alternative splicing (), but how these clones correspond to the ∼85-kD oocyte-specific LAP2 isoform () is presently unclear. It is interesting to note that its biochemical properties make it likely that the 85-kD isoform has a transmembrane domain and is therefore related to LAP2β (). LAP2β is proposed to link chromatin to the NE during interphase (; , ), and at least two lines of evidence support the notion that LAP2β is required for NE reassembly and expansion after cell division: microinjection of recombinant LAP2 mutant proteins into mammalian cells inhibits postmitotic nuclear expansion () and addition of recombinant LAP2 mutant proteins to egg extracts disrupts nuclear morphology and lamina assembly (). In recent years, an increasing number of proteins have been shown to play important and sometimes seemingly unrelated roles in both interphase and mitosis (for review see ). Most prominent among them are the components of the nuclear trafficking machinery, the RanGTPase, and its binding proteins and effectors (for reviews see ; ; ). Several spindle assembly factors are mitotic targets of the RanGTPase pathway, including the targeting protein for XKlp2 (TPX2). TPX2 was originally discovered as a microtubule binding and loading factor for the mitotic kinesin XKlp2 (). Since then, it has been shown to play multiple roles during mitosis: TPX2 is involved in focusing microtubules at mitotic spindle poles (), in bipolar mitotic spindle assembly (; ), in nucleating microtubules near the chromosomes (; ), and in activating the mitotic kinase Eg2 in a RanGTPase-dependent manner (; ). TPX2 localizes predominantly to spindle poles during mitosis and to the nucleus during interphase (; ). Although important functions for TPX2 in mitotic spindle assembly are well documented (; for review see ), a functional role for its accumulation in the nucleus during interphase has not been reported. We show that TPX2 interacts with LAP2 in a cell-free nuclear assembly assay based on extracts made from the eggs of the South African clawed frog, and in vitro. When interphase egg extracts depleted of TPX2 were induced to assemble nuclei by the addition of demembranated sperm chromatin, they formed significantly smaller nuclei than control extracts. Nonetheless, these nuclei were able to perform nuclear functions, including nuclear import, selective permeability, and DNA replication. TPX2 depletion resulted in decreased accumulation of LAP2 at the NE. Together, these data suggest a novel role for TPX2 in LAP2-mediated NE growth. TPX2 accumulates in nuclei assembled in vitro () and in interphase tissue culture cells (; ; ; ; ). TPX2 staining appears mostly reticular but also shows a distinct nuclear rim staining in a subset of nuclei (; see in ). Two explanations can account for the accumulation of TPX2 in nuclei: TPX2 needs to be sequestered from the cytoplasm during interphase to avoid its effects on microtubules or TPX2 plays a functional role in interphase nuclear assembly or architecture. The reticular distribution and nuclear rim staining of TPX2 prompted us to further explore whether TPX2 might serve a functional role in the nucleus. Nuclear assembly can be studied in vitro using well-established nuclear assembly assays based on extracts made from the eggs of the South African clawed frog, . In a process that mimics sperm entry during fertilization, demembranated sperm added to interphase extracts readily attract NE components and assemble functional nuclei in vitro (for review see ). To test whether TPX2 has a role in nuclear assembly or architecture, we used the in vitro nuclear assembly assay to examine the effects on nuclear assembly of altering TPX2 levels (). Adding bacterially expressed recombinant TPX2 to interphase extracts resulted in a dose-dependent reduction in the size of nuclei formed around exogenously added sperm chromatin (). Similarly, depletion of TPX2 from the egg extract resulted in the formation of small nuclei (; we refer to these as TPX2-Δ nuclei), whereas it had no effect on microtubules (). Under conditions where >98% of TPX2 was depleted from the extract () and nuclei were fixed and examined 80 min after the start of the assembly reaction, nuclear size was reduced by ∼80% compared with controls (). Importantly, readdition of endogenous levels (∼100 nM; ) of recombinant TPX2 to TPX2-Δ extract largely rescued the defects in nuclear size (), but only if the exogenous TPX2 was supplied at the beginning of the reaction. Addition of recombinant TPX2 1 h after the start of the nuclear assembly reaction was unable to rescue the defect even after prolonged (>60 min) incubation (). These results suggested that both too much and too little TPX2 disrupted nuclear assembly. They further suggested that TPX2 function is required during the early stages of nuclear reformation. Several possibilities could account for the small size of TPX2-Δ nuclei. For example, TPX2 depletion could directly or indirectly influence the levels of one or more nuclear components that are limiting. These include the building blocks of the nuclear membranes, the nuclear lamina, or the NPCs or factors involved in chromatin remodeling during decondensation. It was also possible that nuclear import was somehow affected. Alternatively, TPX2 could be involved in the overall regulation of nuclear assembly. Furthermore, importin α and β have been implicated in nuclear membrane fusion and NPC assembly (; ; ; ; ; for review see ), and TPX2 is known to interact with both (; ). To distinguish these possibilities and to begin to understand the function of TPX2 in nuclear assembly, we examined nuclei assembled in TPX2-Δ extracts for their ability to support chromatin decondensation and to assemble nuclear membranes and NPCs. Because small nuclei could result from chromatin decondensation defects, we first examined the effect of TPX2 depletion on stage I chromatin decondensation (as defined by ). Both the extent and the timing of decondensation of chromatin incubated in high-speed supernatant depleted of TPX2 were indistinguishable from mock-depleted extracts (Fig. S1, available at ), suggesting that TPX2 was not required for the initial stages of chromatin decondensation. Next, we tested whether TPX2-Δ nuclei were able to assemble nuclear membranes. Staining of TPX2-Δ nuclei with the membrane dye 3,3′-dihexyloxacarbocyanine (DHCC) revealed a smooth and continuous line (), suggesting fused and fully assembled nuclear membranes (; ). To address whether TPX2-Δ nuclei possessed NPCs and a nuclear lamina, we immunostained nuclei assembled in TPX2-Δ extracts for nucleoporins using the monoclonal antibody mAb414 () or for lamin B3 using the monoclonal antibody L6-5D5 (). MAb414 robustly labeled the nuclei assembled in TPX2-Δ extracts, indicating the presence of NPCs ( and Fig. S2, available at ). Similarly, anti-lamin antibody robustly labeled TPX2-Δ nuclei, indicating the presence of a lamina (). Consistent with these observations, thin-section electron microscopy of TPX2-Δ nuclei revealed NPCs embedded in a double nuclear membrane (, b and e) that were indistinguishable from untreated (, a and d) or mock-depleted (, c and f) controls. Interestingly, the overall ultrastructure of TPX2-Δ nuclei was also indistinguishable from controls (). It was possible that, even though NE membranes and NPCs could assemble around chromatin in TPX2-Δ extracts, the resulting structures were unable to perform typical nuclear functions, such as nuclear protein import, exclusion of nonnuclear substrates, and DNA replication. To address these questions, we first added rhodamine-labeled nucleoplasmin () to TPX2-Δ nuclear assembly reactions. Overlap of the rhodamine signal with the chromatin stain indicated that the labeled nucleoplasmin was imported into the depleted nuclei (). When nuclear protein import was experimentally blocked by the addition of the inhibitor wheat germ agglutinin () rhodamine nucleoplasmin failed to accumulate in the nuclei (). These results suggested that TPX2-Δ nuclei are competent for nuclear import. To determine whether TPX2-Δ nuclei provided a permeability barrier, rhodamine-labeled 155-kD dextran was added to TPX2-Δ nuclear assembly reactions. The dextran was excluded from TPX2-Δ nuclei (), indicating that they were able to selectively exclude nonnuclear compounds. Next, we sought to determine whether TPX2-Δ nuclei were competent for DNA replication. To test replication efficiency, we incubated depleted extracts with [P]dCTP and measured the amount of incorporation of the label. The TPX2-Δ nuclei were able to incorporate [P]dCTP (). To distinguish DNA repair from DNA replication, we measured [P]dCTP incorporation in the presence of the inhibitor of DNA polymerase α, aphidicolin. Incorporation of labeled dCTP was sensitive to aphidicolin (, a), indicating that it was not caused by DNA repair. Thus, we conclude that TPX2 is not required for DNA replication. Interestingly, in some experiments, replication proceeded more efficiently in TPX2-Δ nuclei compared with mock-depleted controls (). This is discussed in more detail in the Discussion section. To identify potential binding partners of TPX2 during nuclear assembly and thus begin to understand how TPX2 functions in nuclear assembly, we analyzed GST-TPX2 pull downs from interphase or mitotic egg extracts. SDS-PAGE and Coomassie staining showed that several proteins copurified with GST-TPX2 incubated in interphase or mitotic extracts and reisolated (). MALDI-TOF (matrix-assisted laser desorption ionization–time of flight) mass spectrometry analysis of the ∼40-kD protein that copurified with GST-TPX2 (excised from the gel and processed for mass spectrometry as described in Materials and methods) identified two proteins that were of special interest to us: actin and the inner nuclear membrane protein LAP2. Further analysis by Western blots revealed that both actin (unpublished data) and LAP2 (two isoforms migrating at ∼85 and ∼60 kD, respectively) were indeed present in the TPX2 pull downs (, a and e). Although no obvious differences between interphase and mitotic pull downs were apparent by Coomassie staining, Western blots revealed that LAP2 preferentially copurified with GST-TPX2 incubated in interphase extracts (, a). Endogenous LAP2 and TPX2 also interacted, as shown by coimmunoprecipitation experiments using either TPX2 or LAP2 antibodies as probes (, b and c). To determine whether the interaction between TPX2 and LAP2 was direct or possibly mediated by another protein, purified recombinant GST-XLAP2 nucleoplasmic domain (amino acids 1–471) was mixed with increasing amounts of untagged recombinant TPX2, and the mixture was incubated in vitro. Proteins were then adsorbed to glutathione-agarose beads, and bound and unbound proteins were separated by SDS-PAGE and visualized by Coomassie staining (). TPX2 binding to GST-LAP2 was specific and saturable (), suggesting that LAP2 directly bound to TPX2. If TPX2 and LAP2 interact in vivo, they would be expected to colocalize at least during some parts of the cell cycle. To assess protein distribution, TPX2 and LAP2 were localized in tissue culture. Double stain immunofluorescence microscopy showed that TPX2 and LAP2 overlapped at the nuclear rim in a subset of interphase cells and especially during telophase and early G1 (). These results are consistent with the idea that TPX2 and LAP2 function together during nuclear reformation. LAP2 is involved in attaching nuclear membranes to the underlying lamina and chromatin, and its disruption results in nuclear growth defects (; ). These defects are remarkably similar to the defects observed in TPX2-Δ extracts. We therefore wondered whether the interaction between LAP2 and TPX2 is important for nuclear assembly. To test this hypothesis, we monitored LAP2 localization in TPX2-Δ nuclei. Immunofluorescence revealed that LAP2 localized mainly to the nuclear rim during the early stages of nuclear assembly in mock-depleted extracts and to punctate structures at later stages (, a) but was distributed more diffusely throughout the nuclei at all stages in TPX2-Δ extracts (, b, and Fig. S3, available at ). There also was an overall reduction in the amount of LAP2 associated with TPX2-Δ nuclei. To test whether mislocalization was specific to LAP2 or a more general phenomenon, nuclei were also stained for lamin B3 or for NPC components using mAb414 (Fig. S2). Taking into account the differences in shape and size of TPX2-Δ and mock-depleted nuclei, both lamina and NPC assembly appeared qualitatively similar in TPX2-Δ extracts compared with mock-depleted extracts (Fig. S2). Similarly, TPX2 localization seemed unaffected by the addition of bacterially expressed human LAP2 to nuclear assembly reactions (), a condition that disrupts endogenous LAP2 function (). Together, these results suggest that TPX2 is required for LAP2 localization and that this activity of TPX2 is required for proper nuclear assembly. TPX2 has well-documented functions in mitotic spindle assembly and in the formation of microtubules near the chromosomes (for review see ), although the molecular mechanisms by which it carries out these functions remain obscure. Consistent with its role in spindle assembly, TPX2 localizes to the spindle poles in mitosis, and a variety of mitotic binding partners for TPX2 have been described (; ; ; ; ; ). During interphase, however, TPX2 accumulates in the nucleus, and this nuclear accumulation of TPX2 could serve to stockpile functional protein for use during the next mitosis while preventing it from disturbing the interphase microtubule array. Alternatively, TPX2 could play a role in nuclear assembly or function. The results reported here show for the first time that TPX2 plays a role during nuclear reformation after mitosis. Our most unexpected findings were that depletion of TPX2 from interphase egg extracts severely affected nuclear size and that these effects appeared to be mediated through the interaction of TPX2 with the nuclear assembly protein LAP2. Surprisingly, depletion of TPX2 did not disrupt nuclear function, as TPX2-Δ nuclei were able to specifically import nuclear substrates and replicate their DNA. Indeed, in five out of seven independent experiments, TPX2-Δ nuclei replicated their DNA up to 2.5-fold more efficiently than mock-depleted controls. This increase in DNA replication efficiency is strikingly similar to the phenotype of “overexpressing” a human LAP2β truncation mutant encompassing the lamin and chromatin binding region (amino acids 1–408) in egg extracts (). These results are consistent with the idea that the concentrations of factors inside the nucleus, rather than nuclear structure by itself, are essential for replication competence (), although we cannot yet rule out a direct role for TPX2 in regulating DNA replication. Our finding that TPX2 depletion results in small nuclei is reminiscent of the results reported by , who showed that microinjection of the lamin binding region of rat LAP2 strongly inhibited the threefold nuclear volume increase that takes place during early G1 in HeLa cells (). Similarly, addition of a dominant-negative LAP2 mutant to egg extracts resulted in membrane attachment and nuclear expansion defects (). Two possibilities exist to explain the observed nuclear morphology defects in TPX2-Δ nuclei: TPX2 could be required to regulate NE assembly (presumably through interaction with LAP2) or, alternatively, depletion of TPX2 could result in codepletion of LAP2 from the system. Two arguments suggest that the second possibility is less likely: in extracts from which routinely >98% of TPX2 had been removed by immunodepletion, at least 90% of LAP2 remained in the extract (, d). More important, nuclear size could largely be rescued by readdition of bacterially expressed recombinant TPX2 alone. Addition of exogenous TPX2 also disrupted nuclear morphology under conditions where the levels of LAP2 have not been altered. We therefore favor the hypothesis that TPX2 functions in NE assembly by regulating LAP2. Our finding that LAP2 fails to accumulate in TPX2-Δ nuclei, whereas the localization of TPX2 is unaffected when LAP2 is disrupted, suggests that TPX2 might play a role in the targeting or perhaps anchoring of LAP2. Furthermore, the findings that both reduction and increase in the levels of TPX2 affect nuclear morphology and that TPX2 needs to be present during the early stages of nuclear reformation suggest that the stoichiometry of TPX2 and LAP2 during the early stages of nuclear assembly is important. In this context, it is interesting to note that TPX2 protein levels are tightly regulated during the cell cycle (; ). Although recent studies showed that the bulk of TPX2 is degraded via the anaphase promoting complex/cyclosome and its activator Cdh1 (APC/C) after mitosis in HeLa cells (), analysis of individual cells (both the tissue culture cells used here and HeLa cells; ) shows that detectable levels of TPX2 are still present during telophase and cytokinesis, i.e., during the early stages of nuclear reformation. These results are consistent with the notion that TPX2 acts before the chromatin is enclosed in a NE (). Presumably, its degradation after nuclear enclosure has little consequence for nuclear assembly, although this hypothesis remains to be tested experimentally. Although the presence of TPX2 during nuclear reformation in tissue culture cells does not prove that TPX2 is required for this process, our findings are consistent with the idea that the involvement of TPX2 in nuclear reformation is not restricted to early embryos but may be a feature of somatic cells as well. In summary, four lines of evidence lead us to propose that TPX2 regulates LAP2 function during nuclear assembly: TPX2 interacts with LAP2 in vitro and during interphase but much less during mitosis, and both proteins localize to the nuclear rim; the phenotypes of disrupting LAP2 or TPX2 are remarkably similar; both TPX2 and LAP2 are required during the early stages of NE reformation; and LAP2 distribution is disrupted in TPX2-Δ nuclei. Perhaps our results can help provide an explanation for the different phenotypes observed upon TPX2 depletion from mitotic egg extracts (), tissue culture cells (), or human tissue culture cells (; ), which included spindles with unfocused spindle poles, mitotic cells with two robust asters that are unable to assemble a spindle midzone, and a complete absence of microtubules. Depending on the amount of TPX2 that remains in the extracts or cells, it is possible that one or more of the observed phenotypes is a consequence of defects in nuclear assembly or growth after the previous cell division rather than a mitotic defect. We are currently exploring these possibilities further. TPX2 antibodies (diluted 1:1,000 for indirect immunofluorescence [IIF]) were raised in rabbits against full-length recombinant TPX2 protein and affinity purified (). For the depletion experiments, affinity-purified TPX2 antibodies were cross-linked to protein A beads as described previously (). Rabbit polyclonal antibodies against human LAP2β (2805; raised against amino acids 1–408 of human LAP2β; diluted 1:1,000 for IIF) or against LAP2 (2807; raised against amino acids 1–165 of LAP2; diluted 1:10,000 for Western blot) were provided by K. Wilson (Johns Hopkins Medical School, Baltimore, MD). Guinea pig polyclonal antibodies against LAP2 (diluted 1:500 for IIF and 1:3,000 for Western blots) were described in . Monoclonal antibody L6-5D5, directed against lamin B3 () were diluted 1:400 for IIF and were a gift from R. Stick (University of Bremen, Bremen, Germany) and Bob Goldman (Northwestern University, Chicago, IL). MAb414 antibodies were purchased from Covance and were diluted 1:5,000 for IIF. Alexa 488– or Alexa 594–conjugated secondary antibodies (1:1,000) were purchased from Invitrogen. Alkaline-phosphatase–linked secondary antibodies used for Western blots were purchased from Sigma-Aldrich. TPX2 fused to GST (using the vector pGEX-6P2 [GE Healthcare]; ) was expressed in strain MG1655 ΔKJ (a gift from E. Craig, University of Wisconsin–Madison, Madison, WI). This bacterial strain was used to express TPX2 because chaperones were a common contaminant in our TPX2 preparations and this strain lacks the heat shock proteins DnaJ and DnaK (), thus yielding more highly purified TPX2. GST-TPX2 was purified on glutathione-agarose beads according to the manufacturer's instructions and either it was used directly (in the case of the pull-down experiments; ) or the GST tag was removed from the protein (add-back experiments and in vitro interactions; , , and ) by incubation with PreScission Protease (GE Healthcare) as described by the manufacturer. In either case, the protein was concentrated to ∼5 mg/ml, dialyzed overnight against XB buffer (10 mM K-Hepes, 100 mM KCl, 1 mM MgCl, 0.1 mM CaCl, and 50 mM sucrose, pH 7.6) supplemented with 1 mM DTT, and small aliquots were flash frozen in liquid nitrogen and stored at −80°C. Human 6-His LAP2β (amino acids 1–408 of human LAP2β) was a gift from K. Wilson and was purified as described previously (). The GST LAP2β construct (amino acids 1–471 of LAP2) was a gift from G. Krohne (University of Würzburg, Würzburg, Germany) and was purified on glutathione-agarose beads according to the manufacturer's instructions. The protein was concentrated to ∼5 mg/ml, dialyzed against XB buffer, flash frozen in small aliquots in liquid nitrogen, and stored at −80°C. Crude interphase nuclear assembly extracts were prepared as described previously () and were used fresh for all experiments except for the GST pull downs, for which extracts were prepared by cycling mitotic extracts (; ) into interphase by the addition of 400 μM Ca. To deplete extracts, 200 μl of freshly prepared crude extracts were incubated twice for 1 h each with 50 μl of Affi-Prep protein A beads (Bio-Rad Laboratories) cross-linked to antibodies (anti-TPX2 or nonimmune rabbit IgG) at 4°C. The beads were removed from the depleted extracts by a brief (10-s) centrifugation in a nanofuge. Demembranated sperm chromatin (1,000/microliter final; ) was added to 20 μl of extract on ice, and the reaction was moved to room temperature to initiate nuclear assembly. Reactions were incubated for 80 min unless otherwise indicated. For reconstitution experiments, purified TPX2 (without the GST tag) was added (100 nM final concentration) to extracts that were previously depleted of endogenous TPX2. To visualize nuclear membranes, nuclear assembly reactions were spotted onto a microscope slide and mixed (2:1) with a solution containing 3.7% formaldehyde, 20 μg/ml bisbenzimide H33342 trihydrochloride fluorophore (Hoechst dye; Calbiochem), and 20 μg/ml DHCC (DiOC; Invitrogen) membrane dye in membrane wash buffer (MWB; 250 mM sucrose, 50 mM KCl, 2.5 mM MgCl, 50 mM Hepes, 1 mM DTT, 1 mM ATP, and 1 μg/ml LPC, pH 8.0). Nuclear membranes and DNA were visualized by fluorescence or by differential interference contrast microscopy, as indicated. For immunofluorescence of in vitro–assembled nuclei, nuclear assembly reactions were incubated for 80 min (or the indicated times; ) and processed as described by , with the following modifications. 200 μl of fix solution (4% formaldehyde in nuclear wash buffer [200 mM sucrose, 50 mM KCl, 2.5 mM MgCl, 15 mM Hepes, pH 7.4, and 1 mM DTT]) was added to 20 μl of a nuclear assembly reaction, and the mixture was incubated for 10 min at 22–25°C. The mixture was layered over a 2-ml cushion of 30% sucrose in nuclear wash buffer in a glass tube (Corex) modified to hold a coverslip, as described by . The samples were then spun (3,400 rpm for 7 min at 4°C; JS13 rotor [Beckman Coulter]) onto poly-lysine–coated coverslips, and the coverslips were processed for immunofluorescence (mAb 414, lamin B3, or LAP2) as described (see Immunofluorescence and microscopy). Nucleoplasmin was purified from extracts and labeled with 5 (and 6) carboxytetramethylrhodamine succinimidyl ester (“tetramethylrhodamine”; Invitrogen) as described previously (). Labeled nucleoplasmin was added to nuclear assembly reactions 60 min after initiating assembly, and the mixture was incubated for a further 20 min. A 2-μl reaction sample was then spotted onto a coverslip, fixed with 1 μl of 3.7% formaldehyde in MWB containing 20 μg/ml Hoechst dye, and viewed under the fluorescence microscope. The integrity of the NE membranes was visualized using dextran exclusion. 155-kD TRITC-dextran (0.1 mg/ml final; Sigma-Aldrich) was added to nuclear assembly reactions after 60 min of assembly, and the reactions were allowed to proceed for another 20 min. A 2-μl reaction sample was then spotted onto a coverslip, fixed with 1 μl of 3.7% formaldehyde in MWB containing 20 μg/ml Hoechst dye, and viewed in the fluorescence microscope. For TPX2 overexpression, nuclear assembly reactions were supplemented with 0.2, 1.5, 2.5, 3.5, or 6.0 μM of recombinant TPX2 (without the GST) purified from bacteria. For LAP2 overexpression, reactions were supplemented with 3.0 μM bacterially expressed recombinant human LAP2β. Control reactions were supplemented with XB buffer. After incubation at 22–25°C for 80 min, the reactions were fixed in solution and spun onto polylysine–coated coverslips (described in Probes for NE components). Nuclear size was assessed based on DNA staining with Hoechst dye. XLKWG cells were grown on coverslips (), washed once with PBS, and fixed in 4% paraformaldehyde in PBS for 10 min. Nuclear assembly reactions were fixed in solution and spun onto poly-lysine–coated coverslips (described in Probes for NE components). For IIF, coverslips (of cells or nuclear assembly reactions) were washed with TBS-T (150 mM NaCl, 20 mM Tris-HCl, pH 7.4, and 0.1% Triton X-100), and nonspecific binding was blocked by incubation with Abdil (TBS-T plus 2% BSA and 0.1% sodium azide) for 40 min. Coverslips were then incubated with the appropriate primary and secondary antibodies (diluted in Abdil) for 1 h each in a humidified chamber at 22°C. To visualize DNA, coverslips were rinsed in water, dipped into a solution of 200 μg/ml Hoechst dye in water for 15 s, rinsed again in water, and mounted in mounting medium (Sigma-Aldrich). Samples were photographed with a cooled charge-coupled device camera (Photometrics CoolSnap HQ; Roper Scientific, Inc.) through a 60×/1.4 NA plan apo objective mounted on a fluorescence microscope (Eclipse E800; Nikon). Images were obtained using MetaMorph software and processed using Photoshop (Adobe). Samples were processed for EM by a modification of the procedure described by . 50 μl of mock or TPX2-Δ nuclear assembly reactions were diluted in 500 μl of ice-cold MWB, the mixture was placed on ice, and 550 μl of EM fix (250 mM sucrose, 100 mM Hepes, pH 8.0, 1.5 mM MgCl, 1.5 mM CaCl, 1% glutaraldehyde, and 0.5% paraformaldehyde) was slowly added to the diluted nuclei. The reactions were incubated on ice for 1 h, and the nuclei were collected at top speed in a horizontal rotor for 3 min. The pellets were washed three times (10 min per wash) in 0.2 M cacodylate, pH 7.4, and were incubated in 1% OsO buffered in 0.2 M cacodylate, pH 7.4, for 1 h on ice. The pellets were washed twice (5 min each) in ddHO and were stained overnight on ice with 1% uranyl acetate in ddHO. The pellets were dehydrated through a graded ethanol series (50, 75, 95, and 100% pure ethanol). Dehydrated pellets were embedded in Spurr's low viscosity medium (Ted Pella, Inc.) and hardened at 70°C for 20 h. Blocks were sectioned, poststained with 1% uranyl acetate and lead citrate, and viewed at 75 kV on an electron microscope (H-7000; Hitachi). Nuclear area was measured using the trace region function of MetaMorph to outline individual nuclei. The area enclosed in the outlined region was measured using the region measurements function. At least 50 nuclei were measured per condition in each of five separate experiments. Standard deviations were calculated from the mean nuclear size for each experiment. DNA replication was measured in seven independent experiments by incorporation of [P]dCTP (3,000 Ci/mmol; Redivue [GE Healthcare]) as described by with the following modifications: 20 μl of extract (TPX2-Δ, mock-depleted, or negative control extract) was supplemented with 1,000 sperm/microliter and 10 μCi of [P]dCTP and was incubated for 3 h at 22°C. Two types of negative controls were used with identical results: extract treated with 50 μM aphidicolin (Fisher Scientific) or mitotic extract. After addition of an equal amount of stop buffer (80 mM Tris-HCl, pH 8.0, 8 mM EDTA, 0.13% phosphoric acid, 10% Ficoll, 5% SDS, and 0.2% bromophenol blue) and proteinase K digestion (), the samples were separated on a 0.8% agarose gel. Incorporation of label was detected by autoradiography of the wet gel using a PhosphorImager and was quantitated using Photoshop. To keep the conditions between mitotic and interphase pull downs as similar as possible, we prepared interphase extract from an aliquot of mitotic extract by the addition of 400 μM Ca. GST-TPX2 pull downs from mitotic and interphase extracts were performed as described previously (). In brief, GST-TPX2, GST alone, or buffer were added to extract and incubated at 4°C on a rotator. After 30 min, glutathione-agarose beads (Sigma-Aldrich) were added, and the mixture was incubated for an additional 60 min at 4°C. The beads were retrieved by a brief spin and washed, and bound proteins were eluted by digesting with PreScission protease. Eluted proteins were separated on 10% SDS-PAGE gels and visualized by staining with Coomassie stain or transferred to nitrocellulose and Western blotted according to standard protocols. To immunoprecipitate TPX2 or LAP2, 100 μl of freshly prepared crude extracts were incubated for 1 h at 4°C with antibodies covalently cross-linked to Affi-Prep protein A beads. Antibodies used were rabbit anti-TPX2, guinea pig anti– LAP2, or rabbit IgG against an irrelevant antigen (maskin; ). The beads were collected by a brief (10-s) centrifugation in a nanofuge, washed three times with HB (50 mM Hepes, 1 mM EGTA, and 1 mM MgCl, pH 7.6) plus 100 mM NaCl and once in HB plus 250 mM NaCl. The beads were then resuspended in 25 μl of 2× SDS sample buffer and boiled. Equal fractions of each sample were then separated by 10% SDS-PAGE, transferred to nitrocellulose, and Western blotted according to standard protocols. GST-TPX2 pull downs were concentrated and separated on a 10% gel. Gel bands were excised from the gel and prepared as described previously ().MALDI-TOF mass spectrometry was performed by facility staff on a matrix-assisted laser desorption ionization–time of flight instrument (BIFLEX III; Bruker Daltonics). Increasing amounts of bacterially expressed purified recombinant TPX2 (without the GST tag) were combined with 3.6 μM of bacterially expressed purified GST– LAP2 in a test tube in a total volume of 50 μl. Molar ratios of LAP2/TPX2 were 5:1, 2.5:1, 1:1, 1:2.5, and 1:5. A control reaction contained 3.6 μM of TPX2 but no GST-LAP2. 25 μl of GST-agarose beads were added to each binding reaction, and reactions were incubated for 1 h at 4°C with rotation. The beads were collected by brief centrifugation in a nanofuge, the supernatant was collected, and the beads were washed three times with HB plus 100 mM NaCl and once with 0.1% Triton X-100 in HB plus 250 mM NaCl. Proteins were eluted from the beads by boiling in SDS sample buffer. 4× SDS sample buffer was added to the supernatant to 1×. 20 μl of the beads or 30 μl of supernatant were separated by 10% SDS-PAGE. The gel was stained with Coomassie to visualize protein. Chromatin decondensation was assayed as described by , with the following modifications: 20 μl of untreated, mock-depleted, or TPX2-Δ interphase extract high-speed supernatant (prepared as described by ) was supplemented with ∼1,000 sperm/microliter and incubated at 22°C. 1, 5, 10, 30, and 60 min after initiation of the reaction, 2-μl samples were spotted onto a coverslip, fixed with 1 μl of 3.7% formaldehyde in MWB containing 20 μg/ml Hoechst dye, and viewed in the fluorescence microscope. As a control, sperm chromatin not incubated in extract was spotted directly onto a slide, fixed, and viewed. Fig. S1 shows the results of the sperm chromatin decondensation assay. Fig. S2 shows nuclei assembled in TPX2 or mock-depleted extracts fixed at various time points and stained with mAb414 or antibodies to lamin B3. Fig. S3 shows higher magnification micrographs of selected images from . Online supplemental material is available at .

Eukaryotic cells are able to control gene expression by regulating the movement of transcription factors across the nuclear envelope. Transcription factors can be synthesized as latent precursor forms that are excluded from the nucleus until proper environmental cues activate processes that trigger their targeting and entry into the nucleus (). Several such regulated latent transcription factors have been described, including the extensively studied nuclear factor κB (NF-κB)/Relish, cubitus interruptus, and Notch proteins (; ; ). Notably, the activating signals that induce the translocation of these factors into the nucleus are initiated by receptors at the plasma membrane. Consequently, the movement of the regulated transcription factor provides the means to physically transmit signals from nonnuclear compartments to specific promoter sequences. Fundamental to understanding regulated latent transcription factors is the elucidation of mechanisms that restrict their activity under noninducing conditions. A general mechanism appears to directly regulate nuclear targeting by physically tethering or anchoring latent precursor forms of transcription factors outside the nucleus. The first example of such a mechanism is the sterol regulatory element binding protein (). Sterol regulatory element binding protein is an integral membrane protein that is anchored in the membranes of the early secretory pathway. The cytoplasmically oriented domain possessing transactivation activity is released from membranes in two successive rounds of proteolytic processing by site-specific membrane-bound proteases (). An additional example is NF-κB/Relish signaling; IF-κB sequesters NF-κB in the cytoplasm and prevents nuclear translocation by binding the actin cytoskeleton via ankyrin repeats (; ). However, with respect to nonmembrane factors like NF-κB, little is known regarding the efficiency of cytoplasmic retention. Given the nature and kinetics of protein–protein interactions, the retention of soluble latent transcription factors is expected to be incomplete even under noninducing conditions. Similarly, low level basal processing of soluble or membrane-bound transcription factors under noninducing conditions will generate active proteins that have the potential to inappropriately enter the nucleus. Consequently, to ensure the fidelity of signal transducing pathways, cells are likely to possess other modes of regulation in addition to cytoplasmic retention strategies to maintain the stringency of the inactive state under noninducing conditions. Several of the recognized signaling pathways from the plasma membrane to the nucleus in are involved in sensing nutrient availability and regulating nutrient uptake (; ). In the plasma membrane, the amino acid receptor Ssy1 () functions with two intracellular peripheral membrane proteins, Ptr3 and Ssy5, as the fundamental components of the Ssy1–Ptr3–Ssy5 (SPS)-sensing pathway (). This pathway induces the transcription of amino acid permease genes in response to extracellular amino acids. The homologous zinc finger transcription factors Stp1 and Stp2 are the downstream effectors of the SPS signaling pathway (; ). Stp1 and Stp2 bind to specific upstream activating sequences that are present within SPS sensor–regulated promoters (; ). Both Stp1 and Stp2 are synthesized as latent cytoplasmic factors that are mobilized by receptor-activated processing (, ; ). In response to the addition of amino acids and in a strictly SPS sensor–dependent manner, Stp1 and Stp2 are endoproteolytically cleaved by the endoproteolytic activity of the Ssy5 protease (; ). This event liberates the DNA-binding and transactivation domains from an ∼10-kD NH-terminal fragment. The shorter forms of Stp1 and Stp2 accumulate in the nucleus, where they function to transactivate SPS sensor–regulated genes. Because of the inability to process Stp1 and Stp2, cells lacking a functional SPS sensor exhibit diminished capacities to take up amino acids. Recessive loss of function mutations in (amino acid sensor independent) result in the constitutive expression of SPS sensor–regulated genes, bypass the requirement of a functional SPS sensor, and restore amino acid uptake in SPS sensor–deficient strains (). encodes a novel integral membrane protein. In this study, we examine the intracellular location and membrane structure of Asi1 and whether it is a constituent of the SPS signaling pathway. We report that Asi1 is a glycoprotein component of the inner nuclear membrane that restricts full-length unprocessed forms of Stp1 and Stp2 from binding SPS sensor–regulated promoters. Our findings indicate that Asi1 participates directly within the SPS signaling pathway and that its presence is required to maintain the repressed state of SPS sensor–regulated genes under noninducing conditions. These results reveal a novel role of inner nuclear membrane proteins and illuminate an additional layer of control that is required to establish proper levels of basal gene expression. The observation that SPS sensor–regulated genes are constitutively expressed even under noninducing conditions in mutant cells prompted us to study Asi1 as a potential regulator of transcription factor latency. Primary sequence analysis of Asi1 reveals two equally sized domains: an NH-terminal half with five hydrophobic segments predicted to be membrane spanning and a COOH-terminal hydrophilic half with a conserved Zn-binding RING motif (). To gain a more precise understanding of the significance of these domains, we determined the in vivo topology of Asi1. We created an epitope-tagged allele by inserting a cloning cassette encoding an HA epitope tag just before the stop codon of the open reading frame. The function of the encoded Asi1-HA protein was assessed using a growth-based assay. In brief, leucine auxotrophic () cells lacking a functional SPS sensor () are unable to grow on synthetic complex dextrose (SC) medium (). Such double mutant strains can only import leucine at sufficient rates to support growth in the absence of competing amino acids (synthetic minimal dextrose [SD]) or on SC when is inactive (; compare dilution series 1 with 2; ). Thus, the lack of growth of the triple mutant strain expressing the allele (dilution series 3; ) reflects the presence of a functional allele. Initial experiments using the functional epitope-tagged allele supported the notion that encodes an integral membrane protein. Asi1-HA pellets with membranes and is solubilized only in the presence of detergents (unpublished data). Asi1 has 11 possible NH-linked (NXS/T) glycosylation sites (). We examined whether any of these sites were used in vivo by examining the electrophoretic mobility of Asi1 in extracts before and after treatment with endoglycosidase H (endoH). A distinct endoH-dependent gel mobility shift was observed (, bottom; lanes 1 and 2), indicating that Asi1 is indeed a glycoprotein. As a first step to experimentally determine the topology of Asi1, we constructed the allele, which encodes a mutant protein with three amino acid residue changes (S4 to A, T21 to A, and N29 to Q) that destroy the three potential NXS/T glycosylation sites in the hydrophilic NH terminus. This mutant protein is not glycosylated (, bottom; lanes 3 and 4), indicating that the other potential glycosylation sites are not used and that the NH terminus is oriented toward the ER lumen during biogenesis. The allele (CHO) fully complements -null mutant phenotypes (unpublished data); thus, glycosylation is not required for Asi1 function. The availability of a functional but nonglycosylated CHO form of Asi1 enabled us to use a glycosylation-dependent topological reporter cassette () to further assess the structure of Asi1. The topological reporter cassette was inserted in frame into the hydrophilic COOH-terminal domain and each of the four hydrophilic loops (L1–L4) that separate the five hydrophobic segments of Asi1 (I–IV; ). The resulting gene sandwich fusions with the topological reporter inserted at positions 107, 144, 236, 313, and 624 encode functional proteins (unpublished data). We were unable to obtain functional proteins when the topological reporter was inserted into positions within loop L3. The glycosylation state of the fusion proteins was monitored. The results indicate that the topological reporter was efficiently glycosylated only when introduced into loops L2 and L4, indicating that these loops are oriented toward the lumenal side of the ER membrane during biogenesis (). The topological reporter was not glycosylated when placed in loop L1 or at various positions in the hydrophilic COOH-terminal domain, and, thus, these portions of Asi1 are in an extralumenal orientation (, xtr-l). In summary, the findings that the NH terminus of wild-type Asi1 is glycosylated and that topological reporter cassettes located in the COOH-terminal domain are not demonstrate that the NH and COOH termini are oriented toward opposite sides of a membrane. Consequently, Asi1 must have an odd number of hydrophobic segments that span the membrane. Although our experimental data are consistent with either three or five membrane-spanning segments (the inability to obtain a functional protein with a topological reporter in loop L3 introduces a degree of uncertainty), based on predictive computer algorithms (), it is likely that Asi1 has five membrane-spanning segments. Notably, the data indicate that the conserved RING motif at the extreme COOH terminus of Asi1 () is extralumenally oriented. The RING motif has putative zinc atom–coordinating residues with a spacing typical of C3HC4-type (RING-HC) zinc fingers (i.e., C-x-C-x-C-x-H-x-C-x-C-x-C-x-C; ). To investigate whether the RING domain is required for Asi1 function, we constructed two mutant alleles encoding proteins that lack the ability to bind zinc (, top). Plasmids expressing these mutant alleles ( and A; dilution series 4 and 5; ) did not complement the mutation, and strains were able to grow on SC medium, indicating that the asi1-HA21 and -HA22 proteins are nonfunctional. The levels of Asi1-HA, asi1-HA21, and asi1-HA22 proteins were similar in extracts (), excluding the trivial possibility that the mutations merely affected protein stability. These results indicate that the RING domain is essential for Asi1 function. As Asi1 is an integral membrane protein, we anticipated that knowledge regarding its precise intracellular location would help us understand its role in modulating gene expression. The intracellular location of the functional Asi1-HA epitope-tagged protein was examined by immunofluorescence. The Asi1-HA–dependent fluorescence was restricted to and evenly distributed in the nuclear envelope (, left). The punctate fluorescence associated with nuclear pore complexes (NPCs; , right) was clearly distinct from the continuous rim staining of Asi1. To directly test whether Asi1 associates with NPCs, we expressed Asi1-HA in the strain (; ). Loss of causes the clustering of nuclear pores to distinct areas of the nuclear envelope (, right). In contrast to the single intense spot of NPC fluorescence at the nuclear periphery in mutant cells, Asi1 fluorescence remained evenly distributed in the nuclear membrane (, left). Our observations demonstrate that Asi1 is a component of the nuclear membrane that is not intimately associated with nuclear pores. Next, we examined the localization of Asi1 by postembedding immunoelectron microscopy (). In this method, yeast cells are embedded in the acrylic resin LR White (). Osmium fixation is omitted in these preparations; consequently, the nuclear envelope appears in negative contrast as a light band surrounding the nucleus. This light band includes the inner and outer nuclear membranes. In yeast, the intermembrane space does not have sufficient intrinsic contrast to be visualized. On sections through resin-embedded cells, gold-coupled antibodies label only antigens exposed at the surface; penetration into the depth of sections is limited to a few nanometers at most (; ). We quantified the number of immunogold particles associated with different compartments in the cell (). The bulk of the gold particles (>80%) were associated with the nuclear compartment. Next, we measured the distance of gold particles from the nuclear membrane (). The distance from the inner or outer nuclear membrane was calculated by measuring the shortest distance between gold particles and the border between the negative contrast membrane and more contrast-rich cytoplasm and nucleoplasm. The data clearly demonstrate that immunolabeling was concentrated inside of the nucleus in close proximity (0–10 nm) to the inner nuclear membrane. These results are consistent with Asi1 being a component of the inner nuclear membrane. This finding coupled with our topology studies indicates that the essential RING domain of Asi1 is oriented into the nucleoplasm. Mutations in were identified in a genetic screen based on their ability to restore amino acid uptake in cells lacking a functional SPS sensor (). Two possible explanations exist for the constitutive expression of SPS sensor–regulated genes in mutants. Either Asi1 negatively regulates the activity of the transcription factors Stp1 and Stp2, the only known effectors of the SPS-sensing pathway (), or, alternatively, Asi1 functions independently of these factors. In the latter case, mutations could induce SPS sensor–regulated genes via an independent and perhaps parallel pathway. Accordingly, SPS sensor gene expression should remain constitutive in mutants lacking Stp1 and Stp2 ( ). strains carrying either a P () or P () reporter construct. Cells were grown in the presence or absence of the inducing amino acid leucine. In wild-type cells, the expression of and promoting β-galactosidase was dependent on amino acid availability; low levels of β-galactosidase were detected in uninduced cells (SD), whereas robust activity was detected upon induction (SD + leucine). The induced expression of P and P was strictly dependent on the SPS sensor, and no β-galactosidase was detected in the SPS sensor–deficient mutant. The mutation bypassed the requirement of a functional SPS sensor and the presence of an inducing amino acid. and mutants. mutant; thus, the constitutive transcription of SPS sensor–regulated genes in mutants is strictly dependent on the presence of Stp1 or Stp2. These results indicate that Asi1 is a component of the SPS signaling pathway that negatively modulates the activity of Stp1 and Stp2 under noninducing conditions. The NH-terminal regulatory domains of Stp1 and Stp2 possess two conserved sequence motifs that function independently and in parallel to confer amino acid–induced regulation of an artificial transcription factor (). One motif (Region I) is required for cytoplasmic retention, and the other motif (Region II) mediates SPS sensor–dependent endoproteolytic processing. The finding that loss of Asi1 function constitutively activates the expression of SPS sensor genes in a Stp1- and Stp2-dependent manner raised the possibility that Asi1 somehow affected one or both of these two regulatory activities. We initially considered that Asi1 prevents Stp1 and Stp2 processing under noninducing conditions. If this is so, Stp1 and Stp2 should be constitutively processed in an mutant. Stp1 and Stp2 processing was examined in wild-type () and strains (). In the absence of amino acids, Stp1 and Stp2 were found exclusively in their unprocessed full-length forms. Both factors were processed normally in amino acid–induced cells, and, thus, the processing of Stp1 and Stp2 is Asi1 independent. We monitored the levels of β-galactosidase resulting from P reporter gene expression in the same transformants (). Consistent with our previous results (), expression was constitutive in strains, whereas in wild-type strains, expression was amino acid dependent. To conclusively address the possibility that mutations affect processing, perhaps at levels too low to detect by immunoblot analysis, we assessed the ability of mutations to activate SPS sensor–regulated genes in a strain lacking the Stp1 and Stp2 processing protease Ssy5. An double mutant strain exhibited constitutive gene expression and exhibited growth phenotypes indicative of restored amino acid uptake capabilities (unpublished data). This finding formally rules out the possibility that Asi1 exerts regulatory effects by modulating Stp1 and Stp2 processing. Together, these results demonstrate that in the absence of Asi1, unprocessed full-length forms of Stp1 and Stp2 can enter the nucleus and induce transcription. Thus, the NH-terminal regulatory domains do not appear to grossly interfere with the transactivation potential of Stp1 or Stp2. We have previously shown that Stp1 is efficiently targeted to the nucleus only upon induction by amino acids and in a strictly SPS sensor–dependent manner (). Because mutations enable full-length forms of Stp1 and Stp2 to induce the gene expression of SPS sensor–regulated genes, we examined the possibility that the loss of Asi1 function interfered with the cytoplasmic retention of unprocessed Stp1 and Stp2. In this case, Stp1 and Stp2 should constitutively target to the nucleus in mutants. Using immunofluorescence microscopy, we determined the intracellular location of full-length (uninduced; −leu) and processed (induced; +leu) Stp1 and Stp2 in the wild-type () and strains. In both uninduced and cells, Stp1- and Stp2-dependent fluorescence was barely above background and diffused throughout cells (, C and D; −leu). In cells induced with leucine, intense and highly focused fluorescence that colocalized with DAPI-stained DNA was observed (, C and D; +leu). The finding that Stp1 and Stp2 do not constitutively accumulate in the nuclei of uninduced cells clearly demonstrates that Asi1 does not play a key role in retaining Stp1 and Stp2 in the cytoplasm. The apparent conflicting observations that mutant strains exhibit constitutive Stp1- and Stp2-dependent expression of SPS sensor–regulated genes () without visible nuclear accumulations of Stp1 or Stp2 () could be resolved if mutations enable only low levels of Stp1 and Stp2 to access SPS sensor–regulated promoters. We used chromatin immunoprecipitation (ChIP) to examine this possibility by analyzing the association of Stp1 with two SPS sensor–regulated promoters (i.e., and ). To facilitate the analysis, we used a Stp1 construct that carries a myc epitope at the NH terminus and an HA epitope at the COOH terminus (, schematic diagram); this construct complements –null mutant phenotypes and thus is functional (unpublished data). The use of this doubly tagged Stp1 construct allowed the promoter association of full-length and processed forms of Stp1 to be experimentally distinguished. Anti-myc antibodies can only immunoprecipitate unprocessed Stp1, whereas anti-HA antibodies immunoprecipitate both full-length and processed forms of Stp1. A plasmid encoding native Stp1 without epitope tags was included in the experiment as a control for nonspecific immunoprecipitation. The ability to amplify the promoter was used to control the quality of input DNA and the binding to nonspecific DNA sequences. In cells, we readily detected the association of Stp1 with and promoters, but only after induction with amino acids and only in lysates immunoprecipitated with anti-HA antibody (, compare lanes 2 with 3). The inability of anti-myc antibodies to immunoprecipitate these promoters indicates that only the shorter processed form of Stp1 is able to gain access to promoters in wild-type cells. In lysates prepared from cells, both anti-HA and anti-myc antibodies were able to immunoprecipitate the and promoters (, lane 4), even in cells grown in the absence of inducing amino acids. This finding indicates that in the absence of Asi1, full-length unprocessed Stp1 is indeed able to gain access to SPS sensor–regulated promoters. These results are entirely consistent with our phenotypic analysis of mutants and account for the constitutive expression of SPS sensor–regulated promoters observed in mutants. The NH-terminal fragment of Stp1 comprised of amino acids 1–125 is modular and can be transferred to faithfully regulate the activity of an artificial transcription factor in an SPS sensor–dependent manner (). As previously mentioned, this fragment contains two motifs that are also present in the NH terminus of Stp2 (). These motifs appear to have independent functions. Region I is required to prevent the unprocessed full-length forms of these factors from inducing SPS sensor–regulated genes, whereas Region II is required for amino acid–induced SPS sensor–mediated endoproteolysis. A smaller fragment spanning Region II of Stp1 (Stp1) interacts with Ssy5, the protease component of the SPS sensor (). In contrast, a fragment spanning Region I (Stp1) does not interact with Ssy5. These findings prompted us to directly test whether the NH-terminal regions of Stp1 and Stp2 containing Region I are sufficient to mediate Asi1-dependent promoter exclusion when fused to a nonrelated DNA-binding protein. The NH-terminal domains of Stp1 (amino acids 2–69) and Stp2 (amino acids 2–77) were fused to the COOH terminus of the bacterial DNA-binding protein lexA (). The lexA protein contains an intrinsic nuclear localization signal and, in the absence of additional sequences, is efficiently targeted to the nucleus (). Because of the lack of Region II sequences, the lexA-Stp1 and lexA-Stp2 fusion proteins are not subject to amino acid–induced SPS sensor–dependent processing (unpublished data). The ability of these fusion proteins to access DNA was tested using a reporter plasmid containing lexA operators (OP; schematically depicted in ). The expression of the β-galactosidase gene is regulated in a galactose-dependent manner as a result of the presence of the promoter. However, when lexA is present, it binds to the two OP that are placed between the promoter and the open reading frame, and, consequently, the level of -driven lacZ expression is repressed. A similar repression assay is routinely used to test whether fusion proteins intended to serve as bait in two-hybrid approaches are able to enter the nucleus (). β-galactosidase activity was measured in wild-type and mutant cells expressing the lexA-Stp1 and lexA-Stp2 constructs. Experimental controls included the empty vector and plasmid encoding only the lexA DNA- binding domain. As expected, cells carrying the empty vector and thus lacking lexA exhibited high levels of β-galactosidase activity. Cells expressing lexA lacking Stp1 or Stp2 sequences exhibited very low levels of β-galactosidase activity, clearly demonstrating the repressive effect of lexA. cells expressing either lexA-Stp1 or lexA-Stp2 exhibited high β-galactosidase activity at levels similar to that observed in cells carrying the empty vector control. These results demonstrate that the presence of either the NH-terminal fragments of Stp1 or Stp2 spanning the Region I motif prevent lexA from gaining access to the OP sites. Importantly, the expression of lexA-Stp1 or lexA-Stp2 in mutant cells significantly lowered β-galactosidase activity, indicating that in the absence of Asi1, the fusion proteins are able to access the promoter and reduce the expression of the reporter gene. This repression assay provides independent evidence supporting the view that Asi1 normally functions to prevent unprocessed Stp1 and Stp2 from accessing SPS sensor–regulated promoters. The nuclear envelope is a highly specialized structure that delimits the nucleus of eukaryotic cells. The nuclear envelope is comprised of inner and outer membranes, each with a distinct inventory of resident proteins. Aside from the detailed description of NPCs that provide entry and exit routes through the nuclear envelope, remarkably little is known regarding the role of nonpore proteins (). Importantly, the nuclear envelope is not only a physical barrier that restricts the movement of macromolecules between the cytoplasm and nucleoplasm but is thought to function as a scaffold for proteins required for diverse nuclear functions. Our finding that Asi1 is a component of the inner nuclear membrane that functions to maintain the latent properties of two related transcription factors by preventing them from binding DNA introduces an unanticipated and novel mechanism of gene regulation in eukaryotic cells. We previously reported that loss of function mutations in are recessive and lead to the constitutive expression of SPS sensor–regulated genes (). In this study, we have experimentally addressed Asi1 function. We show that Asi1 is a polytopic membrane protein () and an integral component of the inner nuclear membrane ( and ). Furthermore, the latent precursor forms of transcription factors Stp1 and Stp2 were defined as the targets of Asi1-dependent regulation (), clearly demonstrating that Asi1 is a bona fide constituent of the SPS pathway. Our studies have revealed an additional layer of regulation in this nutrient- induced signaling pathway. The constitutive expression of SPS sensor–regulated genes in mutants was traced to the ability of full-length unprocessed forms of Stp1 and Stp2 to enter the nucleus, bind promoters, and induce transcription ( and ). In contrast, unprocessed forms of Stp1 and Stp2 do not bind SPS sensor–regulated promoters in wild-type cells. During the course of this study, we experimentally addressed and ruled out several possible mechanisms that could explain the constitutive expression of SPS-regulated genes in mutants. First, the loss of Asi1 could facilitate precocious Stp1 or Stp2 processing; amino acid induction and an intact SPS sensor were found to be required for the processing of both factors in mutant strains (). Thus, Asi1 does not negatively modulate the endoprotease activity of the SPS sensor. Second, although Asi1 is not intimately associated with nuclear pores (), Asi1 could indirectly affect transport across the nuclear membrane. The loss of Asi1 did not visibly perturb the intracellular distribution of unprocessed Stp1 or Stp2 (). Therefore, Asi1 appears not to be mechanistically linked with cytoplasmic retention. Third, the presence of the essential RING domain in the nucleoplasmically oriented COOH-terminal region of Asi1 () raised the possibility that Asi1 is a ubiquitin ligase specifically involved in the degradation of nuclear-localized Stp1 and Stp2. Using standard ubiquitylation assays in which positive controls exhibited robust auto- and transubiquitylation activity, Asi1 was not autoubiquitylated, nor did it catalyze ubiquitylation of the purified NH-terminal regulatory domain of Stp1 (unpublished data). Consistent with these latter results, the turnover rates of Stp1 and Stp2 are similar in both wild-type and mutant strains (unpublished data). Our current model for the SPS sensor signaling pathway is schematically presented in . In the absence of inducing amino acids (), newly translated Stp1 and Stp2 are excluded from gaining access to SPS sensor–regulated genes by two parallel activities that converge on the NH-terminal regulatory motif spanning Region I (). The primary mechanism functions to anchor the unprocessed full-length forms to a presently undefined cytoplasmic determinant, which prevents targeting of unprocessed factors to the nucleus. Our current studies reveal that the efficiency of anchoring is not absolute. A second Asi1-dependent mechanism prevents the low levels of full-length Stp1 and Stp2 that enter or “leak” into the nucleus from activating transcription. Therefore, Asi1 is required to maintain the basal uninduced level of SPS sensor gene expression. In the presence of inducing amino acids (), the NH-terminal regulatory domains of Stp1 and Stp2 are proteolytically removed. The SPS sensor–catalyzed processing event depends on the conserved motif designated Region II (). The processed forms of Stp1 and Stp2 target to the nucleus, and because they lack Region I, they bypass Asi1-dependent control and efficiently bind SPS sensor–regulated promoters. Perhaps the simplest mechanism by which Asi1 could restrict Stp1 and Stp2 from accessing promoters would be by directly binding and sequestering them at the inner nuclear membrane. We tested this possibility using the NH-terminal regulatory domain (amino acids 1–125) of Stp1 and the hydrophilic COOH-terminal domain of Asi1 in reciprocal and unbiased genome-wide yeast two-hybrid approaches but failed to detect any interactions. Our lack of success raises the possibility that other proteins may be involved. An obvious candidate is the membrane protein encoded by . Null mutations in this gene give rise to identical phenotypes as mutations (). Asi3 is structurally related to Asi1, it has five putative membrane-spanning segments and a COOH-terminal localized RING domain (), and Asi3 also localizes to the inner nuclear membrane (unpublished data). Using immunoprecipitation approaches, we have found that Asi1 but not Stp1 copurifies with Asi3 (unpublished data). We are currently pursuing alternative strategies to more definitively assess potential Stp1 and Asi protein interactions. It is known that proteins that localize to or associate with the inner nuclear membrane can affect patterns of gene expression (for reviews see ; ). Recruitment of chromatin to the nuclear periphery often correlates with reduced gene expression; however, recent studies in yeast have clearly demonstrated that these events are not obligatorily linked (; ). Our data also provide a clear indication that inner nuclear membrane proteins can regulate gene expression independently of chromatin recruitment. In uninduced cells, Asi1 is required to maintain the repressed state of plasmid-encoded SPS sensor–regulated genes and, importantly, synthetic reporter gene constructs. Consequently, Asi1-mediated repression is independent of chromatin context. Furthermore, our repression and ChIP assays provide incontrovertible evidence that rule out the possibility that Asi1 recruits chromatin to the nuclear periphery indirectly via interactions with the NH-terminal regulatory domains of Stp1 or Stp2. Asi1 negatively controls the repressive activity of lexA-Stp1 or lexA-Stp2, and Stp1 is not bound to promoters when gene expression is repressed. In the nucleus of mammalian cells, lamins are involved in an extensive network of protein–protein interactions, including several known transcriptional regulators, some of which bind DNA (for reviews see ; ). For example, it was recently shown that direct interaction of lamin A/C with c-Fos suppresses AP-1 DNA binding and transcriptional activation presumably by reducing c-Fos/c-Jun heterodimer formation (). Additionally, several lamin-binding proteins (e.g., LBR, LAP2β, emerin, and MAN1) are integral components of the inner nuclear membrane. The ability to bind lamin is believed to be important because it provides the means to repress gene expression by recruiting lamin-associated regions of chromatin to the nuclear periphery. In addition to its ability to bind lamin, MAN1 can antagonize TGF-β and BMP2 signaling by binding Smads (; ; ). Smad proteins are transcriptional regulators that become phosphorylated by activated TGF-β and BMP receptors at the plasma membrane. The phosphorylated forms of Smads target to the nucleus, where they interact with various transcription factors and induce gene expression. The binding of Smads to MAN1 appears to be independent of stimulation by TGF-β or BMP2. However, Smads bound to MAN1 are hypophosphorylated with respect to free Smads (). These observations and the similarity in our findings regarding Asi1 raise the possibility that mechanisms that retain inactive Smads in the cytoplasm are not absolutely efficient, and, consequently, a small quantity of Smads enter the nucleus, where they are sequestered by MAN1 (). Such a mechanism may contribute to restricting the ability of Smads to transactivate gene expression. It seems that the biological regulation of gene expression is not limited to controlling cytoplasmic retention of latent factors. Cytoplasmic anchoring mechanisms that limit nuclear targeting are clearly not absolute, and transcription factors do inappropriately enter the nucleus, where they directly or indirectly affect gene expression. Consequently, eukaryotic cells have apparently evolved additional mechanisms to ensure the fidelity of signaling and to maintain the dormant, or repressed, state in the absence of inducing signals. The regulatory mechanism that we have described represents an additional layer of control that functions to establish the basal expression of genes controlled by a discrete signal transduction pathway. We believe that the insights gained from these studies can be extended to a more general and classic biochemical problem: how to generate a large difference between “off” and “on” catalytic states, which is an important and potentially difficult task in controlling multicomponent biological systems. The Asi1 story provides an intriguing example of how this can be achieved. Standard media, including YPD and ammonia-based SD supplemented as required to enable the growth of auxotrophic strains, were prepared as described previously (). Ammonia-based SC was prepared as described previously (). Where indicated, -leucine was added at a concentration of 1.3 mM to induce the SPS sensor. When required, 5-fluoroortic acid was added to 1 g/l SC. Media were made solid with 2% (wt/vol) Bacto Agar (Difco). Antibiotic selections were made on solid YPD supplemented with 200 mg/l G418 (Invitrogen), 100 mg/l clonNAT (Werner Bioagents), or 300 mg/l hygromycin B (Duchefa). The yeast strains used in data collection are listed in . All strains except nup133 are isogenic descendants of the S288c-derived strain AA255/PLY115 (). Plasmids used are listed in . Details regarding strain and plasmid constructions are provided online in the supplemental material (available at ). β-galactosidase activity was determined with -lauroyl-sarcosine–permeabilized cells (). Cells grown in SD were harvested by centrifugation, resuspended in Z buffer, and the OD was measured. A 250-μl aliquot of cell suspension was mixed with 550 μl of 0.3% (wt/vol) Na -lauroyl-sarcosine Z buffer and incubated at 30°C for 15 min. A 160-μl aliquot of 4 mg/ml ONPG (2-nitrophenyl β-D-galactopyranoside) solution was added, and tubes were mixed by vortexing. The reactions were stopped by the addition of 400 μl of 1 M NaCO, the tubes were centrifuged at 12,000 for 5 min, and the absorbance of the supernatant was measured at 420 nm. Activity was calculated according to the following formula: 10 × A/OD/0.25 (volume of cell suspension added)/time (minutes). Whole cell extracts were prepared under denaturing conditions with NaOH and TCA treatment according to . Extracted proteins were resolved using SDS-PAGE and analyzed by immunoblotting. Immunoblots were incubated with primary antibody (12CA5 ascites fluid; anti-HA monoclonal) diluted 1:1,000 in blocking buffer. Immunoreactive bands were visualized by chemiluminescence detection (SuperSignal West Dura Substrate; Pierce Chemical Co.) of HRP conjugated to a secondary antibody (anti–mouse Ig from sheep and anti–rabbit Ig from donkey; GE Healthcare) and quantified by using a gel documentation system (LAS1000; Fuji). Whole cell protein extracts derived from 1 ml of cultures (OD of 1) were prepared as described previously (). 25 μl of duplicate protein samples (equivalent to an OD of 0.2 cell suspension) were diluted with an equal volume of 100 mM sodium citrate, pH 5.5, and heated for 10 min at 37°C. Three milliunits of endoH (Roche) was added to half of the samples, and all samples were incubated overnight at 37°C (). Proteins were resolved by 7% SDS-PAGE and immunoblotted with monoclonal anti-HA antibody (12CA5). Cells were grown to an OD of 0.8 and processed for indirect immunofluorescence analysis essentially as described previously (). Cells were fixed by the addition of an aliquot of 37% formaldehyde directly to the cultures to a final concentration of 4.5% and incubated for 45 min at 30°C. To detect HA-tagged proteins, the primary antibody used was the 12CA5 or 3F10 anti-HA monoclonal antibody diluted 1:300. To visualize nuclear pores, monoclonal antibody raised against rat Nup153 (monoclonal antibody PF190 × 7A8; gift of V. Cordes, University of Heidelberg, Heidelberg, Germany) was used in a 1:10 dilution (). The secondary antibody was AlexaFluor488 conjugated to goat anti–mouse or donkey anti–rat IgG (H + L; Invitrogen) diluted 1:500. Strain nup133 is temperature sensitive for growth and was grown at 23°C and fixed at RT. Cells were viewed using a microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) with a plan-Apochromat 63× NA 1.40 objective. Digital images of cells examined using Nomarski optics and antibody-dependent and DAPI fluorescence (standard filter sets) were captured using a CCD camera (C4742-95; Hamamatsu) and QED Imaging software (Media Cybernetics). Image files were incorporated into figures using Photoshop CS (Adobe). Cells were grown to an OD of 0.7 and were fixed in the presence of 2% formaldehyde/0.1% glutaraldehyde at 30°C for 1 h with gentle shaking. The fixed cells were washed in SP buffer (1.2 M sorbitol and 0.1 M potassium phosphate, pH 7.5), pelleted by centrifugation, dehydrated in graded ethanol (70–100%), and embedded in LR White Resin (London Resin) by UV polymerization as previously described (). Thin sections were cut on an ultramicrotome (Ultracut; Leica) and picked up on formvar-coated nickel grids. For immunostaining, the sections were blocked with 5% BSA in PBS, pH 7.3, for 30 min and incubated for 120 min with rat anti-HA antibodies (clone 3F10; Roche) diluted 1:50 in PBS/5% BSA. The grids were then rinsed repeatedly with PBS/5% BSA and incubated for 60 min with goat anti–rat IgG conjugated to 12-nm gold particles (Jackson ImmunoResearch Laboratories) that were diluted 1:20 in PBS/0.5% BSA. After repeated washing with PBS/0.5% BSA and PBS alone, the grids were postfixed for 5 min with 2% glutaraldehyde in PBS, washed with PBS and water, and allowed to dry. Contrast staining was performed with saturated uranyl acetate for 5 min and lead citrate for 5 s. Finally, the sections were examined in an electron microscope (CM120; Philips) at 80 kV and photographed using a CCD camera (MegaPlus; Kodak). For quantitative evaluation of the immunogold labeling, successive cells were photographed at high magnification, and the distance of all gold particles from the midline of the nuclear envelope was measured using the analySIS system (Soft Imaging Software). ChIP analysis was performed according to with minor modifications. Cells were grown to an OD of 0.7 and were fixed for 30 min at RT in the presence of 1% formaldehyde. The formaldehyde was added directly to the cultures. Cells were harvested by centrifugation, resuspended in lysis buffer, and disrupted with glass beads by beating six times for 40 s in the beadbeater. The resulting lysate was sonicated twice for 10 s using a sonifier (Sonifier 250; Branson Ultrasonics Corp.) with output control set to 5 (average size of DNA fragments was 0.5 kb). Sonicated lysates were clarified by centrifugation (twice for 10 min at 15,000 ). The protein content was measured, the samples were adjusted to 10 mg/ml in 1,200 μl, and 10 μl of the total lysate was put aside to control input levels. The remaining lysates were split into two equal fractions. Magnetic beads with covalently attached sheep anti–mouse or sheep anti–rat IgG (Dynabeads M-450; Dynal) were incubated with mouse monoclonal anti-myc antibody (clone 9E10; Roche) or rat monoclonal anti–HA antibody (clone 3F10; Roche), respectively. 50 μl of coated beads were used in immunoprecipitation reactions. Immunoprecipitates were sequentially washed in lysis buffer, lysis buffer containing 500 mM NaCl, washing buffer (10 mM Tris-Cl, pH 8.0, 500 mM LiCl, 1% NP-40, 1% Na-deoxycholate, and 1 mM EDTA), and Tris-EDTA. Bound protein was eluated by incubating beads twice for 10 min at 65°C in 75 μl of eluation buffer (50 mM Tris-Cl, pH 8.0, 10 mM EDTA, and 1% SDS). Cross-linking of immunoprecipitates and input samples was reversed by an overnight incubation at 65°C, after which DNA was extracted (PCR Purification Kit; QIAGEN). PCR was performed with primers that amplify promoter regions of (PrMB23/24), (PrMB31/32), and (PrMB41/42). Taq polymerase (Invitrogen) and the corresponding buffer system were used. Hot start was achieved by using TaqStart antibody (BD Biosciences). The appropriate dilution of template DNA and the number of cycles (25–30) were empirically determined. Samples were first incubated for 3 min at 94°C, and the amplification cycle was as follows: 45 s at 94°C, 45 s at 50°C, and 20 s at 72°C. The reactions were stopped in the logarithmic phase of amplification, and the PCR products were separated on 2.3% agarose gel and visualized by ethidium bromide. Plasmid pMB18 is a centromeric version of pJK101 () that contains a -promoted reporter gene (P). Two lexA operators (OP) have been placed between the promoter and the gene; lexA fusion proteins that bind to these operators decrease the level of galactose-induced expression. Yeast strains were grown for 2 d in SC medium containing 4% galactose, 1% raffinose, and 0.2% glucose, and β-galactosidase activity was measured. Supplemental material provides detailed descriptions of strain and plasmid constructions. Table S1 provides data on yeast strains. Online supplemental material is available at .

Phosphoinositides serve as important regulators in cellular membrane dynamics and are concentrated at the specific subcellular locations. For instance, phosphatidylinositol 3′-monophosphate (PI3P), phosphatidylinositol 4′-monophosphate (PI4P), and phosphatidylinositol 4′,5′-bisphosphate are enriched in early endosomes, the Golgi apparatus, and the plasma membrane, respectively (; ). These phosphoinositides contribute to membrane dynamics by recruiting specific downstream factors such as adaptor proteins, lipid transfer proteins, and components of the endosomal sorting complex required for transport (; ). In contrast to their well-known functions in membrane dynamics of authentic organelles, only the PI3P pathway has been implicated in the phosphoinositide signaling of the autophagic processes. PI3P appears to be responsible for two distinct steps during autophagy, the de novo synthesis of the autophagosome and its fusion with vacuole/lysosome (; ; ; ). An autophagy-specific phosphatidylinositol-3-OH kinase (PI3K) complex, containing Atg14 in addition to Vps30 (Atg6) and Vps15, was first identified in budding yeast (). The potential role of other phosphoinositides in the autophagic pathway is not presently known. We examine the selective degradation of peroxisomes through autophagy (termed pexophagy) in . The transient formation of a cup-shaped double-membrane structure (the micropexophagy-specific membrane apparatus [MIPA]) during micropexophagy on the cytosolic side of the peroxisome surface is crucial for the incorporation of the peroxisome into a vacuole (; ). Using fluorescence microscopy together with the MIPA marker protein PpAtg8, we examined the spatiotemporal formation of the MIPA, including its characteristic shape, size, and localization (). As many molecular components are shared between autophagy and pexophagy, pexophagosomes and MIPAs are proposed to be synthesized de novo by a common mechanism. In the present study, we demonstrate that PI4P (generated by PpPik1 and PpLsb6) is necessary for the formation of the membrane structure of the MIPA by revealing a novel PI4P-signaling mechanism necessary for pexophagy. To investigate how PpAtg26, a UDP-glucose:sterol glucosyltransferase having a GRAM (glucosyltransferase, Rab-like GTPase activators, and myotubularins) domain (), contributes to pexophagy, we examined the consequences of GRAM domain–deleted derivatives. The PpAtg26-GRAM domain binds to PI4P. We used these mutants lacking the GRAM domain to implicate this enzyme as an effector of PI4P signaling in pexophagy and to dissect the steps of its action in the membrane formation. The present results indicate that recruitment of PpAtg26 to the nucleation complex is required for the membrane elongation step but not for the nucleation of other Atg proteins. In addition, the sterol-conversion activity of PpAtg26 is essential for membrane elongation and contributes to forming the MIPA. The pathological and clinical importance of GRAM domains is supported by a report documenting that some myopathy patients have mutations within the GRAM domains of myotubularins (; ). GRAM domains from the myotubularins bind to phosphoinositides (; ), but these display low homology to the GRAM domain of Atg26 (; ). First, to explore which phospholipids might bind to the GRAM domain of PpAtg26, we compared the binding properties of the GRAM domains from yeast PpAtg26 (GRAM and truncated GRAM [trGRAM]–PH) with that from human myotubularin (GRAM; ). In a protein lipid overlay assay, both PpAtg26 domains bound to PI4P with the highest affinity (). The binding of GRAM to PI4P was also confirmed by surface plasmon resonance analysis (; K = 6.9 nM). The GRAM domain was found to have less but substantial binding activity to phosphatidylinositol 5′-monophosphate (PI5P). However, the yeast cells do not contain PI5P. Neither GRAM nor trGRAM-PH bound to ergosterol or ergosterol glucoside, the substrate and the product of reaction catalyzed by PpAtg26. GRAM exhibited broader binding specificities than GRAM as reported previously (; ; ). The binding specificity of GRAM was confirmed with a protein-liposome sedimentation assay (unpublished data). These data demonstrate that GRAM binds to PI4P but not to other phosphoinositides present in yeast cells, such as PI3P, phosphatidylinositol 3′,5′-bisphosphate, or phosphatidylinositol 4′,5′-bisphosphate. Next, we generated a mutant version of GRAM that carried a Y57P amino acid substitution. This mutation corresponds to one of the substitutions found in myopathy patients (Mtm1 L59P; ). The GRAM Y57P mutation caused a pexophagic defect (). Both the GRAM Y57P and GRAM L59P substitutions eliminated the high-affinity binding and specificity of the GRAM domains (), thereby linking the biochemical properties of these GRAM domains to the mutant phenotypes. To determine the role of PI4Ks in MIPA formation and pexophagy, we generated three strains, one mutant for each one of the PI4Ks—PpPik1, PpStt4, and PpLsb6. To construct and mutants, serine residues conserved in the putative catalytic site of PI4Ks (S994 for PpPik1 and S1816 for PpStt4) were substituted with phenylalanine (). The and strains exhibited growth defects, whereas cells displayed a normal growth rate. These phenotypes are similar to those reported for the PI4K mutants (; ). Inositol labeling experiments indicated that in comparison to wild-type cells the size of the intracellular PI4P pool was reduced by ∼11, 30, and 45% in , , and mutants, respectively (). We assessed pexophagy in these PI4K mutants by examining peroxisomal alcohol oxidase activity (). This enzyme is degraded through pexophagy along with other components of the peroxisome (). In this assay, persistent alcohol oxidase activity resulting from impaired pexophagy produces a colored colony (). In cells, pexophagy occurred as rapidly as in wild-type cells, whereas the pexophagic activity was abrogated in cells and attenuated in cells (). Next, we followed the formation of the membrane structure of the MIPA in these pexophagy-defected PI4K mutants to explore whether PI4P signaling is essential for the formation of the membrane structure. As PpAtg8 is targeted specifically to the MIPA through a ubiquitin-like conjugation, a YFP-labeled PpAtg8 was used as a marker for the MIPA. 1 h after inducing pexophagy, ∼7% of the wild-type cells exhibited a cup-shaped fluorescent signal characteristic of the MIPA (). The low frequency of observable MIPA is due in part to its transient nature; this membrane structure is formed only just before the incorporation of the peroxisome into the vacuole, after which it is destroyed (). Under these conditions, cells did not exhibit any cup-shaped YFP-PpAtg8 fluorescence; instead, small fluorescent puncta were proximal to the peroxisomes (). mutant cells showed cup-shaped MIPA fluorescent signals, but the percentage of cells containing these structures was lower than that seen in the wild-type cells (). These results correlate well with the data obtained from the alcohol oxidase assay (), demonstrating a novel and critical role for PI4Ks in the formation of the membrane structure during pexophagy. YFP-GRAM was used as a probe to visualize the intracellular distribution of PI4P. In a CHO cell line, YFP-GRAM produced a typical Golgi-like fluorescence that overlapped with the fluorescence pattern of the Golgi marker, BODYPY TR C5-ceramide (Fig. S1, available at ). As GRAM specifically binds to PI4P and the Golgi is a PI4P-containing organelle (), our observation verified that the expressed GRAM probe recognizes the intracellular PI4P in a heterologous host. In , YFP-GRAM generated peculiar ring-shaped fluorescence under pexophagic conditions and interfered with the localization of CFP-PpAtg8 to the MIPA (). The shape and localization of the YFP-GRAM fluorescent signal, together with subcellular fractionation experiments (), support the conclusion that YFP-GRAM localizes to the Golgi apparatus or Berkeley body. Mislocalization of CFP-PpAtg8 to MIPA may be due to the secondary effect of disordered function of the Golgi apparatus. We speculated that the inhibition of MIPA formation resulted from the masking of functional PI4P by the overexpressed GRAM. To eliminate this inhibition, we performed subcellular fractionation experiments of the MIPA-formed cells and probed the PI4P pool with a GRAM domain fusion protein (). Purified GST-GRAM-CFP proteins (, WT or Y57P) were incubated with membrane fractions isolated from pexophagy-induced cells expressing HA-tagged PpAtg8. Sucrose density gradient ultracentrifugation in combination with immunoblot analyses was performed on the subsequent fractions. The bulk of GRAM protein (, closed rectangles) was present in two peaks within the range of the high-density fractions (fractions 1–4) and the low-density fractions (fractions 10–12; ). HA-PpAtg8 (a MIPA marker; , open circles) signal exclusively distributed to the high-density fractions (fractions 1–4) and colocalized with one of the GRAM peaks. The peaks in the low-density fractions overlapped with the Kex2-positive fractions (, asterisks), consistent with PI4P in the Golgi apparatus (). We conclude that GRAM targeting to the membrane fraction was PI4P specific and that the high-density PpAtg8-positive fractions also contained PI4P. The GRAMY57P mutant protein (, open rectangles) was not enriched in the high-density fractions, consistent with the notion that the GRAM targets to the high-density fractions through binding PI4P. The GRAMY57P mutant protein was present in the low-density fractions, but its distribution was broader than that of the GRAM and included higher density fractions. As the GRAMY57P mutant protein does not bind PI4P specifically but retains weak and nonspecific PI binding activities (), the distribution of the GRAMY57P mutant protein may result from less specific binding to various PI pools in the cellular membranes. The enrichment of PI4P in PpAtg8-positive structures also was confirmed by in situ application of CFP-GRAM to permeabilized cells expressing YFP-PpAtg8 (unpublished data). Thus, the formed PpAtg8-positive structures, i.e., MIPAs, are rich in PI4P. Both PI4Ks (PpPik1 and PpLsb6) involved in the membrane formation did not colocalize with YFP-PpAtg8, the MIPA marker (). CFP-PpPik1 exhibited a punctate distribution that did not overlap with the cup-shaped YFP-PpAtg8 fluorescence of the MIPA. CFP-PpLsb6 resided on the vacuolar membrane (and lumen), as reported for ScLsb6 (; ). These observations are consistent with PI4P migrating from the sites of their production to the nascent MIPA. Studies of the localization of Atg proteins in reveal that multiple Atg proteins localize to a single puncta in the vicinity of the vacuole before autophagosome formation that is referred to as the preautophagosomal structure (PAS; ; ). The formation of PAS appears to be a prerequisite for autophagic membrane formation. Our studies have also identified a punctuate structure containing several Atg proteins (PpAtg8, PpAtg5, PpAtg16, and PpAtg26) before the MIPA formation (unpublished data). Lipidation of PpAtg8 through a ubiquitin-like conjugation system (PpAtg4, PpAtg7, and PpAtg3) is necessary for PpAtg8 to localize to punctuate structures (), consistent with these PpAtg8 puncta–containing protein–lipid complexes. In the cells, the YFP-PpAtg8 puncta colocalized with PpAtg16- and PpAtg5-CFP (), two Atg proteins concentrated in PAS in (). In contrast, the YFP-PpAtg8 puncta in the cells did not overlap with CFP-PpAtg26 (). Thus, PI4P recruits PpAtg26 but is dispensable for the localization of at least three Atg proteins (PpAtg8, PpAtg5, and PpAtg16) to these punctate structures. Disruption of another responsible PI4K, PpLsb6, caused a less prominent phenotype (). The percentage of the cells possessing MIPAs was reduced in comparison to the wild-type strain (), but the observed MIPAs contained CFP-PpAtg26, unlike cells. These results are consistent with our findings in the pexophagy and MIPA formation assay (), where PpPik1 was a major contributor to pexophagy but PpLsb6 was not. In mutant cells (Δ), YFP-PpAtg8 fluorescence did not develop into a cup-shaped MIPA structure but rather colocalized in puncta with PpAtg16-CFP (). Thus, PpAtg26 is required for MIPA formation. In addition to a phosphoinositide binding domain (PBD) comprising GRAM, trGRAM, and a pleckstrin homology (PH) domain, PpAtg26 contains a catalytic region including a UDP-sugar binding domain (UBD) that is necessary for the catalysis of ergosterol and UDP-glucose to ergosterol glucoside (). To analyze further the role of PpAtg26 in MIPA formation, we introduced constructs into Δ cells that encode one of several CFP-PpAtg26 variants, including a full-length PpAtg26 (CFP-PpAtg26FL; ), a PpAtg26 mutant protein lacking the PBD (aa 198–652; CFP-PpAtg26ΔPBD; ), and a PpAtg26 variant missing the UBD (aa 1085–1113; CFP-PpAtg26ΔUBD; ). CFP-PpAtg26FL rescued the pexophagic defect of the Δ mutant cells; it colocalized with cup-shaped YFP-PpAtg8 fluorescence of MIPAs (). In contrast, expression of CFP-PpAtg26ΔPBD did not yield the cup-shaped YFP-PpAtg8 fluorescence. CFP-PpAtg26ΔPBD was dispersed throughout the cytosol (). We conclude that PI4P binding by the PBD is essential for the PpAtg26 localization to the Atg8-positive puncta proximal to peroxisomes, a prerequisite for the membrane elongation necessary to form the MIPA. Deletion of each domain within the PBD (trGRAM, PH, or GRAM) produced a similar pattern of PpAtg8 fluorescence to the overall PBD deletion; i.e., only fluorescent puncta were observed (Fig. S2, available at ). The PH domain–deleted mutant had the most severe phenotype, as we did not observe any colocalization with YFP-PpAtg8 (Fig. S2). The CFP-PpAtg26 variants lacking either the GRAM domain or trGRAM colocalized with PpAtg8 puncta less frequently than CFP-PpAtg26FL ( and Fig. S2). These deletion constructs confirm the important role for each domain in MIPA formation, consistent with previous experiments (). In addition, we conclude that these PI4P binding regions function cooperatively to recruit PpAtg26 to punctual structures. Overexpression of CFP-PpAtg26ΔUBD did not restore the sterol-converting activity to the Δ mutant cells as determined by thin-layer chromatography (Fig. S3, available at ). This strain did not exhibit the cup-shaped YFP-PpAtg8 fluorescence but did have YFP-PpAtg8 puncta (). In contrast to CFP-PpAtg26ΔPBD expression (), these puncta colocalized with CFP-PpAtg26ΔUBD fluorescence (). We conclude that PpAtg26-dependent ergosterol conversion at the assembly site of multiple Atg proteins is necessary for maturation and elongation of the membrane structure to form the MIPA. The present study examines a novel PI4P-signaling pathway for the formation of the membrane structures required for pexophagy. Although our previous study identified PpAtg26 as a factor necessary for pexophagy and demonstrated the functional importance of its GRAM domain in pexophagy (), neither the role of Atg26 in the formation of the membrane structure nor the biochemical function of the GRAM domain was made clear. Here, we demonstrate that the GRAM domains of PpAtg26 bind specifically to PI4P. We propose that the GRAM domain is indeed a phosphoinositide binding motif that is conserved from yeast to higher eukaryotes. This finding prompted us to study PI4P signaling during pexophagy. Indeed, among three PI4Ks (PpPik1, PpStt4, and PpLsb6) present in yeast cells, mutations in PpPIK1 and PpLSB6 impair pexophagy. We consider this direct evidence that PI4P signaling is involved in pexophagy, especially in the formation of the membrane structure. The formation of the membrane structure during pexophagy involves three distinct steps (): PI4P (predomi- nantly synthesized by PpPik1) concentrates in the nucleation complex, where multiple Atg proteins (including lipidated PpAtg8) assemble; PpAtg26 is recruited to the nucleation complex through an interaction between PI4P and the GRAM domain; and sterol conversion at the nucleation site triggers elongation and maturation of the membrane structure. Although the maturation and elongation of the membrane structure are defective in the PI4K or PpAtg26 mutant strains, many Atg proteins, except PpAtg26, assembled at the peroxisomal surface normally. In contrast, in strains deficient in the PI3 kinase (Vps34) complex (), the nucleation step of autophagosome formation is impaired, and there is a marked deficiency in the localization of multiple Atg proteins at the PAS (; ). For example, we observed multiple YFP-PpAtg8 puncta dispersed in the cytosol of the mutant (unpublished data) in contrast to a single punctum in the mutant, suggesting that proper formation of the MIPA precursor at the peroxisomal surface requires PI3P signaling. These findings support a nucleation and elongation/maturation model of autophagosome assembly (; ). We propose that each step of the de novo membrane formation during pexophagy is regulated by two distinct phosphoinositides—PI3P and PI4P. To date, most of the molecular events exerted by PI4P signaling in yeast cells are related to the Golgi apparatus (; ). The absolute requirement of PpPik1 for MIPA formation implicates lipid/membrane flow from Golgi-related compartments during the de novo membrane structure synthesis. Previous studies on other autophagic pathways demonstrate the contribution of the early secretory pathway (involving ER and Golgi) in autophagosome formation (; ). In addition, Tlg2, a syntaxin homologue Tlg2 that shuttles between the late Golgi and the endosomal membrane, is involved in a specific autophagic cytosol to the vacuole targeting pathway for aminopeptidase I biogenesis (). Our results also support the possibility that PI4P (possibly together with other lipids) flows from the early secretory pathway to the nucleation site to form membrane structure. In contrast to the evidence implicating PpPik1 in pexophagy, the role of PpLsb6 is not established. The mutant has no abnormalities in the growth rate or endocytosis (deduced by FM4-64 uptake). How and where PI4P synthesis by PpLsb6 contributes to pexophagy is not clear. However, considering that the formation frequency of MIPA was reduced in the mutant (), we propose that PI4P produced by PpLsb6 is transported to the nucleation complex and that this PI4P helps to recruit PpAtg26 alongside PI4P produced by PpPik1. Although the expressed GRAM domain localizes to the Golgi apparatus as well as to the MIPA (), CFP-PpAtg26FL exclusively localized to the MIPA (). We assume that this difference results from the other PpAtg26 regions contributing to the localization of this protein, especially the PH domain. Our previous study demonstrated that PpAtg26 resides in a membrane fraction that is resistant to extraction by Triton X-100, and this property of detergent insolubility is partially dependent on the PH domain (). Thus, the PH domain may recognize “lipid” rafts within the target membrane. The concerted actions of both the GRAM and PH domains may mediate the specific recruitment of PpAtg26 to the MIPA. PI4P signaling recruits PpAtg26 to the nucleation complex and culminates with ergosterol conversion at a specific site. The exact functions of ergosterol conversion in the phenomenon of membrane elongation are not yet understood. However, our finding reveals a novel physiological role of sterol as a reaction substrate required for the formation of a membrane structure. Also, it is worthwhile to point out the sterol-free characteristics of autophagosome-like membranes in mammalian cells. Unlike endosomal membranes, mammalian Atg8 (LC3)-positive isolation membranes, which engulf bacteria, are resistant to the cholesterol binding toxin streptolysin O (). Although mammalian Atg26 homologues have not been identified as yet, has an Atg26 orthologue that contains a phosphoinositide binding FYVE domain (). These observations may reflect a conserved mechanism of de novo membrane formation during the selective autophagy that is dependent on phosphoinositide signaling and site-specific sterol conversion. PpAtg26 regions corresponding to aa 196–248 (trGRAM-PH), aa 586–642 (GRAM), and the HsMtm1 region (aa 29–97; GRAM) were generated by PCR amplification. The amplified fragments were cloned into the pGEX6P-1 vector (GE Healthcare) to produce fusion proteins with GST and YFP. Point mutations of the tyrosine residue at position 57 of GRAM with a proline residue (Y57P) and the leucine residue at position 59 of GRAM with a proline residue (L59P) were made with the QuikChange PCR kit (Stratagene). Protein expression in Rosetta DE3 cells (Novagen) was induced with 0.5 mM IPTG at 20°C for 6 h. Expressed proteins were purified using glutathione Sepharose 4B (GE Healthcare) according to the manufacturer's instructions, except that DTT and Triton X-100 were omitted from the procedures. The glutathione-eluted samples were dialyzed against HBS buffer (10 mM Hepes and 150 mM NaCl, pH adjusted to 7.4) and used as a purified fusion protein in the subsequent assays. The protein lipid overlay assays were done with PIP Strips (Echelon Biosciences) using 1 μg/ml of the purified proteins in accordance with the manufacturer's instructions. Surface plasmon resonance analysis was done using a BIACORE 2000 system and Sensor chip L1 (Biacore). Liposome samples (1 mg/ml total lipid concentration) containing 60% (wt/wt) phosphatidylcholine, 19% (wt/wt) phosphatidylserine, 19% (wt/wt) cholesterol, and 2% (wt/wt) phosphoinositide were prepared by sonication (the lipids were purchased from Sigma-Aldrich). The sensor chip was coated with 3,000 RU of liposome and then coated with another 3,000 RU of liposome that lacked phosphoinositides. Samples containing 10–40 ML/ml of the purified GRAM fusion protein were applied to the chip, and the results were analyzed with BIAevaluation 3.0 (Biacore). The dissociation constants (K values) were determined from three independent experiments. To generate the and mutant strains, 1-kb fragments within both kinase genes (available from GenBank/EMBL/DDBJ under accession nos. and ) encoding their COOH-terminal regions were cloned into pPICZ-A (Invitrogen). Point mutations were then introduced using the QuikChange PCR kit. The plasmids were linearized and introduced into wild-type YAP0004 genome (), resulting in the genomic replacement of PpPIK1and PpSTT4 by the mutated ORFs, the COOH-terminal region of PpPik1S994F and PpStt4S1816F, respectively. To generate the strain, the PstI–BglII fragment of the PpLSB6 ORF (available from GenBank/EMBL/DDBJ under accession no. ) was cloned into pPICZ-A (Invitrogen). The resultant plasmid was digested with SnaBI and introduced into the YAP0004 strain genome to produce cells carrying two truncated forms of PpLSB6. To induce pexophagy, all of the prepared mutant strains were cultured for 15 h in synthetic methanol medium (0.75% methanol, 0.75% yeast nitrogen base without amino acids [Difco], and 100 mg/l auxotrophic amino acids) containing 0.37 MBq/ml H-labeled -inositol (GE Healthcare) and shifted to glucose medium (2% -glucose, 0.75% yeast nitrogen base without amino acids, and 100 mg/l auxotrophic amino acids) for 1 h. After the cells were harvested, total lipid was extracted, deacylated with methylamine, and analyzed by HPLC according to the procedure described in a previous study (). 3 μg of each of the purified GRAM fusion proteins used in the protein–lipid binding experiments was incubated at 4°C for 14 h with a 100,000 pellet fraction (equivalent to 0.4 mg protein) of the PPM5011 strain () cell homogenate expressing HA-tagged PpAtg8. Subsequent ultracentrifugation was done essentially as described previously (), except that 40 mM NaF was added to the lysis buffer. Each fraction was subjected to immunoblot analysis. The primary antibodies used for immunoblotting were as follows: the GRAM domains were detected with an anti-GFP rabbit polyclonal antibody (1:3,000; Invitrogen), the PpAtg8 with an anti-HA mouse monoclonal antibody (1:1,000; Boehringer), and Kex2 with an anti-Kex2 goat polyclonal antibody (1:1,000; Santa Cruz Biotechnology, Inc.). The ECL system was used for secondary detection (GE Healthcare), and bands were analyzed with MetaMorph imaging software (Universal Imaging Corp.). The alcohol oxidase assay and fluorescence microscopy were done as described previously (). In order to observe the localization of GRAM in mammalian CHO-K1 cells, a plasmid derived from pTRE2hyg vector (CLONTECH Laboratories, Inc.) harboring a DNA fragment encoding YFP-GRAM was introduced into CHO-K1 Tet-On Cell Line (CLONTECH Laboratories, Inc.). After successive selection with hygromycin for transfected cells, doxycycline (CLONTECH Laboratories, Inc.) was applied at 1 μg/ml and incubated for 48 h to induce the expression of the fusion protein. The cells were subsequently incubated for 1 h with BODYPY TR C5-ceramide complexed to BSA (Invitrogen) in the presence of 5 μM ceramide. A confocal microscope (LSM510 META; Carl Zeiss MicroImaging, Inc.) equipped with a multiline (458, 477, 488, and 514 nm) argon ion visible lasers along with a 63× Plan-Apochromat NA 1.4 oil immersion (Carl Zeiss MicroImaging, Inc.) was used for observation of CHO cells. The frequency of MIPA formation was determined by counting the number of cup-shaped images after YFP-PpAtg8 labeling. These values were normalized to the number of peroxisome grains labeled with CFP in the same observation area. The presented values are the mean from three separate experiments (where each count included >100 cells). CFP and NH-terminal 1-kb PpPik1 fragments were cloned into the pIB2 expression vector that carried a (Ala)Gly-Ser linker sequence between the two ORFs to express the CFP-PpPik1 under glyceraldehyde-3-phosphate dehydrogenase (GAP) promoter. The resultant plasmid was excised with EcoRV and then introduced into the SA1017 strain (containing YFP-PpAtg8) to replace the NH-terminal region of the genomic PpPIK1. The full-length PpLSB6 ORF fused to CFP in the pIB2 expression vector was also introduced into the SA1017 strain to visualize PpLSB6. PpAtg5-CFP was expressed under the control of its own promoter cloned into pHM100 (), and PpAtg16-CFP was expressed under the control of the GAP promoter. The plasmids encoding CFP-fused PpAtg26 and its domain-deleted derivatives were constructed in the CFP-cloned pIB4 expression vector (with alcohol oxidase promoter) as follows: PpAtg26FL, full-length ORF; PpAtg26ΔPBD, ORF lacking the region corresponding to aa 196–652; PpAtg26ΔUBD, ORF lacking the region for aa 1085–1113 (all the domain deletions were generated with the QuikChange PCR kit). Each deletion construct was expressed in a YFP-PpAtg8–harboring strain whose endogenous had been disrupted by the method described previously (). The lipid extraction from 100 OD unit cells of each strain was done by the Bligh-Dyer method, and the thin-layer chromatography was performed as described previously (). Sterol glucoside was detected by spraying 2% (wt/vol) 5-methylresorcinol in 2N sulfuric acid. Fig. S1 shows the image of YFP-GRAM fluorescence along with that of a Golgi-specific marker. Fig. S2 shows the images of fluorescence from CFP-PpAtg26 variants along with YFP-PpAtg8 fluorescence. Fig. S3 provides the result of thin-layer chromatography of the total lipid extract from the CFP-PpAtg26 variant-harboring strains used in . Online supplemental material is available at .

Many pathways of intermediary metabolism, macromolecular biosynthesis, energy transduction, motility, protein trafficking, and intracellular signaling require collaboration among organelles. Examples include the conversion of fatty acids to gluconeogenic substrates (; ), sterol biosynthesis (; ), membrane lipid synthesis (), the unfolded protein response (), vesicular transport (; ), the retrograde pathway (), and calcium signaling (; ). Intermediates en route between organelles generally pass through the cytoplasm, but this does not always have to be the case. Organellar compartments are usually densely packed within the cytoplasm, and organelles often associate with one another, raising the possibility of communication that bypasses the cytoplasmic compartment. Intracellular lipid bodies store neutral lipids, mainly triglycerides and sterol esters (). The organelles are derived from the ER, although the details of this pathway are not well understood (; ; for review see ). They are bounded by a phospholipid monolayer () into which are embedded a specific subset of proteins. Many of these have been identified through recent proteomic efforts (; ; ). This compartment is comprised of enzymes that promote the synthesis or mobilization of lipids, structural proteins, and signaling molecules. Lipid bodies can be in close proximity to mitochondria or ER (; ). The ER can virtually surround lipid bodies, and mitochondria have been detected in the cortical layer surrounding lipid bodies. There are also reports of the close proximity of peroxisomes to lipid bodies in etiolated cotyledons, mammalian cells, and yeast (; ; ). Because lipid bodies are an obvious source for fatty acids, the substrates for mitochondrial and peroxisomal oxidation, which are contacts between lipid bodies and peroxisomes or mitochondria, could indicate the direct transfer of fatty acids across organellar boundaries. The yeast is an apt model system to explore peroxisome–lipid body interactions. This organism can be cultured on oleic acid, which generates large lipid bodies and causes peroxisomes to become engorged with enzymes of fatty acid oxidation and the glyoxylate cycle (; ). In this study, we show that peroxisomes stably adhere to lipid bodies when grown in oleic acid and that they can extend processes into their core. Enzymes of peroxisomal β oxidation are selectively enriched in purified lipid bodies. Our data suggest that peroxisomal contact may stimulate neutral lipid breakdown in lipid bodies: fatty acids accumulate within lipid bodies if peroxisomes are present but are unable to metabolize them, generating novel structures we term “gnarls.” To determine the extent to which peroxisomes associate with lipid bodies in , Pot1p (peroxisomal 3-ketoacyl-CoA thiolase) was tagged with GFP, and cells were stained for lipid bodies with oil red O. Cells cultured in glucose had relatively small lipid bodies and few peroxisomes (), although peroxisomes associating with lipid bodies could easily be seen (, arrows). Cells cultured in oleic acid contained large clusters of lipid bodies and more peroxisomes, many of which seemed to adhere to the lipid bodies (). To determine whether the same pattern was seen in living cells, cells containing Pot1p-GFP were also tagged with Erg6p-mDsRed (); Erg6p is an abundant protein of lipid bodies (). Because Erg6p is on the periphery of the organelle, the large lipid bodies from oleate cultures often appear hollow. Again, peroxisomes were seen to decorate lipid bodies, especially in cells grown on oleic acid. To quantify this effect and compare it with the extent of association of other organelles with lipid bodies, we obtained matched strains () in which markers of peroxisomes, late Golgi, or endosomes (Pex3p, Sec7p, or Snf7p, respectively) were tagged with GFP. We then labeled their lipid bodies with Erg6p-mDsRed as before. Random images from single focal planes (such as those in ) were used to score the organelles that appeared to be associated with lipid bodies. We found in glucose cultures that 64.0 ± 2.9% of peroxisomes associated with lipid bodies compared with 49.5 ± 1.2 and 43.3 ± 3.7% of late Golgi and endosomal particles, respectively. Although a fraction of these interactions undoubtedly reflect adventitious proximity, these data may also reflect a role of lipid bodies in providing membrane lipids to all of these organelles. In oleate cultures, >90% of all three organelles seemed to associate, although much of this reflects the large volume within the cytoplasm that is taken up by the lipid bodies and the limited spatial resolution of this technique. To compare the dynamics of these organellar associations, we tracked the dissociation of individual organelles from lipid bodies using spinning disk confocal microscopy, and the results are shown for glucose and oleate cultures in (G and N, respectively). In glucose culture, both peroxisomes and endosomes dissociated from lipid bodies more slowly than did Golgi, although only 20% of peroxisomes remained bound after 4 min. In contrast, all observed peroxisomes in cells from oleate culture remained bound during the experiment, whereas 74% of endosomes and 82% of Golgi dissociated (see Videos 1–3 of representative cells from the three tagged strains, available at ). To visualize the physical association of peroxisomes and lipid bodies at higher resolution, cells were imaged by transmission electron microscopy. The large fraction of cytoplasmic volume taken up by lipid bodies is obvious in these cells (). Lipid bodies had extensive contacts with each other. In , all nine lipid bodies in the section are connected in a linear array. Less often, branches were observed; the lipid body in the center of contacts three other lipid bodies. The individual units were connected through novel structures that resemble nipples or valves (, arrows). We frequently observed whisps or ribbons of electron-dense material within lipid bodies either close to the periphery or more internally (). The composition of this material is unknown, although they may represent structures induced by free fatty acids (see Fatty acids accumulate…in ). As expected from the fluorescently tagged cells, contacts between peroxisomes and lipid bodies were easily detected (). The contact sites usually were accompanied by an increase in electron density on the lipid body surface (; arrows), suggesting an interface more complex than the three phospholipid leaflets of the apposing organelles. Occasionally, lipid bodies were detected with extensions of peroxisomes that we term pexopodia, which penetrate into the core of the organelles. shows a lipid body that is both enwrapped by a peroxisomal tail (bottom rectangle, enlarged in ) and is also penetrated by a separate peroxisome (top square, enlarged in ). The tail contains a midline structure that is similar to peroxisomal extensions that extend from the ER in mouse dendritic cells (). The pexopodium extends beyond the periphery of the organelle, appearing to make contact with the lipid core. In other lipid bodies, we see inclusions that may be the result of pexopodial invasion (an example is shown in ). Pexopodia occurred in 3.2% of cell sections in which both lipid bodies and peroxisomes were visible. We saw a general correlation between the number of pexopodia and the nutritional state of the cells. For example, when cells growing in oleate were starved of carbon source for 8 h, the occurrence of pexopodia rose to 9.1%. An intermediate response was seen when cells growing on oleate were switched to low (0.1%) raffinose: 7.2% of cells had pexopodia. These results suggest that pexopodial formation was stimulated when cells were forced to use fatty acids from lipid bodies. As reported elsewhere for other systems (; ), we also observed frequent contacts of lipid bodies with ER and mitochondria (unpublished data). To examine with more certainty the physical interactions of peroxisomes and lipid bodies, we used antibodies to Pot1p and Pox1p (acyl-CoA oxidase), two abundant core enzymes of peroxisomal β oxidation, in an immunogold analysis (). Both antibodies stained peroxisomes, although antithiolase typically stained throughout the peroxisomal matrix (), whereas antioxidase more often stained closer to the membrane (). Thiolase often appeared extensively around the periphery of lipid bodies, representing peroxisomes bound to these organelles. In contrast, acyl-CoA oxidase staining also was seen in the lipid body inclusions (). Our images indicate that peroxisomes frequently associate with lipid bodies in , suggesting a close metabolic relationship. The contacts between lipid bodies and peroxisomes was more than mere apposition; peroxisomes extended processes into the cores of lipid bodies. The immunogold evidence raises the possibility that the contents of pexopodia and lipid body inclusions may be somewhat different to that of intact peroxisomes and are perhaps specialized structures for the rapid and efficient mobilization/oxidation of stored lipid. Biochemical evidence provided further support for lipid body–peroxisome associations. For these experiments, we tagged lipid bodies by introducing the myc epitope into the chromosomal copy of . The binding of peroxisomes to lipid bodies was confirmed by gently floating lipid bodies from a postnuclear supernatant (PNS). The Erg6p-myc marker was clearly enriched in this fraction compared with the lipid body–depleted cytoplasm underneath (). Similarly, three peroxisomal markers, Pox1p, Pot1p, and Pex11p (an abundant peroxisomal membrane protein), were also enriched in this fraction. In contrast, the cytoplasmic marker Zwf1p was not enriched. (The lipid body fraction was not purified further to remove contaminating cytoplasm for fear of disrupting this association). Next, lipid bodies were highly purified by centrifuging the organelles from a PNS through a buffer cushion and washing them several times, which is similar to the procedure previously described for mammalian lipid bodies (). The quality of the preparation was assessed by staining with oil red O. Virtually all particles stained with this dye, suggesting a high level of lipid body purity and integrity (). These organelles, along with the cytosol and organellar pellet (containing peroxisomes), were analyzed by immunoblotting (). Zwf1p was almost exclusively localized to the cytosol fraction, whereas Pex11p was found exclusively in the organellar pellet. Pox1p and Pot1p were also localized to the pellet, although in this experiment, there was leakage of Pot1p into the supernatant. However, a small fraction of Pox1p (10% or more of total Pox1p in some experiments) clearly migrated with the purified lipid bodies in all experiments, which is consistent with our immunogold results. We next compared the localization of Pox1p with that of two peroxisomal proteins that function outside the β-oxidation pathway: Mls1p and Mdh3p (both myc tagged). Neither enzyme associated with lipid bodies (). We considered the possibility that Pox1p traffics to lipid bodies independent of its targeting to peroxisomes. The localization of Pox1p to peroxisomes depends on the shuttling receptor Pex5p. We found that Pox1p associated with neither peroxisomes nor lipid bodies but remained in the cytosol in a strain (unpublished data), indicating that Pox1p trafficking to lipid bodies is Pex5p dependent. However, because Pox1p binds to a location on Pex5p far removed from the normal PTS1 (peroxisomal targeting signal 1)-binding site (), we considered the possibility that Pex5p might shuttle Pox1p to lipid bodies in some unusual way that bypassed peroxisomes. To address this possibility, we forced Pox1p to traffic via the PTS1 pathway by attaching a PTS1 to Pox1p and eliminating the normal site of interaction on Pex5p for Pox1p binding (see Materials and methods; ). With these modifications, Pox1p still localized to peroxisomes and lipid bodies as it did in control cells (unpublished data). We conclude that the binding of Pox1p to lipid bodies occurs subsequent to its targeting to peroxisomes. However, the binding of Pot1p, Pex11p, Mls1p, or Mdh3p to these organelles is somehow excluded or is at least much weaker so that they are removed upon fractionation. To identify other peroxisomal proteins that were tightly associated with lipid bodies and to detect other tightly associated organelles, highly purified lipid bodies from oleate cultures were subjected to proteolysis and mass spectrometric analysis of peptides. The identified proteins corresponding to the peptides are listed in . All but three of them fall into the following three categories: previously identified proteins of lipid bodies (); proteins of mitochondria, peroxisomes, and/or ER (organelles that we and others have seen associating with lipid bodies; ; ; unpublished data); and cytosolic proteins (assignments of localization based on ). We did not consider the three remaining identifications, Pma2p (the plasma membrane hydrogen-exporting ATPase), Hnf1/2p, and Htb1/2p (histones), of likely physiological significance. Among cytoskeletal elements, only actin was identified. No proteins of other cellular compartments (including vacuoles or other endocytic or exocytic organelles) were found, suggesting that adventitious binding of cellular elements during organelle isolation was minimal. We assume that the ER and mitochondrial proteins identified in the lipid body isolation were derived from tightly attached organelles, as is the case for mammalian lipid bodies (). A region of the ER is known to tightly associate with mitochondria (). Nearly all of the identified mitochondrial proteins were derived from membranes or nucleoids (25 of 27), suggesting leakage of soluble proteins during fractionation. In contrast, all of the eight peroxisomal proteins in the fraction were matrix components. Furthermore, all eight proteins function in the β-oxidation pathway of oleic acid (). Besides the core enzymes of β oxidation (including Pox1p), two proteins involved in the rearrangement of the double bond in oleic acid (Sps19p and Idp3p) and one protein that detoxifies HO (Cta1p) were identified. In fact, the entire β-oxidation pathway was represented in the lipid body proteome except Dci1p and Eci1p. Interestingly, although Pot1p was not detected (consistent with our biochemical and immunogold data), the other peroxisomal ketoacyl-CoA thiolase, Tes1p, was identified even though it is much less abundant in isolated peroxisomes than Pot1p (unpublished data). Other abundant matrix proteins, such as the glyoxylate cycle enzymes Mls1p and Mdh3p, were not identified, which is consistent with our fractionation results, nor were any peroxisomal membrane proteins, including Pex11p, the most abundant of them. These data suggest that pexopodia and lipid body inclusions are enriched in the enzymes of fatty acid oxidation (with Tes1p catalyzing the thiolase reaction) and not those proteins with other peroxisomal functions. As a control, we also performed identical analyses on lipid bodies purified from a glucose culture. Only one peroxisomal protein was detected in lipid bodies from this source, Pex30p, and it ranked 74th by logE score (only one Pex30p peptide was confirmed; ), suggesting a very low abundance in lipid bodies. This result is consistent with the more transient binding of peroxisomes to lipid bodies in glucose cultures, which is shown in . To confirm that the peroxisomal identifications were caused primarily by the binding of organelles rather than random cytosolic adsorption, we also repeated the analysis in the strain, which lacks the PTS1 transporter. Three peroxisomal proteins were identified (Pox1p, Fox2p, and Faa2p; all contain PTS1), but all were less abundant than in the wild-type strain based on ranking by logE scores. Thus, Pox1p dropped from second to the 42nd position (immunoblots of total cell lysates showed similar levels of Pox1p in the two strains; unpublished data), Fox2p dropped from fourth to 12th, and Faa2p dropped from 19th to 46th. The other five peroxisomal proteins identified in wild-type lipid bodies were not detectable at all in lipid bodies from the sample even though more proteins were identified in lipid bodies in this strain (176 compared with 105 for wild type). We conclude that the localization of peroxisomal proteins with lipid bodies shown in principally represents the binding of peroxisomal particles rather than adventitious binding. The presence of mitochondrial, ER, and peroxisomal proteins (but virtually no proteins of other organelles) confirms the morphological observations that we and others have reported (; ; unpublished data) and suggests that these organelles can be tightly coupled for metabolic or ontogenic reasons. The simplest interpretation is that these organelles dock on lipid bodies to either deliver lipids to this compartment or withdraw lipid metabolites from it. Although lipid bodies usually have a simple morphology (), we sometimes observed electron-dense material within the core (), and a few tubular elements at the periphery were rarely seen. Electron-dense inclusions were much more abundant in the strain using our laboratory strain MMY011α (McCammon et al., 1990) as background. Lipid bodies in 10–20% of cell sections from this strain contained a series of elongated, curled, and often tangled electron-dense tubules (). The diameter of the tubules varied but was usually between 10 and 30 nm. The tubules radiated from the periphery, extending outward from the lipid body () or inward (, central lipid body). Lipid bodies that were connected to each other were often filled by similar structures (). We termed these lipid body inclusions gnarls. The lumen of the tubules often emptied into the core of the lipid body. There was usually an electron-dense ground substance associated with the tubules (). In certain sections, the dense material resolved into an ordered pattern of electron-dense rows 7 nm apart (, arrow). We attempted to isolate gnarled lipid bodies from using density gradient centrifugation to identify specific proteins associated with the structure, but the profile of lipid bodies (identified with Erg6p-myc) in gradients was indistinguishable from wild-type organelles (unpublished data). There were no abundant proteins found on SDS gels from lipid body fractions that were specific to the strain, and the aforementioned proteomics data failed to show the presence of any peroxisomal membrane proteins, ruling out the possibility that the gnarls represented peroxisomal membranes. We also tried to visualize gnarls using different stains but found that potassium permanganate, which stains polar lipids (), was essential to visualize the gnarls. We concluded that the gnarls were probably not proteinaceous but might instead represent arrays of lipid-derived material such as free fatty acids. We reasoned that gnarls may accumulate in the strain because an important Pex5p substrate failed to be imported into peroxisomes and was therefore inactive. Both Pox1p and Fox2p (encoding the two initial enzymes of the core β-oxidation pathway) are imported via Pex5p. We found that gnarls accumulated in both Δ and Δ strains as well as in a strain with a point mutation (E488Q) in Pox1p that should render it catalytically dead (; ). Gnarls were present but less developed in Δ (probably because Tes1p can substitute as a thiolase). Thus, several single peroxisomal enzyme deficiencies were sufficient to promote gnarl formation. Because peroxisomes and lipid bodies can associate extensively, we determined whether the presence of peroxisomal membranes was necessary to promote gnarls. Peroxisomes do not form in a strain because is essential for formation of the peroxisomal membrane (). Indeed, gnarls were rarely observed if was disrupted (unpublished data). We next decided to use as the genetic background because gnarls are abundant in it and disrupted in it. As a control, we also disrupted , which controls peroxisomal fission (; , ), in the background. We scored the resulting strains growing in oleic acid medium for gnarls, clear lipid bodies, or lipid bodies that contained up to a few tubules at the periphery, which is probably a precursor to gnarls (). Gnarls and tubules were present in wild-type cells but were quite rare (1.9% of cells had them, and another 3.8% had lipid bodies with a few tubules or less). In contrast, 50% of cells had either gnarls or tubules (17.3% gnarls and 32.7% tubules). Introducing Δ in the background diminished the frequency of gnarls to 7.4%, although the number of tubules was slightly higher (35.2%). A much more dramatic effect was seen upon introducing in the background. Gnarls were completely suppressed in this strain, and the frequency of tubules was greatly diminished (to 9.6%). We conclude that gnarls form as a result of peroxisomes that are unable to metabolize fatty acids. However, gnarls require peroxisomal membranes (present in ), suggesting that contact with peroxisomes stimulates gnarl formation. The relatively minor effect of Δ may be explained by the presence of fewer peroxisomes to stimulate gnarl formation (; ). A simple explanation for the nature of gnarls is that they are free fatty acids, the synthesis of which is stimulated by contact with peroxisomal membranes. If they are not metabolized by peroxisomes, they accumulate and form structures that are visualized with permanganate. Experiments are underway to identify the lipase that may be stimulated by contact with peroxisomes; there are several in lipid bodies, including the newly identified Tgl4p and Tgl5p (; ). To determine whether an increase in free fatty acids within lipid bodies correlates with the development of gnarls, we determined their levels in various yeast strains, normalizing the data to the amount of lipid body triacylglycerols () or to lipid body protein (). Similar to our detection of gnarls, levels of free fatty acids were highest in and in disruptions of the core β-oxidation genes. They were the lowest in wild type and (absence of peroxisomes) and also low in Δ, where the peroxisomal fatty acid oxidation pathway is intact. The relatively lower level of fatty acids in ( encodes the PTS2 receptor and is required for Pot1p import) may be caused by the sharing of thiolase activity by Pot1p and Tes1p. Thus, these biochemical results are consistent with our hypothesis that gnarls result from the accumulation of fatty acids within lipid bodies, as illustrated in (top and left). We predict that fatty acids generated by the interaction of lipid bodies and peroxisomes can directly enter peroxisomes, perhaps by diffusing into the pexopodia, where they can be quickly oxidized without passing through the cytosolic compartment. In the absence of peroxisomal catabolism, they form gnarls. In this study, we report novel interactions between peroxisomes and lipid bodies. These two organelles physically associate to a degree not previously seen outside of germinating oil seeds, where extensive contacts between glyoxysomes and lipid bodies presumably allow the efficient conversion of fatty acids to TCA cycle substrates (). We provide evidence that the association of peroxisomes and lipid bodies results in the synthesis of free fatty acids from neutral lipids, providing substrate for peroxisomal oxidation. Transfer of fatty acids from the lipid body to the peroxisome may occur across the extensive lipid body–peroxisome interface, including the pexopodia. Our work shows that peroxisomes are not solitary organelles; many associate with lipid bodies even when cells are cultured on glucose, and peroxisomes are not required for growth. It is known that peroxisomes can acquire their membrane lipids from lipid bodies (). As our images indicate, contact may involve a large surface area. What attracts these organelles to each other? It is reasonable to think that surface proteins interact to provide the apparently high affinity. In glucose culture, peroxisomes dock on lipid bodies for a short time and then disengage; in oleate culture, interactions are more permanent. The molecular basis for this difference probably resides in the nature of the interacting surface proteins or intervening factors. We also report the formation of pexopodia that extend into lipid bodies. (box) illustrates our hypothesis that this occurs as the result of a hemifusion of the single leaflet of the lipid body membrane and the outer leaflet of the peroxisomal membrane. A stalk forms according to current fusion theory (; ), and its enlargement leads to direct contact of the inner peroxisomal leaflet with the core of the lipid body. This is energetically favorable because the hydrophobic tails of the leaflet would interact with the neutral lipids of the core. Moreover, fatty acids from the lipid body should be able to easily diffuse across the monolayer to be available for peroxisomal oxidation. What is the nature of these peroxisomal insertions? Evidence that they are physiologically important is provided by the list of peroxisomal proteins that are associated with purified lipid bodies (). The group comprises nearly the entire β-oxidation pathway and nothing else even though there are several other peroxisomal proteins that are at least as abundant as some of the proteins identified (unpublished data). Therefore, it is tempting to speculate that these inclusions consist of sacs of β-oxidation enzymes that somehow are segregated from the other peroxisomal enzymes. Whether the pexopodia are physiologically important or are simply fragments of peroxisomes that tightly adhere to isolated lipid bodies, it is clear that the β-oxidation enzymes have a high affinity for lipid bodies compared with other peroxisomal proteins. The lack of peroxisomal membrane proteins in the preparation may reflect active exclusion from the lipid body–peroxisome interface. Complementary proteomic results were recently obtained from purified peroxisomes, where the lipid body proteins Faa1p, Erg1p, and Erg6p were identified (). It is unclear whether this reflects residual lipid body monolayer that is tightly associated with purified peroxisomes. These investigators also detected a significant amount of Rho1p, as did we with lipid bodies. We do not yet know if Rho1p, which was shown in other work to be important for peroxisomal biogenesis (), is appearing in our analysis as a result of peroxisomal (or other endomembrane) attachment or if has a separate function for lipid body dynamics. Our entry into this project was the visualization of gnarls in . We originally hypothesized that gnarls were the result of aberrant peroxisomal membrane assembly in association with lipid bodies, but we looked in vain for peroxisomal proteins associating with these structures. We then considered that they were lipid structures. This is consistent with the increase in free fatty acids that we see in these aberrant lipid bodies. The gnarl tubules could represent inverted fatty acid structures where the polar head groups have self-assembled (perhaps through cation bridges provided by potassium permanganate) within the lipid core. We also see gnarled lipid bodies associating with vacuoles that may represent lipid hydrolysis during autophagy (unpublished data). Although these structures have not been previously reported in yeast cells, they are somewhat similar to the whorls of lipid lamina that penetrate cholesteryl ester droplets in mouse macrophages (). The gnarls in our study were not seen without the use of permanganate stain, and their delicate nature may depend on early prefixation steps described by the method () and the extended infiltration time we used. There may also be differences in their frequency in various background strains. In fact, we observed them more frequently in our MMYO11α background than in BY4741 or BY4742, two other frequently used background strains. However, gnarled lipid bodies occur as usual in an Δ strain (in which autophagy does not occur) that is also disrupted in . Thus, autophagy cannot account for gnarls in the peroxisomal mutants. Gnarls are scarce in wild-type cells. In this sense, they are unphysiological. They represent a failure of peroxisomes to metabolize fatty acids, the synthesis of which was stimulated by their physical interaction with lipid bodies. We also observed several interesting elements of lipid body organization. These organelles are often clustered with connections between them, across which core and peripheral material may flow. This may explain the frequent appearance of gnarls in adjacent organelles. The valvelike structures we see between them may represent lipid body communication either as a passive filtration system or a more active and regulated barrier. Perhaps they are related to the stringy material observed when lipid bodies are isolated from mammalian cells (). Regardless, it is evident that lipid bodies are not independent of each other but form a larger structure—an adiposomal reticulum. Although lipid bodies are often depicted as isolated intracellular fat globules, it is clear that they are at the center of metabolism in many ways. Our work focused on their interactions with peroxisomes, but the proteomics data suggest that they also have essential interactions with mitochondria and ER. Lipid intermediates certainly can flow through the cytoplasm (in vesicles or bound to transfer proteins), but physical interactions may be essential for the efficient transfer of some metabolites, as suggested for other organellar interactions (). The lipid body surface may be considered as a source for diffusible substrates and also as a dock receiving the heavy traffic of cargo going in both directions and on which organelles can affect the availability of cargo from the lipid body core. Organelles such as peroxisomes are likely to collaborate with lipid bodies in a controlled manner (for example, in regulating fatty acid production), and the rules governing this regulation will be interesting to work out. Restriction enzymes, DNA ligases, and other reagents for DNA manipulation were purchased from New England Biolabs, Inc., Roche Diagnostics, and Bio-Rad Laboratories. All other chemical reagents were obtained from Sigma-Aldrich or Fisher Scientific. Antisera against Pox1p was generated in rabbits from SDS gel–purified protein from isolated peroxisomes; antisera against Pot1p was a gift of R. Rachubinski (University of Alberta, Edmonton, Canada). The expression and cloning plasmids consisted of pRS313-316 () or pBluescript KS(−) (equivalent to pBluescript II KS(−) in the current catalog except the bases ATT immediately upstream from the T7 promoter has been replaced by GCGCGC in the II series; Stratagene). The promoter and/or terminator used in many plasmid constructions was cloned from pRS315-PGK (). Standard recombinant DNA techniques were used (). Changes to chromosomal genes were performed using PCR products containing ∼50 bases of chromosomal sequence at each end for homologous recombination. DNA manipulations resulting in changes to the coding sequence as well as changes to the genome were verified by sequencing performed by the McDermott Center for Human Growth and Development on campus (University of Texas Southwestern Medical School, Dallas, TX). Yeast strains were transformed by the lithium acetate method (). The bacterial strain DH10β (Invitrogen) was used for all bacterial transformations. For yeast experiments, ATCC 201388 (American Type Culture Collection) was used as the background strain for the organelle dynamics experiment shown in (G and N), and BY4741 (Open Biosystems) was used for the thin-section EMs shown in . For all other yeast experiments, MMYO11α, which was developed for growth in oleic acid (McCammon et al., 1990), or derived mutants or transformants (see below) were used. The disruption strain in MMYO11α was a gift of D. Klionsky (University of Michigan, Ann Arbor, MI). Cells were cultured either in 2% glucose or in oleic acid as previously described (). xref #text The PGK promoter and terminator were cloned from pRS315-PGK such that the 5′ and 3′ ends of the promoter contained new SacII and XbaI sites, respectively, and 5′ and 3′ ends of the terminator contained HindIII and XhoI sites, respectively. These fragments were then inserted into pRS315 at these sites, generating pRS315-PGKII. The open reading frame (from genomic DNA) and the sequence for monomeric DsRed (mDsRed, from pDsRed-monomer-N1; a gift from B. Glick, University of Chicago, Chicago, IL) were used in an overlap extension PCR using primers that inserted the sequence for a two-glycine spacer in between the protein elements and conferred BamHI and HindIII ends to the ERG6-mDsRed fusion sequence. This fragment was then inserted between the promoter and terminator of pRS315-PGKII. The resulting vector is termed pERGmDsRed. The reading frame and 652 bp of upstream sequence was subcloned into pRS316 at the SacII site. The terminator was subcloned downstream at the Xho1 and Kpn1 sites, generating Pox1-316, which was used for mutagenesis (see the following paragraph) and to transform yeast for the production of Pox1p at normal levels as a control. The E488Q mutation and the GGAKL addition were performed using a Pfu polymerase-based method. The E488Q mutation should yield a catalytically dead acyl-CoA oxidase by comparison with other acyl-CoA oxidases that have been crystallized and characterized more fully (; ), and the addition of GGAKL to the carboxy terminus of Pox1p adds a peroxisomal targeting signal-1 upstream of the stop codon. Primers and their complements containing the sequences for both the E488Q mutation and the GGAKL addition were used in a PCR reaction on Pox1-316 plasmid using Pfu polymerase. The extension time was determined by the size of the Pox1-316 plasmid (>2 min/kb). The PCR reactions were then treated by direct addition of the restriction enzyme DpnI at 37°C for 3 h to cleave the template DNA (the Pox1-316 plasmid). DpnI only cleaves its recognition site, GATC, when it is methylated, and this site is methylated in the template DNA because it was produced in DH10β, a strain that is dam+. After treatment with DpnI, the PCR reaction was used to transform the bacteria. Plasmid DNA was harvested, the combination of mutation and addition was screened by PCR, and the correct plasmid was used for yeast transformations. For the construction of myc-tagged Pox1p, a plasmid copy of the promoter (652 bp of upstream sequence) and open reading frame was constructed with an XbaI site between them and was subcloned into the Pox1-316 plasmid at the SacII and XbaI restriction sites (which removed the sequences originally present in Pox1-316) to create the plasmid Pox1-2. The open reading frame was then amplified and subcloned into Pox1-2 at the XbaI and Xho1 restriction sites to create Pox1-3. Double-stranded DNA encoding the myc epitope were then subcloned into Pox1-3 using XbaI to create plasmid myc–Pox1-316, encoding Pox1p with two copies of myc (MEQKLISEEDLEQKLISEEDLSR) at its amino terminus. The open reading frame and 608 bp of downstream sequence (to serve as a terminator) was subcloned into pRS316 at the EcoR1 and XbaI restriction sites to create plasmid Mdh3-1. The promoter region (582 bp upstream of the ORF) was then subcloned into Mdh3-1 using the Kpn1 and Xho1 restriction sites to create plasmid Mdh3-2. The myc tag was prepared similarly as in the preceding paragraph and subcloned into Mdh3-2 using Xho1 and EcoR1 to create plasmid myc–Mdh3-316 such that the myc ×2 sequence MEQKLISEEDLEQKLISEEDLEF was inserted in frame before the open reading frame. A similar strategy was used to generate myc-Mls1p. The ORF and 318 bp of downstream sequence were subcloned into the pRS316 plasmid at the EcoR1 and SacII restriction sites to create plasmid Mls1-1. The promoter region (793 bp upstream of the open reading frame) was amplified and subcloned into Mls1-1 using the Kpn1 and Xho1 restriction sites to create plasmid Mls1-2. The myc tag was added exactly as described for the myc–Mdh3-316 plasmid. For the acquisition of static images, the MMY011α strain was transformed either with pPOT1GFP alone if they were to be stained with oil red O or with pPOT1GFP along with pERG6DsRed or pERG6mDsRed. For cells expressing both GFP- and DsRed/mDsRed fusion proteins, cells in log phase were removed from growth medium (either SD or oleate medium), concentrated by centrifugation (3,000 for 5 min at room temperature), and suspended on microscope slides in growth medium containing 1% agar (). Images were acquired in Slidebook (version 4.1.0.3; Intelligent Imaging Innovations) using a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a 100× 1.3 NA oil immersion objective lens (plan-Neofluar) equipped with a digital camera (Sensicam; Cooke). GFP images were acquired using the fluorescein isothiocyanate filter set, and DsRed/mDsRed images were acquired with the CY3 filter set (Chroma Technology Corp.). For (A, B, G, and H), z-axis stacks were obtained at 0.1-μm spacing, and the Slidebook commands of deconvolution (nearest neighbor option) and projection were used to build images through the cell. Further processing of images used Slidebook and Photoshop software. For cultures transformed only with the GFP-containing plasmid, lipid bodies were visualized with oil red O. To prepare the stain, a few milliliters of a stock of 1 g oil red O (Sigma-Aldrich) in 100 ml isopropanol was diluted with water (3:2 vol/vol) and filtered through 0.45- and 0.22-μm syringe filters connected in series. Yeast were pelleted by centrifugation (3,000 for 5 min at room temperature), washed with water, and resuspended in the freshly filtered oil red O solution (1 A unit of yeast/200 μl of filtered oil red O solution) in a microfuge tube. The yeast were vortexed briefly and incubated in the dark at room temperature for 10 min. The stained yeast were washed twice with 1 ml of water, resuspended in 25 μL of water, and mounted on slides for microscopy as in the previous paragraph. For z stacks of oil red O–stained cells, 0.5-μm spacing between images was used. A detailed protocol for processing yeast cells for transmission EM () was followed closely except that the infiltration procedure was extended. After washing the samples in 100% ethanol as described previously (), samples were subjected to 1% Spurr (in ethanol) overnight, 2% Spurr for 8 h, 10% overnight, 33% for 8 h, 66% overnight, and 100% for 36 h with three to four changes. After embedding, samples of 70–90-nm thickness were placed on 200 mesh copper Formvar grids and poststained. The thin sections were observed in a transmission electron microscope (1200 EX; JEOL) at 80 kV using a CCD camera (C4742-95; Hamamatsu) and AMT410 software (Advanced Microscopy Techniques) for image capture. For immunogold analysis, cells cultured on oleic acid were high pressure frozen, freeze substituted, and embedded in LR White (London Resin Company; ). Thin sections (70–90 nm) were cut and placed on Formvar nickel grids. The grids were placed in an automated EM immunogold labeler (Leica) such that samples were exposed to primary antibody for 4.5 h and protein A–gold (GE Healthcare) for 90 min, both at room temperature. Grids were contrast stained with 2% aqueous uranyl acetate. Yeast cells were converted to spheroplasts, osmotically lysed, and a PNS was prepared as previously described (). To determine peroxisomal association with crude lipid bodies, the PNS was centrifuged at 2,500 for 45 min at 4°C in a SW60 rotor (Beckman Coulter). Approximately 100–250 μl containing the lipid bodies was removed from the top of the gradient. An identical volume of infranatant below the lipid bodies was also removed. Cytosol fractions were TCA precipitated, and all samples were subjected to SDS gel electrophoresis and immunoblotting. For experiments involving purified lipid bodies, the organelles were purified from PNS as described previously (). In brief, PNS prepared as above (in the previous paragraph) was overlaid with 3–5 ml Hepes buffer and centrifuged at 198K for 50 min in a SW41 rotor (Beckman Coulter). Lipid bodies were removed from the top of the tubes and washed several times in Hepes buffer. They were visualized by embedding in collagen and staining with oil red O (). To compare organelle markers among fractions, the pellet and the cytosol from the SW41 spin were harvested. Equal amounts (typically 2%) of purified lipid bodies, cytosol (concentrated by TCA precipitation and dissolved in SDS sample buffer), and pellet fractions were analyzed by PAGE and immunoblotting. Dried protein extracts from yeast lipid droplets were dissolved in 50 μl of 50 mM NHCO buffer. Porcine modified trypsin (Promega) was added to the solution at the ratio of 1:50 (wt/wt) followed by overnight incubation at 37°C. The solution was then dried in SpeedVac (ThermoFinnigan), and tryptic peptides were desalted by μ-C18 Ziptip (Millipore) according the manufacturer's instructions before HPLC/mass spectrometry analysis for protein identification. Data from tandem mass spectra were searched against a nonredundant yeast protein sequence database from the National Center for Biotechnology Information by an in-house X! Tandem database search engine (The Global Proteome Machine Organization, ; ). Peptide mass spectra identified with a log (expectation value) less than or equal to −2.0 were manually evaluated (). Frozen lipid bodies purified from 500 or 1,000 ml of cultures were thawed, adjusted with HO to 500 μl, and extracted by a modified Bligh and Dyer method () using hot isopropanol (70°C) instead of methanol (). Lipids were extracted into chloroform and washed with 1 M KCl. The purified lipids were then dried under N and dissolved in a small volume, typically 50 μl of chloroform. Samples were separated by thin layer chromatography on 20-cm K6 silica gel plates (Whatman) in 80:20:1 hexane/diethyl ether/acetic acid (vol/vol/vol). Lipids were visualized after staining with I and quantified after scanning the plate with ImageJ software (National Institutes of Health [NIH]) using a standard curve and quantitative reference standards (Nu-check Prep, Inc.). In the experiment shown in , the plate was charred to visualize lipids. Samples from at least three independent cultures of each strain were analyzed. Protein concentration of lipid body samples was determined by amido black staining (). Strains expressing Erg6p-mDsRed and GFP-tagged Pex3p, Sec7p, or Snf7p were monitored over a 4-min period by spinning disk confocal microscopy at room temperature. Online supplemental material is available at .

Microtubules are hollow nanoscale biopolymers () that, together with actomyosin filaments and intermediate filaments, form the composite cytoskeleton that controls cell shape and mechanics (). Microtubules are the most rigid of the cytoskeletal biopolymers, with a bending rigidity ∼100 times that of actin filaments and a persistence length on the order of millimeters (). Cytoplasmic microtubules are particularly critical for stabilization of long cell extensions, such as nerve cell processes (), and their disruption results in process retraction and production of regular polygonal cell forms (). Microtubules also orient vertically in cells when they become columnar, as observed during neurulation in the embryo (), and they can physically interfere with cardiac myocyte contractility in certain heart conditions (, ; ). Despite their long persistence length, microtubules are also often highly curved in cultured cells, suggesting that they experience large forces within the cytoplasm. Indeed, both slow retrograde flow of the actin cytoskeleton at the cell periphery and stimulation of cell contraction appear to cause compressive buckling of microtubules (; ; ), whereas disruption of microtubules results in increased transfer of cytoskeletal contractile stress to extracellular matrix adhesions (). These observations suggest that microtubules can bear compressive loads, which is consistent with models for cellular mechanics in which microtubule compression helps stabilize cell shape by balancing tensional forces within a prestressed cytoskeleton (; , ; ; ). However, other biophysical studies suggest that individual microtubules cannot bear the large-scale compressive forces generated by the surrounding cytoskeleton in a whole living cell. Moreover, under compressive loading, isolated microtubules exhibit a classic Euler buckling instability, resulting in the formation of a single long-wavelength arc that is completely unlike the highly curved appearance of microtubules observed in living cells (; ). These results suggest that microtubules may not support compressive loads because they should buckle at much larger wavelengths and should be unstable to very small forces (; ). The structural role of microtubules in whole living cells thus remains highly controversial. Despite the importance of understanding the forces associated with the microtubule cytoskeleton for control of cell shape and mechanics, there have been very few studies that quantitatively measure these forces, and the precise physical basis of the microtubule bending seen throughout the cytoplasm remains unknown (; ). In this paper, we address the question of whether or not microtubules bear large-scale compressive forces in living cells. Our results reveal that individual microtubules can and do bear levels of compressive force that are one hundred times greater in whole cells than in vitro. This is possible because of lateral mechanical reinforcement by the surrounding elastic cytoskeleton. To illustrate this principle, we present a macroscale model composed of a plastic rod embedded in an elastic gel, which mimics the short-wavelength curvature observed in microtubules in living cells. We show that a reinforced buckling theory accounting for the surrounding elastic network can quantitatively predict the wavelengths of buckling induced by compression at both the macro- and microscales. This simple reinforcement principle, which appears to be widely used by nature to enhance the structural stability of cells, provides an explanation for how microtubules can bear large compressive forces within the cytoskeleton of living cells. We first explored whether microtubules bear large-scale compressive loads in living cells by addressing the question of why cytoplasmic microtubules exhibit highly curved forms, whereas isolated microtubules undergo single long-wavelength Euler buckling. The Euler buckling instability observed in vitro with isolated microtubules occurs at a critical compression force given by(see supplemental discussion, available at ), where κ is the bending rigidity and is the length of the microtubule (; ). Within a cell, microtubules are typically even longer than those studied in vitro, suggesting that they should buckle easily under small loads of order 1 pN; forces larger than this can be generated by even a single kinesin or myosin motor protein (). To investigate the response of microtubules to endogenous polymerization forces, Cos7 epithelial cells and bovine capillary endothelial cells were either transfected with tubulin labeled with EGFP or microinjected with rhodaminated tubulin and then analyzed using real-time fluorescence microscopy. The dynamic ends of growing microtubules that polymerized toward the edge of the cell consistently buckled when they hit the cell cortex ( and Videos 1 and 2, available at ), as observed in a previous study (). Given the well defined (end-on) loading conditions that were visualized with this real-time imaging technique, these results strongly suggest that these particular microtubules are compressively loaded. These microtubules did not, however, exhibit the expected long-wavelength Euler buckling, but, instead, consistently formed multiple short-wavelength (λ ≈ 3 μm) arcs near the site of contact ( and Videos 1 and 2), which were similar to the microtubule shapes that were previously observed in various different cell types (; ; ; ). Short-wavelength buckling has not been reported in studies with isolated microtubules under compression (). Therefore, we directly tested whether compressive forces are indeed the cause by imposing an exogenous compressive load on intracellular microtubules in living cells. To accomplish this, a glass microneedle controlled by a micromanipulator was used to compress the cell membrane and underlying microtubules at the cell periphery, while simultaneously analyzing their structural response (). When initially straight microtubules that were aligned along the main axis of force application were compressed in this way, their proximal regions buckled with a short wavelength (2.8 ± 0.5 μm; mean ± SD) that was nearly identical to that naturally exhibited by the ends of growing microtubules (3.1 ± 0.6 μm; and Videos 3 and 4, available at ). Therefore, we conclude that the short-wavelength buckling of microtubules is indeed a mechanical response to compressive loading caused by axial forces. Short-wavelength buckling forms are also observed in microtubules located deep within the cytoplasm of these same cells (). Hypercontraction of the cytoskeleton in nonmuscle cells can induce buckling of cytoplasmic microtubules, and normally curved microtubules straighten when tension is released (). Myosin-based force generation also appears to drive microtubule compression under retrograde flow of the lamellar actin network (), and overproduction of microtubules can impair cardiac cell contraction in dilated heart muscle (). These observations suggest that the contractile actin cytoskeleton may also compressively load and buckle cytoplasmic microtubules (). To test this possibility under physiological conditions, we analyzed microtubule behavior in beating cardiac myocytes where actomyosin-based contraction is temporally periodic and well defined. Microtubules in these cells cyclically buckled and unbuckled with each wave of contraction (systole) and relaxation (diastole) of the cardiac muscle cell ( and Videos 5 and 6, available at ). The wavelength was again short; λ ≈ 2.9 ± 0.6 μm (). Thus, in addition to polymerization forces and external loads, internally generated actomyosin contractile forces can also compressively load microtubules under physiological conditions. The microtubules buckle at identical short wavelengths in each case, suggesting that the same physics governs their behavior. A clue to the origin of this high-curvature buckling comes from observations of the dense microtubule network in beating cardiac myocytes, where we found that neighboring microtubules often buckled in a coordinated manner, both temporally and spatially in phase (Video 7, available at ). This suggests that they are mechanically coupled to each other because of the intervening elastic cytoplasm. Thus, the short-wavelength buckling of individual microtubules could reflect constraints on microtubule bending that are caused by the need to deform lateral structural reinforcements. This mechanical coupling likely reflects the fact that microtubules in eukaryotic cells are embedded in a surrounding elastic network of cytoskeletal actin microfilaments and intermediate filaments, as demonstrated by high-resolution microscopy (; ). This composite network of cytoskeletal filaments is largely responsible for the elastic response of the cytoplasm (), and may also act as a reinforcing lateral constraint that prevents long-wavelength buckling of microtubules (). To test the biophysical hypothesis that lateral mechanical constraints lead to short-wavelength buckling in microtubules within living cytoplasm, we first performed studies using a macroscopic model system in which a thin plastic rod (∼0.1-mm-diam) was used to mimic a single microtubule. When this rod was compressed in aqueous solution, it yielded the expected long-wavelength Euler buckling mode (). To mimic the effect of a surrounding elastic network, we embedded the rod in gelatin: when the rod was compressed, the long-wavelength mode was suppressed. Instead, shorter-wavelength buckling resulted (), which was strikingly similar to that displayed by microtubules in living cells (– ). This finding suggests that similar lateral constraints may lead to the short-wavelength buckling behavior observed in living cells. This short-wavelength buckling behavior can be described quantitatively using a constrained buckling theory. For a constrained rod to bend, it must push into and deform the surrounding network. Because of the energetic cost of this deformation, short-wavelength buckling will be preferred because the same degree of end-to-end compression is possible with smaller lateral motion (smaller deformation). This can be described with the following equation showing the total energy of deformation, which is a sum of integrals along the length of the rod (; ):where is the transverse displacement of the rod as a function of the axial coordinate , and α is proportional to the elastic modulus (see supplemental discussion). The first term represents the bending energy of the rod, the second term expresses the axial compression energy released by the buckling, Δ, where Δ is the change in axial length of the rod, and the third term is the elastic deformation energy associated with pushing laterally into the surrounding medium. The Euler buckling theory comprises the first two terms; in contrast to the long wavelength Euler buckling result, the additional deformation energy represented by the third term leads to buckling on a short wavelength. The wavelength is set by the ratio of the bending rigidity of the rod and the elastic modulus of the surrounding network in the following equation: This constrained buckling theory provides a quantitative description of our macroscopic observation. Using measured values of the elastic modulus of the gelatin ( = 1.5 kPa), and the bending rigidity of the plastic rod (κ ≈ 10 Nm), the theory predicts a buckling wavelength of λ ≈ 1 cm, which is in excellent agreement with the experimental result (λ ≈ 1.1 cm). Moreover, by using an array of plastic rods with bending rigidities varying by over an order of magnitude, we confirmed that the theoretical predictions closely matched the measured wavelengths in all cases (). The biological relevance of this theory was also examined quantitatively by testing its ability to predict the buckling wavelength of microtubules in living cells. Using the reported microtubule bending rigidity (), κ ≈ 2 × 10 Nm, and the elastic modulus of the surrounding cytoskeletal network (; ), ≈ 1 kPa, the theory predicts that compressively loaded microtubules should buckle with a wavelength of λ ≈ 2 μm; this is in good agreement with the measured wavelength λ ≈ 3 μm (). We note that the effect of intracellular microtubule-associated proteins on the microtubule bending rigidity is unclear (), and the rigidity also appears to depend on the speed of their polymerization (); indeed, measurements of microtubule bending rigidity have varied by an order of magnitude. The elastic modulus of the cytoskeleton is also locally heterogeneous within the same cell, and measurements of this quantity have varied. However, the one-quarter power dependence of the wavelength on both the bending rigidity and the elastic modulus makes the predicted wavelength relatively insensitive to uncertainty in these values. The close agreement between the predicted and observed wavelengths shows that this constrained buckling theory can quantitatively explain both the macroscale model and the microscale buckling of microtubules in living cells (). Despite the fact that these two systems differ by more than four orders of magnitude in spatial scale, with bending rigidity differing by over 16 orders of magnitude, the same physics governs their behavior. A central component of the elastic cytoskeleton of the cell is the actin filament network (; ; ), which surrounds and is connected to intracellular microtubules (; ); therefore, we hypothesize that this actin network plays an important role in microtubule reinforcement. To test this hypothesis, we pretreated Cos7 cells with 2 μM cytochalasin D for 30 min to disrupt the surrounding actin filament network, and used a microneedle to compress microtubules. We found that the wavelength increased to 4.3 ± 1.0 μm in cytochalasin-treated cells, compared with 2.8 ± 0.5 μm in untreated cells (P < 0.0005; paired test; ). The wavelength of naturally buckled microtubules also increased to a similar degree in these cells. This increase, albeit small, supports the hypothesis that the actin network plays a role in reinforcing microtubules. The relatively small change is consistent with the weak dependence of the buckling wavelength on the elasticity of the surrounding cytoskeleton (λ ≈ ). Indeed, the larger buckling wavelength corresponds to a decrease in the elastic modulus of the surrounding cytoskeleton by a factor of ∼5, which is consistent with previous measurements of cytochalasin-treated cells (; ). Thus, the lateral structural reinforcement of microtubules responsible for their enhanced compressive load-bearing capacity appears to be at least partly attributable to the surrounding actin cytoskeleton in living cells. In studies with both muscle and nonmuscle cells, we found that localized regions of single microtubules repeatedly underwent short-wavelength buckling at the same sites when analyzed over many minutes, whereas intervening regions of the same microtubule, and also neighboring microtubules, remained straight ( and ; and Videos 6 and 8, available at ). Nonbuckled microtubules in cells may experience a large, but subcritical compressive force, even when other regions of the same microtubule exhibit short-wavelength buckling, perhaps because of local weak spots in the surrounding elastic network. We observed similar behavior in the macroscopic experiment when a localized region of the surrounding gelatin network was disrupted; the rod preferentially buckled in this same localized region when it was compressed a second time, even though adjacent segments of the same rod remained straight (). These observations suggest that the location of high curvature microtubule buckling may be linked to local variations in the stiffness of the surrounding cytoskeletal network. Interestingly, when microtubules were compressed by forces acting at their tips, the buckling typically did not extend along the entire microtubule; instead, the buckling was localized to a region close to the site of force application (, , and ). We found similar behavior when we performed the macroscopic buckling experiment in tall sample chambers; the amplitude of the short-wavelength buckling decayed with distance from the site of force application (). This localization of the buckling results from longitudinal mechanical coupling (sticking) of the rod to the network; this attenuates the compressive force along the axis of the rod (see supplemental discussion). Thus, there is likely a similar coupling of the microtubule to the surrounding network, causing the attenuation of its buckling. This is consistent with studies showing that microtubules are physically linked to the surrounding cytoskeleton through protein cross-linkers (; ). Thus, the strength and kinetics of these cross-linkers determine the distance over which forces are mechanically transmitted through the cell. The structural organization and mechanical behavior of microtubules are believed to play a central role in the determination of polarized cell shape and directional motility that are critical for tissue development. Microtubules also appear to contribute to some heart diseases by physically interfering with the contraction of hypertrophied cardiac muscle cells (, ). Yet, studies of isolated microtubules suggest that they should not be able to bear more than ∼1 pN of compressive force, and thus, they should not contribute significantly to the mechanical stability of the whole cell. This apparent discrepancy reflects the lack of information about the mechanical behavior of microtubules within the normal physical context of the living cytoplasm. In this study, we directly addressed the question of whether individual microtubules can bear the levels of compressive forces necessary to influence overall cell mechanical behavior by studying and modeling microtubule buckling behavior in the living cytoplasm. Our results show that microtubules exhibit similar buckling responses, with nearly identical short wavelengths and correspondingly high curvature, whether compressed by endogenous polymerization or contractile forces or by direct application of end-on compression using a micropipette. This buckling wavelength could be increased by weakening the reinforcement provided by the cytoskeletal actin network. Moreover, similar buckling behavior can be mimicked using a macroscale model of a plastic rod embedded in an elastic gelatin network. A constrained buckling theory provides a quantitative description of this behavior at all size scales. The finding that microtubules buckle in the living cytoplasm implies that they are under a minimum level of compressive loading because buckling is a threshold phenomenon; it only occurs once the compressive force reaches a critical value. However, lateral reinforcement ensures that a microtubule can remain structurally stable and continue to support a compressive load even after it buckles. Within this picture, we can calculate the critical force usingthis expression is similar to that for Euler buckling, except that the relevant length scale is now λ, which is the shorter wavelength of buckling (see supplemental discussion). This critical force depends linearly on the bending rigidity and, therefore, is sensitive to the large uncertainties in the microtubule bending rigidity. Nevertheless, using the measured wavelength, which is λ ≈ 3 μm, and the bending rigidity of microtubules, the minimum compressive force experienced by microtubules that exhibit short-wavelength buckling can be estimated, and we obtain ≈ 100 pN. Interestingly, this is about 10 times larger than the microtubule polymerization forces measured in vitro (), which could reflect larger forces in the cell caused by the complex molecular environment at the microtubule tip (; ). Short-wavelength shapes similar to those we describe have also been seen in microtubules that were buckled by retrograde flow of the actin network (; ), and can be seen in microtubules in various other cell types and species (; ; ). Some of this high curvature microtubule bending may result from transverse shear stresses (). For example, the active viscoelastic flow of the cytoplasm generates a slowly evolving stress field () that can cause both longitudinal compression and transverse shear stresses, depending on the details of the local stress field. Indeed, microtubules also display bending on longer length scales that appears to result from this complex stress field ( and ). However, it is highly unlikely that the observed multiple short-wavelength bending could be caused by effects of transverse stresses alone. Instead, our results suggest that this ubiquitous highly curved form of microtubule deformation reflects the generic nature of reinforced microtubule compression in living cytoplasm. Our results suggest that microtubules can be used to probe the local mechanical environment within cells. Although we have focused on the cytoskeleton of interphase cells, the mitotic spindle is another important microtubule-based structure. This may provide additional insight into the poorly understood mechanical behavior of mitotic spindles (; ; ; ). Microtubules within spindles have been observed to buckle at somewhat longer wavelengths under natural conditions (), or after mechanical or pharmacological perturbations (; ), which suggests that spindle microtubules also experience compressive forces. This long-wavelength buckling may reflect an increased effective stiffness of microtubules caused by reinforcement by intermicrotubule bundling connections within the complex structure of the spindle. However, in the absence of bundling, these results suggest that the elasticity of any surrounding matrix cannot be very large. Another cell in which longer-wavelength buckling is observed is the fission yeast, where nuclear positioning is thought to occur by compressive loading of microtubules (). This again suggests that the elasticity of any surrounding network must be considerably less than that of the interphase animal cells we studied. Thus, in these particular microtubule arrays, structural reinforcement may be either unnecessary, or mediated by other mechanisms, such as microtubule bundling. An important implication of this work is the demonstration that cytoplasmic microtubules are effectively stiffened when embedded in even a relatively soft (elastic modulus ∼1 kPa) cytoskeletal network; e.g., a reinforced 20-μm-long cytoplasmic microtubule can withstand a compressive force (>100 pN) >100 times larger than a free microtubule before buckling. Consequently, individual microtubules can withstand much larger compressive forces in a living cell than previously considered possible (). Moreover, as demonstrated by our results with cytochalasin-treated cells, the lateral reinforcement is robust; even disruption of the surrounding actin network only slightly increases the buckling wavelength, with a corresponding decrease in the critical force by a factor of ∼2. This is likely caused by the presence of other sources of elasticity, such as intermediate filaments, which have been previously shown to both connect laterally to microtubules (), and to contribute to whole cytoskeletal mechanics (). As illustrated with the macroscopic model, this reinforcement is a robust phenomenon that is insensitive to the specific molecular details; the only requirement is that the surrounding matrix must be elastic. Mechanical reinforcement by the surrounding cytoskeleton may therefore provide a physical basis by which the microtubule network can bear the large loads required to stabilize the entire cytoskeleton and thereby control cell behavior that is critical for tissue development, including polarized cell spreading, vesicular transport, and directional motility. These data also suggest that these are often large compressive forces; this is consistent with mechanical models of the cell that incorporate compression-bearing microtubules which balance tensional forces present within a prestressed cytoskeleton (; ; ). Compressive loading of reinforced microtubules may also have important implications for specialized cell functions, such as in cardiac myocytes, where elastic recoil of compressed microtubules may contribute to diastolic relaxation or interfere with normal contractility in diseased tissue. These results represent a first step toward a quantitative understanding of how living cells are constructed as composite materials and mechanically stabilized at the nanometer scale. Cos7 cells (African green monkey kidney–derived) were obtained from the American Type Culture Collection and cultured in 10% FBS DME. Bovine capillary endothelial cells were cultured as previously described (). For EGFP studies, confluent monolayers of cells were incubated for 24–48 h with an adenoviral vector encoding EGFP-tubulin (). Cells were sparsely plated onto glass-bottomed 35-mm dishes (MatTek Corp.) and allowed to adhere and spread overnight. For some studies, cells were microinjected with ∼1 mg/ml rhodamine-labeled tubulin (Cytoskeleton, Inc.) using a Femtojet microinjection system (Eppendorf) and allowed to incorporate fluorescent tubulin for at least 2 h. For some experiments, cells were incubated with 2 μM cytochalasin D (Sigma-Aldrich) for 30 min before imaging. Microtubules were buckled using Femtotip needles controlled with a micromanipulator (both Eppendorf). Cardiac myocytes were isolated from 2-d-old Sprague-Dawley rats (Charles River Laboratories). In brief, whole hearts were removed from killed animals and subsequently homogenized and washed in HBSS, followed by trypsin and collagenase digestion for 14 h at 4°C with agitation. Once trypsinized, cells were resuspended in M199 culture medium supplemented with 10% heat-inactivated FBS, 10 mmol/liter Hepes, 3.5 g/liter glucose, 2 mmol/liter -glutamine, 2 mg/liter vitamin B-12, and 50 U/ml penicillin at 37°C and agitated. Immediately after purification, cells were plated on glass-bottomed Petri dishes (MatTek Corp.). For some studies, the dishes were spin coated with polydimethylsiloxane silicone elastomer (Sylgard; Dow Corning) that was treated with 25 μg/ml human fibronectin in ddH20 for 1 h. Immediately before incubation with the protein solution, culture substrates were treated in a UVO cleaner for 8 min (Jelight Company, Inc.). Cells were kept in culture at 37°C with a 5% CO atmosphere. Medium was changed 24 h after plating to remove unattached and dead cells, followed by changes with supplemented M199 medium containing 2% FBS 48 and 96 h after plating. Cells were transfected with EGFP-tubulin or microinjected with fluorescent tubulin, as with nonmuscle cells. For some studies, beating was stimulated with 1 μmol/liter epinephrine immediately before imaging. Fluorescent images were acquired on an inverted microscope (DM-IRB; Leica) equipped with an intensified charge-coupled device camera (model C7190-21 EB-CCD; Hamamatsu) and automated image acquisition software (MetaMorph; Universal Imaging Corp.). Images were analyzed using custom-built filament tracking software to extract the contours of single microtubules as a function of time. The shape at each time point was then analyzed using a Fourier decomposition (). The macroscopic rods were either hollow capillary-loading pipette tips (Eppendorf) or solid fishing line (Stren). The rods were placed in a 1-cm cuvette cell that was then filled with liquid gelatin (Sigma-Aldrich) and allowed to gel before applying a compressive force from the top. For the capillary-loading pipette tips, we used tabulated values for the elastic modulus of the polypropylene plastic ( = 1.9 GPa) and measured the outer ( ≈ 116 μm) and inner radius ( ≈ 93 μm) of the rod. For the fishing line, we determined the elastic modulus using a tensile test ( = 0.9 GPa) and measured the radius. We tested samples in the range ≈ 103–227 μm. We estimated the bending rigidity () using the following equation:The elastic modulus of the gelatin was determined using a stress-controlled rheometer (model CVOR; Bohlin Instruments) equipped with a 4°C 40-mm cone and plate tool. Video 1 shows EGFP microtubules buckling into sinusoidal shapes when they polymerize into the edge of a Cos7 cell. Video 2 shows an EGFP microtubule buckling into a sinusoidal shape as it polymerizes against the edge of a capillary endothelial cell. Video 3 shows an initially straight EGFP microtubule at the edge of a Cos7 cell that is compressively loaded with a glass microneedle and undergoes buckling. Video 4 shows another example of an EGFP microtubule compressively loaded with a microneedle at the edge of a Cos7 cell. Video 5 shows a rhodamine-labeled microtubule repeatedly buckling into a short-wavelength shape with each cycle of contraction in a beating cardiac myocyte. Video 6 shows an example of a rhodamine-labeled microtubule in a beating cardiac myocyte that repeatedly buckles in the same spot, even while adjacent microtubules remain straight. Video 7 shows two microtubules within an EGFP-tubulin–transfected cardiac myocyte that can be seen to cyclically buckle in a coordinated manner with each contractile beat. Video 8 shows a microtubule locally buckling within the cytoplasm of an EGFP-tubulin–transfected capillary cell, whereas adjacent regions of the same microtubule remain straight. Online supplemental materials are available at .

An increase in cytosolic [Ca] ([Ca]) is a ubiquitous intracellular signal that controls various cellular processes, including fertilization, proliferation, development, learning and memory, contraction, and secretion (). [Ca] rises evoked by extracellular stimuli are often observed in the form of pulsatile Ca spikes that result from the transient opening of Ca channels located either in the plasma membrane or on the cytosolic Ca stores (). The spatial counterpart of Ca spikes are Ca waves, which are produced when an initial localized [Ca] elevation leads to the propagation of the rise in [Ca] throughout the cytoplasm (; ). The frequency of the occurrence of Ca spikes is correlated with the stimulus intensity (; ; ; ), and the time course of an individual Ca spike depends on the type of receptor stimulated but not on the stimulus intensity (; ), indicating that the information of both the stimulus species and its intensity is encoded within the temporal pattern of cytosolic Ca. Stimulus-induced cytosolic Ca spikes usually form as a result of an initial slow pacemaker rise in [Ca] followed by a rapid rise in [Ca] (; ; ). The rate of the rapid [Ca] rise remains relatively constant regardless of the stimulus intensity (; ), suggesting that a regenerative process is involved in the generation of the abrupt upstroke (). Such regenerative processes require a positive-feedback element (), and some molecules that are critical to the regenerative process have been proposed. PLC catalyzes the formation of inositol 1,4,5-trisphosphate (IP) and diacylglycerol from phosphatidylinositol 4,5-bisphosphate and has been hypothesized to act as a positive-feedback element (; ) because its activity is stimulated by cytosolic Ca (). According to their hypothesis, the positive-feedback regulation of PLC by cytosolic Ca allows generation of IP spikes that result in Ca spikes through activation of the IP receptor (IPR)/IP-gated Ca release channel (). Another hypothesis is that the positive-feedback regulation is attributable to the intrinsic property of the IPR. The biphasic dependency of channel activity on Ca () may create the rapid upstroke of Ca spikes even at constant levels of IP (; ; ; ; ; ). These mechanisms for the generation of Ca spikes, however, are not necessarily exclusive, and combined models are also plausible. Observation of cytosolic IP dynamics during Ca spikes should help us better understand the mechanism responsible for the Ca spike generation (). The GFP-tagged pleckstrin homology domain (PHD) of PLC-δ1, which interacts with phosphatidylinositol 4,5-bisphosphate and/or IP, has been used to monitor IP production and its oscillatory translocation from plasma membrane to the cytosol synchronously with Ca oscillation was observed in MDCK cells (), suggesting that each Ca spike accompanies IP production. However, there is no simple correlation between its translocation from the plasma membrane to the cytosol and actual cytosolic [IP] ([IP] ; ; ). Fluorescent IP sensors based on the IP binding domain of IPR have recently been developed (; ; ), but no one has ever used them to investigate the mechanism underlying the Ca spike generation in living cells. In the present study, we developed low IP binding affinity, high signal-to-noise ratio cytosolic IP sensors based on the IP binding domain of mouse type 1 IPR (IPR1) and used them to analyze the mechanism responsible for the generation of Ca oscillations. Simultaneous imaging of [Ca] and [IP] in living cells exposed to extracellular stimuli provided us with a novel insight into the mechanism of generation of intracellular Ca spikes. To develop cytosolic IP sensors, we constructed tandem fusion proteins of a variant of yellow fluorescent protein, Venus (); an IP binding domain of IPR1 (; ; ); and an enhanced cyan fluorescent protein (ECFP; ). We used amino acid residues 224–569, 224–575, 224–579, 224–584, and 224–604 of mouse IPR1 as an IP binding motif and found that all fusion proteins except the protein composed of residues 224–569 showed IP-dependent decrease of fluorescence resonance energy transfer (FRET) between Venus and ECFP (). The largest FRET change was obtained in the fusion protein composed of residues 224–575 of mouse IPR1, and the 475-nm (ECFP) to 525-nm (Venus) emission ratio of the fusion protein increased by 25.1 ± 8.0% ( = 3) after the addition of 60 μM IP into cell lysates. We designated this fusion protein as IPR-based IP sensor 1 (IRIS-1). The emission ratio of IRIS-1 changed depending on the concentration of IP applied, but its apparent IP sensitivity (K = 549 ± 62 nM; = 3) was significantly lower than those of fusion proteins composed of residues 224–604 (K = 107 ± 41 nM; = 3), 224–584 (K = 105 ± 0 nM; = 3), and 224–579 (K = 95 ± 38 nM; = 3; ). We further characterized IRIS-1 in vitro. The protein was expressed in Sf9 cells and was purified as described in Materials and methods. shows emission spectrum of IRIS-1 when excited at 420 nm. Addition of 100 μM IP slightly increased the 475-nm (ECFP) emission and decreased the 525-nm (Venus) emission of IRIS-1. We also constructed IRIS-1–Dmut, in which two critical amino acid residues (Thr267 and Lys508) for IP binding have been replaced from IRIS-1, and found that the emission spectrum of IRIS-1–Dmut was unaltered by the addition of 100 μM IP (), indicating that FRET between the flanking fluorescent proteins decreased in response to IP binding to the IP binding domain of IRIS-1. The emission change of purified IRIS-1 exhibits an IP sensitivity (K = 437 ± 30 nM; = 6) that is slightly higher than that measured in COS7 cell lysates (). IRIS-1 can discriminate IP from its natural metabolites, inositol 1,3,4,5-tetrakisphosphate (IP) and inositol 1,4-bisphosphate (IP), with >50 and >400 times the sensitivity, respectively (). The IP sensitivity of IRIS-1 was not influenced by the addition of 1 μM Ca (). IRIS-1 did not reveal severe pH sensitivity, at least in the range examined (). IRIS-1 was uniformly distributed within the cytosol of intact HeLa cells (), and ∼80% of IRIS-1 was released from the cells after 5 min of treatment with 0.1% saponin (). ECFP-tagged IPR1, which is an ER resident protein, was hardly released from the cells with the same treatment (). These results indicate that IRIS-1 is mainly localized in the cytosol of HeLa cells. To monitor cytosolic IP and Ca dynamics simultaneously, Indo-1 was loaded into IRIS-1–expressing HeLa cells. shows FRET changes of IRIS-1 accompanied by Ca transients () elicited by sequential stimulations with 1, 5, and 10 μM of histamine. We did not detect FRET changes in IRIS-1–Dmut–expressing cells even when Ca transients were observed (), and the expression level of IRIS-1–Dmut was indistinguishable from that of IRIS-1 (). FRET changes of IRIS-1 were completely blocked by the addition of 10 μM of PLC inhibitor U73122 but not by its inactive analogue U73343 in (Fig. S1, A and B, available at ). These results indicate that IRIS-1 can monitor [IP] changes in living cells. To monitor cytoplasmic IP dynamics during Ca oscillations, we introduced both IRIS-1 and metabotropic glutamate receptor 5a (mGluR5a) cDNAs into HeLa cells. The frequency of Ca oscillations mediated by mGluR5a is known to depend on the extent of mGluR5a expression (), and the Ca oscillation frequency varied from 19 to 72 mHz (45.6 ± 16.0 mHz; = 25) in the cells transfected with mGluR5a cDNA alone (no IP sensor proteins). IP sensors that can bind IP may perturb IP dynamics in living cells to some extent with their potential to function as an IP buffer. IRIS-1–expressing cells, however, exhibited a Ca oscillation frequency (36.9 ± 7.4 mHz; = 23; ) that was not significantly different from the frequency observed in the IRIS-1–Dmut–expressing cells (32.5 ± 11.2 mHz; = 17; ; P > 0.05, test). The decay times of each Ca spike were also indistinguishable between IRIS-1–expressing cells and IRIS-1–Dmut–expressing cells stimulated with 100 μM glutamate (unpublished data). These results indicate that the expression of IRIS-1 did not have a marked influence on Ca dynamics evoked by mGluR5a stimulation. We noticed that the expression of ECFP-fused PHD (C-PHD) and Venus-fused PHD (V-PHD) significantly reduced the Ca oscillation frequency (21.5 ± 10.3 mHz; = 38; ) in comparison with the cells expressing mGluR5a alone (P < 0.05, test) or mGluR5a plus IRIS-1–Dmut (P < 0.05, test), but the expression level of mGluR5a was not reduced in C/V-PHD–expressing cells (), indicating that the exogenous expression of PHD perturbs the intracellular Ca dynamics evoked by mGluR5a stimulation. shows the IP dynamics monitored by IRIS-1 during Ca oscillations in mGluR5a-expressing HeLa cells stimulated with 100 μM of glutamate. Cytosolic IP rapidly increased after the addition of glutamate and was progressively accumulated during Ca oscillations evoked by the continuous presence of the stimulus. [IP] did not return to its basal level within the interspike interval of the Ca oscillations in 28 out of 29 cells that exhibited Ca oscillation frequencies between 14 and 45 mHz. After the removal of glutamate from the extracellular solution, the [IP] slowly returned to its basal level, with time constants of 46.8 ± 14.4 s ( = 31). The characteristic temporal pattern of the IRIS-1 signals was significantly different from that of the C/V-PHD signals that showed oscillatory changes synchronous to the Ca oscillation (), as previously reported in MDCK cells () and in mGluR5a-expressing CHO cells (). The accumulation of cytosolic IP was also observed in HeLa cells stimulated with 3 μM of histamine (), indicating that the IP dynamics that were observed are not specific to mGluR5a-expressing cells. IRIS-1 signals in the cells stimulated with histamine did not show significant fluctuations even when Ca oscillations occurred (), indicating that the small fluctuations of [IP] observed in the cells stimulated with 100 μM glutamate () are not required for the generation of Ca spikes in HeLa cells. We could not detect any significant PHD signal changes in HeLa cells stimulated with 3 μM of histamine because of the low signal-to-noise ratio of the C/V-PHD signals (unpublished data). We analyzed the mean IRIS-1 signal at the time of the onset of each Ca spike in mGluR5a-expressing HeLa cells stimulated with 100 μM of glutamate and in HeLa cells stimulated with 3 μM of histamine. shows that [IP] at the time of Ca spike generation continuously changes during the period of Ca oscillations in both cells, indicating that a constant threshold of IP is not involved in the generation of repetitive Ca spikes. We did not detect IP spikes, i.e., rapid upstrokes of [IP] with a constant amplitude followed by a gradual decrease of [IP] to its basal level, even when Ca spikes were triggered by glutamate () or histamine application (). However, if the IP sensor was saturated with IP in the cells and/or the FRET change of IRIS-1 upon IP binding was too slow to detect [IP] changes, we would have missed the IP spikes, so we conducted the following experiments to investigate these possibilities. IRIS-1–expressing HeLa cells were permeabilized with 60 μM β-escin for 3 min and then exposed to solutions containing various concentrations of IP. IRIS-1 exhibited a maximal response of 28.0% after addition of an excess amount of IP under the conditions used (Fig. S1 C). The maximal response observed was greater than that observed in intact cells stimulated with 100 μM glutamate (∼20%; ) and 3 μM histamine (∼7%; ), indicating that the dynamic range of changes in the IRIS-1 signals is greater than the changes in [IP] evoked by the stimuli when the resting level of [IP] is sufficiently low (<100 nM). The possibility of saturation of the IRIS-1 signals was further investigated with IRIS-1.2, a low-affinity mutant of IRIS-1 in which lysine 249 is replaced by glutamine (K249Q). IRIS-1.2 exhibited approximately sevenfold lower affinity for IP than IRIS-1 (Fig. S1 D). FRET signals observed with IRIS-1.2 revealed temporal patterns similar to those observed with IRIS-1 during Ca oscillations evoked by mGluR5a stimulation (Fig. S1, E and F). The IRIS-1.2 signals rapidly rose after the addition of glutamate, and they remained at the elevated level and underwent fluctuations (but not baseline spikes) during Ca oscillations. These results indicate that the IRIS-1 signals in intact cells were not saturated when the mGluR5a-expressing HeLa cells were stimulated with 100 μM glutamate. The rate of changes in IRIS-1 signals was evaluated directly by measurements of the fluorescent changes of IRIS-1 after rapid mixing with IP using stopped-flow fluorescence spectrometry (see the supplemental text and Fig. S2, available at ). The analysis of the kinetics of the fluorescence changes indicates that there are two conformations of IRIS-1 with a different FRET efficiency and only IRIS-1 with a low FRET efficiency is able to bind to IP (see the supplemental text). The kinetic parameters were estimated from both the kinetics data (Fig. S2 B) and the equilibrium data (Fig. S2 C) and were used to calculate the FRET changes during and after the addition of a brief pulse of IP. The resting level of [IP] was configured to be 40 nM based on measurements in oocytes (). Fig. S2 D shows the dynamics of the fraction of IRIS-1 with low FRET efficiency evoked by a 1-s pulse of various concentrations of IP (from 200 nM to 12.8 μM). The fraction of IRIS-1 with low FRET efficiency increased more than twofold 1 s after the onset of the IP pulses and returned to its basal level within 3 s after the IP pulses with all concentrations applied (Fig. S2 D). The frequency of the Ca oscillations in the IRIS-1–expressing cells stimulated with 100 μM glutamate was 36.9 ± 7.4 mHz (), and the mean duration of the interspike interval was 27.1 s. If [IP] actually returns to it basal level during the period of each Ca spike, IRIS-1 would be able to monitor it. We therefore concluded that IRIS-1 possesses a sufficient temporal property to detect [IP] changes during Ca oscillations evoked in HeLa cells. Spatiotemporal profiles of both [Ca] and cytosolic [IP] at the onset of initial Ca spikes in HeLa cells after mGluR5a stimulation were monitored with the fast acquisition equipment (see Materials and methods). Images were acquired every 246 ms. shows the spatiotemporal patterns of [Ca] and [IP] when the first Ca spike occurred. The rapid [Ca] rise occurred in HeLa cells stimulated with 100 μM glutamate with little or no pacemaker Ca rise (). No clear initiation site of the [Ca] rise was detected in the cell shown (). The [IP] rise was found to precede the abrupt [Ca] rise () in 48 of 52 cells, and [IP] gradually rose almost homogeneously in the cells stimulated (). The onset of the [IP] rise, which was identified as an increase greater than twice the SD of the baseline signals, preceded the onset of the [Ca] rise by 1.11 ± 1.75 s ( = 52). The rate of [IP] rise did not accelerate during the rising phase of the Ca spike (, between c and e). If the regenerative IP production mediated by PLC drives the rising phase of Ca spikes, the rate of [IP] rise should peak when the rate of [Ca] rise is at its maximum. To test this possibility, the fluorescent signals of both Indo-1 and IRIS-1 were differentiated and aligned at the time when the rate of [Ca] rise reached its maximum (). As shown in , the rate of [IP] rise did not peak when the rate of [Ca] increase was at its maximum, indicating that the steep rise in [Ca] occurred without any acceleration of the [IP] rise. A similar relationship was observed between [Ca] and [IP] during the rising phases of the subsequent Ca spikes (). As reported in histamine-stimulated HeLa cells (), the shape of the first and the subsequent Ca spikes was different in mGluR5a-expressing HeLa cells stimulated with glutamate. The subsequent spikes had more extended pacemaker activity, and the rate of the [Ca] rise after the pacemaker activity of the subsequent spikes was slower than that of the first spike (). As shown in , [IP] started to increase before the onset of the pacemaker [Ca] rise of the second, third, and fourth Ca spikes (filled and open arrowheads), and the [IP] increase did not accelerate during the period of the rapid [Ca] rise of these Ca spikes (between the thin vertical lines). The number of cells that showed [IP] rises preceding [Ca] increases and the mean interval between the onset of the [IP] rise and the onset of the [Ca] rise are summarized in Table S1 (available at ). These results suggest that regenerative IP production does not drive the abrupt Ca increase of either the first or the subsequent Ca spikes in mGluR5a-stimulated HeLa cells. Ca-dependent stimulation of PLC activity has been proposed to be a positive-feedback element that causes the steep rising phase of Ca spikes (). As shown in and , however, the rapid increase in [Ca] in the rising phase of Ca spikes was unaccompanied by any acceleration of the [IP] rise. To evaluate the component of the Ca-dependent IP production in HeLa cells, [IP] was monitored in the cells in which [Ca] increased without activation of G protein–coupled receptors. HeLa cells were treated with 1 μM thapsigargin for 5 min to deplete intracellular stores and then exposed to 2 mM Ca. In these cells, a [IP] increase that exceeded twice the SD of the baseline signal was detected with a much greater delay (66.6 ± 13.6 s; = 9) than the onset of the [Ca] increase mediated by capacitative Ca entry (). IP signals were well fitted with a monoexponential function having a time constant of 85.7 ± 25.1 s ( = 10). Histamine stimulation alone in the absence of extracellular Ca induced a rapid but small increase in [IP] (). When thapsigargin-treated cells were stimulated with 3 μM histamine in the presence of 2 mM extracellular Ca, [IP] increased in a complex pattern consisting of a rapid transient increase (, arrowheads) followed by a slow sustained increase (). The amplitude of the initial rapid transient was far below the maximal IP concentration observed at the end of the stimulation (). The rapid transient increase in [IP] was never observed in the thapsigargin-treated cells stimulated with histamine alone ( = 14; ) or with extracellular Ca alone ( = 19; ). There were no differences in rate of Ca increase and maximal Ca concentration reached between the cells stimulated with extracellular Ca alone () and the cells stimulated with histamine in the presence of extracellular Ca (). Similar temporal patterns of [Ca] were obtained with Fura-4F (K = 770 nM; unpublished data), indicating that Indo-1 was not saturated under the conditions used. These results indicate that the cytosolic Ca increase itself induces IP production in HeLa cells, but the process is relatively slow compared with agonist stimulation–evoked IP production, and receptor activation and the [Ca] increase synergistically induce the rapid, transient IP production in HeLa cells, but the amplitude of the IP transient is relatively small even when [Ca] is persistently elevated. sup xref #text An anti-GFP antibody and an anti-mGluR5 antibody were purchased from MBL International Corporation and Upstate Biotechnology, respectively. The IP binding core cDNA was obtained by PCR from pBact-ST-neoB-C1 (; ). Venus () and ECFP cDNAs were fused to the 5′ and the 3′ end, respectively, of the IP binding domain to produce IRIS proteins. Mutations of T267A and K508Q or K249Q were introduced as described elsewhere () to create an IP binding–deficient mutant and a low IP binding–affinity mutant, respectively. IP sensor cDNAs were cloned in BamHI and XhoI sites of pcDNA3.1 zeo(+) (Invitrogen) that contains 6× Histidine tag () in NheI to HindIII site for the expression in mammalian cells. The multicloning site of pcDNA3.1 zeo(+) was amplified by PCR and cloned into SalI site of pFastBacI (Invitrogen). IRIS-1 and IRIS-1–Dmut were digested by NheI and XhoI from pcDNA3.1 zeo(+) and inserted to the modified pFastBacI for the expression in Sf9 cells. Venus or ECFP cDNA was fused to 5′ end of cDNA corresponding to amino acid residues 11–140 of rat PLCδ1 to create V-PHD (provided by M. Hattori, Nagoya City University, Aichi, Japan) and C-PHD, respectively. The resulting cDNAs were cloned in pcDNA3.1 zeo(+). Rat mGluR5a cDNA inserted into pME18S () was a gift from S. Nakanishi (Osaka Bioscience Institute, Osaka, Japan). In vitro characterization of IRIS proteins were performed by using lysates prepared from COS7 or Sf9 cells expressing IRIS proteins and IRIS proteins purified from Sf9 cells. COS7 cells expressing IRIS proteins were lysed with the solution containing 10 mM Hepes, pH 7.4, 100 mM NaCl, 1 mM EDTA, 1 mM 2-mercaptoethanol, and 0.5% NP-40. His-tagged IRIS-1 and IRIS-1–Dmut proteins were expressed by means of Sf9 baculovirus–expressing system (Invitrogen) and were purified as follows. Sf9 cells were suspended in homogenization solution containing 10 mM Hepes, pH 7.8, 100 mM NaCl, 5 mM 2-mercaptoethanol, 0.1% NP-40, and proteinase inhibitors (0.1 mM PMSF, 10 μM Leupeptin, 10 μM Pepstatin A, and 10 μM E60) and were homogenized at 1,000 rpm for 10 strokes at 4°C. Homogenates were applied to Probond resin (Invitrogen), and His-tagged proteins were eluted with 300 mM imidazole. Proteins were then dialyzed for 30 min against the solution containing 10 mM Hepes, pH 7.4, and 100 mM KCl. The dialyzed proteins were loaded to HiTrap heparin columns and eluted with the high KCl solution containing 10 mM Hepes, pH 7.4, and 250 mM KCl. Proteins were diluted by adding two volumes of the solution containing 10 mM Hepes, pH 7.4, and 100 mM KCl and concentrated with centrifugal filter devices (30,000 MWCO; Amicon Ultra 15 [Millipore]). For the evaluation of the rate of changes in IRIS-1 signals, IRIS-1–expressing Sf9 cells were suspended in 10 mM Hepes, pH 7.4, 100 mM NaCl, 1 mM EDTA, 1 mM 2-mercaptoethanol, 0.01% NP-40, and proteinase inhibitors and were homogenized at 1,000 rpm for 10 strokes at 4°C. The homogenates were centrifuged at 20,000 for 30 min, and the supernatants were used for stopped-flow fluorescent measurements. The number of IRIS-1 molecules in the supernatants was quantified by the equilibrium IP binding analysis using [H]IP as described previously (). IRIS signals were measured with a spectrofluorometer (FP-750; Jasco). IRIS cDNAs inserted in pcDNA3.1 zeo(+) were transfected into HeLa cells with transfection reagents (TransIT; Mirus). After 12–36 h, cells were used for imaging. After loading the cells with 5 μM Indo-1 AM (DOJINDO), imaging was performed under the constant flow (2 ml/min) of the balanced salt solution containing 20 mM Hepes, pH 7.4, 115 mM NaCl, 5.4 mM KCl, 1 mM MgCl, 2 mM CaCl, and 10 mM glucose as an imaging media at 37°C. Imaging was performed at 37°C through an inverted microscope (IX-71 or IX-81; Olympus) with a cooled charge-coupled device camera (ORCA-ER; Hamamatsu Photonics) and a 40× (NA 1.35) objective. A 333–348-nm excitation filter and pair of 420–440- and 460–510-nm emission filters and a 425–445-nm excitation and pair of 460–510-nm (cyan) and 525–565-nm (yellow) emission filters were used for fluorochromes (Indo-1 and IRIS, respectively). Beam-splitter mirrors (400 and 450 nm) were alternately inserted into the light path for Indo-1 and IRIS, respectively. Images were acquired at 0.25 Hz, with an exposure time of 100 or 150 ms. For the fast acquisition (4.07 Hz) of fluorescent images of IRIS-1 and Indo-1, an emission splitter (W-view; Hamamatsu Photonics) was used with a fast light source exchanger (DG-4; Sutter Instrument Co.) on the IX71 inverted microscope. Sequential excitation of IRIS-1 and Indo-1 was performed by using a 450-nm dichroic mirror and two excitation filters (a 425–445-nm filter for IRIS-1 and a 333–348-nm filter for Indo-1). Dual emission at 460–490 nm (for IRIS-1 and Indo-1) and >520 nm (for IRIS-1) split with a 460–490-nm filter, a long-path 520-nm barrier filter, and two 505-nm dichroic mirrors equipped in W-view. Images were acquired at 4.07 Hz, with an exposure time of 100 or 150 ms. Acquisition was performed with the custom software TI Workbench (written by T. Inoue). Off-line analysis was performed with TI Workbench combined with Igor Pro software (WaveMetrics). Spectral analysis of Ca oscillations was performed as described previously (). The change in the intensity of Venus fluorescence (525 ± 20 nm) of IRIS-1 after the addition of various concentrations of IP was monitored using the FP-750 spectrofluorometer with a stopped-flow rapid mixing accessory (RX2000; Applied Photophysics). The IRIS-1–containing supernatants (final 2.5 nM of IRIS-1) were mixed with a 24-fold–excess volume of 10 mM Hepes, pH 7.4, 100 mM NaCl, 1 mM EDTA, 1 mM 2-mercaptoethanol, and 0.01% NP-40 containing various concentrations of IP at 25°C, and the change of the intensity of Venus fluorescence of IRIS-1 excited at 440 ± 20 nm was monitored at 1,000 Hz. At least 10 traces were averaged and were used for the nonlinear regression analysis with Igor Pro software. The IP dependency of FRET changes of IRIS-1 at equilibrium was monitored using the same batches of the supernatants as used for the kinetic measurements with a 1-cm cuvette and the FP-750 spectrofluorometer. Fig. S1 shows effects of PLC inhibitors on emission changes of IRIS-1 elicited by 10 μM of histamine and evaluation of IRIS-1 signals observed in living HeLa cells. Fig. S2 shows the rate of reaction of IRIS-1. Fig. S3 shows simulation of IP and Ca dynamics. Table S1 shows numbers of cells that showed [IP] rises preceding [Ca] increases, the mean intervals between the onset of [IP] rises, and the onset of [Ca] rises. Table S2 shows the parameters used to calculate IP and Ca dynamics. The supplemental text shows evaluation of the effect of the rate of FRET changes of IRIS-1 upon IP binding. Online supplemental material is available at .

Cell adhesion and migration are pivotal to many biological and pathological processes, including development, inflammation, repair, and cancer metastasis. The regulation of cell adhesion and motility is very complex. It requires the coordination of adhesion receptor–matrix interactions on the cell surface, trafficking of the receptors to and from the sites of adhesion (adhesion site turnover), and cytoskeletal reorganization inside the cell. Moreover, all these events have to be precisely orchestrated not only spatially but also temporally. α/β-Integrin heterodimers are key molecules involved in cell adhesion and migration. Binding of cell surface–expressed integrins to specific ligands on the extracellular matrix is relatively well understood (). In contrast, much less is known about the turnover of integrin-containing adhesion sites and the intracellular traffic of integrins, although their importance in cell adhesion and migration is being increasingly recognized (; ). In particular, the mechanisms that target integrins to specific endocytic compartments remain unknown. Eukaryotic cells internalize cell-surface receptors by endocytosis. Integrins, like several other proteins lacking the AP-2 localization signal are internalized from the cell membrane via nonclathrin-derived structures, and some integrins have been shown to subsequently fuse with compartments containing clathrin-derived cargo proteins (; ; ; ). The endocytosed receptors are subsequently recycled back to the cell surface or targeted for degradation. Depending on the cell type and the stimulus, integrins have been shown to be transported through caveolin-1–positive structures or early endosomes, either directly or via the perinuclear recycling compartment, back to the plasma membrane (). Rab proteins are small GTPases that regulate both endocytosis and exocytosis. Several members have been implicated to function on the endocytic pathway, in which the function of Rab5 is understood in more detail (). Rab5 regulates membrane traffic into and between early endosomes as well as vesicle transport along microtubules (). Integrin traffic is regulated by several kinases (; , , ; ; ) and motor proteins (). However, very little is known about the mechanisms that target integrins to specific intracellular compartments and how this may be controlled. Although Rab proteins are known to bind a multiplicity of diverse effectors (), only a few examples demonstrate an interaction between a Rab GTPase and a cargo molecule (; ). We identify Rab21 and Rab5 as integrin-associated proteins and positive regulators of integrin traffic. To identify novel integrin-interacting proteins, we screened a mouse embryonic cDNA library using α2-integrin cytoplasmic domain as bait in a yeast two-hybrid screen. The bait comprised the conserved membrane-proximal sequence shared by most α-integrin subunits followed by the α2-specific segment. Several positive clones encoded the COOH-terminal part of Rab21. Rab21 is a ubiquitously expressed and poorly characterized member of the Rab family that has recently been shown to function on the endocytic pathway (). The ability of Rab21 to associate with integrins was further confirmed in human cells. GFP-tagged Rab21 coprecipitated with α2β1-integrin, a collagen binding molecule (), in cells plated either on collagen or on plastic (), indicating a constant, rather than a matrix adhesion–inducible, association between the two proteins. Furthermore, Rab21 was found to associate with several α/β1-integrin heterodimers. Rab21 coprecipitated with chimeric integrins containing the extracellular domain of α2-integrin fused to the cytoplasmic domains of either α1- or α5-integrin () and endogenous integrins that were immunoprecipitated with antibodies against α1, α2, α5, α6, and β1 subunits (; and Table S1, available at ). These data suggest that the association involves the shared conserved membrane-proximal segment present in most α-subunit cytoplasmic domains. To get evidence for the association with an independent method, we performed yeast two-hybrid studies with α-tail mutants. In remating tests, we found that the COOH-terminal part of Rab21 (amino acids 95–222) was able to associate with the cytoplasmic tails of α2- and α11-integrin (). Introduction of a proline residue adjacent to the conserved membrane-proximal sequence (α2P, to create a conformational change into the α2 tail) markedly weakened the association between the α2 cytoplasmic domain and Rab21, whereas mutagenesis of a conserved positive charge in α2A (K1160) and α11A (R1170) tails had no effect (). To further characterize the Rab21–integrin association, we generated several mutants of α2-integrin. All of these mutants were expressed and mediated adhesion to collagen in transiently transfected CHO cells ( and not depicted; see ). Mutagenesis of residue R1161 (in α2AA and α2AAKYA, removing another conserved charged residue) to alanine significantly reduced α2 tail association with Rab21 judged by yeast mating tests and immunoprecipitations (). In addition, F1159A mutation (α2AARA, possibly creating a conformational change) showed reduced association in the yeast assays, whereas in immunoprecipitations the reduction was not significant. These data on the mutant integrins suggest that the conformation of the cytoplasmic domain and residue R1161 of α2-integrin are important for the Rab21 association. Finally, endogenous β1-integrins readily associated with endogenous Rab21 and to some extent Rab5 proteins in vivo. These associations were specific, as Rab7 and Rab11 failed to coprecipitate β1-integrin from these cells, even though separately all proteins were efficiently immunoprecipitated (). Next, we studied the enzyme activity dependence of the association between integrins and Rab21. A Rab21 mutant () that has defects in GTP hydrolysis (Rab21 Q76L, designated Rab21GTP) showed a 10–73% increase in its ability to associate with integrins in different cell lines when compared with Rab21 wild type (WT). In contrast, a mutant having defects in the binding of GTP (Rab21 T31N, designated Rab21GDP) showed a 26–35% decrease (; and Table S1). In addition, Rab21 with a deletion in the COOH terminus (Rab21c-del; residues 1–144) demonstrated inefficient association with the β1-integrins in vivo (37 ± 10% decrease; and Table S1). The cellular localization of endogenous β1-integrins, Rab5A, and Rab21 was studied in MDA-MB-231 cells. β1-Integrin was detected by the NH2-terminally binding antibody in the lumen of endogenous Rab5- and Rab21-positive large vesicles (, arrowheads). In addition, these cells also harbor large endogenous EEA1-positive endosomes that stained positive for Rab5A but showed very limited overlap with Rab21 (, arrows). α/β1- Integrin heterodimers can be expressed in active (extracellular domain detected by a monoclonal antibody HUTS21; active-β1 antibody) and inactive forms on the cell surface (extracellular domain of both forms detected by P5D2; pan-β1 antibody; ; ). Expression of Rab21 in MDA-MB-231 breast cancer cells altered the subcellular localization of the total pool of β1-integrins. Cells expressing GFP-Rab21 contained numerous β1-integrin−positive vesicles (compare the nontransfected cell [, asterisk] and the transfected cell []). The majority of these had GFP-Rab21 on their limiting membrane (72 ± 8% of pan-β1–positive vesicles and 68 ± 10% of active-β1–positive vesicles; = 20 cells; ). The recruitment of integrins to Rab21-positive intracellular structures was confirmed using two additional antibodies recognizing active-β1-integrin (12G10 [] and P4G11 []; ). Similar to previous observations on breast cancer cells (), the ECM proteins collagen and fibronectin were also detected in the endocytic structures ( and not depicted). Interestingly, both Rab21 mutants induced marked alterations in the cellular distribution of β1-integrins. GFP-Rab21GTP showed a tubular- vesicular staining pattern that largely overlaps with the ER (; and not depicted), and active-β1-integrin was detected in smaller and more irregularly shaped endocytotic integrin vesicles, whereas the pan-β1 antibody detected β1-integrin at the membrane, diffusely in the cytosol and accumulated in the large vacuolar structures. Strikingly, expression of GFP-Rab21GDP induced localization of active-β1 into large focal adhesions (, arrows), which are rarely detected in these highly motile cells endogenously (). GFP-Rab21GDP variant but not GFP-Rab21GTP was also seen at membrane ruffles (, arrows). Proper membrane targeting of Rab21 was also mandatory for the regulation of integrin localization in vivo. Mutagenesis of the putative COOH-terminal prenylation motif (CCXXX; ) in Rab21 resulted in a complete loss of vesicular localization of Rab21 (GFP-Rab21CCSS; ) and a 55–62% ( = 2) less efficient association with β1-integrins as compared with the Rab21WT (). Cells transfected with GFP-Rab21CCSS showed prominent focal adhesions (active-β1 antibody; , arrows) when compared with the vesicular pattern observed in the nontransfected cells (, asterisk). Interestingly, these effects on integrin distribution were almost indistinguishable from that of Rab21GDP mutant (see the previous paragraph). To analyze the dynamics of integrin internalization, we investigated the ability of different Rabs to coprecipitate surface-labeled integrin before and after internalization. Interestingly, Rab5 and Rab21 antibodies very weakly coimmunoprecipitated biotinylated ∼160-kD proteins located on the plasma membrane (, no internalization). After 15 min of internalization, Rab5/21, but not Rab7, associated with biotinylated plasma membrane–derived ∼160-kD protein (, internalization 15 min, lanes 1 and 2). Reimmunoprecipitation revealed that β1-integrin was present, possibly with other plasma membrane–derived proteins, in the biotinylated Rab5/21 coprecipitating fraction (, internalization 15 min, lane 3), indicating that integrins associate with Rabs mainly after internalization. Further colocalization and internalization studies showed that internalized β1-integrins originally traverse through the same compartment as transferrin but, subsequently, the integrin traffics into separate, mainly GFP-Rab21–positive vesicles that can also contain caveolin, whereas transferrin remains colocalized extensively with GFP-Rab5 (Fig. S1, available at ). In these cells, internalized integrin did not colocalize with GFP-Rab11 (not depicted). We also found that Rab5 and Rab21 regulated integrin internalization/recycling. The most rapid integrin internalization was observed in GFP-Rab5–expressing cells, in which 82–100% of the surface-labeled receptor internalized within 5 min (). GFP-Rab21 but not -Rab11 induced integrin internalization as well (74–86% after 10 min). Importantly, transferrin endocytosis was unaltered by Rab21 expression, suggesting that Rab21 does not influence the traffic of endosomally translocated receptors in general (Fig. S2 A, available at ). We are aware that we have not excluded the effects of Rab21 expression for all types of unspecific internalization (e.g., macropinocytosis), but at least its closest homologues (Rab5 and Rab22) have not been shown to affect macropinocytosis (; ). Interestingly, the internalized receptor recycled rapidly back to the cell surface in Rab5- and Rab21-expressing cells with >50% of the labeled pool being recycled during a 15-min chase (detected by the reduction in the amount of biotinylated receptor in these cells where cell surface biotin is cleaved). Integrin internalization was reduced by expression of GFP-Rab21GDP, whereas GFP-Rab21GTP induced a steady accumulation of the internalized receptor when compared with GFP-, Rab5-, Rab11-, and Rab21GDP-expressing cells (, 30-min chase). Importantly, knock down of endogenous Rab21 reduced the amount of rapidly internalized integrin (, min). In density gradient fractionations, expression of GFP-Rab21 shifted the integrins toward the denser Rab-positive fractions (), and GFP-Rab21 cofractionated with α2-integrin in fractions 3–9 (). A further shift in the endogenous integrin pool (to fractions 5–11) was observed upon expression of GFP-Rab21GTP, and GFP-Rab21GDP was also observed in the denser fractions (). In the lighter fractions (3–5), GFP-Rab21 and integrin were found to cosediment with the Golgi-marker GM130, whereas in the denser fractions, cosedimentation was observed with the ER marker P115 (fractions 6–8) and EEA1 (fractions 7–9; Fig. S3 A, available at ). This data, together with the abundance of integrin vesicles observed in Rab21-expressing cells (), suggests that Rab21 targets integrins to the endocytic fraction in human cells. Further characterization of the large intracellular structures induced by GFP-Rab21 overexpression was performed by electron microscopy and immunogold labeling of GFP (Fig. S3, B and C). Overexpressed GFP-Rab21 was found predominantly on the limiting membranes of large multivesicular body (MVB)–like structures and in numerous vesicles surrounding it (Fig. S3, B and C), thus resembling the doughnut-shaped structures observed by immunofluorescence (). In addition, GFP-Rab21 staining was observed in structures with autophagic morphology (<20% of the labeled structures; unpublished data). We used time-lapse imaging to visualize the kinetics of Rab21 in real time. GFP-Rab21 was detected in large, swollen endosome-resembling structures, as well as in motile, smaller vesicles, which moved bidirectionally between the cytosol and the cell edges at velocity ranging from 0.75–2.25 μm/s (, vesicles B and C; and Video 1, available at ). Small vesicles were observed interacting with each other and associating with bright Rab21-containing domains on the larger Rab21-positive structures (, vesicle A) identical to the β1-integrin–containing vesicles seen in our immunofluorescence stainings ( and and Fig. S1). The short-range motion of the Rab21 vesicles at the cell periphery was sensitive to the disruption of actin filaments with 20 μM cytochalasin (unpublished data), and the long-range curvilinear movements observed in the cell body were abolished upon treatment with 5 μM nocodazole (unpublished data), suggesting a role for actin and microtubules. Combined total internal reflection fluorescence microscopy (TIRFM; pseudocolored green) and conventional widefield epifluorescence analysis (pseudocolored red) showed GFP-Rab21 vesicles emanating from the membrane into the cell (changing from green to red/yellow) and back or vice versa (although formally this technique does not exclude the possibility that some vesicles will by chance come closer and further away from the plasma membrane). The intensity plots show rapid changes in the intensities of the signal in the green channel, whereas the signal in the red channel remains constant ( and Video 2). No obvious diffusion of the bright Rab21 structures was seen using TIRFM, indicating either that the Rab21-labeled structures remain as distinct domains on the plasma membrane or that they do not fuse with the membrane. This former possibility correlates well with the observation that upon binding the small vesicles seem to remain as bright patches on the limiting membranes of the intracellular large Rab21-positive structures (). Real-time imaging and TIRFM confirmed that the intact GTP/GDP cycling of Rab21 was crucial for its vesicular traffic because motile small vesicles were not detected in GFP-Rab21GDP–transfected cells and Rab21 was associated with rather static structures at the cell membrane ( and Video 3). These resemble YFP-talin–positive structures detected by TIRFM () and may represent adhesion structures that have reduced turnover because of the expression of GFP- Rab21GDP and possibly its association with β1-integrin. We also addressed whether Rab21 expression alters the motility of integrins in live cells. We engineered NH-terminally GFP-tagged α2-integrin (GFP-α2), which is dynamic in vivo, mediates cell adhesion to collagen, is recognized by α2I-domain binding monoclonal antibodies, and is expressed on the cell surface (see ; not depicted). Upon cotransfection, Renilla luciferase (Rluc)–tagged Rab21 and GFP-α2 associated efficiently (). Rab21 expression induced motile GFP–α2-integrin–labeled vesicles ( and Video 4, available at ), whereas cells expressing GFP–α2-integrin alone or with Rluc-Rab21GDP, showed a vesicular-tubular staining pattern with no obvious vesicles and that partially overlaps with ER tracker stain in live cells (, Video 5, and not depicted). Collectively, these data demonstrate that enzymatically active Rab21 is able to target integrins to endocytic vesicles in live cells. To investigate the importance of Rab21-mediated integrin traffic in β1-integrin–dependent cell adhesion, we studied MDA-MB-231 breast and PC3 prostate cancer cells having relatively high and low endogenous Rab21 expression, respectively (unpublished data). Of all the Rabs tested, overexpressed Rab21 was most efficient in increasing cell adhesion to collagen (86 ± 8% in PC3 cells and 22% in MDA-MB-231 cells; ) during the initial 30 min of adhesion. Rab21 expression did not enhance the initial adhesion step of cells plated on fibronectin or vitronectin (Fig. S2 B). Expression of Rab5 increased adhesion by 48 ± 13% in PC3 cells. The GTP/GDP cycling of Rab21 was found to be important for supporting integrin function because the Rab21 mutants were unable to induce cell adhesion to collagen (). The lack of dominant effects of the mutants is probably due to the fact that their ectopic expression is not sufficient to interfere with the high number of endogenous associations between integrins and Rabs involved in regulating adhesion/migration, whereas introduction of Rab21WT supports adhesion to collagen by further increasing the endosomal traffic of α/β1-integrin heterodimers. Overexpression of GFP-Rab5, -Rab11, -Rab21, or -Rab21GTP did not influence the amount of β1-integrin detected on the surface of MDA-MB-231 () and PC3 cells (not depicted), suggesting that efficient transport of endocytosed integrins to newly formed sites of adhesion, rather than a change in the steady-state expression of integrins on the cell surface, is the basis of Rab21-induced cell adhesion. Interestingly, GFP-Rab21GDP caused a modest increase in β1-integrin surface expression in MDA-MB-231 cells (). Thus, Rab21GDP both inhibits integrin traffic and increases cell surface levels of integrins. This induction of two counteracting forces further explains the lack of dominant-negative effects on adhesion. Most important, specific silencing of Rab21 with two different RNAi oligos () resulted in a 30% reduction in the adhesion of MDA-MB-231 cells to collagen. As no down-regulation of β1-integrin was observed, the effect of Rab21 knockdown is most likely due to inefficient integrin traffic. Finally, to demonstrate that Rab21–integrin association is involved in Rab21-induced cell adhesion, we used the mutant α2-integrins with reduced Rab21 binding (). CHO cells lack endogenous collagen receptors. Transient expression of α2WT, α2AA (deficient Rab21 association), or α2A (unaltered Rab21 association), together with Rluc, enabled equivalent adhesion of these cells to collagen (, bottom). Rluc-Rab21 induced adhesion to collagen when coexpressed with α2WT- or α2A-integrins but failed to do so with α2AA-integrin mutant (, top). Furthermore, coexpression of Rluc-Rab21 induced vesicular localization of GFP-α2WT but had no effect on the cellular distribution of GFP-α2AA (). The endo/exocytic cycle of integrins has been suggested to facilitate focal complex assembly as well as cell motility (). We observed that the number of motile Rab21 vesicles close to the cell edge in the protruding lamellae was high in cells that were actively spreading on collagen or migrating at the edge of a scratch wound (, – h after plating; and not depicted) when compared with cells that had been adherent overnight or were located within a confluent monolayer ( and not depicted). We also found that cells expressing GFP-Rab21WT migrated on plastic in a scratch wound assay to close the wound almost completely, whereas GFP-, GFP-Rab21GDP–, and GFP-Rab21GTP–expressing cells migrated less efficiently, covering only ∼70–75% of the wound area (). Conversely, stable Rab21-shRNA (short hairpin RNA)–transfected cells with reduced Rab21 expression () migrated poorly, covering only sim;59% of the wound area, whereas Scr-shRNA–transfected control cells closed the wound almost completely (). During this time, there were no obvious differences in the proliferation of the stable transfected cells (not depicted). These data suggest that expression of Rab21 regulates migration in these cells. Prostate and breast cancer cells are known to metastasize to bone (). Because fibrillar collagen is one of the main components in human bone, we finally analyzed the effects of Rab expression on cancer cell adhesion to bone matrix. Interestingly, overexpression of Rab21 in PC-3 prostate cancer cells resulted in a modest but constant and statistically significant (20 ± 8%; = 7; P < 0.03) increase in adhesion to bone (). Together, our data show that Rab21 activity regulates integrin-dependent adhesion to collagen and human bone as well as the motility of cancer cells on cell-secreted and serum-derived matrix components other than collagen. We report the discovery of a previously unknown association between integrins and Rab proteins. We show that Rab5 and Rab21 associate with internalized integrins. Furthermore, Rab21 expression is shown to regulate integrin-containing focal adhesions and adhesion and migration of breast and prostate cancer cells. Live-cell imaging revealed that Rab21-positive vesicles move between the plasma membrane and the cell body and that expression of Rab21 modulates vesicular motility of integrins in vivo. Although it is clear that the endocytic traffic of integrins involves a complex machinery of kinases, motor proteins, and members of the Rab family of GTPases, how integrins are actually targeted to the intracellular vesicles has remained enigmatic. The identification of integrin association with the putative early endosomal Rab21 led us to identify a functional relationship between integrin association with Rab21 and Rab5 and the regulation of cell adhesion. Several studies have demonstrated integrin endocytosis into intracellular vesicles and their recycling back to the membrane (, ). In serum-starved cells stably adhering to tissue culture plastic, integrins have been shown to be transported to a perinuclear recycling compartment. Upon growth factor stimulus, internalized integrins either recycle very rapidly back to the plasma membrane from the perinuclear compartment (β1-integrins; ) or are completely rerouted to a short-loop trafficking pathway directly back to the membrane (αvβ3-integrin; ). The existence of a Rab–integrin association, which positively regulates cell adhesion, provides a missing mechanistic link into the orchestration of this complex process. We show on one hand that the motility of Rab21 vesicles close to the plasma membrane requires the actin cytoskeleton and, on the other hand, that expression of Rab21 mutants with impaired GTP binding or membrane localization induce integrin targeting to the membrane and the formation of exaggerated adhesion sites. This is in agreement with the recent finding that integrins in newly forming protrusions travel on actin cables associated to motor protein Myosin X and that normal cell adhesion and spreading during the initial stages of adhesion seem to require the efficient motility of integrins close to the plasma membrane (). There is only limited data available on the dynamics of integrins in cells. One study demonstrates the internalization of GFP-tagged integrin in live cells (). However, no integrin movements back to the plasma membrane were seen. Although the role of Rab proteins was not addressed in the study of , it is intriguing to note that the large vesicles they described are remarkably similar to the structures induced here by the overexpression of GFP-Rab21. We show here that these large structures resemble MVBs. MVBs are generally thought of as being major protein sorting stations in the endocytic pathway. Proteins at their limiting membrane can be packaged into transport vesicles destined to the cell surface or to the TGN. Alternatively, proteins can be targeted to inwardly budding vesicles of the MVBs. These latter are considered to be the exosomes secreted after fusion of MVBs with the plasma membrane (). The observation that overexpressed GFP-Rab21 is seen mainly in the limiting membrane and vesicles surrounding the MVBs points to an involvement of the MVBs in recycling vesicles back to the plasma membrane. The enlargement of the MVBs and accumulation of the vesicles most likely is the result of overexpression of the Rab21 protein, indicating its role in traffic to the MVBs. Our data does not unambiguously show whether integrins and Rabs interact directly. The facts that the association was detected from a yeast two-hybrid screen and that it is abrogated by mutagenesis of the α-cytoplasmic domain suggest that it may be direct. However, we cannot rule out the possibility that as-yet-unknown proteins that are capable of binding to the integrin may serve as linkers between Rab21 and integrins. We were unable to detect direct interaction of GST-Rab21 with synthetic integrin cytoplasmic tail peptides. However, this does not necessarily mean that the interaction is not direct. It is possible that the peptides are not presented in a correct conformation for the interaction to occur. Evidence from detailed biochemical and nuclear magnetic resonance studies (; ) indicates that upon integrin activation the membrane-proximal regions of the integrin cytoplasmic tails move out of the membrane into the cytoplasm, revealing the highly conserved residues to the cytoplasmic face and involving substantial structural changes in the cytoplasmic tails. We show that Rab21–integrin association occurs downstream of the internalization step and that a substantial portion of the internalized integrin is in an active conformation as detected by the activation epitope–specific antibodies. This is in line with our data showing that the association is mediated via the membrane proximal–conserved segment of the α-subunit COOH terminus, with conserved residue R1161 being especially important, and that the association seems to be α-tail conformation sensitive. Prenylation-dependent membrane targeting of Rabs is crucial for Rab function as regulators of vesicle fusion in intracellular protein trafficking (). We show that the mutagenesis of the putative prenylation motif in Rab21 results in complete loss of its vesicular localization. At the same time, this variant shows reduced association with β1-integrins and upon overexpression induces integrin localization on the plasma membrane and in large focal adhesions (by targeting the integrins or by blocking the internalization). Antibodies against the following antigens were used: EEA1, Rab5A, Rab7, Rab11, caveolin-1 (all from Santa Cruz Biotechnology, Inc.), Rab21 (), β1-integrin (P5D2, P4G11, and AIIB2), α5-integrin (BIIG2), EGFR (151-IgG; all from the Drosophila Studies Hybridoma Bank), α2 (mAb MCA2025; Serotec), pAb AB1934 (Chemicon), α1 (MAB1973; Chemicon), α6 (MAB699; Chemicon), β1 (HUTS-21 [BD Biosciences] and MAB2252 [Chemicon]), collagen type 1 (RAHC11; Imtek), GFP polyclonal antibody, fluorescently conjugated secondary antibodies, Cell Tracker dyes, and labeled transferrin (all from Invitrogen). Full-length Rab21 was subcloned from Rab21 murine cDNA (clone 6490069; IMAGE) by PCR amplification and ligated into pRluc-C2 (PerkinElmer) and in pEGFP-C2 (CLONTECH Laboratories, Inc.). Rab21GTP (Q76L), Rab21GDP (T31N), and CCSS (residues 218 and 219) mutants were generated using QuikChange Site-directed mutagenesis kit (Stratagene). Rab21 COOH-terminal–deletion mutant was generated by introducing a stop codon after E144. Plasmids encoding GFP-Rab5a, GFP-Rab7, YFP-Rab9, and GFP-Rab11 have been described (; ; ; ), and YFP-mouse talin was provided by D. Critchley (University of Leicester, Leicester, UK). α2-Integrin was subcloned from the α2 cDNA in pawneo2 vector () into pEGFP-C2 vector. The signal sequence of α2-integrin (annealed synthetic oligos corresponding to nucleotides 43–129 in the published sequence [] was inserted to the NheI site in the vector to enable correct targeting to the plasma membrane. The GFP–α2-integrin cytoplasmic tail mutants were generated by using QuikChange Site-directed mutagenesis kit. All clones were verified by sequencing. Rluc-tagged Rab21 constructs alone or with GFP-α2 variants (CHO cells that lack endogenous collagen binding integrins) were transfected into 95% confluent cells using Lipofectamine 2000 and incubated for 18 h. For immunoprecipitations with endogenous proteins, confluent MDA-MB-231 cells (20 × 10) were collected from plastic plates with cold PBS. For analysis of association with cell surface–labeled integrin, MDA-MB-231 cells were plated on collagen-coated dishes for 1 h and surface biotinylated with cleavable biotin (0.5 mg/ml EZ-sulfo-NHS-SS-biotin in HANKS buffer) for 30 min on ice. After washings, cells were either lysed immediately or warmed for 15 min in HANKS +37°C to allow internalization. Cells were lysed in IP buffer (PBS with 1% octylglycoside, 0.5% BSA, 1mM CaCl, 1mM MgCl, and protease inhibitor cocktail [Roche]) on ice for 15 min. Postcentrifugation supernatant was precleared with BSA-blocked (IP buffer) protein G–agarose beads and divided into five aliquots for immunoprecipitations with different anti-Rab antibodies or a control antibody and protein G beads (90-min rotation at 4°C). After three washings (IP buffer containing 0.3% octylglycoside), SDS sample buffer was added and the proteins were separated by SDS-PAGE (4%/10%) and immunoblotted for β1-integrin (MAB2252). Reimmunoprecipitations were performed as described earlier (). For the luminescent immunoprecipitations, the beads were transferred into white microtiter plate wells (96-well) and treated with Rluc substrate (5 μg/ml coelenterazine [Nanolight Technologies]), and the luminescence was measured with a multilabel HTS counter (VictorV; PerkinElmer). These were performed as described previously (; ) with some modifications. After 1 h of adhesion to collagen-coated dishes, the transfected cells were placed on ice, washed once with cold PBS, and surface labeled with 0.5 mg/ml cleavable NHS-SS-biotin (Pierce Chemical Co.). After washings, prewarmed (+37°C) HANKS medium was added, and protein traffic (internalization and recycling) was allowed to occur for the times indicated. Biotin was removed from cell surface proteins by MesNa reduction and iodoacetamide quenching on ice. The cells were lysed (200 mM NaCl, 75 mM Tris, 15 mM NaF, 1.5 mM NaVO, 7.5 mM EDTA, 7.5 mM EGTA, 1.5% Triton-X-100, and Complete), and the amount of biotinylated integrin was assayed using the anti–β1-integrin antibody AIIB2 to capture the integrins and HRP anti-biotin antibody for ELISA detection. As control, the cells were lysed after the labeling to determine the amount of total biotinylated integrin. Stable MDA-MB-231–expressing GFP-Rab21 cells were fixed in 1% PFA with or without 0.01% glutaraldehyde in 100 mM phosphate buffer, pH 7.0, for 2 h at RT. Next, cells were pelleted in 10% gelatine and postfixed in 1% PFA for another 24 h. Ultrathin cruosections were prepared on a cryochamber (EM FCS; Leica), and thawed sections were incubated with a polyclonal antiserum raised against EGFP followed by incubation with protein A complexed to 5-nm gold particles according to standard procedures. Sections were observed in an electron microscope (model 1010; JEOL) operating at 80 kV. HeLa cells were transiently transfected with GFP, GFP-Rab21, or GFP-Rab21GTP using Lipofectamine 2000 as described in the Immunoprecipitations section. 48 h after transfection, the cells were harvested and fractionated on a sucrose density gradient and analyzed by Western blotting as described previously (). The α2-integrin COOH-terminal tail (28 residues) Gal4 DNA binding domain fusion (pGBKT7 vector) was used to screen a mouse E17 Matchmaker cDNA library (CLONTECH Laboratories, Inc.) as described previously (). In yeast mating tests, pGADT7-Rab21 (95–222) prey was transformed in Y187 host strain and cytoplasmic tails of α2- and α11-integrin (pGBKT7-α2 and -α11) and their variants in AH109 host strain. Point mutants were generated with the QuikChange Site-directed mutagenesis kit and confirmed by sequencing. The negative and positive controls in yeast mating tests were pGBKT7-53/pGADT7-T and pGBKT7/pGADT7, respectively. MDA-MB-231 cells (American Type Culture Collection) were grown in DME + 1% nonessential amino acids and 10% FBS. Saos-2, HeLa, HT1080, and HEK293T cells (American Type Culture Collection) were grown in DME + 10% FBS, and PC3 cells (American Type Culture Collection) were grown in F12 medium + 10% FBS. CHO cells (American Type Culture Collection) were grown in MEM Alpha Medium + 5% FBS. Saos-2 cells express no endogenous α2 (). Stable Saos-2 cells expressing equal levels of chimeric integrins (extracellular domain of α2 fused with α1 or α5 cytoplasmic tails; ) have been described (). Two different annealed siRNAs targeting Rab21 (sense, ggcaucauucuuaacaaagtt and ggucaagagagauuccaugtt; Ambion) or scramble control siRNA (Silencer Negative control #1 siRNA; Ambion) were transfected at a 100-nM concentration to MDA-MB-231 or PC3 cells using Oligofectamine (Invitrogen) according to the manufacturer's protocol (48-h culture). pSilencer 4.1-CMV hygro vector (Ambion) was used to express shRNAs. Annealed DNA oligos (Scr sense strand, gatcccgcgaatcctacaagcgcgcttgatatccggcgcgctttgtaggattcgttttttccaaa; Rab21 sense strand, gatccggtcaagagagagettccatgttcaagagacatggaatctctcttgacctga) were ligated to the vector between BamHI and HindIII sites. Plasmids were verified by sequencing. shRNA plasmids and transfected into MDA-MB-231 cells using Lipofectamine 2000 (Invitrogen), and stable cell clones were generated with hygromycin selection. Cells were plated on acid-washed glass coverslips coated with 5 μg/ml collagen type I, allowed to adhere for 1 h, washed in PBS, and PFA fixed. After permeabilization (PBS/0.02% saponin/10% FBS, 15 min), cells were stained with primary antibodies (in the same buffer) for 1 h at RT. After three washings, Alexa 488–, Alexa 555–, or Alexa 647–conjugated secondary antibodies were added (in the same buffer). Slides were examined using an inverted fluorescence microscope (Carl Zeiss MicroImaging, Inc.) or a confocal laser-scanning microscope (Axioplan 2 with LSM 510; Carl Zeiss MicroImaging, Inc.) equipped with 100×/1.4 Plan-Apochromat oil-immersion objectives. Confocal images represent a single z section of ∼1.0 μm. β1-Integrin and transferrin internalization were studied as described previously (). A multilaser microscope (IX81; Olympus) equipped with a 488-nm TIRF condensor and a 60×/1.4 Plan-Apochromat oil-immersion objective was used for TIRFM. TIRFM was combined with conventional widefield epifluorescence microscopy and time-lapse series (frame rate ∼2/s) Widefield images were pseudocolored red and TIRFM images green. Transiently transfected GFP-Rab21 cells were plated on acid-washed glass-bottomed dishes (MatTek Corporation) coated with 10 μg/ml collagen type I and allowed to adhere for 1 h before microscopy. Clear medium with 2.2 g/l NaHCO3 was used for imaging in heat (37°C) and CO (5%) stable environment box. The Axioplan 2 microscope equipped with Plan-Apochromat 63× (NA 1.4) objective and a camera (Orca 2; Hamamatsu Photonics) was used for widefield epifluorescence time-lapse imaging at a rate of 2 frames/s. GFP-Rab21 and its mutant variants (or Rluc-Rab21 and GFP–α2-integrin in the cotransfection studies) were transfected to MDA- MB-231 adenocarcinoma cells. Clear DME 4500 supplemented with 1% -glutamine, 0.5% BSA, and 30 mM Hepes was used as imaging medium. Microscopy was performed in a heat-stable environment for no longer than 1 h. MetaMorph imaging software (Universal Imaging Corp.) was used in image analysis. Fluorescence intensities for TIRFM and widefield epifluorescence microscopy were measured and analyzed with MetaMorph software. Vesicle intensities from time-lapse series were background corrected in each time point with the formula (I − I) × (A/[A − A]), where I is integrated intensity for region area A. B stands for background and V for vesicle. Region for vesicle (A) was created just around vesicle and region for background (A) just around the vesicle region. Results from two groups were compared using a test, and statistical significance was set at P < 0.05. Fig. S1 shows the cellular localization of endogenous β1-integrin, caveolin-1, Rab21, and EEA1 in MDA-MB-231 cells expressing GFP-Rab5 or -Rab21 and internalization of β1-integrin antibody and labeled transferrin in GFP-Rab5– and GFP-Rab21–expressing cells. Fig. S2 shows that overexpression of Rab21 does not influence the traffic of labeled transferrin in cells or the adhesion of cells to matrixes other than type I collagen. Fig. S3 shows the localization of organelle markers on the sucrose gradient–fractionated GFP-Rab21–expressing HeLa cells and immunogold electron micrographs of GFP-Rab21–positive structures in MDA-MB-231 cells. Table S1 demonstrates the association of Rab21WT and its variants with α/β1-integrin heterodimers in HT1080 cells. Video 1 shows MDA-MB-231 cells expressing GFP-Rab21, adhering to collagen recorded on GFP channel. Video 2 shows a combined widefield epifluorescence and TIRFM analysis of MDA-MB-231 cells expressing GFP-Rab21, adhering to collagen. Video 3 shows a combined widefield epifluorescence and TIRFM analysis of MDA-MB-231 cells expressing GFP-Rab21GDP mutant, adhering to collagen. Video 4 shows MDA-MB-231 cells cotransfected with GFP–α2-integrin and Rluc-Rab21WT, adhering to collagen recorded on GFP channel. Video 5 shows MDA-MB-231 cells transfected with GFP–α2-integrin alone adhering to collagen recorded on GFP channel. Online supplemental material is available at .

Morphogenesis and function of epithelial tissues involve the coordinated regulation of proliferation, apoptosis, differentiation, and migration. In many cases, these physiological processes are orchestrated by a combination of signals from the ECM through integrins and soluble factors including steroid or peptide hormones and growth factors (). One tissue that has been used to understand the molecular basis of epithelial differentiation is the mammary gland. This tissue develops in a temporal and spatially regulated manner so that the epithelial cells only produce their differentiation products, such as milk proteins, at the right time and place (i.e., during lactation and in cells that are spatially restricted to acini). Although endocrine signals such as prolactin (Prl) control differentiation in a temporal fashion, adhesion to basement membrane (BM; a specialized form of the ECM) is also required for lactation. Thus, to respond to the biological requirements of the organism, the epithelial cells need to integrate signals from both soluble factors and the ECM. Our laboratory has used the mammary gland system as a paradigm to dissect the molecular basis of signal integration by soluble factors and ECM, and, in the present study, we demonstrate a novel and key role for Rho family GTPases. The ECM control of mammary epithelial cell (MEC) differentiation occurs at two distinct levels. First, matrix specificity is critical because the BM protein laminin-1 supports Prl-dependent activation of the Jak2–Stat5 signaling pathway and the transcription of Prl- and Stat5-regulated milk protein genes (e.g., β-casein), whereas adhesion to the stromal protein collagen I does not (). Second, β1 integrins are actively required for Prl signaling both in culture and in vivo because function-perturbing anti–β1 integrin antibodies block MEC differentiation (), a dominant-negative (DN) β1 integrin transgene compromises Stat5 activation and milk production (), and Prl cannot activate Stat5 in β1 integrin–null MECs (). Thus, integrins regulate Stat5 transcription factor activation and expression of tissue-specific genes, but the mechanism underpinning the requirement for adhesion receptors is not yet known. Rho GTPases are good candidates to relay the adhesion-mediated signals provided by integrins. These enzymes are molecular switches that are turned on by guanine nucleotide exchange factors and have a broad function in cell division, survival, migration, and polarity (). They coordinate various cellular responses through specific effector proteins to regulate focal adhesion complexes, cell–cell junctions, actin dynamics, and the generation of reactive oxygen species (; ; ), but their role in differentiation and gene expression has not been studied widely. Because Rho GTPases can affect the activity of receptors within the plasma membrane (e.g., epidermal growth factor receptor; ), we reasoned that they might provide a mechanistic link to integrate ECM and Prl signals and, thus, control epithelial cell differentiation. Rho GTPases have a role in the morphogenesis and differentiation of some cell types; for example, Rac and Cdc42 regulate lumen formation in endothelial capillaries, the establishment of apical-basal polarity and tubulogenesis in kidney epithelia, and keratinocyte terminal differentiation (; ). In the mammary gland, Rho GTPases have been studied in cancer cells, where it has been shown that Rac1 and Cdc42 mediate motility, whereas Rho is important for the tubulogenesis of T47D cells. Rac1 also influences survival through nuclear factor κB in transformed HMT-3522 cells, and Rac1B contributes to the genomic instability of breast cancer (; ; ; ). In this study, we uncover a key role for Rac1 in the differentiation of normal, untransformed MECs. We have demonstrated that laminin and β1 integrins are essential for Prl signaling and milk protein gene expression and now show that Rac1 provides a mechanism for their integration. This study is the first to demonstrate the involvement of Rho family GTPases in the expression of tissue-specific genes during the process of glandular epithelial differentiation. Cultured mammary epithelia organize themselves into 3D acini when they are plated onto reconstituted BM matrix (Matrigel) and secrete milk proteins into the inner lumina when stimulated with lactogenic hormones (). They are morphologically and functionally similar to lactating acini in vivo (). To analyze the effects of Rho GTPases on milk protein synthesis and, thus, lactational differentiation, primary MECs were infected with adenoviruses expressing DN forms of RhoA, Rac1, and Cdc42 (Ad-mycN19RhoA, Ad-mycN17Rac1, or Ad-mycN17Cdc42) or a control adenovirus (Ad–β-galactosidase) for 1 h in suspension at an MOI of 50 (which yielded >90 ± 5% infection). Infected cells were plated on BM matrix to assemble into 3D acini for 24 h and stimulated with Prl for a further 24 h. Prl induced the synthesis of β-casein, a mammary-specific differentiation marker, in uninfected cells; however, synthesis was either completely abolished or substantially decreased in cells expressing N17Rac1 or N17Cdc42, respectively (). In contrast, β-galactosidase and N19RhoA had no discernable effects on β-casein synthesis even though it did prevent stress fiber formation (unpublished data). The prevention of milk protein synthesis was confirmed by immunofluorescence staining (). The effects of N17Rac1 and N17Cdc42 were not caused by apoptosis (Fig. S1, available at ). These data indicate that DN Rac1 and DN Cdc42 specifically block milk protein synthesis, suggesting that these GTPases have a key role in regulating epithelial cell differentiation. Prl induced Stat5 tyrosine phosphorylation in 3D acini () and nuclear translocation of Stat5a in ∼75% of monolayer cells treated in the overlay assay (). In virally infected cells, N17Rac1 and N17Cdc42 but not control β-galactosidase abolished Stat5 phosphorylation and reduced its nuclear translocation by 70–80%. N19RhoA did not affect Stat5 phosphorylation, but, interestingly, it slightly inhibited Stat5a nuclear translocation compared with uninfected controls. To determine whether Rac1 or Cdc42 is required for signaling upstream of Stat5, we examined Prl receptor (PrlR) tyrosine phosphorylation. PrlR has no intrinsic kinase activity, and Jak2 phosphorylates its tyrosine residues in response to ligand binding. Although N17Rac1 and N17Cdc42 had no effect on the total levels of PrlR, both substantially diminished PrlR phosphorylation by its ligand, Prl (). Rac activity was dependent on cell interactions with ECM because it was unable to bind PBD efficiently in suspension (), as seen in other cell types ().Rac activity was not dependent on the milieu of lactogenic hormones (). In contrast, Cdc42 did not require adhesion to the BM to be active but instead was sensitive to the hormones, hydrocortisone, insulin, and Prl (). To establish whether Prl can regulate Cdc42 independently, cells cultured in the continuous presence of hydrocortisone and insulin were stimulated with Prl for 15 min. A twofold increase in Cdc42 activity was observed in Prl-treated acini (). Because Rho GTPases regulate the actin cytoskeleton, cell–cell adhesion, cell polarity, and focal adhesion dynamics, it is possible that they contribute to mammary differentiation through one or more of those mechanisms. To determine their role in 3D acinar configuration, we investigated the effects of N17Rac1 and N17Cdc42 on the distribution of basal (β1 integrin) and lateral (E-cadherin) proteins as well as apical markers (actin and ZO-1, which localizes at the apical region of lateral surfaces). Small acini ranging between 15 and 25 μm in diameter were analyzed because these polarize within 2 d of plating onto BM matrix and lumen formation does not require apoptotic clearance of cells, which is in contrast to those that form larger acini (). In primary cultures, acini form by reorganization of cells rather than proliferation from a single cell. Stacks of fluorescent images were deconvolved to reveal protein localization within the center of the acini, where lumina are visible, and are shown as individual () or multiple acini (Fig. S2, available at ). Polarized control acini were contained as a single layer of cells surrounding a hollow lumen (; and Fig. S2, A and B). β1 integrins were localized to the basal domain of each cell, which faced the exterior and contacted the matrix (, and G; and Fig. S2, E and F). E-cadherin was located at intercellular junctions (). The apical domain was opposed to the lumen, where ZO-1 tight junctions () and a concentration of actin structures assembled (). Small lumens were visible in 2-d-old acini (), but these expanded during extended culture of >4 d (), and milk proteins were secreted into the lumen when stimulated with Prl (). Acini expressing N17Rac1 and N17Cdc42 lost their polarized structure, and nuclei filled the luminal space. No visible lumina were evident by phase-contrast microscopy in acini expressing N17Rac1 or N17Cdc42 compared with controls (; and Fig. S2, C and D). N17Rac1-expressing acini displayed disorganized apical F-actin in 89% of cases () and ZO-1 in 92% of acini (); E-cadherin was disrupted at intercellular regions in 68% of acini (). However, β1 integrins were still retained at the basal plasma membrane, with perturbation in only 22% of cases ( and Fig. S2 G). Similar to N17Rac1, the arrangement of F-actin () and ZO-1 ( and Fig. S2 H) in N17Cdc42-expressing acini was disrupted in 78 and 72% of cases, respectively; E-cadherin was only partially disrupted at intercellular junctions (42%; ). However, in contrast to N17Rac1, the basal distribution of integrins was perturbed in 80% of acini ( and Fig. S2 H). To establish whether altered acinar organization is a mechanism by which these GTPases affect differentiation, we inhibited Cdc42 and Rac1 function in acini after they had formed and become polarized. Under these conditions, F-actin still localized apically and small lumens were present (; long arrow), although some cells were visible in the lumens (; short arrows). In contrast to infection before acinar formation (), N17Cdc42 expression in postpolarized acini did not disrupt the discrete basal distribution of β1 integrins (), and, importantly, it did not block β-casein synthesis, as shown by immunofluorescence () and blotting (). This indicates that N17Cdc42 blocked mammary differentiation in the previous experiments () primarily by interfering with the organization of acini. Therefore, we did not further examine the role of Cdc42 in this study. We reasoned that Rac1 operates through a distinct mechanism from Cdc42 to influence differentiation because β1 integrin was retained at the basal acinar surface in the presence of N17Rac1 ( and Fig. S2 G). Direct infection of polarized 3D acini with Ad-mycN17Rac1 also prevented β-casein synthesis () while leaving acinar structure relatively intact (). This indicates that Rac regulates Prl signaling downstream of integrins and that its effects are not coupled to acinar organization. Adhesion of MECs to collagen I is integrin mediated, and the cells form 2D monolayers on this ECM. However, this matrix is not permissive for Prl-induced signaling even though the cells retain their differentiation potential if subsequently provided with laminin (). Therefore, we tested whether V12Rac1 could rescue this defect in cells cultured in two dimensions on collagen I (with no BM overlay). Control cells failed to differentiate in response to Prl; however, under the same culture conditions, V12Rac1-infected cells synthesized β-casein (). Moreover, V12Rac1 restored the ability of Prl to translocate Stat5a to the nucleus (). Thus, when the ECM signals do not have the specificity to allow Prl signaling, constitutively active Rac1 can bypass the defect. The mechanism for the rescue of differentiation is not caused by the induction of laminin synthesis and assembly, which can occur in MDCK cells (), because V12Rac1 did not cause any noticeable laminin deposition on the surface of MECs (unpublished data). This suggests that the effect of V12Rac1 lies downstream of integrins. Thus, we examined the effect of V12Rac1 in β1 integrin–null acini. Such acini can be generated by the Cre-mediated deletion of integrin alleles harboring LoxP sequences, an approach we used previously to demonstrate the importance of β1 integrins for Prl signaling and milk protein synthesis both in culture and in vivo (). Cells isolated from Itgβ1 mice were infected in monolayer with Ad-CreM1 virus, leading to a significant reduction in β1 integrin levels within 24 h (). The deletion of β1 integrin also abolished Rac1 activity in 3D acini on BM matrix, indicating that Rac1 is regulated by β1 integrins (). β1 integrin–null cells or control cells were then reinfected with either Ad–β-galactosidase or Ad-mycV12Rac1 and replated onto BM matrix to assemble into 3D acini for 48 h. The deletion of β1 integrin largely prevented Prl-induced synthesis of β-casein (, lane 5). However, V12Rac1 but not β-galactosidase expression in β1-null acini rescued milk protein synthesis (, compare lanes 6 with 7). In a previous study, we demonstrated that the inhibition of Prl signaling in cells cultured without BM proteins (i.e., in two dimensions on collagen I) occurs through protein tyrosine phosphatases (PTPs; ). Therefore, we reasoned that the effects of Rac1 on differentiation might correlate with the altered function of phosphatases. SHP2 binds the PrlR–Jak2 complex directly and regulates Jak–Stat signaling both positively and negatively (; ; ; ). Phosphorylation of SHP2 on Y542 triggers its phosphatase activity by binding and displacing the N-SH2 domain from the catalytic cleft (). In MECs cultured on collagen I, the phosphorylation of SHP2-Y542 was elevated considerably compared with cells cultured with BM, either as acini on BM matrix () or as 2D cultures with BM overlay (not depicted). This form of SHP2 was also detected as a slower migrating band when immunoblotted with a total SHP2 antibody (, arrow). In coimmunoprecipitation experiments, SHP2 complexed with Jak2, but the slower migrating phosphorylated form only associated with Jak2 in cells on collagen 1 (, lanes 1 and 2; arrow). This suggests that a possible mechanism for the inability of Prl to signal on this substratum is that Jak2-bound SHP2 is phosphorylated on Y542 and, therefore, is active as an inhibitory PTPase. To confirm a link between SHP2 dephosphorylation and differentiation, we examined the effect of Ad-mycV12Rac1 on SHP2-pY542 in MECs cultured on collagen with no BM overlay. Strikingly, V12Rac1 inhibited the phosphorylation of this residue () and prevented phospho-SHP2 from complexing with Jak2 (). Thus, the ability of V12Rac1 to rescue the differentiation defect in cells cultured on collagen I () correlates with SHP2-pY542 dephosphorylation. #text Primary mouse MECs were isolated from 15.5–17.5-d pregnant ICR mice and cultured in F-12 medium supplemented with 10% FBS, 5 μg/ml insulin, 1 μg/ml hydrocortisone, and 5 ng/ml EGF (growth medium) and were plated on collagen I–coated dishes in growth medium to allow them to dissociate from in vivo acini and form cell monolayers. For some experiments, cells were isolated from Itgβ1 transgenic mice (). Differentiation medium consisted of DME–Hams F-12 medium containing insulin and hydrocortisone with or without 3 μg/ml Prl. Mouse anti–β-casein, rabbit anti–β1 integrin, and rabbit anti-PrlR antibodies have been described previously (). Commercial primary antibodies were as follows: Cdc42, Stat5a, Erk, and SHP2 (Santa Cruz Biotechnology, Inc.); Rac (23A8), phosphotyrosine (4G10), phospho-Stat5 (Y694/Y699), and Jak2 (Upstate Biotechnology); myc (9E10; Roche); β-galactosidase (Promega); SHP2 (Y542), paxillin, and FAK (Cell Signaling Technology); paxillin (Y31) and FAK (Y397; Biosource International); E-cadherin (ECCD2; Takara Bio Inc.); ZO-1 and Cre (Chemicon); and calnexin (Bioquote). Secondary antibodies were HRP-conjugated anti–mouse and anti–rabbit IgG, Cy2-conjugated anti–mouse and anti–rat IgG, Cy5-conjugated anti–mouse and anti–rabbit IgG, and conjugated rhodamine RX (Jackson ImmunoResearch Laboratories). Amino-terminal myc-tagged N19RhoA, N17Rac1, N17Cdc42, V12Rac1, and Ad–β-galactosidase adenoviruses were gifts from A. Ridley (Ludwig Institute for Cancer Research, London, United Kingdom; ), and Ad-CreM1 was obtained from Microbix Biosystems. Recombinant adenoviral DNA was amplified in E1-competent 293T human embryonic kidney cells. Infected cells were harvested, and viable adenoviral particles were purified on a caesium chloride gradient as previously described (). Adenovirus titre was determined by scoring cytopathic effect in 293T cells using the tissue culture infectious dose 50 method. Primary cells cultured on collagen I dishes for 3 d were trypsinized, and single cells in suspension were infected (for 1 h at 37°C) and plated onto either collagen I (7 × 10 cells/cm) or BM matrix dishes (2 × 10 cells/cm) saturated with growth medium as described previously (). Cells were infected with adenoviruses at an MOI of 50, which produced a >90% rate of infection. Adenoviral proteins were expressed for 24 h in cells cultured in growth medium before incubation in differentiation medium for up to 24 h. Single cells attach to the BM matrix and assemble into 3D acini over 2 d; over this time frame, only small acini form. For rescue experiments, cells were infected for 2 h in monolayer with Ad-CreM1. 24 h later, cells were trypsinized and reinfected in suspension with Ad–β-ga1actosidase or Ad-mycV12Rac1 viruses and replated onto BM matrix. For infection directly in three dimensions, small acini (15–25 μm) were assembled by plating 8 × 10 cells per 35-mm dish onto BM matrix for 48 h. Acini were infected by incubating with 200 MOI viruses for 3 h in serum-free medium. For milk protein analysis, Prl was added for 24 h in differentiation medium. Expression and distribution of various proteins were visualized by indirect immunofluorescence. 48 h after plating, cells were fixed for 10 min in PBS/4% (wt/vol) PFA and permeabilized for 7 min using PBS/0.2% (vol/vol) Triton X-100. Nonspecific sites were blocked with PBS/10% goat serum (for 1 h at RT) before incubation with antibodies diluted in PBS/2% goat serum (for 1 h at RT each). F-actin was detected by incubating cells with TRITC-phalloidin (Sigma-Aldrich) or Texas red phalloidin (Invitrogen) for 1 h at RT, whereas nuclei were stained using 4 μg/ml Hoechst 33258 (Sigma-Aldrich) for 2 min at RT. Cells were washed in PBS before mounting in either DAKO (DakoCytomation) for monolayers or prolong antifade (Invitrogen) for 3D acini. Immunostained cells were visualized with a microscope (Axioplan2; Carl Zeiss MicroImaging, Inc.) using plan-Apochromat 100× and 63× NA 1.40 lenses with Immersol 518F oil at RT. 0.2-μm z sections were captured with a camera (Orca ER; Hamamatsu) and analyzed with Volocity 3.5.1 software (Improvision) using iterative deconvolution. Quantification of acini displaying a particular phenotype was scored by analyzing 100 acini for each condition. Where approximately half the cells within an acinus displayed a change in morphology, these acini were scored as positive for that change. Nonbiased cell counts were performed by concealing the identity of each slide. Milk proteins and Stat5 phosphorylation were analyzed as described previously (). For PrlR and Jak2 immunoprecipitations, cells were scrape lysed in 2× NP-40 lysis buffer for BM matrix or 1× for monolayer (10% [wt/vol] glycerol, 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 1% [wt/vol] NP-40, 2 mM MgCl, and fresh protease/phosphatase inhibitors). Lysates were rotated for 30 min at 4°C, homogenized, and ultracentrifuged at 70,000 for 15 min at 4°C to remove both detergent-insoluble proteins and the dense part of the BM matrix. Antibodies and protein A–Sepharose beads were incubated with lysates for 2 h each at 4°C. Primary cells were plated onto collagen I–coated coverslips in growth medium for 48–72 h. Adenoviruses at an MOI of 50 were added to attached cells for 2 h in growth medium. Virus-containing medium was removed, and cells were overlaid with diluted BM matrix (1:50) in differentiation medium for 48 h followed by stimulation with Prl for 15 min. Nuclear translocation of Stat5a was detected by immunostaining, and nonbiased cell counts were performed by concealing the identity of each slide. Cells plated on factor-reduced BM matrix or collagen I were rapidly scrape lysed into 2× or 1× NP-40 lysis buffer and ultracentrifuged at 70,000 for 15 min at 4°C. Cells in suspension were pelleted before lysing in an equivalent volume of NP-40 lysis buffer. 25 μg of recombinant GST alone or GST-PAK PBD purified from the bacterial expression vector system and coupled to glutathione agarose beads was used to precipitate GTP-bound Rac and Cdc42 from cell lysates for 40 min at 4°C (). Active Rac and Cdc42 were detected by immunoblotting with anti-Rac or anti-Cdc42 antibodies and quantified using ID Software (Kodak). Fig. S1 shows that the inhibition of differentiation by N17Rac1 and N17Cdc42 is not caused by apoptosis. Fig. S2 presents multiple acini showing the effects of N17Rac1 and N17Cdc42 on acinar morphology. Online supplemental material is available at .

Tetraspanins are medium-sized (∼250 amino acids) membrane proteins that contain cytoplasmic NH and COOH termini and two extracellular domains separated from each other by a short inner loop. The mammalian family of these evolutionarily conserved proteins contains 32 members. Tetraspanins are expressed in a wide range of tissues and cell types, and members of this protein family have been implicated in regulating various biological functions, including antigen presentation, cell adhesion and migration, cell–cell fusion, cell activation, and proliferation (for reviews see ; ; ; ; ; ). Tetraspanins associate specifically with distinct integrins, various Ig superfamily members, and other tetraspanins, thus establishing a scaffold for various cellular functions. Numerous biochemical analyses and functional studies predicted the existence of tetraspanin-enriched microdomains (TEMs) that together form the so-called tetraspanin web (; ). TEMs are thought to organize the plasma membrane and intracellular membranes, where some tetraspanins are primarily located, by selectively concentrating specific membrane proteins and membrane-peripheral signaling molecules. Such TEM-based concentration/exclusion of proteins involved, for example, in adhesion or in intracellular signaling is thought to dynamically segregate molecules, similar to how lipid rafts are proposed to laterally organize cellular membranes. Although some studies have documented colocalization of individual tetraspanins with various membrane receptors and costimulatory molecules, e.g., in adhesion complexes (; ), the concept that different members of the tetraspanin family associate at membranes, thus forming distinct microdomains, is based largely on coimmunoprecipitation and protein cross-linking data. Neither the mean size of TEMs nor their overall distribution at the plasma membrane of these microdomains has been determined. Human immunodeficiency virus type 1 (HIV-1), like other enveloped viruses, exits from cells by budding through membranes, a process that does not lead to disintegration of the cell. For its budding, HIV-1 uses the host cell machinery that is responsible for the formation of intralumenal vesicles in multivesicular bodies (MVBs), components of the endosomal compartment (for review see ). Nevertheless, HIV-1 primarily buds through the plasma membrane of T lymphocytes and other cell types. Only in macrophages is HIV-1 known to bud exclusively into MVBs. Viruses sequestered in these late endosomes (LEs) are thought to exit from macrophages upon fusion of the limiting membrane of MVBs with the plasma membrane (; ). HIV-1 produced in macrophages specifically incorporates the tetraspanin CD63, compatible with the finding that this antigen largely resides at the limiting membrane and on intralumenal vesicles of LEs/MVBs (; ; ; ; ). However, despite its low abundance at the plasma membrane of cells, CD63 is also specifically incorporated into HIV-1 particles produced in primary and transformed T lymphocytes and in other nonmacrophages where this virus buds mainly through the cell cortex (; ). Furthermore, we previously reported that we occasionally observed colocalization of HIV-1 Gag and CD63 at the periphery of T lymphocytes and melanocytes, though it was difficult to distinguish with certainty between the small fraction of CD63 associated with the plasma membrane and the vast majority of this antigen residing on intracellular membranes (). Here, we tested the hypothesis that the small fraction of CD63 that resides at the plasma membrane is concentrated at distinct microdomains, i.e., in TEMs, together with other members of the tetraspanin family. Fluorescence microscopy analysis of cells in which we selectively stained the surface fractions of tetraspanins combined with EM of immunolabeled sheets of plasma membrane allowed us to visualize TEMs containing the tetraspanins CD9, CD81, CD82, and CD63. Our data also document the size and distribution of these TEMs in membranes. Furthermore, the data presented in this paper suggest that these newly visualized plasma membrane microdomains can function as exit gateways for HIV-1. Although the tetraspanin CD63 is known to cycle via the cell surface, at steady state, the vast majority (>98%) of this antigen resides in perinuclear LEs/MVBs of HeLa cells (). The overall distribution of CD63 is thus identical to what is seen in melanocytes and T lymphocytes () and other cell types (; ). To examine specifically the cell surface distribution of CD63, we incubated HeLa cells with the antibody H5C6, whose epitope is located either in the small or in the large extracellular loop (EC1 or EC2, respectively) of this tetraspanin (). After fixation and incubation with a secondary, fluorophore-conjugated antibody, cells were analyzed by widefield fluorescence deconvolution microscopy (). Such selective staining of surface CD63 in live cells revealed that the small fraction of this antigen that resides at the plasma membrane, rather than being diffusely distributed, clusters in discrete microdomains. To exclude the possibility that CD63 clustering was due to antibody cross-linking, cells were also stained after fixation in some experiments. visualizes that discrete CD63 assemblages exist at the cell surface before antibody binding. These microdomains are not restricted to the adhesion structures but are present in the entire cell periphery, as shown in (bottom). Video 1 (available at ) shows a 3D-rendering of a cell surface stained for CD63 and visualizes that clusters of this antigen are equally abundant in those areas of the cell surface that are not contacting other cells or the plastic of the tissue culture dish. documents that the plasma membrane of an individual cell contains several hundred CD63-enriched microdomains (mean number of domains per 25 μm = 18 ± 4). Through biochemical studies, CD63 is known to associate with other members of the tetraspanin family (; ). We were interested in investigating the localization of CD9, CD81, and CD82, three tetraspanins that were previously shown to coprecipitate with CD63 in preB cells () and two of which (CD81 and CD82) were shown to associate with HIV-1 produced in MVBs of macrophages (). First, we sought to compare the overall distribution of these tetraspanins. Cells were incubated with the respective antibodies after permeabilization. Compared with CD63, the three other members of the tetraspanin family showed significantly higher surface expression ( and Fig. S1 A, available at ). Although some colocalization at intracellular membranes was observed for these tetraspanins (Fig. S1 B), at steady state, substantial fractions of those antigens apparently reside at different intracellular membranes. Next, cells were stained for the plasma membrane fraction of the selected tetraspanin, as described in , and the relative colocalization of surface CD63 with CD9, CD81, or CD82 was assessed. Substantial colocalization of CD63 with any one of the three tetraspanins was observed (Fig. S2, available at ), suggesting that relatively large fractions of CD9 and CD82 reside in plasma membrane TEMs together with CD63, whereas a small fraction of total CD63 contributes to these surface microdomains. The concept that various tetraspanins contribute dissimilarly to the formation of TEMs is based on biochemical studies only (for review see ). The data presented in and Fig. S2 support that concept and provide an unprecedented visualization of bona fide surface TEMs. At the plasma membrane, CD63 typically colocalizes with one or more of the other three tetraspanins (Fig. S2). Also, CD9, CD81, and CD82 partially colocalize with each other (unpublished data). It follows that a subset of the TEMs visualized in Fig. S2 contains at least three of the analyzed tetraspanins. In , we demonstrate that this is indeed the case. A detailed visual inspection of the three triple stainings revealed that more than two thirds of TEMs contain at least two of the three analyzed tetraspanins and about half of them contain all three antigens. Although biochemical analyses (i.e., coprecipitation experiments) allow determination of how much an individual tetraspanin contributes to an average TEM, such analyses of the bulk population do not permit determination of whether those relative contributions are uniform, i.e., whether tetraspanins accumulate at TEMs in defined ratios. The analysis of triple immunostainings, however, can provide such information. The magnifications in (right) and visualize that the composition of TEMs at steady state varies widely, and the statistical analysis shown in documents that the tetraspanins analyzed in this study do not exist in TEMs in distinct, uniform ratios. Note that TEMs containing only one or two of the three analyzed tetraspanins were excluded from this quantification. TEM heterogeneity may thus be even larger than suggested by the graph shown in . The overall presence of CD63 at the limiting membrane of MVB and on intraluminal vesicles of MVBs has been well documented previously by immuno-EM in various cell types (; ). We used immuno-EM to analyze CD63-enriched TEMs at the plasma membrane. To visualize the lateral arrangement of these domains with higher spatial resolution and to determine the approximate size of these domains, we used a technique that allows an en face view of the cytosolic face of the plasma membrane and that preserves submembrane structures such as clathrin lattices and cortical cytoskeleton (; ). After incubation with the anti-CD63 antibody 1B5, HeLa cells, transfected or not with a CD63-YFP expressor plasmid, were allowed to sediment on poly--lysine–coated grids. Subsequently applied hypotonic shocks, followed by sonication, disrupted the cells and removed the cytoplasm. The plasma membrane, still adherent to the grid, was processed for visualization by EM. shows a representative micrograph. It confirms the fluorescence microscopy data ( and ) by visualizing that CD63 expression at the plasma membrane is restricted to distinct small islands that are separated from each other by membrane that is virtually free of this antigen. The size and shape of the CD63-enriched domains, as well as the distance between these microdomains, vary considerably (mean area = 0.2 μm; mean distance between CD63-enriched domains = 0.65 μm; see ). No significant differences were observed between CD63-enriched microdomains in cells expressing only the endogenous version of this tetraspanin () and cells expressing also CD63-YFP (not depicted). Interestingly, many CD63 molecules appear to be situated along cytoskeletal elements and adjacent to clathrin lattices and/or clathrin-coated pits. As shown in by immunofluorescence, despite displaying an overall distribution that differs dramatically from the distribution of CD63, CD9 shows substantial colocalization with CD63 in surface TEMs. We thus sought to confirm the features of surface TEMs revealed in by analyzing CD9 clusters at the plasma membrane. Live cells were incubated with the anti-CD9 antibody KMC8.8, and CD9 localization was visualized by EM () as described for CD63. CD9-containing TEMs display the same characteristics with regard to size and their 2D distribution within the plasma membrane as those enriched in CD63. Also, as expected given the fluorescence microscopy data displayed in and Fig. S2, double labeling of cells with antibodies against CD9 and CD63 reveals microdomains containing one or both tetraspanins (, compare C and D, displaying separate CD9 and CD63 clusters and colocalizing tetraspanins, respectively). As for the analysis of TEM distance (see Materials and methods), the relative colocalization of the two antigens was quantified by positioning a grid consisting of squares with 200-nm length onto the images. CD9- and CD63-associated gold particles residing in the same or adjacent squares were scored as colocalizing. This analysis revealed that about a third (29.7 ± 4.7%; = 20 micrographs) of the CD63-associated gold particles are situated proximal to CD9, in good agreement with the data shown in Fig. S2. A recent ultrastructural study analyzed the surface distribution in HAb2 fibroblasts of the influenza viral envelope glycoprotein HA. HA, a membrane protein that is sometimes used as marker for fluid-ordered raft domains, was demonstrated to cluster at distinct plasma membrane microdomains if expressed independently or together with the other viral proteins (; ). To investigate the spatial relationship between HA and the tetraspanins CD9 and CD63, HeLa cells expressing HA were surface stained with the respective antibodies. (left) documents the 2D distribution of HA at the plasma membrane of these cells. As in HAb2 cells, HA is restricted to discrete microdomains of variable size. Although the overall appearance (size and distribution) of HA clusters resembles that of CD9- and CD63-containing TEMs (, middle), very little colocalization (<10%) was detected between HA and either CD9 or CD63 (right). Despite its low abundance at the plasma membrane of cells, the LE marker CD63 is one of few cellular antigens that are specifically enriched in HIV-1 particles even if this virus is produced in cells where newly assembled particles bud primarily through the plasma membrane (; ). We thus reasoned that HIV-1 Gag, the major structural protein of this virus, must assemble at areas of the plasma membrane that are enriched in CD63. The results shown in demonstrate that this is indeed the case. In HeLa cells transiently transfected with a plasmid carrying the HIV-1 provirus and giving rise to the expression of the complete set of viral proteins and subsequent particle release, a large fraction of viral Env at the plasma membrane colocalizes with surface CD63 (), supporting the hypothesis that HIV-1 Gag is specifically accumulating at these microdomains. We previously showed that, in a melanocyte cell line, about one third of viral Gag, the major viral structural protein, colocalizes with CD63 (). However, unlike in the current analysis, in those studies we looked at total Gag and total CD63 and thus were visualizing primarily LE/MVB-hosted CD63. As documented in Fig. S3 (available at ), many of these CD63-enriched surface TEMs must also contain other tetraspanins because significant proportions of viral Gag also colocalize with the tetraspanins CD9, CD81, and CD82. To confirm that viral Gag is indeed colocalizing with CD9- or CD63-containing TEMs at the plasma membrane and not, for example, at tetraspanin-enriched endosomes or MVBs situated right beneath the plasma membrane, we used the same EM technique as in . To achieve high expression levels for the viral proteins, we cotransfected cells with Gag and Env expressor plasmids, which gives rise to the production of virus-like particles (VLPs). confirms that tetraspanin surface expression is limited to distinct, often cytoskeleton-associated plasma membrane microdomains, as shown also in . Further, it documents that Gag can indeed associate with surface assemblages of either tetraspanin analyzed here. Colocalization analysis showed that 21.1 ± 5.0% of Gag colocalized with CD63-containing TEMs and 41.8 ± 4.8% of Gag colocalized with CD9-containing TEMs ( = 27/29 micrographs, respectively; TEMs were defined as in ). To determine whether viral particle morphogenesis and egress are likely to take place at surface TEMs, we analyzed if TSG101 and VPS28, constituents of endosomal sorting complex required for transport (ESCRT1) and as such known to be essential components of the HIV-1 particle–forming machinery (; ; ; ; ), are recruited to Gag-containing TEMs. Cells transfected with expressor plasmids for YFP-tagged TSG101 (TSG101-YFP) and CFP-tagged VPS28 (VPS28-CFP) and cotransfected with a HIV-1 Gag expressor plasmid, which leads to the formation and release VLPs even in the absence of Env expression, were analyzed for the localization of TSG101 and VPS28. A strong shift toward surface localization was immediately apparent for both ESCRT1 components ( and Fig. S4, available at ), similar to what was previously observed in 293T cells producing HIV-1 (). Upon closer examination, it became evident that TSG101 and VPS28 are recruited to small but distinct Gag-containing patches at the plasma membrane (). To evaluate whether these Gag-containing microdomains are surface TEMs, cells coexpressing HIV-1 Gag, TSG101-YFP, and FLAG-tagged VPS28 were double stained with anti-Gag and anti-CD63 antibodies and were analyzed for the localization of TSG101, Gag, and CD63. As demonstrated in , in cells expressing HIV-1 Gag and thus producing VLPs, a significant fraction of the ESCRT1 component TSG101 relocates to the surface, where it colocalizes with surface CD63, strongly suggesting that surface TEMs can serve as sites for the production and the release of viral particles. Finally, a similar recruitment to the plasma membrane and colocalization with viral Gag was found for VPS28, another ESCRT1 component required for HIV-1 budding (unpublished data), providing further evidence for the notion that surface TEMs can provide exit gateways for HIV-1. The data shown in and suggest that a considerable fraction of HIV-1 Gag, which can form VLPs if expressed in the absence of other viral components, is targeted to CD63-containing surface TEMs. However, as previously noted (), Gag's colocalization with membranes carrying the late endosomal markers major histocompatibility antigen type II and/or CD63 is increased if it is expressed in the context of a full-length virus. We thus sought to quantify the colocalization of Gag and Env with tetraspanins if the viral proteins were expressed individually or together. and Fig. S5 (available at ) document that very little Env and relatively modest amounts of Gag colocalize with CD9- and CD63-containing surface TEMs if either of the viral antigens is expressed alone. Colocalization of viral Gag with the two tetraspanins is barely augmented in cells that coexpress viral Env, whereas clearly more Env colocalizes with the two tetraspanins in cells cotransfected with plasmids carrying the gene. Note that TEM association of either viral component is most pronounced if Env and Gag are expressed from full-length provirus. Importantly, and in further support of the hypothesis that TEMs can function as egress sites, we found that if we restricted our analysis to the subpopulation of Gag that colocalizes with Env, an even higher fraction of Gag was found to accumulate at surface TEMs (almost 50% colocalization with CD63-containing TEMs and almost 70% colocalization with CD9-containing TEMs). Though nonlymphocytes such as HeLa and 293T cells are model systems in which many of the cellular components of the HIV-1 assembly/release functions were identified and characterized, it was necessary to determine whether viral release is routed though surface TEMs in T lymphocytes. However, as in HeLa cells, the overwhelming majority of CD63 at steady state resides in LEs/MVBs of Jurkat T lymphocytes (). Live, nonpermeabilized cells thus had to be stained with anti-CD63 antibodies to visualize the distribution of this antigen in the plasma membrane. (left) documents that CD63 clusters in plasma membrane TEMs that display sizes and overall distribution comparable to those analyzed in HeLa cells (). Also, as in HeLa cells, there is substantial colocalization of surface CD63 and CD9. Most important, (B and C) demonstrates that virtually all HIV-1 Env expressed at the cell surface in the context of the full-length virus colocalizes with surface CD9 and surface CD63, signifying that plasma membrane TEMs can serve as exit gateways for HIV-1 in T lymphocytes. Judging by Env localization, both nonpolarized and polarized viral release was observed, the latter probably being triggered by cell–cell contact (; ). Despite its very low abundance at the cell surface, the tetraspanin CD63 is enriched in HIV-1 particles produced in cells where this virus buds primarily through the plasma membrane. This prompted us to investigate whether CD63 is concentrated at those sites. We demonstrate that the surface fraction of CD63, together with other tetraspanins, clusters at discrete sites, thus forming surface TEMs. Besides providing a map of these surface TEMs, we also show that the structural components of HIV-1, together with elements of the host cell machinery responsible for viral budding, are recruited to precisely these microdomains. Thus, TEMs likely function as exit gateways for HIV-1. Various biochemical and functional studies have linked specific cellular processes to the association of distinct proteins and tetraspanins. However, few studies have also analyzed the localization of tetraspanins involved in these processes, and physical aspects of TEMs, particularly size and distribution of these microdomains, have remained enigmatic. We hypothesized that CD63-enriched microdomains exist at the cell surface despite the fact that this tetraspanin is a resident primarily of LEs/MVBs. documents for HeLa cells that this is indeed the case. These surface microdomains at first sight appear morphologically featureless; i.e., they are not restricted to pseudopods, for example, but they are distributed over the entire plasma membrane. The same holds for the other three tetraspanins analyzed in this study. Not only do they colocalize to a significant extent with CD63, surface TEMs containing these other tetraspanins are also spread over the entire cell surface and are not apparently engaged in any adhesion/migration-related functions (), though, as noted below, they clearly appear to be enriched in microvilli. The relative contributions of tetraspanins to the formation of TEMs was assessed by selecting fluorescence thresholds for each channel followed by counting the individual clusters of each tetraspanin to gauge the presence or absence of one, two, or all three channels in a given cluster. The relative contributions of tetraspanins to the formation of triple TEMs () was calculated by selecting TEMs containing signal for three tetraspanins and quantifying the relative intensity and area of each channel in a given TEM. Cells were washed with PBS and fixed with 3.7% paraformaldehyde for 10 min. Without permeabilization, cells were incubated for 1 h with primary antibody at 37°C in 1% BSA in PBS. Alternatively, live cells were precooled on ice and incubated for 1 h on ice at 4°C with primary antibody in complete medium supplemented with 1% Ca. Cells were then washed extensively with PBS and fixed for 10 min with 3.7% paraformaldehyde. After washing with PBS, cells were blocked for 10 min with 1% BSA in PBS and incubated with the appropriate fluorophore-conjugated secondary antibody for 30 min at room temperature. After extensive washing with PBS, cells were overlaid with 1% BSA in PBS and examined by fluorescence microscopy. Jurkat cells were washed with PBS and incubated in suspension for 1 h on ice at 4°C with primary antibody in complete medium supplemented with 1% Ca. The cells were then washed with PBS and fixed with 3.7% paraformaldehyde for 10 min. The cells were washed, blocked for 10 min in 1% BSA in PBS, and transferred to chambered coverglasses precoated with CellTak (BD Biosciences) and incubated with the secondary antibodies at 37°C, simultaneously immobilizing the cells on the coverglass. In case multiple antibodies of the same species were used, the antibodies were conjugated with a fluorophore using the Zenon technology, and the stainings were done in series. CD63- or CD9-enriched microdomains were visualized at the EM ultrastructural level on the inner face of the plasma membrane of HeLa cells as previously described (; ). In brief, HeLa cells were incubated for 1 h at 4°C with either anti-CD63 antibodies (clone 1B5 of mouse anti-CD63 IgG2b [] and antibody H5C6) or anti-CD9 antibodies (clone KMC8.8 of rat anti-CD9 [Santa Cruz Biotechnotolgy, Inc.] or K41 [Bachem]) in cold 1% BSA in PBS. After antibody binding, cells were washed twice in cold PBS to remove excess antibody and incubated with secondary species-specific gold-conjugated antibodies (10- or 20-nm colloidal gold) for 1 h at 4°C. Unbound secondary antibody was removed by washing twice with cold PBS. Cells were detached from the Petri dishes and allowed to sediment on formvar-coated EM nickel grids previously coated with poly--lysine. Adherent isolated plasma membranes were obtained by incubating the cells attached to the grids with hypotonic (0.65×) PBS for 30 s and then sonicating the cells at a weak power. This procedure disrupts the cells but allows a large portion of plasma membranes with conserved internal structures, such as clathrin-coated membranes and cytoskeleton elements, to stay adherent to the poly--lysine–coated grids. Adherent membranes were next washed with cold PBS and prefixed for 15 min at 4°C with 1% - (3-dimethylaminopropyl)-′-ethylcarbodiimide hydrochloride/0.2% glutaraldehyde, followed by 15 min at room temperature with 4% glutaraldehyde in PBS. Adherent membranes were subsequently fixed in 2% osmium tetroxide in PBS for 8 min and 30 s at room temperature, washed three times for 5 min with PBS, incubated with 1% aqueous tannic acid for 10 min, washed twice for 5 min in distilled water, incubated with 1% uranyl acetate for 10 min, and washed twice for 1 min with distilled water before air-drying. For CD63–CD9 double-labeling experiments, CD63 gold labeling was performed as described above, but CD9 was localized using a rabbit polyclonal anti-CD9 antibody (clone H-110; Santa Cruz Biotechnology, Inc.). Membranes considered well conserved were then photographed on an electron microscope (CM10; Philips). The mean size of CD63-enriched microdomains, as well as the mean distance separating these domains, was determined by analyzing 23 EM micrographs illustrating >50 well-delimited CD63-enriched microdomains. For this analysis, a grid consisting of squares with 200-nm length was positioned onto the images, and the numbers of gold particles per square were evaluated. Adjacent squares containing a sum of at least five particles were considered one TEM, and the mean size of these domains was calculated. The mean distance between TEMs was calculated by measuring the distances between the outermost squares defining the individual TEMs. To quantify either colocalization of CD63 and CD9 () or Gag and tetraspanins (), the numbers of tetraspanin-associated gold particles situated either within the same square or in adjacent squares were determined (), or the numbers of Gag-associated gold particles overlapping with or directly adjacent to TEMs (as defined above) were determined. The percentages ± SEM are listed in the Results section. For immunogold labeling of ultrathin cryosections, cells were detached and fixed for 1 h in phosphate buffer (100 mM NaPO, pH 7.4) containing 4% paraformaldehyde (EMS) and 0.1% glutaraldehyde (EMS). Thereafter, the fixative was rinsed out three times with phosphate buffer and the cells were processed for cryosectioning as described previously (). In brief, the cell pellet was infiltrated with sucrose and frozen in liquid nitrogen. Frozen sections (45-nm thickness) were cut with a cryotome (FCS; Leica), transferred to grids, and incubated with antibodies against CD63 (clone 1B5 of mouse anti-CD63 IgG2b). Grids were examined with a transmission electron microscope (Tecnai G-12; FEI Company). Fig. S1 shows the overall distribution of the tetraspanins CD9, CD81, CD82, and CD63 when the cells were stained for only one of the antigens or dual labeled for two tetraspanins. Fig. S2 identifies surface TEMs containing two tetraspanins. Fig. S3 documents accumulation of viral Gag at plasma membrane TEMs containing CD9, CD81, or CD82. Fig. S4 demonstrates that TSG101 and VPS28, rather than accumulating at surface TEMs, are concentrated at intracellular compartments throughout the cell in the absence of viral Gag expression. Fig. S5 provides an immunofluorescence analysis of viral Env and Gag colocalization with tetraspanins representative of some of the data shown in . Video 1 shows untransfected HeLa cells that were surface stained for CD63, as described in . Online supplemental material is available at .

The Rho, Rac, and Cdc42 GTPases of the Rho protein family integrate environmental signals to regulate the organization and dynamic of the actin cytoskeleton (; ). This confers to Rho proteins a direct role in controlling cell shape and adhesion, which are two key determinants in the regulation of the barrier function of the endothelium and epithelium (). Hence, the major role played by Rho proteins accounts for the existence of a large number of bacterial virulence factors targeting them (). Rho proteins oscillate between a GDP-bound form sequestered in the cytosol, in association with the cellular factor Rho guanine nucleotide dissociation inhibitor (RhoGDI; ), and a GTP-bound form that is found in specific membrane locations, which bind and activate effector proteins (). Transitions between both forms of Rho are primarily regulated by guanine nucleotide exchange factors for activation and GTPase-activating proteins for inactivation (). Rho protein isoforms specifically regulate the architecture and dynamic of the actin cytoskeleton (; ). The activation of Rac or Cdc42 leads to actin filament polymerization, forming actin-rich lamellipodia or filopodia, respectively. RhoA induces actomyosin contraction and the formation of actin stress fibers by controlling the phosphorylation status of the myosin light chain (MLC). Phosphorylation of MLC is controlled by MLC kinases and Rho kinases (ROCKs; ; ). Activation of ROCKs by RhoA primarily leads the phosphorylation/inactivation of the regulatory subunit of the myosin-specific phosphatase MYPT1 (). Thus, Rho regulates different aspects of the organization of the actin cytoskeleton in differentiated cells that impact their morphology, as well as both intercellular and cell–matrix adhesion (; ). For instance, the activation of Rho by vasoactive factors such as thrombin induces actomyosin contractions that are responsible, in part, for destabilizing the endothelial intercellular junctions and promoting the formation of intercellular gaps through cell contraction ( ). Hence, recent findings indicate that Rho proteins of endothelial cells participate in leukocyte transmigration across the endothelium (; ). In columnar epithelial cells, Rho promotes the formation of actin filaments associated with apical tight junctions and in so doing contributes to epithelium cohesion (). epidermal cell differentiation inhibitor (EDIN; ) and EDIN-like factors () belong to a family of ADP-ribosyltransferases that are expressed both by human and animal Gram-positive pathogenic bacteria (). They consist of a single polypeptide chain, which penetrates host cells by an undefined molecular mechanism. Upon reaching the host cell cytosol, these factors catalyze the preferential ADP-ribosylation of RhoA (; ) and, to a lesser extent, other isoforms of Rho proteins (). Posttranslational modification of Rho by ADP-ribosylation leads to the tight association of RhoA with RhoGDI, leading to Rho sequestration into the cytosol (; ). In addition, ADP-ribosylation blocks RhoA activation by the guanine nucleotide exchange factor lbc (). The inhibitory effects of RhoA ADP-ribosylation by EDIN-like factors lead to the disruption of actin stress fibers (; ; ). Major progress has been made in understanding how bacterial Rho ADP-ribosylating factors interfere with immune cells (; ). To gain more insights on the biological activity of EDIN-like factors, we have investigated the effects of purified recombinant EDIN, as well as EDIN-producing , on endothelial cells and on the endothelium. We have observed that intoxication of human umbilical vein endothelial cells (HUVECs) with purified recombinant EDIN results in the formation of large transcellular tunnels that we have named MAs, in addition to the previously described dislocation of actin stress fibers (; ; ). We noticed the presence of membranes extending from the edge of MAs (, inset, arrowhead). Analysis of the intoxication of HUVECs by EDIN, using time-lapse video microscopy, revealed the transient nature of MAs (; and Video 1, available at ). Complete cycles of MA opening and closure took 2–20 min, with a mean time of 12 min ( = 45 MAs, at 24 h of intoxication). In contrast, no MAs were observed in control HUVECs. EDIN also produced MAs in other potential targets of , such as primary human microvascular endothelial cells and fibroblasts, as well as the myoblast and osteoblast cell lines L6 and Saos-2, respectively (). A low occurrence of MAs was observed in EDIN-treated human keratinocytes and columnar epithelial T84 subconfluent cells (MAs ≤ 2% at 48 h; unpublished data). We next investigated the characteristics of MA induction by EDIN in subconfluent endothelial cells. MAs were detected 3 h after cell intoxication and progressively affected 33% of the cell population after 24 h of intoxication (). The efficiency of MA formation was compared with that of Rho ADP-ribosylation, as well as inhibition of Rho activity. Correlative analysis of these kinetics indicated that MAs induced by EDIN began 3 h postintoxication concomitantly with ADP-ribosylation of 75% of endogenous Rho and 58.2% inhibition of Rho activity (). In further studying this newly described effect of EDIN, we determined that the number of cells displaying MAs increases as a function of the concentration of EDIN (). Previous studies had shown that cells recover their actin stress fibers upon Rho resynthesis, which occurs after Rho ADP-ribosylating toxin clearing (). Consistently, we also observed that endothelial cells recovered their initial status within 48 h after EDIN clearing (). Induction of MAs by EDIN is, thus, a time- and dose-dependent phenomenon. In addition, no effect was observed with the catalytically inactive EDIN mutant (; ). Further analyzing these aspects on endothelial cells transfected with either EDIN or EDIN expression plasmids, we observed a disruption of actin cables and the formation of MAs, specifically in cells expressing functional EDIN (). This allowed us to establish, after cell fixation and actin cytoskeleton staining, that 24 h of expression of EDIN leads to a loss of actin cables in 91% of the cells and formation of MAs in 49% of these cells ( = 200). Thus, consistent with the transient formation of MAs, which was observed using video microscopy, after cell fixation we found that a maximal effect of EDIN produces MAs that are visible in nearly half of HUVECs. Using video microscopy, we next analyzed the formation of MAs in cells transfected with EDIN together with GFP-caveolin1. Plasma membrane was labeled with TRITC-conjugated wheat germ agglutinin (WGA). MAs induced by EDIN appeared to result from the enlargement of a pore associated with the formation of retraction filaments (; and Video 2, available at ). Membrane retraction produces a local and transient accumulation of GFP-caveolin1 at the edge of the aperture. Rho ADP-ribosylation factors have been extensively studied using cell lines displaying thick actin cables, such as VERO cells. We noticed the formation of curved membrane “retractions” in VERO cells expressing EDIN (, arrows), which appeared to occur at the cell periphery (Video 3). Thus, at the difference with HUVECs in which EDIN induced MAs in VERO cells, membrane retractions occur at the edges of cells, leading to cellular retractions. We next investigated whether MAs were initiated by membrane ruptures. Induction of membrane wounding by pathogens or mechanical stress leads to lysosomal exocytosis for membrane repair (). In contrast with what was reported for wounded cells (), the lysosomal marker Lamp1 was not detected at the cell surface of EDIN-intoxicated cells (). Finally, we observed that membrane waves, which formed at the edge of apertures, contained Rac, cortactin, and Arp3 machinery of actin polymerization (), as observed in membrane ruffles (). EDIN and other isoforms of the C3 exoenzyme family ADP-ribosylate RhoA and, to a lesser extent, different combinations of other Rho proteins (). For this reason, we examined the effects of expression of three other members of the C3 exoenzyme family, i.e., EDIN-B and -C of , as well as the C3 exoenzyme of . Expression of these isoforms of C3 exoenzyme also induced the formation of MAs (), in contrast with catalytically inactive mutants (Fig. S1, available at ). MAs produced by C3 exoenzyme of also formed transiently (Video 4). Collectively, these results suggested that RhoA inhibition may account for a general mechanism by which MAs are produced. To address this question we knocked down RhoA using RNA interference. The cellular depletion of RhoA was quantified by immunoblotting, and its level was found to decrease to 25% within 24 h after cell transfection (). Under these conditions, we observed that 24 h after transfection of a RNAi expression plasmid, endothelial cells displayed MAs and a disruption of actin cables (). The effects of RNAi-mediated RhoA knockdown on the dynamic of the actin cytoskeleton was further analyzed in endothelial cells expressing a GFP-actin chimeric molecule. In control cells, the GFP-actin signal shows actin cables and cortical actin cytoskeleton (Fig. S2 a). In endothelial cells cotransfected with RNAi and GFP-actin expression plasmids, MA formation appeared to be rapidly followed by the recruitment of GFP-actin around the aperture (; Video 5). GFP-actin next concentrated into lamellipodia-like structures extending from the edge of the aperture. The absence of effects of control RNAi on the actin cytoskeleton was verified (Fig. S2 b). The dynamics of the actin cytoskeleton during a cycle of MA formation by EDIN appeared similar to those observed after RNAi-mediated RhoA knockdown (Video 6). Collectively, this demonstrated that RNAi-mediated RhoA knockdown recapitulated the effects of EDIN, resulting in transient formation of MAs in endothelial cells. Our results favor the hypothesis that the destruction of cytosolic actin cables after RhoA inhibition leads to the formation of MAs. To test this hypothesis, HUVECs were treated with different drugs that destabilize the actin cytoskeleton, using mild conditions to minimize cell retractions. Under these conditions, MAs were observed upon cell treatment with cytochalasin D and latrunculin B, as well as with the ROCK inhibitor Y-27632 (). Finally, we verified that Y-27632 did not produce RhoA inhibition (not depicted), but, instead, dynamic MAs similar to those produced by EDIN (Fig. S2 c). In addition, we observed that the overexpression of the Rho-family inhibitor RhoGDI produced cellular retractions instead of MAs (), as does the RhoA, Rac, and Cdc42 inhibitory toxin-B of (Video 7; ). In conclusion, we show that the preferential inhibition of RhoA by EDIN and the disruption of RhoA-regulated actin cytoskeleton account for MA formation. We next investigated MA induction by EDIN in endothelial cell monolayers. Intercellular junctions were visualized by staining VE-cadherin, which is a major component of endothelial adherens junctions (; ). When HUVECs were engaged in intercellular junctions, we measured a 20% higher level of active Rho, as compared with that measured in subconfluent cells ( = 3, mean value). Under these conditions, we measured that the level of active Rho dropped to 0.2% of total Rho at 48 h, a value similar to that obtained with subconfluent cells intoxicated for 24 h (). In HUVEC monolayers, at 48 h of intoxication by EDIN we measured that 5.9 ± 0.5% ( = 400 cells) of cells displayed MAs (). MAs were also observed in human microvascular endothelial cell (HMVEC) monolayers that were intoxicated for 48 h in similar conditions (; 6.9 ± 1%; = 300 cells). In contrast, no MAs were observed in HUVEC and HMVEC monolayers treated with EDIN (unpublished data). Similar results were obtained with the C3 exoenzyme of (unpublished data). The VE-cadherin signal was not detected at the edge of MAs (; and Video 8, available at ). MAs also formed transiently in monolayers (Videos 8 and 9). We next investigated the effects of EDIN on endothelial monolayer barrier function. Previous studies had established that RhoA activation by vasoactive factors, such as thrombin, leads to stress fiber formation and actomyosin contractions, which are responsible for intercellular gaps opening (; ; ). These effects of thrombin are blocked by C3 exoenzyme. Consistent with previous findings, we measured that short periods of intoxication by EDIN or C3 exoenzyme efficiently blocked the thrombin- induced activation of RhoA, the formation of intercellular gaps and an increase of monolayer permeability, whereas it did not efficiently inhibit the level of active-Rho (Fig. S3). Longer periods of intoxication were necessary to lower the level of active Rho required to produce MAs and to compromise monolayer barrier function (). Collectively, our results were indicative of a role of EDIN in producing MAs in the endothelium. We addressed this question by analyzing the effects produced by purified recombinant EDIN on the endothelium of rat arteries. The lumenal surface of arteries was analyzed by scanning electron microscopy 7 h after intoxication with either purified recombinant EDIN or with the catalytic-inactive mutant EDIN (). We specifically observed MAs at the surface of the endothelium of EDIN-intoxicated arteries (). EDIN-intoxicated arteries had 400 ± 50 MAs/mm at their surface. Consistent with our observations of endothelial cells, we observed a high heterogeneity of the size of these MAs with a mean value of 7 ± 3 μm ( = 50 MAs, at 7 h of intoxication). MAs were found to unmask the fibrous structure of the extracellular matrix visualized underneath the endothelium (). In parallel, we observed by transmission electron microscopy that EDIN disrupted the endothelial cell layer integrity (). Finally, investigating the effects of Rho ADP-ribosylating toxins on vascular permeability using a classical Evans blue dye extravasation assay in mice, we found that EDIN and C3 exoenzyme induced dye diffusion (; and Fig. S3, d and e). Collectively, our results show that Rho inactivation produces MAs in the endothelium and compromises the endothelium barrier function. We next addressed the question of whether the effects of EDIN might confer to a physical mechanism by which the endothelium basement membrane is exposed to the bacteria. To address this question, S25- and - strains were engineered from the EDIN-producing human pathogenic strain S25. We verified in these strains the presence of an EDIN-encoding gene, as well as the RhoA ADP-ribosylation activity (Fig. S4, available at ). After MA formation in HUVECs that were infected by these different strains of , we determined that EDIN was specifically required for actin stress fiber disruption and MA induction (). The transient formation of MAs in HUVECs infected by wild-type S25 was also verified (Video 10). We went on to determine the biological effects of EDIN on the endothelium of rat arteries infected by . We observed that both wild-type S25 and S25- strains, in contrast to S25- , produced MAs in the endothelium (). The surface of the endothelium also appeared smoother after infection by EDIN-producing . Consistent with our hypothesis, EDIN-producing bacteria were observed in close proximity to the endothelium basement membrane (). We report that inactivation of RhoA, either by ADP-ribosylating factors or by cellular depletion, leads to the formation of large and transient transcellular tunnels within endothelial cells that we have named MAs. We show that these structures compromise the integrity of the endothelium barrier. The observation that EDIN induces both actin cable disruption and a formation of MAs favor a scenario in which the specific inactivation of RhoA by EDIN first leads to actin cable disruption, which in turn allows formation of transcellular MAs. Consistently, we show that MAs are formed through RhoA depletion by RNAi knockdown and ROCK inhibition, as well as actin filament disassembly with cytochalasin D and latrunculin B. EDIN did not affect the levels of active Rac or Cdc42 (unpublished data). The importance of the specificity of RhoA inhibition in inducing the formation of MAs is reinforced by the observation that inhibitors that affect several Rho proteins, such as RhoGDI or toxin-B, produced cell retractions instead of MAs. We have noticed that high concentrations of the ROCK inhibitory molecule Y-27632 are required to produce MAs. Thus, we cannot exclude the implication of other RhoA-regulated pathways in EDIN-induced MA formation. One likely hypothesis is that destruction of thin actin stress fibers through inactivation of the RhoA–mDIA pathway contributes to MA formation (). We demonstrate that EDIN produces a specific destruction of the actin cytoskeleton network regulated by RhoA, which accounts for the formation of MAs in endothelial cells. One question, which remains to be addressed, concerns the initial membrane events leading to the formation of MAs. One hypothesis is that MAs are initiated by ruptures of the plasma membrane. This is unlikely, considering that EDIN did not trigger Lamp1 localization at the plasma membrane like membrane wounding does (). Alternatively, formation of MAs could be caused by either a fusion of vesicles and/or plasma membranes or to the enlargement of a preexisting pore. In line with this last hypothesis, endothelial cells are extremely thin and are rich in interconnected vesicles referred to as vesiculo-vacuolar organelles (VVO), which provide a route of extravasation of macromolecules that is induced by the vascular permeability factor VEGF, for example (for review see ). Formation of MAs through enlargement of VVO structures seems unlikely, considering that cotreatment of EDIN-intoxicated endothelial cells with VEGF did not increase the rate of MA formation (unpublished data), and that MAs form in cell types other than endothelial cells. Finally, it is possible that MAs initiate through luminal and abluminal membrane fusion. Notably, plasma membrane fusions occur during daughter cell partitioning, which is a biological phenomenon in which the completion of cytokinesis may require the down-regulation of RhoA (). Bridging luminal and abluminal plasma membranes might favor formation of MAs. In line with this latter hypothesis, high intravascular pressures lead to the formation of intracellular gaps in endothelial cells (). Hence, it has been previously reported that RhoA ADP-ribosylation induces cell spreading, which is a phenomenon that may favor the contact between luminal and abluminal plasma membranes (). In line with this last hypothesis, we have measured that EDIN produced less MAs when endothelial cells are buried in monolayers, as compared with subconfluent cells. Different types of structures account for endothelium permeability (for review see ). first observed that the increase in microvascular permeability is achieved by formation of gaps between endothelial cells. Nevertheless, ultrastructural studies have shown that other types of gaps, which are referred to as openings, pass through, rather than between, endothelial cells (; ). Finally, recent works suggest that endothelium permeability also results from the formation of VVO structures, in the absence of either inter- or intracellular gap formation (). RhoA activation and the downstream contraction of the actin cytoskeleton has been recognized as a central element of intercellular gap formation in response to vasoactive mediators such as thrombin (; ; ). Consistent with this notion, blocking the thrombin-induced activation of RhoA using Rho ADP-ribosylating C3 exoenzyme impairs the increase of endothelial monolayer permeability (). We report that a higher threshold of inactivation of RhoA allows MA formation in endothelial cells and impacts endothelial monolayer barrier function. Collectively, these findings point to the activation or inactivation of RhoA, leading to either inter- or intracellular gap formation. The relative contribution of inter- and intracellular gaps in the regulation of the endothelium permeability is under debate (). One hypothesis is that the formation of MAs represents a favorable pathway for inflammatory cell emigration from the blood stream. This idea is supported by the observations that inflammatory cells can migrate across thinner areas of endothelial cells that are distant from intercellular junctions during their extravasation from the blood stream (; ). In vitro studies indicate that during their diapedesis inflammatory cells trigger the formation of membrane projections or transmigratory cups in endothelial cells (; ). Transmigratory cups show similarities with the MAs surrounded by membrane waves induced by EDIN. In line with possible similarities between both phenomenons, membrane projections formed during leukocyte transmigration were not blocked upon RhoA ADP-ribosylation, whereas they were blocked upon inhibition of Rho proteins by toxin-B (). Thus, it will be interesting to determine whether the complex crosstalk of leukocytes with endothelial cells triggers a local inhibition of RhoA for transmigratory cup formation, and whether EDIN has exploited this regulation to induce the formation of MAs. Little is known concerning the role of EDIN with regard to human infection by , except that pathogenic have a higher prevalence of genes, as compared with nasal isolates (). This Gram-positive bacterium colonizes the epithelium in 30–50% of healthy adults worldwide. Its ability to produce bacteremia on accidental or surgical wounds can lead to severe endovascular and metastatic infections (). Development of such serious infections requires numerous bacterial virulence factors, comprising microbial surface components recognizing endothelial cell–adhesive matrix molecules (; ). The mechanism by which luminally located accesses components of the endothelial cell basement membrane remains an outstanding question. Based on our observations, it can be hypothesized that MA formation may provide EDIN-producing with a specific mechanism to induce a discontinuity of the endothelium barrier. The DNA encoding EDIN (NCBI M63917) was cloned using EcoRI–XhoI into pcDNA4 (Invitrogen) after PCR amplification on the E1 strain of (provided by M. Sugai, Hiroshima University, Hiroshima, Japan; ) using the following oligonucleotides: E1r, 5′-CCCGAATTCATGAAAAACAAATTACTTTTTAAAATTTTTTTG-3′ and E1f, 5′-CCCCTCGAGCTATTTTTTAAAAACAATAGCTGTTATTATG-3′. For EDIN production, the DNA was cloned using BamHI–EcoRI into pET28a after PCR amplification using oligonucleotides 5′-CCGGATCCGCTGATGTTAAAAATTTCACTGATTTAG-3′ and 5′ GGGAATTCCTATTTTTTAAAAACAATAGCTGTTATT-3′. For expression in , was cloned using BamHI–PstI into pMK4-pPROT () after PCR amplification using oligonucleotides E1r, 5′-CCCGGATCCATGAAAAACAAATTACTTTTTAAAATTTTTTTG-3′ and E1f, 5′-CCCCTGCAGCTATTTTTTAAAAACAATAGCTGTTATTATG-3′. Recombinant bacterial strains were generated from an EDIN-producing human pathogenic strain of (labeled S25) isolated from a 75-yr-old man with a spondylodiscitis-associated bacteremia (P. Boquet and P. Dellamonica, Hôpital ARCHET II, Nice, France). The S25- strain was generated by introducing the pMK4-pPROT- into the S25- strain corresponding to wild-type S25 cured of EDIN-encoding plasmid, using standard protocols. These strains were checked for the presence of an -encoding gene (using oligonucleotides E1r and E1f), as well as for ADP-ribosylation activity of RhoA (Fig. S4). The pEF-C3 was used for expression into cells of C3 exoenzyme (). The DNAs encoding EDIN-B (NCBI AJ277173) and EDIN-C (NCBI NC_003265) were isolated after PCR amplification of pathogenic strains of isolated at the Hospital of Nice (Nice, France; strains 7256 and Stal36, respectively). The DNAs of -B and -C were cloned using BamHI–EcoRI into pcDNA4 (Invitrogen) after being amplified using the following oligonucleotides: B, 5′-CCGGATCCATGGCCGAGACTAAAAATTTTACAGACTTAGTT-3′ and 5′-GGGAATTCTTATTTTTTAAAAACTACAGCAGTTATAAT-3′ or C, 5′-GGGGATCCATGAAAAGAAAATTATTTTTTAAAATT-3′ and 5′-GGGAATTCTTACTTTTTGAAAACGATACCTTCGAT-3′. Mutation R185E in the conserved sequence GRPI of EDIN, and the equivalent in EDIN-B, -C, and the C3 exoenzyme were introduced by changing AGA into GAA (EDIN), AGG into GAG (EDIN-B), AGA into GAA (EDIN-C), and AAG into GAG (C3 exoenzyme) using the QuickChange site directed mutagenesis kit (Stratagene). A previously described RNAi sequence (5′-AAGGCAGAGAUAUGGCAAACA-3′) or a corresponding RNAi scrambled sequence (control RNAi; 5′-GAAGTGACGACAGGACAAATA-3′) were cloned in the pSIRNA vector (Invivogen) according to the manufacturer's instructions. Human β-actin cDNA was cloned using XhoI–ApaI in pEGFP-C1 (CLONTECH Laboratories, Inc.). pCMV-RFP-actin was provided by E. Fuchs (The Rockefeller University, New York, NY). Other DNA constructs used were pDsRed2-Nuc (CLONTECH Laboratories, Inc.), canis Caveolin1 cloned using XhoI–BamHI in pEGFP-C1, rattus Lamp1 cloned using XhoI–BamHI into pEGFP-N2, human RhoGDI-α 6His-tagged cloned using BglII–PstI in pCMV (pRhoGDI), and VE-cadherin cloned using EcoRI–BamHI in pEGFP-N3 (provided by D. Gulino, Institut de Biologie Structurale Jean-Pierre Ebel, Grenoble, France). HUVECs (PromoCell) were grown in serum-free medium (SFM) supplemented with 20% FBS, 20 ng/ml basic FGF, 10 ng/ml EGF (Invitrogen), and 1 μg/ml heparin (Sigma-Aldrich). HUVECs were electroporated as previously described (), using 5 μg pDsRed2-Nuc together with 10 μg of pEDIN for toxin expression. For subconfluent HUVEC experiments, cells were plated at 15,000 HUVECs per well in 12-well plates. Human primary keratinocytes and fibroblasts were provided by L. Turchi (Faculté de Médecine, Nice, France). HMVECs (provided by A. Orecchia and G. Zambruno, Instituto Dermopatico dell'Immacolata–Instituto di Ricovero e Cura a Carattere Scientifico, Rome, Italy) were isolated and grown as previously described (). Other cells used were Saos-2 epithelial cells derived from osteosarcoma (HTB-85), L6 myoblast cells (CRL-1458), and VERO epithelial cells derived from kidney (CCL-81; all from American Type Culture Collection). C3 exoenzyme and EDIN were purified by His-tag affinity chromatography. EDIN was further dialyzed against 25 mM Tris and 50 mM NaCl, and then purified on a CM–Sepharose fast flow (GE Healthcare). The antibodies used were monoclonal anti-RhoA, anti-Cdc42, and anti-Rac1 (BD Biosciences), anti-GFP (clones 7.1 and 13.1; Roche), anti–β-actin (clone AC-74; Sigma-Aldrich), anti-cortactin (clone 4F11; Upstate Biotechnology), polyclonal anti-cadherin5 (Bender MedSystems), and allophycocyanin-conjugated anti-Lgp120/Lamp1 (clone H4A3; Abcam). Arp3 polyclonal antibodies were provided by P. Cossart (Institut Pasteur, Paris, France). Anti– serum was raised against the E-1 strain () using standard rabbit immunization protocols (Eurogentec). For immunofluorescence analysis, primary antibodies were visualized using Texas red–conjugated anti–mouse antibodies (Vector Laboratories), Texas red–conjugated anti–rabbit antibodies (Jackson ImmunoResearch Laboratories), or FITC-conjugated anti–rabbit antibodies (Biosys). For immunoblotting, primary antibodies were visualized using goat anti–mouse or anti–rabbit horseradish peroxidase–conjugated secondary antibodies (DakoCytomation), followed by chemiluminescence detection ECL (GE Healthcare). Biochemical products were purchased from Sigma-Aldrich, with the exception of latrunculin B and cytochalasin D (Calbiochem). The toxin-B of was provided by I. Just (Hannover Medical School, Hannover, Germany). For permeability assays, HUVECs were grown on gelatin-coated polyester filters (3-μm pore size; 12-mm diam; Greiner Bio-One). Cells were plated at 2 × 10 cells/well and grown for 4 d in supplemented SFM. For intoxication, HUVEC monolayers were treated with 100 μg/ml of EDIN or EDIN in supplemented SFM. For thrombin treatment, monolayers were incubated 1 h with thrombin 1 U/ml in SFM. Variations of permeability of each monolayer were followed at different periods of time (0, 24, and 48 h). In brief, at each time point the medium in the top chamber was replaced by supplemented SFM containing 0.5 mg/ml FITC-BSA (Invitrogen). Samples were collected after 10 min in the bottom compartments. Monolayers were washed once and incubated again in supplemented SFM containing toxin up to the next measurements. Levels of FITC-BSA in the bottom chamber were determined with a fluoroscan Ascent (Thermolab System), using an excitation wavelength of 485 nm, and detecting emission at 538 nm. Animal vascular permeability was assessed using a classical Evans blue dye extravasation assay. Groups of 6-wk-old BALB/c mice were injected into the tail vein three times at 12 h intervals, with 5 or 10 mg of toxins/Kg (for C3 or EDIN). Evans blue dye (30 mg/Kg in 100 μl PBS; Sigma-Aldrich) was injected into the tail vein 36 h after the first injection, and the Evans blue dye was extracted 1 h later from ears and quantified, as previously described (). Animals used during this study were maintained and handled according to the regulations of the European Union and the French Department of Health and Agriculture. For electron microscopy, 6–8 wk old male Wistar rats (150–200 g; Janvier Laboratories) were anesthetized with pentobarbital (30 mg/kg i.p.) and ketamine (100 mg/kg i.p.) before i.v. injection of 10 mg/kg heparin (Sigma-Aldrich). 5-mm-long aortic arches were excised and rinsed in Celsior medium (Imtix-Sangstat) supplemented with 0.1% wt/vol heparin. Arteries were incubated for 7 h at 37°C in Celsior supplemented with 0.2 μg/ml of purified recombinant EDIN or EDIN or for 1 h with strains. Samples were removed and immersed in 1.5% glutaraldehyde in 100 mM phosphate buffer, pH 7.4, for at least 18 h at 4°C, and then processed in parallel for transmission electron microscopy (CM12; Philips) and scanning electron microscopy (6340F; Jeol), using standard techniques (). Animals used during this study were maintained and handled according to the regulations of the European Union and the French Department of Health and Agriculture. For ADP-ribosylation assays, control or intoxicated HUVECs (10 cells/time point) were homogenized in 0.25 ml cold BSI buffer (3 mM imidazole, pH 7.4, and 250 mM sucrose) supplemented extemporaneously with 1 mM phenylmethylsulfonyl fluoride. Cells were lysed by passing 40 times through a 1-ml syringe equipped with a 25G × 5/8" needle (U-100 Insulin; Terumo Medical Corporation). Nuclei were removed by centrifugation for 10 min at 10,000 at 4°C. Protein concentrations of the postnuclear supernatants were determined using DC protein assay (Bio-Rad Laboratories). ADP-ribosylations were performed for 30 min at 37°C on 10 μg of intoxicated cell lysates supplemented with 0.5 μCi [P]NAD (800 Ci/mmol) and 1 μg of EDIN. Levels of active Rho were determined by GST-rhotekin RBD pull-down that was modified as described previously (). Fig. S1 shows actin cytoskeleton disruption and MA formation by EDIN-isoforms. Fig. S2 shows the effect of EDIN on HUVECs expressing GFP-actin. Fig. S3 shows the interference of EDIN on thrombin-induced increase of monolayer permeability. Fig. S4 is a characterization of S25-derived strains. Video 1 shows endothelial cell intoxication by EDIN. Video 2 shows EDIN intoxication of HUVECs expressing GFP-caveolin1. Video 3 shows VERO cell intoxication by EDIN. Video 4 shows endothelial cell intoxication by C3-exoenzyme of . Video 5 shows actin dynamics in RNAi transfected HUVECs. Video 6 shows actin dynamics in EDIN-intoxicated endothelial cells. Video 7 shows HUVEC intoxication by toxin-B of . Video 8 shows the effect of EDIN on endothelial cells expressing VE-cadherin–GFP and RFP-actin in the monolayer. Video 9 shows the effect of EDIN on the endothelial cell monolayer. Video 10 shows HUVEC infection by . Online supplemental material is available at .

A new factor, hepatoma up-regulated protein (HURP), was identified as having a role in chromosome congression (; ; ). HURP had previously been identified as an Aurora A substrate up-regulated in hepatomas (). In each of the current studies, HURP was identified using a different approach. The Mattaj laboratory biochemically fractionated MAPs from egg extracts to identify factors required for spindle assembly. The Nigg laboratory used a proteomics approach to purify and identify spindle components, while the Fang laboratory mined microarray data to identify proteins whose expression was induced during G2 or G2/M of the cell cycle and that were also coregulated with known mitotic proteins. All three labs identified HURP as a MAP that could bundle MTs in vitro and which localized predominantly to the portion of K-fibers (bundles of spindle MTs that attach to kinetochores in metazoans) closest to the chromatin. Interestingly, Sillje and coworkers found that HURP did not localize to astral MTs in HeLa cells (), suggesting that HURP's role is specific to K-fibers in somatic cells. Consistent with this observation, the loss of HURP resulted in misaligned chromosomes at the metaphase plate, suggesting a role for HURP in chromosome congression. This misalignment stemmed from a failure, in many cases, of K-fibers to attach to kinetochores (). Further experiments demonstrated a role for HURP in increasing K-fiber stability, suggesting that either the bundling of MTs to form a K-fiber stabilizes them or that HURP has additional roles in stabilizing MTs. However, HURP's role in congression may not be restricted to stabilizing and bundling K-fibers. Using an MT regrowth assay, Wong and Fang found that HURP is required for de novo MT production from chromosomes in a manner similar to TPX2 (). Interestingly, despite chromosome misalignment in HURP-depleted cells, the cells entered into anaphase after a prolonged period in prometaphase. Wong and Fang examined this event more closely and found that the spindle checkpoint was activated despite progression into anaphase. However, checkpoint activation was not just due to a lack of MT attachment to kinetochores, but also a lack of tension between amphitelically attached sister kinetochores. The reduced tension resulted from reduced MT stability within the K-fibers (). However, Wong and Fang showed that HeLa cells, used by all groups, can override the spindle checkpoint when MTs are partially destabilized through a variety of means. Therefore, the spindle checkpoint override may not be a direct consequence of HURP activity, but may rather be due to a general destabilization of MTs, in HeLa cells at least. How the checkpoint is over-ridden remains unclear, but once deciphered will provide considerable insight into how the checkpoint operates. Recent modeling and experimental studies suggest a role for the Ran gradient in directing MT growth toward chromosomes (; ). HURP localizes to and stabilizes K-fiber ends closest to chromosomes, suggesting that it could be the long sought after Ran-dependent MT stabilizing factor (). HURP localization appears to be particularly sensitive to RanGTP concentrations. Upon overexpression of an allele of Ran locked in the GTP bound form, RanQ69L, which should elevate RanGTP levels in the cell, HURP relocalizes to regions of the spindle closest to the poles. Assuming then that HURP is only active at the highest concentrations of RanGTP within the cell, it would be active close to the chromosomes, thereby facilitating the final run-in of MTs to the kinetochore. Bipolar spindle assembly requires the balance of plus- and minus-end–directed motor activities. Previous studies using egg extracts have shown a correlation with RanGTP-dependent changes in the dynamics of the mitotic kinesin Eg5, on astral microtubules and the initiation of bipolar spindle assembly (). These data suggested that RanGTP could stimulate bipolar spindle assembly by regulating the balance of motor protein activity. However, the biochemical connection of Ran to Eg5 remained unclear. Using an in vitro assay, fractionated MAPs isolated from egg extracts and identified a large molecular weight complex that stimulated the reorganization of asters into bipolar spindles. This complex contained two of the usual suspects known to be directly regulated by Ran: Aurora A and TPX2 (; ; ). Intriguingly, the complex also contained Eg5, XMAP215, and HURP, thus providing a biochemical link between Eg5 and Ran that opens up exciting new avenues to further define the mechanism by which RanGTP stimulates bipolar spindle assembly. How the complex serves to regulate Eg5 remains to be determined. Eg5 activity could be modulated by virtue of its assembly into the complex, either through a conformational change and/or its phosphorylation by Aurora A (). The discovery of this complex suggests the interesting possibility of a relationship between microtubule stability and the balance of motor activity required for bipolar spindle assembly. Another fascinating problem is how Ran exerts its effect on the HURP complex and how the complex functions (). Ran may regulate targeting of the complex to MTs by regulating the MT binding activity of HURP (). Indeed, the recruitment of Aurora A and KLP61F (the Eg5 homologue) to MTs is also dependent on Ran in vivo (). In addition, assembly of the complex is dependent upon Aurora A activity (), suggesting that Ran could exert its affect through the characterized activation of Aurora A by TPX2, which is known to depend on MTs (; ). Intriguingly, the spindle localization of HURP complex components does not completely overlap, raising the possibility that the complex is transitory within the cell. This transient interaction may facilitate the Aurora A–dependent phosphorylation of the complex before dispersal. Our understanding of this exciting complex is in its early stages, but it raises the possibility that several processes in spindle assembly are regulated through one complex. However, we do not know the full extent of the complex yet: are there more components? What are the interactions within the complex? Is the full complex required for each process? Defining answers to these questions will shed more light on the underlying mechanisms behind spindle assembly and reveal in greater detail how Ran regulates mitosis.

The CPC can be regarded as a complex similar to the cyclin/CDK kinase complexes, in which the binding of a nonenzymatic protein to its enzymatic partner is essential for functioning of the kinase. Instead of one nonenzymatic/regulatory subunit in a cyclin/CDK complex, the CPC contains three nonenzymatic subunits, all of which are essential for the function of Aurora-B. These subunits determine activity, localization, stability, and possibly also substrate specificity of Aurora-B. In human cells, these subunits are Survivin, the inner centromere protein (INCENP), and Borealin/Dasra-B (hereafter referred to as Borealin; ). These proteins are conserved among species, as in all investigated model organisms similar proteins have been identified. Borealin forms an exception, because orthologues have not been identified so far in and . Also, the putative functional orthologue of Borealin in , CSC-1, is only distantly related to other Borealin proteins (). Interestingly, Bir1p, the yeast homologue of Survivin, is much larger than its mammalian orthologues, raising the intriguing possibility that Survivin and Borealin are combined in a single protein in yeast and that Bir1p has diverged into different polypeptides during evolution. In all studied organisms, the CPC proteins function in multiprotein complexes, and in mammalian cells, complex formation between the CPC proteins is needed for protein stability (; ). Recent evidence suggests that two distinct passenger complexes exist during mitosis: one containing all four CPC members and another consisting of INCENP and Aurora-B (). Of these two complexes, the quaternary CPC functions during chromosome alignment and cytokinesis, whereas the INCENP–Aurora-B complex might be responsible for modifying Histone H3 (). How do these proteins dictate Aurora-B function? The CPC proteins show a very dynamic localization during mitosis that gave them their name (): they initially paint the entire chromatin during the onset of mitosis, move from the chromosome arms toward the inner centromeric chromatin (in between the kinetochores, the sites of microtubule attachment) during prometaphase, relocalize to the microtubules of the central spindle at the metaphase–anaphase transition, and finally concentrate at the midbody during telophase/cytokinesis. This localization correlates with the diverse functions of the CPC during mitosis: modifying histones at the chromatin, correcting misattachments while at the centromere, and regulating cytokinesis at the central spindle (and later midbody). Given the correlation between localization and function, it is apparent that timely and proper localization of the CPC is the key to allowing Aurora-B to exert its diverse functions during mitosis. To allow timely CPC localization, one or multiple CPC subunits should recognize a docking site (i.e., receptor) on chromosome arms, centromere, or central spindle. Because of the molecular differences between these structures (e.g., centromeric chromatin versus microtubules on the central spindle), it is likely that different receptors exist on these structures and that different CPC members are involved in the specific targeting of Aurora-B. The mechanism by which the CPC is targeted to the chromosome arms in unclear, but a plausible possibility would be via interaction with HP-1, as INCENP has been described to interact with this chromatin-associated protein (). HP-1 displacement from chromosome arms is mediated by Aurora-B (; ), which could explain the transient localization of the CPC at chromosome arms during prometaphase. Also, the CPC receptors at the centromeres are unknown, but there are clues to the mechanisms by which the CPC interacts with its centromeric receptors (). INCENP-deletion studies identified an NH-terminal domain needed for centromere localization of the CPC (). Survivin interacts with INCENP via this domain, and replacement of this domain with Survivin suffices for targeting a functional CPC to the centromeres (). When Survivin is linked covalently to INCENP, a functional CPC can be targeted, albeit less efficiently, to the centromeres and central spindle in the absence of Borealin (). Thus, Borealin appears to play only a minor role in centromere targeting when Survivin and INCENP are forced into a complex. However, Borealin is essential for centromere localization of the endogenous proteins, suggesting it plays a major role in promoting interaction between Survivin and INCENP. Indeed, depletion of Borealin disrupts the interaction between exogenously expressed Survivin and INCENP (). Similarly, efficient in vitro interaction between Survivin (BIR [baculovirus IAP repeat] 1) and INCENP (ICP-1) depends on the Borealin orthologue CSC-1 (). Interestingly, recent data showed that Borealin also interacts with the NH terminus of INCENP and that Borealin can interact with double-stranded DNA in vitro (), suggesting that, in addition to facilitating the Survivin–INCENP interaction, the contribution of Borealin to centromere targeting is mediated via direct interaction with chromatin (). Collectively, this allows for a model in which Survivin and Borealin cooperatively mediate centromere targeting of the CPC through multiple docking sites, including the chromatin itself. By interacting with the NH terminus of INCENP, these proteins can then recruit INCENP and Aurora-B to centromeres. Because Survivin and Borealin can oligomerize in vitro (; ; ; ), it is possible that a heterooligomer of Borealin and Survivin assembled on the NH terminus of INCENP forms the centromere binding interface of the CPC. Within Survivin, the BIR domain is the most likely domain to interact with putative CPC receptors at the centromere, as disruption of this domain impaired CPC centromere function (but not Borealin interaction; ). Interestingly, Bir1p, the Survivin orthologue, interacts with Ndc10, a subunit of the centromere binding factor-3 complex (), making this protein a good candidate for a CPC centromere receptor. However, a mammalian orthologue of Ndc10 has not been identified. Clearly, identifying Survivin (BIR domain) and Borealin interactors is necessary to further elucidate the mechanisms of CPC centromere targeting. Most of our knowledge regarding CPC centromere targeting is based on immunofluorescence data in fixed cells. However, it is clear that association of the CPC to centromeres is highly dynamic. For example, Survivin localizes dynamically to the centromere (). Inhibition of Aurora-B or depolymerization of microtubules greatly reduces Survivin turnover at centromeres (), suggesting that CPC localization and microtubule attachment are linked during prometaphase/metaphase. Indeed, recent evidence showed that proper dynamics of Survivin (and presumably the entire CPC) at the centromeres is essential for proper chromosome alignment. identified ubiquitination as a posttranslational modification required for proper targeting and dynamics of Survivin at centromeres. Interference with this process, by removing either a deubiquitinating enzyme (hFAM) or a ubiquitin binding protein (Ufd1) impaired localization and turnover of Survivin at the centromeres and, as a consequence, disturbed chromosome alignment (). Moreover, Survivin is also an Aurora-B substrate, and mimicking constitutive phosphorylation impairs centromere localization (). It will be interesting to see whether these modifications are interdependent and/or influence each other and how posttranslational modifications on Survivin can influence the function of the entire CPC. To function during cytokinesis, Aurora-B needs to translocate from the centromeres to the central spindle at the metaphase–anaphase transition. In , this translocation is negatively regulated by cyclin B/Cdk1–dependent phosphorylation of INCENP. Dephosphorylation of residues within the coiled-coil domain of INCENP by Cdc14 triggers translocation of the CPC to the central spindle (). However, a recent phosphoproteomics analysis failed to identify phoshosites in the coiled-coil domain of human INCENP. Yet, multiple putative cyclin B/Cdk1 phosphosites were identified in a region in INCENP previously shown to interact with HP-1 (; ). Together with the observation that expression of a nondegradable cyclin B mutant prevented spindle transfer of Aurora-B during anaphase in human cells (), this suggests that phospho-dependent regulation of CPC spindle transfer is conserved but that different domains in INCENP might be involved. Besides the phosphosites found within the HP-1 binding domain, several additional residues in INCENP were found to be phosphorylated during mitosis, suggesting complex phospho-dependent regulation of the human CPC (). Relocalization of the CPC from centromeres to the central spindle at the metaphase–anaphase transition also requires dynamic microtubules, as treatment of anaphase cells with the microtubule-stabilizing drug taxol impaired central spindle targeting (). INCENP interacts directly with polymerized microtubules via its coiled-coil domain (). Additionally, a small domain in the NH terminus of INCENP interacts with β-tubulin and is essential for efficient spindle localization of the CPC (; ). Survivin can also interact in vitro with polymerized microtubules, and mutation of the coiled-coil domain in Survivin impaired this interaction (), suggesting a dual interaction of the CPC with microtubules (i.e., via INCENP and Survivin). However, it is unclear whether this interaction is crucial for proper CPC localization during anaphase. Besides microtubules, central spindle localization of the CPC also depends on Mklp2, a mitotic kinesin. In human cells depleted of Mklp2, the CPC fails to relocalize to the central spindle during anaphase (). Aurora-B directly interacts with Mklp2 in human cells (), and similarly, Aurora-B functionally interacts with the related kinesin Zen-4 (). In this case, Aurora-B itself targets the CPC to the central spindle by interacting with Mklp2. Additionally, Mklp2 interacts with and is required for central spindle localization of human Cdc14a, a homologue of Cdc14. Altogether, relocalization of the CPC to the central spindle depends on the orchestrated actions of at least spindle microtubules, Mklp2, and Cdc14 (). Is centromere localization a prerequisite for central spindle localization? Initial experiments in which disruption of centromeric localization also impaired anaphase spindle transfer indicated that it might be (). However, mutants undergoing meiosis without chromosomes can execute cytokinesis and concomitantly localize Aurora-B to the central spindle (). Additionally, recent experiments revealed that the COOH-terminal region of Survivin (the region containing the coiled-coil domain that binds microtubules in vitro []) is sufficient to direct a functional CPC to the central spindle without prior centromere concentration (). It will be interesting to investigate whether this domain in Survivin interacts in vivo with one of the known central spindle CPC receptors or with as-yet-unidentified central spindle receptors. Taken as a whole, it seems that, although in normal cells centromeric and central spindle localization are tightly linked, they can be uncoupled and involve different targeting mechanisms. In vitro experiments have demonstrated that INCENP is critically needed to activate Aurora-B. INCENP interacts with Aurora-B via its conserved COOH-terminal IN-box, and incubation of this domain with Aurora-B causes an increase in kinase activity (; ). Addition of Borealin does not activate Aurora-B in vitro (), whereas conflicting data exist regarding the ability of Survivin to activate Aurora-B. In extracts, Survivin is needed for full Aurora-B activity (), but in vitro experiments with human proteins did not reveal a role for Survivin in activating Aurora-B, whereas INCENP could activate Aurora-B in this in vitro setup, suggesting that INCENP is the major Aurora-B activator in human cells (). Alternatively, in vivo regulatory mechanisms might exist (e.g., additional proteins and/or posttranslational modifications) that are needed for additional Survivin-dependent activation of Aurora-B. INCENP, Survivin, and Borealin are subject to phosphorylation by Aurora-B (). INCENP phosphorylation at a TSS motif close to the IN-box induces a conformational change in Aurora-B, causing full activation of Aurora-B (). This phosphorylation is essential for in vitro (; ) and in vivo (unpublished data) functionality of Aurora-B. Survivin is phosphorylated on threonine-117 by Aurora-B in vitro (), and this phosphorylation is involved in regulating localization (see Regulation of Aurora-B localization by the CPC proteins). The COOH terminus of Borealin is phosphorylated by Aurora-B, but the functionality of this phosphorylation is unknown (). Further research regarding these phosphorylations in CPC function (e.g., on localization and dynamics) will be necessary to deepen our understanding of CPC regulation. Additionally, it will also be interesting to explore whether INCENP, Survivin, and Borealin also influence substrate specificity and recognition of Aurora-B. #text

Homologous pairing during meiotic prophase is essential for homologous recombination and for chromosome segregation during meiosis I; yet, it is largely unknown how homologues approach each other. A chromosomal rearrangement called the telomere bouquet in early meiotic prophase is thought to play a key role in homologous pairing (; ). It forms as telomeres cluster to a small region on the nuclear envelope (NE). Though the telomere bouquet is conserved in most organisms, the mechanism of its formation is not well understood. Fission yeast, , serves as a good model organism to study telomere bouquet formation because it has a conspicuous method for clustering telomeres. Telomeres are in subclusters scattered on the inner NE in mitotic cells and reorganized upon pheromone sensing (), as telomeres start to cluster on the NE adjacent to the SPB. Such reorganization persists during premeiotic S phase and meiotic prophase, as the horsetail-shaped nucleus is driven back and forth by the microtubule arrays attached to the SPB. Therefore, the telomere bouquet could help the homologues approach each other, and the horsetail movement stretches them to facilitate their alignment. Although many proteins have been identified in the telomere complex and the SPB, the linking components between them for bouquet formation are not well understood, and the forces that cluster telomeres have not been described. Rap1p is a telomere binding protein that is essential for telomere clustering (; ). Sad1p, a spindle pole body (SPB) protein () could function as a transmembrane linker for the bouquet formation. It has a transmembrane domain and a Sad1/UNC-like domain, which is an essential part of the UNC-84 protein for nuclear migration in (). These two domains are also in the SPB half-bridge protein Mps3p/Nep98p (), the reciprocal best hit with Sad1p in budding yeast. Furthermore, Sad1p interacts with Kms1p (), a protein that is required for telomere clustering (), and with dynein light chain Dlc1p, which is required for the horsetail movement (). Cytoplasmic microtubules are thought to be involved in clustering telomeres in fission yeast. However, this may not be the case in other organisms. To identify more genes for bouquet formation, we developed a visual screen by monitoring heterochromatin reorganization upon meiosis. One mutant we found, , was defective in bouquet formation and was an allele to that was identified in genome-wide screens by systematically deleting up-regulated genes during meiotic prophase independently in two other groups (; ). Using multiple GFP- and mCherry-tagged markers for telomeres, the SPB, and the NE, we cytologically proved that Bqt2p specifically functions as a linkage component between telomeres and the SPB. It plays a key role in initiating telomere clustering by generating Sad1p foci proximate to the telomere foci upon pheromone sensing. The dynamic reorganization of heterochromatin during sexual development in the fission yeast can be monitored using GFP-tagged Swi6p, a homologue of heterochromatin protein 1, which binds to all telomeres and centromeres as well as the silent mating type locus (which cannot be detected as a separate identity by light microscopy). During the horsetail stage, telomeres cluster at the leading edge of the nucleus adjacent to the SPB (, arrows), whereas centromeres are released into the interior of the nucleus. The attachment of telomeres to the SPB transmits mechanical forces from microtubule arrays to chromosomes, causing stretching and alignment of the homologues (). Both telomere clustering and horsetail movement are important for homologous pairing (). To identify genes involved in telomere clustering, we developed a visual screen to search for mutants with aberrant GFP-Swi6p pattern during the horsetail stage. A strain expressing GFP-Swi6 () was mutagenized by random insertion of fragments (see Materials and methods). Because haploid cells have three chromosomes, 4–7 GFP-Swi6p foci are expected during the horsetail stage, depending on the extent of centromere pairing. From 18,000 insertional mutants (i.e., mutants), we found several with defective telomere clustering, including , , , , , , , , and . Among these, showed the most severe phenotype, as more than seven GFP-Swi6p dots were observed in >60% of the cells during meiotic prophase and stayed in the middle of the cells with little movement (). Furthermore, in >90% of the cells, the chromosome mass was not distended (). Therefore, both the telomere bouquet and the microtubule-driven chromosome stretching were severely disrupted in cells. The insertion in was mapped to (). This is an allele to (hereafter ; ; ). The insertion replaced the 11th to 14th nucleotides of exon 2. The same meiotic defects in were observed in , and plasmids carrying this gene rescued the phenotype (unpublished data). Therefore, the insertion in caused the observed defects. We wanted to determine how lack of Bqt2p leads to defective telomere organization. The phenotype is not due to defects in sister chromatid cohesion or DNA replication because mutants found in our screen defective in these functions ( and , respectively) have stretched chromosomes and moving leading edge. To determine whether the link between telomeres and the SPB was disrupted in , we labeled the SPB with Sad1p tagged with mCherry () and telomeres with Taz1p tagged with GFP (). In wild type, Sad1p and Taz1p colocalized as a single spot at the leading edge of the elongated nuclei (). In , however, multiple Taz1p foci were scattered in the middle of the cell, whereas a single Sad1p-mCherry spot moved to the cell tip (). Unlike in wild type (), the chromosome mass in is no longer stretched along with the single Sad1p foci driven by the microtubule arrays (), which was labeled with the GFP-tagged α-tubulin subunit Tub1p (). Therefore, the meiotic SPB and telomeres were no longer connected in and, thus, the force generated by the microtubules failed to stretch chromosomes. To see if the disassociation of telomeres to the SPB in cells depends on the vigorous movement of the SPB, was used because there is only subtle movement in this mutant (). The double mutants () showed a similar Taz1p-GFP pattern to that observed in () but not in (). Therefore, even without vigorous movement of the SPB, the connection between telomeres and the SPB is disrupted in Bqt2p-depleted cells. After meiotic prophase, telomeres were detached from the SPB, and the telomere foci and DNA segregates equally in wild type (). However, nearly 40% of the showed abnormal distribution of chromosomes and Taz1p foci (). This phenotype is typical of mutants deficient in telomere clustering. Telomeres start to cluster upon pheromone sensing before conjugation (). Therefore, we analyzed telomere clustering upon pheromone response in cells using deletion, which does not affect the response of to the pheromone release by cells but blocks pheromone response of cells and, consequently, the conjugation between and cells. When homothallic cells are starved, the cells switched to cells can sense the pheromone released by cells, but not vice versa. Pheromone sensing induces a tip projection, called schmoo, necessary for cells to approach each other and conjugate. Schmoo and telomere clustering were blocked in starved heterothallic cells because of the lack of pheromone (). In nearly 30% of the starved homothallic population, cells showed either a single fully colocalized Sad1p and Taz1p foci (; 50 out of 193 cells) or more than one Sad1p spot that fully (; 6 out of 193 cells) or partially colocalized with Taz1p foci (; 26 out of 193 cells). This suggests that pheromone sensing generated Sad1p foci next to telomeres, which were pooled together by the microtubule arrays (). The incomplete association of telomere foci and Sad1p foci suggests that the link between them is dynamic or not stable. In contrast, without Bqt2p, no telomere spots colocalized with the Sad1p foci, even in schmooing cells (; 137 cells checked). These results demonstrate that Bqt2p is essential for generating Sad1p foci next to the telomeres to initiate telomere clustering. Although Bqt2p is essential for the association of telomeres and the SPB, it could function in one of two ways, either serving as a component of the linkage or coupling either telomeres or the SPB to the NE. In fission yeast, the SPB were observed to be free floating in some mutant cells (). In budding yeast, telomeres are detached from the NE specifically upon meiosis entry in the mutant, which results in defective telomere clustering (). To determine whether there was similar defect in during meiotic prophase, we used a membrane marker, D817, which is the NH-terminal 275 amino acid peptide of NADPH-cytochrome P450 reductase (). In both wild type and , a single Sad1p spot remained at the tip of the NE (), but the NE adjacent to the tip in was stretched to form a thin line. From images of single middle z axis slices, all telomere foci were observed to attach to the inner NE in both wild type () and (). Therefore, telomeres and the SPB were attached to NE membrane independent of Bqt2p during meiotic prophase. The disconnection of telomeres to the SPB must be due to a failure to establish a mechanical linkage between them. The endogenous Bqt2p was COOH-terminally tagged with GFP to check its expression and localization. Its expression was not observed in starved heterothallic cells (). But upon sensing pheromone, when Bqt2p foci fully colocalized with Sad1p-mCherry in ∼40% of the population (), most of them were not schmooing yet. Therefore, pheromone induced Bqt2p expression. These results were consistent with the finding that Bqt2p was required to initiate telomere clustering. Compared with data showing partial colocalization of Taz1p and Sad1p in , these results also suggest that the colocalization of Bqt2p to Sad1p does not guarantee stable connection of all telomeres to the Sad1p foci. As expected, Bqt2p-EGFP colocalized with the meiotic SPB during the horsetail stage. However, Bqt2p remained at the meiotic SPB at early meiosis I (8 out of 9 cells with same level of signal as that during meiotic prophase, and 1 with less intense signal) and until the end of meiosis II (23 out of 30 cells; few with same level of intense signal, and most with less intense signal; ). Therefore, the disassociation of Bqt2p from the meiotic SPB is not required for dissolving telomere clustering at the end of meiotic prophase. Ectopic expression of meiosis-specific proteins sometimes mimics meiotic cellular activity in vegetative cells. This is not the case for Bqt2p, as it was preferentially located in the nucleus but did not concentrate at the SPB when expressed in vegetative cells (). Therefore, other meiosis-specific factors may be required for docking Bqt2p to the SPB. We analyzed the localization of Bqt2-GFP in several different mutants, including (), (; ), and (; ), all of which are defective in telomere clustering. Without Kms1p, multiple Sad1p foci are present, and Bqt2p colocalized to all the Sad1p foci (). Also, Bqt2p localized to the single Sad1p foci in the (). These data demonstrate that Bqt2p localization to the Sad1p foci is independent of Kms1p, Rap1p, or telomeres. However, Bqt2p was no longer localized to the single Sad1-mCherry spot in (). Therefore, colocalization of Bqt2p to the meiotic SPB requires Bqt1p, and Bqt1p may link and/or stabilize the Bqt2p association with the Sad1p foci. We have shown that Bqt2p is specifically essential for linking telomeres to the SPB and for transmitting force from the microtubule arrays to the chromosomes. Bqt2p is essential to initiating telomere clustering upon pheromone sensing, and it remains so after telomeres leave the SPB. Our observations agreed with the model for telomere clustering recently proposed by . First, upon pheromone sensing, Bqt1p-Bqt2p binds to Sad1p. Once telomere binding proteins, such as Rap1p, interact with Bqt1p-Bqt2p, more Bqt2p-Bqt1p-Sad1p and other factors aggregate and attract Kms1p as well as other components of the microtubule cytoskeleton. However, other factors may be required to stabilize the connection between telomeres and the Sad1p foci. Kms1p may also function as one of these stabilizers because, in mutants, there are multiple Sad1p foci not localized with the telomere foci (), which is totally different from the pattern in mutants (). Finally, the telomere foci are pooled together. Other forces, together with the forces generated by the cytoplasmic microtubules, are required to cluster telomeres. If this were not the case, we would expect that the Sad1p spot at the mitotic SPB would not colocalize with any telomere spots in , but this colocalization was observed (). Our hypothesis is consistent with the telomere bouquet formation in other organisms, which can occur without nuclear movement in meiotic prophase. At the end of meiotic prophase, the telomere bouquet is dissolved possibly by regulating the stabilizers for the linkage between telomeres and the SPB. Methods and media for fission yeast were as described previously (). Molecular cloning was done as described previously (). The following plasmids were used in this study: , , , , , , , , , , , , , , , , , , , , , , , , , and . The following fission yeast strains were used in this study: , , , , , (provided by J.P. Cooper, Cancer Research UK, London, UK), , , , , , , , , , , , , , , , (provided by T. Toda, Y. Hiraoka, and D.Q. Ding, Kansai Advanced Research Center, Kobe, Japan), , and . Fragment was amplified by PCR using as the template and two primers (5′-gaaagatatcggatccagacaaatgttggactagtggtttcttagacgtcaggtg-3′ and 5′-gaaagatatccagctgaattccaccatggaaatggtagatctggtaccatgtgagcaaaaggccagca-3′), restricted with EcoRV, and self-ligated to generate . Fragment was amplified by PCR using as the template and two primers (5′-gaaaagatctccatggaattcagctggtttcttagacgtcaggtg-3′ and 5′-gaaaagatctactagtaaaaagatatctggtaccatgtgagcaaaaggccagca-3′), restricted with BglII, and self-ligated to generate . Plasmid had a polylinker site (SpeI–BamHI–EcoRV–PvuII–EcoRI–BglII–KpnI), and had another one (PvuII–EcoRI–BglII–SpeI–EcoRV–KpnI). The fragment from (Invitrogen) was amplified and cloned into the KpnI site of , generating . PCR was used to amplify the following fragments: , , , , , (CLONTECH Laboratories, Inc.), (provided by R.Y. Tsien, University of California, San Diego, La Jolla, CA), and (hygromycin B–resistant gene; provided by C. Rasmussen, University of California, Berkeley, Berkeley, CA). If the fragments had some restriction sites that were the same as in the polylinker of , the two-step PCR overlap extension method was used to change one nucleotide and disrupt those sites but without altering amino acids. These PCR fragments were cloning into using EcoRV and EcoRI, generating the following plasmids: , , , , , , , and . All these plasmids have common sequences and restriction sites at the two sides. The SpeI–PvuII fragments carrying and from and were inserted to the SpeI–EcoRV site of , generating and , respectively. The SpeI–PvuII fragments carrying and from , were inserted to the SpeI–EcoRV sites of and , respectively, generating . The SpeI–PvuII fragments carrying and from and were inserted to the SpeI–EcoRV sites of , generating and , respectively. The fragment together with its 5′ upstream promoter (∼1.5 kb) was amplified from genomic DNA, restricted with BglII, and cloned into the BamHI site of and , respectively, generating and . The gene fragment was amplified, restricted with EcoRV and BglII, and inserted into the PvuII–BglII sites in , with the stop codon deleted, generating . The fragment in was cut out with BamHI and BglII and inserted into the BamHI site in to make , in which was COOH-terminally tagged with and under the control of the inducible . Two primers with 18 bp specific to the gene and 30-bp random nucleotides were used to amplify the fragment. . Transformants were selected on Edinburgh minimal media without uracil. Stable insertional mutants were selected by replicating them, back and forth for twice, onto yeast extract plus supplement and Edinburgh minimal media without uracil. The colonies were replicated onto synthetic sporulation agar media to induce meiosis. GFP-Swi6p pattern was checked one by one after 20–24 h. Genomic DNA was isolated from the mutants and ligation-mediated suppression PCR () was used to map the insertions. DNA fragments, such as , , and , constructed in the plasmids were all flanked with two common cassettes (upstream, 5′-tgtctggatccatcggga-3′, and downstream, 5′-tcgctgaattccaccatg-3′; the three nucleotides underlined indicate the reading frame for reporter genes, in which the stop codons were deleted). The terminator had a stop codon next to the left cassette. Selection markers and combinations of reporter gene––selection marker between these two cassettes were used for gene disruption and COOH-terminal gene tagging, respectively. Two-step fusion PCR was used. One ∼600-bp fragment upstream from the targeting site was amplified and was added with the 21 nucleotides of the upstream cassette at the 3′-end that was fused to the cassette. Another ∼600-bp fragment downstream from the targeting site was amplified and was added with the 21 nucleotides of the upstream cassette at the 5′-end that was fused to the cassette. These two PCR products together with the corresponding modular vector for gene tagging and disruption were used as the templates to do fusion PCR. The fusion PCR products were transformed using the method described previously (). The endogenous was popped out using the method described previously (). Strain carrying the plasmid was grown on appropriate minimal medium with 10 μM of thiamine and then streaked out to the same medium without thiamine to induce Bqt2p-GFP expression ectopically in mitotic cells (). Live imaging was performed as described previously () with subtle modifications. Fresh cells grown on yeast extract plus supplement plates were induced into mating and meiosis by streaking cells onto synthetic sporulation agar plates. Cells were scraped off the plates 16–20 h later for checking pheromone response, 18–24 h later for checking cells during the horsetail stage, or 22–26 h later for checking cells during meiosis I and II. They were then suspended in 8 mg/ml Hoechst solution for at least 10 min and spotted onto a thin layer of 1% agarose (in 1/2 synthetic sporulation agar liquid medium) on slides. Then cells were covered and spread by an 18- × 18-mm coverslip. Images were collected using a widefield inverted fluorescence microscope (IX70; Olympus) with a UPlanApo 100×, NA 1.35, oil-immersion objective (Olympus) using a charge-coupled device camera (CH350; Roper Scientific, Inc.) cooled to −35°C with DeltaVision image acquisition software (Applied Precision, Inc.). Serial sections in the z axis were acquired at 0.3-μm intervals, and data stacks were deconvolved using softWoRx deconvolution software (Applied Precision, Inc.). For presentation purposes, 2D projections were created from the 3D datasets using the DeltaVision image analysis software. To observe horsetail movement, images were taken at ∼3-min intervals. Brightfield images were taken with six sections and 0.4-μm intervals. The images were processed and merged using Photoshop version 7.0 (Adobe).

The lumen of the ER enables proteins to fold properly before they are transported along the secretory pathway (). When proteins misfold, the ER quality control system ensures that they are retained in the ER to prevent them from reaching their final destination and/or to allow for their refolding. Irreversibly misfolded proteins are eliminated by retrotranslocation to the cytosol, where they are ubiquitinated and degraded by the proteasome (for review see ). The ER factors facilitating these opposing reactions are largely unknown. We used cholera toxin (CT), which is secreted by the bacterium to study retrotranslocation. CT consists of a receptor-binding homopentameric B subunit that is noncovalently associated with a single catalytic A subunit. Once CT is secreted from the A subunit is cleaved into the A1 toxic domain and the A2 domain, which are connected by a disulfide bond and other noncovalent interactions. To intoxicate cells, the holotoxin is endocytosed and travels from the plasma membrane to the ER lumen (). In the ER, the A subunit is disguised as a misfolded protein and hijacks the retrotranslocation machinery so that the A1 chain reaches the cytosol, where it is resistant to proteasomal degradation (), whereas the B subunit remains in the ER (). In the cytosol, the A1 peptide activates a cAMP-dependent signal cascade that results in chloride and water secretion, leading to diarrhea (). Elucidating the ER–cytosol transport mechanism of CT will not only clarify a decisive step in toxin trafficking but will also clarify the retrotranslocation mechanism of misfolded proteins. Previous in vitro analysis found that the ER oxidoreductase protein disulfide isomerase (PDI) unfolds the A and A1 chains of CT (), a reaction we believe prepares the toxin for retrotranslocation. The PDI-like protein ERp29 has also been implicated in protein unfolding reactions (). However, PDI family proteins have also been shown to facilitate protein folding (for review see ). Thus, it is possible that certain PDI-like proteins are dedicated to the retention and refolding of misfolded polypeptides, whereas other PDI family members function to unfold misfolded proteins in preparation for their retrotranslocation. In this study, we developed a semipermeabilized cell system that monitors the ER–cytosol transport of CT and found that PDI facilitates the toxin's retrotranslocation, whereas ERp72, a PDI-like protein, mediates its ER retention. Furthermore, these activities were found to operate on endogenous ER misfolded proteins, indicating the generality of this mechanism. These results identify PDI family members as playing opposite roles in ER quality control and establish an assay to elucidate the retrotranslocation process of CT. To study CT retrotranslocation, we developed an assay that monitors the transport of the A and A1 subunits from the ER into the cytosol, taking advantage of a semipermeabilized cell assay that efficiently separates cytosolic from ER proteins (). CT-intoxicated HeLa cells were treated with 0.04% digitonin to permeabilize the plasma membrane and were fractionated by centrifugation. The supernatant should contain cytosolic proteins as well as ER–cytosol-transported CT, whereas the pellet should contain the plasma membrane, intracellular organelles (including the ER), and toxin that did not undergo retrotranslocation. We tested the purity of these fractions and found the ER resident protein ERp57 to be entirely in the pellet (, second panel from bottom; lanes 2, 4, and 6) and the cytosolic protein Hsp90 to be mostly in the supernatant (, bottom; lanes 1, 3, and 5). When cells were intoxicated with CT at 37°C, a portion of the A1 subunit was found in the supernatant (, top; compare lane 6 with 5), whereas the B subunit was absent in this fraction (, second panel from top; compare odd with even lanes) as expected. Typically, 15–30% of toxin and <0.01% of ER resident proteins was detected in the supernatant (calculation not depicted), indicating that the presence of the toxin in this fraction is not caused by ER leakage. This range is likely caused by the variable efficiency of the low level of detergent used in this study in permeabilizing the plasma membrane. When cells were incubated with CT at 4 or at 37°C in the presence of brefeldin A (BFA), which are conditions shown previously to block the arrival of CT to the ER (), the A1 chain did not appear in the supernatant (, top; compare lane 5 with 1 and 3). A fraction of the A1 peptide in the pellet was generated after lysis, as cell permeabilization in the presence of the alkylating reagent -ethylmaleimide (NEM) diminished the appearance of the A1 chain (, compare lane 10 with 8). However, the level of A1 peptide in the supernatant is similar regardless of whether NEM is present in the lysis buffer (, compare lane 9 with 7). Thus, postlysis reduction of CT in the pellet does not trigger the release of the A1 chain to the supernatant. To further verify the assay, we tested the fractionation pattern of a CT mutant whose A chain cannot be cleaved because of a mutation at the cleavage site (R192H). Chloride secretion triggered by the R192H mutant is attenuated dramatically (); thus, we anticipated that less of this toxin would arrive to the cytosol when compared with the wild-type toxin. Indeed, when cells were incubated with the R192H toxin, no toxin was found in the supernatant (, compare lane 3 with 1). CT-stimulated chloride secretion was previously shown to be resistant to proteasome inactivation (), suggesting that the toxin escapes proteasomal degradation in the cytosol. As expected, proteasome inactivation with MG132, which was confirmed by the accumulation of polyubiquitinated proteins (, compare lane 2 with 1), did not significantly alter the toxin level in the cytosol (, top; compare lane 5 with 3). The consistency of these results () with previous chloride secretion studies (; ) validates the use of this semipermeabilized assay as a tool to study CT retrotranslocation directly. We next tested whether PDI, an ER chaperone that unfolds the A and A1 chains in vitro (), plays a role in the ER–cytosol transport of CT. Our approach was to down-regulate PDI and other PDI-like proteins in cells using RNAi and test their effect on CT retrotranslocation. Of the 14 known human PDI-like proteins in the ER, we chose to down-regulate ERp72 and ERp57 as controls because they are expressed at similar levels as PDI (unpublished data). PDI, ERp72, and ERp57 are characterized by the presence of a CxxC sequence within their thioredoxin domain (). PDI (PDI), ERp72 (ERp72), and ERp57 (ERp57) were down-regulated in cells separately (, top); in each case, the expression of other PDI-like proteins (, bottom three panels) was unaffected. These results indicate that PDI, ERp72, and ERp57 expression can be reduced efficiently and specifically. We asked whether down-regulation of the PDI-like proteins elicited the unfolded protein response, a cellular stress response in which ER misfolded protein accumulation triggers the expression of ER chaperones such as BiP to alleviate protein misfolding. We found that BiP expression was only up-regulated marginally in PDI, ERp72, and ERp57 cells compared with wild-type cells (, compare lanes 3, 5, and 7 with 1), indicating that the lack of these proteins did not significantly induce the unfolded protein response. Moreover, the addition of tunicamycin, a drug that blocks -glycosylation and, thus, causes the accumulation of misfolded proteins, induced BiP expression in wild-type, PDI, ERp72, and ERp57 cells (, compare lane 2 with 4, 6, and 8). We conclude that down-regulation of the PDI family proteins did not globally disrupt ER function. We assessed the effect of PDI down-regulation on the transport of CT to the cytosol by examining the toxin level in the supernatant after cell fractionation. When cells were incubated with CT for 45 or 90 min, we found a decreased level of both the A and A1 chains from PDI cells when compared with wild-type cells (, top; compare lane 2 with 1 and lane 4 with 3; quantified as shown in the graphs). These data indicate that PDI facilitates toxin transport from the ER to the cytosol, which is consistent with our hypothesis that PDI-dependent unfolding of CT mediates the toxin's retrotranslocation (). We note that a low A subunit level also appeared in the supernatant. This is unlikely to be caused by ER leakage, as ER markers and CTB (CT B subunit) were absent from this fraction. Thus, it is possible that although the A1 subunit is the preferred substrate for retrotranslocation, the A chain is also transported, albeit with lower efficiency. In contrast, the A and A1 peptide level in the supernatant increased in ERp72 cells when compared with wild-type cells (, top; compare lane 6 with lane 5 and lane 8 with 7; quantified as shown in the graphs), suggesting that ERp72 acts to retain CT in the ER, a reaction that opposes PDI's function. The effects of down-regulating PDI and ERp72 on toxin transport were also observed when NEM was present in the lysis buffer (unpublished data). In ERp57 and wild-type cells, the toxin level in the supernatant was similar (, top; compare lane 10 with 9 and lane 12 with 11). We conclude that the opposing effects of PDI and ERp72 on CT transport are specific and are unlikely the result of a general disruption of ER chaperone functions. To show that PDI and ERp72 did not affect CT trafficking to the ER, we used a variant of CT harboring consensus motifs for -glycosylation on the B subunit (glycosylated CT [CT-GS]). Modification by -glycosylation, detected by a molecular mass increase in the B subunit, indicates the toxin's arrival to the ER. This toxin was previously used to show that CT is transported as a holotoxin from the plasma membrane to the ER (). Thus, -glycosylation of the B subunit indicates the arrival of the A subunit to the ER. When cells were incubated with CT-GS at 37°C and the cell lysate was analyzed by SDS-PAGE followed by immunoblotting with an antibody against the B subunit, a band larger than the nonglycosylated B subunit was observed (, lane 2). This band was shown previously to be -glycanase sensitive () and, thus, represents glycosylated B subunits. Glycosylated B subunits were not detected in cells incubated with the toxin at 4 or 37°C in the presence of BFA (, lanes 1 and 3). These experiments validate the use of CT-GS in monitoring A-subunit arrival to the ER. Significantly, the similar level of glycosylated B subunits found in wild-type, PDI, and ERp72 cells (, lanes 4–6) indicate that PDI and ERp72 act on the toxin only after it reaches the ER. Furthermore, PDI and ERp72 down-regulation does not affect CT reduction (, compare lane 2 with 1 and lane 4 with 3). Upon reaching the cytosol, the catalytic A1 subunit ADP ribosylates the Gαs protein, activating adenylate cyclase that, in turn, generates cAMP. Therefore, we measured the CT-induced cAMP response in PDI and ERp72 cells. The CT-triggered cAMP level in PDI cells was 50% lower than in wild-type cells (, left). Forskolin, which stimulates adenylate cyclase directly, elicited a similar cAMP level in wild-type and PDI cells (unpublished data), indicating that PDI down-regulation did not affect adenylate cyclase. In contrast, the CT-induced cAMP level in ERp72 cells was 40% higher than in wild-type cells (, right). This result was normalized against the forskolin-induced cAMP response, as forskolin also triggered a higher cAMP response in ERp72 cells. The higher forskolin-induced cAMP response in these cells is likely caused by a concomitant increase in cell surface expression of adenylate cyclase (unpublished data). Thus, the cAMP data are consistent with the ER–cytosol transport assay—namely that PDI retrotranslocates CT, whereas ERp72 retains the toxin in the ER. PDI's ability to unfold CT () is consistent with its role in toxin retrotranslocation. ERp72's role in facilitating ER retention suggests that it may recognize CT as a misfolded protein and attempt to “refold” the toxin's structure. Incubation of purified PDI but not the control protein BSA with CTA (CT A subunit) was shown previously to render the A and A1 chains sensitive to trypsin digestion (, compare lane 2 with 1; ), indicating that PDI unfolded the toxin. Purified ERp72 () did not cause the toxin to become protease sensitive (not depicted). We now find that regardless of PDI's presence, a higher trypsin concentration rendered the A and A1 chains sensitive to degradation (, compare lanes 3 and 4 with 1). However, under this condition, ERp72 protected the toxin from degradation (, compare lane 5 with 3 and 4), indicating that ERp72 alters CT's conformation to render it more folded and compact. Alternatively, it is possible that ERp72 protects the toxin against protease digestion by either stabilizing the native conformation of the toxin or interacting with CT in a similar manner as PDI but with reduced efficiency. Nonetheless, PDI and ERp72's opposing effects on CT's conformation likely reflect their roles in facilitating retrotranslocation and ER retention during ER quality control. Do the opposing functions of PDI and ERp72 represent a general mechanism operating in the ER? We developed a method to measure the bulk arrival of misfolded proteins from the ER to the cytosol. Upon reaching the cytosolic surface of the ER membrane, ER misfolded polypeptides are polyubiquitinated before being targeted for proteasomal degradation (for review see ). Consequently, proteasome inactivation ought to lead to the accumulation of polyubiquitinated ER misfolded proteins as well as cytosolic proteins. The effect of ER resident protein down-regulation on the accumulation of total polyubiquitinated proteins should, therefore, reflect their role in the ER–cytosol transport process. We measured the level of polyubiquitinated proteins and found that incubation of cells with the proteasome inhibitor MG132 resulted in an increase of polyubiquitinated proteins (, compare lane 3 with 1 and lane 7 with 5). However, in PDI cells, less MG132-induced polyubiquitinated proteins appeared when compared with wild-type cells (, compare lane 4 with 3; quantified as shown in the graphs), whereas in ERp72 cells, more polyubiquitinated proteins were present (, compare lane 8 with 7; quantified as shown in the graphs). Although the source of the polyubiquitinated proteins is unclear, they likely originated from the ER. Quantification showed that although PDI down-regulation reduced the total level of polyubiquitinated proteins by ∼40%, a significant portion (∼60%) was unaffected; the unaffected polyubiquitinated proteins are presumably cytosolic proteins. We then asked whether cytosolic ubiquitination reactions are affected nonspecifically by PDI and ERp72 down-regulation. IκBα is a cytosolic protein that is polyubiquitinated in response to TNFα stimulation before being degraded by the proteasome (for review see ). Indeed, we found that IκBα was degraded with similar efficiency upon TNFα treatment in wild-type, PDI, and ERp72 cells (, compare lanes 4–6 with 1–3), suggesting that IκBα ubiquitination was not disrupted by PDI and ERp72 down-regulation. These findings indicate that PDI facilitates the retrotranslocation of some ER misfolded proteins, whereas ERp72 mediates their ER retention. We further characterized this effect on a specific ER misfolded substrate. Thyroglobulin (Tg), the thyroid prohormone, is synthesized and folded in the ER before being secreted. A mutant form of Tg, cogTg, was shown previously to undergo retrotranslocation and proteasomal degradation (). PDI overexpression in CHO cells stably expressing cogTg (CHO-PDI) stimulated the rate of cogTg degradation when compared with parental CHO cells expressing cogTg (CHO-P; , compare open with closed squares). In contrast, ERp72 overexpression in CHO cells (CHO-ERp72) decreased the rate of cogTg degradation (, compare open squares with circles). These results suggest that PDI stimulates the ER–cytosol transport of cogTg, whereas ERp72 retains it in the ER. We conclude that PDI and ERp72's opposing roles operate not only on CT retrotranslocation but more generally for ER misfolded proteins. How do chaperones that belong to the same family serve opposite functions? PDI possesses two thioredoxin domains containing the redox-active CxxC motif (, CxxC box) and two thioredoxin-like domains without the CxxC motif (, white rectangles), whereas ERp72 contains three redox-active and two redox-inactive domains (). Their major difference is that PDI contains an additional c domain that is absent in ERp72. It is possible that this domain participates in the unfolding of polypeptides. Interestingly, PDI but not ERp72 was implicated in the unfolding of the nonnative structure of bovine pancreatic trypsin inhibitor during disulfide bond rearrangement (; ), supporting our conclusion that PDI specifically unfolds misfolded substrates. PDI has also been shown to facilitate the proteasomal degradation of ER misfolded proteins in yeast (), a process that presumably requires substrate unfolding and retrotranslocation. Our findings that ERp72 mediates the ER retention of CT and misfolded proteins are consistent with recent data demonstrating that ER retention of a misfolded secretory () and a transmembrane protein () is coincident with binding to ERp72. ERp72 and PDI's dedicated roles in ER quality control appear analogous to the cytosolic chaperone system that functions in protein folding and degradation (; ). In this case, the chaperones Hsp70 and TriC assist in protein folding, whereas the Hsp70–Hsp90–Sti1–Sse1 complex mediates protein degradation. As PDI can promote protein folding and unfolding reactions, it is possible that PDI cofactors exist to control these opposing processes, which is similar to Hsp70 regulation. Antibodies against PDI, BiP, and Hsp90 were purchased from Santa Cruz Biotechnology, Inc. Antibodies against ERp72 were obtained from StressGen Biotechnologies, antibodies against ubiquitin were purchased from Zymed Laboratories, and antibodies against IκBα were obtained from Biolegend. CT A and B antibodies and CT-GS were provided by the Lencer laboratory, the Tg antibodies were obtained from the Arvan laboratory, and the CT R192H mutant was provided by R. Holmes (University of Colorado, Boulder, CO). Hsp27, ERp57, and ERp29 antibodies were gifts from M. Welsh (University of Michigan, Ann Arbor, MI), S. High (University of Manchester, Manchester, England), and S. Mkrtchian (Karolinska Institutet, Stockholm, Sweden), respectively. CT, CTA, and PDI were purchased from Calbiochem. Mouse ERp72 cDNA was subcloned into pQE30, and the protein was purified using a Ni–nitrilotriacetic acid agarose column. PDI-specific (5′-GACCTCCCCTTCAAAGTTGTT-3′) siRNA was synthesized by Ambion, and ERp72- (5′-CAAGCGUUCUCCUCCAAUUTT-3′) and ERp57-specific (5′-UGAAGGUGGCCGUGAAUUATT-3′) siRNAs were synthesized by Invitrogen. 10 nM duplexed siRNA was transfected into HeLa cells using Oligofectamine (Invitrogen) according to the manufacturer's protocol. Protein expression was assessed by SDS-PAGE and immunoblot analysis. Experiments were initiated 48 (ERp57) or 72 h (PDI and ERp72) after transfection. Cells were incubated with 10 nM CT in HBSS at 37°C for 45 or 90 min. For permeabilization, 2 × 10 cells were resuspended in 100 μl of 0.04% digitonin in HCN buffer (50 mM Hepes, pH 7.5, 150 mM NaCl, 2 mM CaCl, and protease inhibitors) with or without 10 mM NEM, incubated on ice for 10 min, and centrifuged at 16,000 for 10 min. The supernatant was removed, and the pellet was washed with PBS and resuspended in 100 μl of the original buffer. Fractions were analyzed by nonreducing SDS-PAGE and immunoblot analysis. Where indicated, cells were treated with 2 μM MG132 for 90 min. HeLa cells were incubated with 50 nM CT-GS in HBSS for 3 h at 4 or 37°C in the presence or absence of 5 μg/ml BFA. Cells were harvested in TN lysis buffer (1% Triton X-100, 1.75% n-octyl-B-/d-glucopyranoside, 10 mM Tris, pH 7.4, 150 mM NaCl, 5 mM EDTA, and protease inhibitors), and the lysates were analyzed for the presence of glycosylated CTB. A cAMP enzyme immunoassay system (GE Healthcare) was used to quantify cAMP synthesis induced by 10 nM CT or 50 μM forskolin in HBSS. Samples were assayed in duplicate, and the mean optical density was used to calculate the cAMP level/well. The cAMP response was determined by dividing the cAMP level in CT- or forskolin -treated cells by the cAMP level in unstimulated cells. Forskolin induced a sevenfold higher cAMP response in ERp72 cells than in wild-type cells. Results are reported as a percentage of the wild-type CT-induced cAMP response normalized to the forskolin-induced cAMP response. Purified CTA was incubated with 1 mM glutathione, 2 μg/ml BSA, 2 μg/ml PDI, or 0.2 μg/ml ERp72 for 30 min at 37°C. 100 or 300 μg/ml trypsin was added to the samples for 30 min at 4°C. Samples were analyzed by nonreducing SDS-PAGE followed by immunoblot analysis. Cells were incubated with or without 20 μM MG132 for 30 min and lysed in a buffer containing 1% Triton-X, 10 mM NEM, 30 mM Hepes, pH 7.4, 150 mM NaCl, 5 mM EDTA, and protease inhibitors. Cleared lysates were analyzed by 4–20% SDS-PAGE followed by immunoblot analysis. To monitor ubiquitin-dependent degradation of IκBα, cells were treated with 10 ng/ml TNFα for 15 min, and the cell lysate was analyzed for the presence of IκBα. The cogTg degradation rate was analyzed as previously described ().

polarizes cell growth to the bud during cell replication and to the mating projection when cells are induced by pheromones to change their shape to form shmoos. This polarization process is characterized by a hierarchy of steps. First, the site on the cell surface is selected by intrinsic and extrinsic cues. This site is marked by the deposition of landmark proteins. Second, cell polarity is established by the activation of small GTPases with CDC42 as the major player. Last, a multiprotein machine is assembled that spools out actin cables to direct post-Golgi traffic to the site of polarized cell growth (; ; ; ). During budding, membrane traffic is directed by actin cables to the bud, and the septin ring at mother-daughter cell neck region functions as a physical barrier, preventing diffusion of membrane components from the bud to the mother cell (; ). During mating, the biosynthetic transport is directed to the shmoo tip (). However, there is no diffusion barrier like the septin ring and most proteins diffuse laterally over the entire cell surface. Nevertheless, a specific subset of proteins required for mating is clustered at the tip of the mating projection. We have shown previously that the polarized distribution of membrane components involves raft lipids (sphingolipids and ergosterol) and the actin cytoskeleton (). Based on these observations, we proposed that lipid rafts are clustered at the tip of the mating projection and that this process is important for the retention of the associated molecules at the mating projection. Recently, it was demonstrated that cycles of endocytosis combined with polarized membrane delivery into the mating projection are able to restrict protein localization to the tips of shmoos (). Here, we report that polarized localization of Fus1p, a type I transmembrane protein involved in cell fusion (), does not require endocytosis. Instead, the protein is retained at the tip of the mating projection through the interaction of its cytosolic tail with a multiprotein scaffolding machinery. Additionally, we provide evidence that the lipid bilayer at the tip of the mating projection is more ordered than over the cell body and that sphingolipids are required for this specific lipid organization. To revisit the kinetic recycling model we have analyzed the role of polarized delivery and endocytosis in polarizing Fus1p, a type I transmembrane protein involved in cell fusion (; ), to the tip of the mating projection. We first analyzed shmoo tip delivery of Fus1p and compared it to another marker protein that is distributed all over the plasma membrane of mating cells, Mid2p (). Mid2p is a cell wall integrity sensor, and similarly to Fus1p, it is a type I transmembrane protein (). 1 h after induction of expression, the marker proteins were delivered to the shmoo tip where both were localized at this point. However, 2 h later Mid2p had diffused over the entire plasma membrane, whereas Fus1p remained at the tip. We then analyzed the effect of endocytosis on the process of Fus1p polarization. We also used Snc1p, a yeast v-SNARE involved in post-Golgi plasma membrane transport, as a second marker protein that is tip localized. It has been shown that Snc1p polarization was abolished after inhibition of endocytosis (). In endocytosis-deficient cells, Snc1p was no longer polarized but was distributed over the plasma membrane of the shmooing cells. In contrast, the polarization of Fus1p to the tip of the mating projection remained normal in cells (). Thus, there must be an alternative mechanism that maintains biosynthetically delivered Fus1p at the shmoo tip irrespective of ongoing cycles of endocytosis and exocytosis. Important also to note is that most mutants that inhibit endocytosis mate with similar efficiency as wild-type cells (). These findings confirm that the kinetic polarization model using endocytosis and polarized exocytosis is involved in local concentration of membrane proteins such as Snc1p. However, this is not the only mechanism used by shmooing cells to polarize their mating machinery. One reason why Fus1p is retained at the mating tip could be due to interaction with the cell wall, as was demonstrated for glycosylphosphatidylinositol-anchored proteins (). Thus, we constructed fusion proteins between Fus1p and Mid2p where we swapped the extracellular, the transmembrane, and the cytosolic domains of the two proteins (schematically shown in ). Analysis of their surface distribution demonstrated that the information for mating tip retention was localized to the cytosolic tail (). The Mid2 protein carrying the cytoplasmic domain from Fus1p was localized to the mating projection. PAGE and Western blot analysis confirmed that this protein displayed a pattern of glycosylation, typical for mature Mid2p ( and unpublished data). We then analyzed how this chimeric protein behaved in mutants in which endocytosis was inhibited, both in cells and at the nonpermissive temperature in cells. This chimeric protein behaved like Fus1p and maintained its polarization in endocytosis-deficient cells ( and Fig. S1, available at ). Thus, we concluded that the cytosolic tail of Fus1p mediates protein retention at the tip and that interactions with the cell wall cannot explain the polarization. The cytosolic tail of Fus1p is 416 amino acids long and contains an Src kinase homology 3 (SH3) domain close to its COOH terminus, followed by a proline-rich domain, both known to be responsible for protein–protein interactions (). We deleted the SH3 domain from the chimeric protein Mid-Fus (used in ) or Fus1p (not depicted) and saw no effect on polarization. At this time, a report from appeared, in which a detailed analysis of the cytoplasmic domain of Fus1p was described. They showed that both domains were important for mating efficiency, but even the double-mutant protein was polarized normally in wild-type cells. Because mutations in these domains prevented protein interaction with the scaffolding machinery () we considered the possibility that the double mutant of Fus1p could be polarized via the endocytic recycling mechanism. To test this possibility we expressed the mutated Fus1p in the endocytosis-deficient strain and found that protein polarization was still normal (). We concluded that additional sites on the cytoplasmic tail of Fus1p might contribute to Fus1p retention to the mating tip. In a detailed two-hybrid analysis it was demonstrated that the cytosolic tail of Fus1p interacts with several key players in mating polarity, including the GTP-bound form of CDC42, components of the polarisome Pea2p and Bni1p, Fus2p, and Ste5p, the scaffold protein for MAP kinase signaling (). We analyzed the Mid-Fus protein in , , , , , and deletion mutants. The localization was normal in all mutants accept in cells, which did not form shmoos and in , which exhibits a subtle defect in protein polarization (Fig. S2). From these data we propose that Fus1p is directly embedded in a dynamic network of protein–protein interactions that is responsible for scaffolding and localization of Fus1 to the shmoo tip as part of the mating machinery. Based on the findings that polarization of the mating machinery to the shmoo tip is inhibited in and in cells; that mutations that severely affect the synthesis of the major raft lipids in yeast reduced mating efficiency; and on the polarized distribution of filipin, a molecule that has high affinity for sterols, we postulated previously that raft lipid clustering plays a role in establishing and maintaining mating tip polarization (). Because partitioning of filipin does not directly correlate with lipid ordering in the bilayer, we took advantage of a dye that does. Laurdan is an environmentally sensitive dye that has a peak of emission shifting from ∼500 nm in liquid-disordered membranes to ∼440 nm in ordered membrane domains (). We simultaneously recorded the Laurdan fluorescence intensity in two channels. By expressing a normalized ratio of the two emission regions—the general polarization (GP; see Materials and methods)—Laurdan fluorescence provides a relative measure of lipid order in cell membranes. Importantly, Laurdan does not preferentially partition into a specific lipid phase, nor do GP values depend on the local probe concentration within the membrane (). The GP images revealed that the membrane at that mating projection is more condensed and ordered than the domain on the opposite site of the cell (). Hence, the membrane at the mating projections displayed the biophysical characteristic that is expected for raft clustering. The coalescence of condensed membrane at the tips of shmoos was also found to occur in the endocytosis-deficient strain but not in the sphingolipid mutant (). These data clearly demonstrated the asymmetric organization of the lipids in the plasma membrane of yeast cells during mating. More detailed analysis is needed to understand the molecular mechanisms responsible for the formation and maintenance of the mating projection. Also, mammalian cells use raft clustering to polarize their cell surfaces during cell migration or cell–cell contacting during immune recognition. In migrating neutrophils it was demonstrated that lipid raft clusters are localized to the rear of the cells in an actin-dependent manner (). In migrating T-lymphocytes, showed that two types of raft clusters are assembled at opposite poles, at the leading edge and at the uropod. Recently, it was also demonstrated that when the T cell receptor is activated, a condensed raft cluster is formed at the activation site (). Each raft-clustering process is specific in that a subset of raft components is included in the assembly, associating and disassociating from the cluster dependent on their raft-partitioning characteristics (; ) and the kinetics of the protein–protein interactions (; ). This mechanism could also drive the surface polarization during yeast mating (). The scaffolding of proteins would occur mainly through protein–protein interactions. Nevertheless, coming together of proteins with a condensed lipid domain at the mating tip could lead to activation of the mating machinery spatially and temporally by specific lipid–protein interactions (). These interactions could involve integral proteins binding to raft lipids in the bilayer. For instance, the EGF receptor has been shown to be activated by interactions with the ganglioside Gd1a and the glutamate receptor by raft-cholesterol (; ). Recently, it was demonstrated that yeast Ste5p, a protein that plays a crucial role in pheromone signaling and interacts with the cytoplasmic tail of Fus1p, has a phospholipid binding domain that is necessary for protein localization and signaling (). The generation of cell surface polarity during yeast mating is thus a complex process involving on one hand endocytosis and recycling and on the other hand establishment of the site where the mating machinery is scaffolded and the membrane reorganized. It is our contention that complex membrane processes such as cell surface polarization are driven by protein–protein and protein–lipid interactions. However, only future work directed specifically toward analysis of these issues will unravel the mechanisms involved. In this study the following yeast strains were used: RH690-15D [wild-type] () was obtained from H. Riezman (University of Basel, Basel, Switzerland), and RH1965 [end4Δ] () and RH268-1 [en4-1 ts] () were obtained from C. Walch-Solimena (MPJ-CBG, Dresden, Germany). 1302-WT (BY4742), deletions are in BY strains derived from S288C () and were obtained from EUROSCARF. Cells were grown overnight in yeast extract/peptone (YP) medium containing 2% raffinose (YPRaf) as a carbon source at 24°C. For the induction of expression from the GAL-S promoter, 2% galactose was added. To induce a mating response, 5 μM α-factor (Sigma-Aldrich) was added and cells were incubated for 3 h at 24°C (or as indicated). Plasmids used in this study are listed in . All constructs created in our lab are based on the centromeric plasmid p416 () and expression was driven from the inducible GAL-S promoter. Plasmids p4269, p4580, and p4667 containing mutants of under control of its own promoter were obtained from the C. Boone lab (University of Toronto, Toronto, Canada; ). The plasmid containing GFP-SNC1 under control of constitutive TPI promoter was obtained from the H. Pelham lab (MRC Laboratory of Molecular Biology, Cambridge, UK; ). Construction of plasmids MBQ30, MBQ35, TPQ53, and TPQ55 was described previously (; , ). Plasmids TPQ63, TPQ65, TPQ72, and TPQ57 were created by triple ligation method, where two PCR-amplified fragments of DNA are introduced into a vector (for details see ). To generate TPQ63, a DNA fragment coding the extracellular domain of Fus1p linked to the transmembrane domain (TMD) from Mid2p (amplified from plasmid TPQ55 with primers containing XbaI and BamHI sites) and a DNA fragment coding the cytoplasmic tail of Fus1p fused to GFP (amplified from TPQ53 with primers containing BglII and HindIII sites) were co-ligated to the MBQ1 vector digested with XbaI–HindIII. To create TPQ65, a fragment of DNA coding the extracellular domain of Mid2p linked to the TMD from Fus1p (amplified from plasmid TPQ53 with XbaI and BamHI sites added on the primers) and a DNA fragment containing the cytoplasmic tail of Mid2p fused to GFP (amplified from MBQ35 with BglII and HindIII sites added on the primers) were coligated into MBQ1 (XbaI–HindIII). To make TPQ72, a DNA fragment coding the extracellular domain and TMD of Fus1p (amplified from plasmid MBQ30 with primers containing XbaI and BamHI sites) and a DNA fragment containing the cytoplasmic tail of Mid2p fused to GFP (prepared as for TPQ65) were co-ligated to the XbaI–HindIII-digested vector MBQ1. Primers TPQ94 and TPQ97 were constructed by homologus recombination in RH690-15D cells. To generate TPQ94, a DNA fragment (obtained from TPQ63 with BglII–HindIII digestion) coding the TMD from and the cytoplasmic tail from , followed by the GFP, was cotransformed with MBQ35 (linearized with BamHI). To create TPQ97, a DNA fragment coding the truncated cytoplasmic tail of Fus1p followed by the GFP coding sequence (PCR amplified from TPQ57) was cotransformed with the NheI-linearized TPQ97. The successful recombination was verified by observation of fluorescence in microscope and sequencing of the plasmids. For Laurdan microscopy, cells were treated with 5 mM sodium azide and 4% paraformaldehyde for 10 min at 24°C; 250 μM Laurdan (Molecular Probes) was added for a further 5 min at 24°C and cells were washed twice and imaged in water. Laurdan fluorescence was excited at 800 nm with a Verdi/Mira 900 multi-photon laser system and intensity images were recorded simultaneously in the range of 400–445 nm and 445–530 nm for the two channels (Bio-Rad Laboratories), respectively. The generalized polarization GP, defined aswas calculated for each pixel using the two Laurdan intensity images. GP images were pseudocolored in Adobe Photoshop. The GP values were determined at the mating tip or opposite the tip in a region measuring ∼1.2 × 0.2 μm, and each data point (or symbol) in the scatter plots represents derivatives from one individual cell. GP values were corrected using the G-factor obtained for Laurdan in DMSO for each experiment. Means and standard deviation of multiple comparisons were compared with one-way ANOVA with Tukey's post-testing assuming Gaussian distributions (PRISM) (). Fig. S1 shows polarized distribution of chimeric Mid(cyt-FusΔSH3) protein in end4-ts cells. Fig. S2 shows localization of Mid(cyt-FusΔSH3) in bni1Δ and WT cells. Online supplemental material is available at .

Centrins are a family of EF-hand–containing proteins most closely related to calmodulins and, like calmodulin, probably have multiple unrelated functions in association with other proteins. At present, the best-characterized functions of centrins are in microtubule organizing centers (MTOCs), such as centrosomes and basal bodies (), where they are often present in different parts of the MTOC (; ) and probably have specialized functions associated with these various locations. Two clear functions in MTOCs have been established: one is in the duplication of the MTOC (; ; ) and the other is as constituents of filaments within and attached to the MTOC. Some of these filaments can contract in response to changes in Ca concentration (). However, the molecular basis for these centrin-based functions in MTOCs has not yet been established. Budding yeast has a single simple MTOC, the spindle pole body (SPB), responsible for the organization of both the spindle and cytoplasmic microtubules (see diagram in ). The SPB is a multilayered structure embedded in the nuclear envelope, which remains intact during yeast mitosis. Attached to one side of each SPB is a specialized area of the nuclear envelope called the half-bridge, which has a critical role during SPB duplication () and is where the single centrin in budding yeast, Cdc31p, is localized (). The half-bridge consists of a densely stained rectangular area of the nuclear membrane together with a cytoplasmic outer layer. The assembly of cytoplasmic components of the daughter SPB initiates from a satellite structure at the distal end of the bridge () and continues to form a duplication plaque on the cytoplasmic side of the bridge (). The SPB is then inserted into the nuclear envelope, nuclear SPB components are added, and the two SPBs separate by dissociation or cleavage of the bridge, leaving a half-bridge with each SPB. Centrin has an essential function during SPB duplication, and temperature-sensitive (ts) mutants arrest with a single large SPB (). Centrin binds to three proteins in the half-bridge, Kar1p (; ), Mps3p (), and Sfi1p (). Kar1p and Mps3p have transmembrane domains and are probably associated with the lipid bilayers of the half-bridge. Sfi1p, which has an essential function during SPB duplication, does not have a transmembrane domain; however, it has ∼20 continuous conserved repeats in the center of the protein. Five of these repeats were tested, and four were found to bind centrin in pull-down assays (). This suggests a model for the Sfi1p–centrin complex where ∼20 molecules of centrin bind continuously to the repeats on a molecule of Sfi1p, possibly producing a filamentous structure. This paper describes structural studies on the Sfi1p–centrin complex, its arrangement on the half-bridge and bridge, and a model for its role during SPB duplication. Sfi1 repeats that bind centrin have a consensus AXLLXF/LXW (). There were only a small number of sequences available when Sfi1 repeats were first analyzed, but since then more sequences have become available and these show a similar consensus (), and there are also biases in sequence at other positions in the repeat (see legend and the supplemental text, available at ). Fungi have a variable repeat length, whereas most repeats in other eukaryotes are 33 amino acids long (). All the complete sequences analyzed in have between 20 and 24 continuous repeats. Sfi1p was originally given only 17 repeats with gaps (). These gaps were close to exact multiples of repeats, and inspection of the gap sequence, particularly when aligned with other fungal species, has identified potential repeats in the gaps (). To determine the structure of the Sfi1p–centrin complex, we coexpressed fragments of Sfi1p containing two to three repeats from as GST fusion proteins together with yeast centrin (Cdc31p) in bacteria. All constructs tested gave stable complexes that could be purified by gel filtration and Q Sepharose chromatography in 1 mM EGTA–containing buffers, confirming that the Sfi1p–centrin interaction is stable at low Ca concentrations (). One construct (crystal 1; ) containing two repeats (K643-E710) gave crystals without Ca (see Materials and methods) that could be solved at 3.0 Å (, Table S1, and the supplemental text, available at ). Another construct (N218-H306; crystal 2; ) had three repeats, was crystallized with 0.1 M calcium acetate, and was solved at 3.2 Å (, Table S1, and the supplemental text). High Ca concentrations were necessary because concentrations <50 mM did not give suitable crystals. In these crystals, Ca was not present at positions other than EF-hands, as an anomalous difference map (see Materials and methods) showed Ca only in EF-hands 1, 3, and 4; EF-hand 2 is probably unable to bind Ca because it lacks essential ligands (). Both of these crystal structures show the Sfi1p fragments as α helices with centrin in an extended conformation bound to each repeat. The centrin N-terminal domains bind to the N-terminal half of the Sfi1 repeat containing the conserved alanine, whereas the centrin C-terminal domains bind to the more conserved C-terminal half of the Sfi1 repeat (). Parts of the C-terminal domain of centrin E, and that part of the Sfi1 repeat it probably binds to (R295-W304), were not visible in the electron density map. Surprisingly, despite the absence of Ca in crystal 1 and the presence of Ca in crystal 2, the conformations of the N- and C-terminal domains of centrins A–E are very similar (). The root mean square deviations (rmsd's) for a comparison of the main chain of the N-terminal domain (C L18-K91) of centrin A with the same domains of centrins B–E are 0.5, 0.7, 0.7, and 1.0 Å, respectively, and for the C-terminal domains (C L95-C158) the rmsd's are 1.2, 0.5, and 0.3 Å (centrin E was excluded from this comparison). The N-terminal domain is in the closed conformation, whereas the C-terminal is open (see the supplemental text). EF-hand proteins such as calmodulin in the apo form have closed or semi-open conformations of their N- and C-terminal domains, respectively. On binding Ca, the domains open up, exposing hydrophobic surfaces that are able to bind other proteins (). However, in the case of yeast centrin bound to Sfi1p, and despite the rather high Ca concentration in the crystal, there is little influence of Ca on the conformation. The conformationally near identical N-terminal domains interact with the N-terminal part of the Sfi1p repeat through parts of a hydrophobic patch between centrin helices I and II containing L31, F32, V47, the side chain of K50, and A51. Centrins A and C interact with bulky residues Y651 and F224 in the alanine position 10 of the Sfi1 repeat (), and these push the Sfi1p helix away from the N-terminal domain (). In contrast, centrins B and D interact with the less bulky alanines A682 and A255. These allow the Sfi1p helix closer to the N-terminal domains () so that the carbon atoms of the side chains of residues K685 and R686 () and F258 and R259 () can interact with a second overlapping hydrophobic patch containing A28, L31, A51, and L52. In addition, R686 and R259 can form a potential salt bridge with E97. These extra interactions may explain the preference for alanine at position 10 of the Sfi1 repeat. The N-terminal domain of centrin E also interacts with Sfi1p similarly to centrins B and D, but, probably because of the shorter Sfi1 repeat, this occurs one turn of the helix further down. The alanine interaction is now with A285 (), which is in the equivalent interaction position to A682 and A255 (), and allows F289 to interact with the second hydrophobic patch, similar to the side chain of R686. Unfortunately, part of the C-terminal domain of centrin E is not visible, but that part that is visible does make normal interactions with F294 (). This centrin has to be in a less extended conformation to make these interactions, which seems to necessitate a bend in the Sfi1p helix between the N- and C-terminal domains of centrin E (). The centrin C-terminal domains also have a similar conformation and have mainly hydrophobic interactions with Sfi1p (). They interact mainly with the more conserved C-terminal part of the Sfi1p repeat and, as expected, these amino acids are involved in a large part of the interactions. These interactions are not described in detail, as they are similar to interactions of Kar1p with the C-terminal domain of centrin (). This domain is very similar to the centrin A C-terminal domain (rmsd 0.9 Å) and to the open form of the calmodulin C-terminal domain (). However, although the interactions of centrin with Kar1p and Sfi1p are similar, the two binding sequences are reversed (). This ability to bind the same motif in reverse has previously been found for calmodulin () and other helical motifs (). In conclusion, these results show the N- and C-terminal domains of centrin as relatively rigid structures, unaffected by Ca when they are bound to Sfi1p but able to recognize heterogeneous sequences of variable lengths as α helices. One of the most interesting features of the two crystal structures is that similar centrin–centrin interactions are made (). These are between the C-terminal domain of one centrin and the N-terminal domain of the next, with the second centrin rotated ∼65° clockwise around the Sfi1p helix. These interactions are mainly potential hydrogen bonds between K58 and the main chain carbonyls of E139 and D142, a potential salt bridge between H43 and E140, and hydrophobic interactions between F141 and H43 and also L62 and L143 or a potential salt bridge between R59 and D144. All of these amino acids were mutated to alanine to determine whether any phenotype was associated with the loss of these interactions. Earlier studies showed that the mutation F141A is lethal () and D142A gave a ts phenotype (). We found H43A also gave a ts phenotype, but single mutations in the other amino acids gave no clear phenotype, although the double mutations H43A, K46A and H43A, K58A were lethal (not all combinations were tested). These results suggest that some of these interactions are important, though we cannot exclude an indirect conformational effect. Several of the amino acids involved—H43, L62, and F141—are specific to the centrin 3 family (), suggesting that only this subfamily can form this particular type of filament in association with Sfi1p. The most obvious model for the structure of the Sfi1p-repeat region based on the two crystal structures solved here is a filament, with the 20 or so repeats as a single continuous α helix wrapped with centrins, also in a semiordered helical array. We assume that all the centrins within Sfi1p make the centrin–centrin interactions described in the previous paragraph, which sets the angle between adjacent centrins at ∼65°. This leads to a problem because of the geometry of the α helix, where each amino acid is rotated by 100° along the helix. Only one repeat length, 33 amino acids, will rotate the corresponding amino acids in the next repeat by 60°, close to the correct angle, which presumably accounts for why this repeat length is favored in higher eukaryotes (). Other repeat lengths spread the corresponding amino acids at 20° intervals through the circumference of the helix. An example of how other repeat lengths are accommodated can be seen between centrins A and B and centrins C and D, where the repeats are both 31 amino acids (measured between the conserved tryptophan positions), predicting a rotation of 220°. This large angle is compensated for by partially unwinding the Sfi1p helix between the repeats in both cases and probably also by bending this helix and variations in the linker angles between the centrin N- and C-terminal domains. There is little unwinding in the Sfi1p helix between centrins D and E, probably because the repeat length here is 26 amino acids, predicting a rotation of 80°, close to the 65° needed. This shorter repeat is accommodated by the N-terminal domain interacting with the Sfi1p helix one turn later (), putting centrin E into a less extended form and probably bending the Sfi1p helix so that the C-terminal domain can make normal interactions (). This ability of the N-terminal domain to interact with different α helix turns probably accounts for the peak distribution of repeat lengths in fungi (), which are separated by distances corresponding to helical turns. The model for the Sfi1p–centrin complex as a single Sfi1p α helix with centrins wrapped continuously around in a helical arrangement was tested by coexpression of ∼15 Sfi1 repeats (K246-E677; ) with yeast centrin (). The molecular mass of this complex was measured by nanospray mass spectrometry under nondenaturing conditions and gave a value of 336,290 D (). This is reasonably close to a theoretical value of 332,713 D for a 1:15 Sfi1p–centrin complex, with retention of ∼3.6 kD of buffer molecules in the gas phase ions, as is often found for large complexes analyzed under conditions gentle enough to preserve quaternary interactions (). A minor component with a mass of 317,308 D was also observed. This is consistent with a 1:14 stoichiometry (theoretical mass = 314,093 D), suggesting that partial dissociation is able to occur in solution. This complex was examined by EM shadowing () and showed filaments 59 ± 7 nm ( = 74) in 1 mM EGTA and 59 ± 8 nm ( = 82) in 2 mM CaCl. Thus, there was no discernable difference in filament length between low and high Ca. This is in agreement with the structures () solved from crystal 1 (no Ca) and crystal 2 (0.1 M Ca), where Ca has little effect on the conformations. If the Sfi1p fragment examined here was a single continuous α helix covered by centrins, then the 432 amino acids in this construct would predict a length of 65 nm, close to the 59 nm observed. The bridge structure of the SPB, where Sfi1p is localized (), has been reported to change in length during the cell cycle from 90 nm in single SPBs to 150 nm in satellite-bearing SPBs (); however, the numbers of cells examined in this study were small because the appropriate images occur rarely in thin sections. To confirm this change in length and measure the bridge length in paired SPBs, we reexamined all of our EM data collected over the years (see Materials and methods). We found the following lengths: for half-bridges from single SPBs in haploid (K699) mitotic cells, 57 ± 5 nm ( = 20), and for bridges from satellite-bearing and paired SPBs, 117 ± 9 nm ( = 53). The same numbers for a diploid strain (K842) were 60 ± 5 ( = 6) and 112 ± 9 nm ( = 9). There was no significant difference in bridge lengths between paired SPBs and satellite or duplication plaque-bearing SPBs. Another diploid strain (NCYC74) gave 109 ± 11 nm ( = 12) for paired SPBs, and a tetraploid strain () showed 118 ± 9 nm ( = 9) for satellite-bearing SPBs. These relative distances confirm the differences found earlier (), though there is a discrepancy in the absolute value of the numbers that we cannot explain. There may be some ambiguity in the absolute value of the numbers because of different fixation conditions; however, it seems probable that differences between the numbers under the same fixation conditions are valid. The numbers suggest an approximate doubling in bridge length between single SPBs in mitotic cells and cells undergoing SPB duplication and indicate that, in contrast to SPB size, which increases with ploidy (), bridge length remains constant as ploidy increases. When the localization of Sfi1p was examined by immuno-EM in yeast, it was noticed that it appeared to have a restricted distribution on the bridge: at the distal end of the half-bridge in single SPBs and in the center of the bridge in paired SPBs (). This staining used anti-GFP with Sfi1p labeled with GFP at the C terminus; thus, the staining pattern may reflect the vicinity of the C-terminal region of Sfi1p. We have shown that a fragment of the Sfi1p–centrin complex containing ∼15 repeats forms a filament 60 nm long, which suggests that the full complex containing 21 repeats could be 90 nm long. This is long enough to span the 60-nm half-bridge with the rest possibly in the central plaque, and it agrees with the staining of centrin spread along the mainly cytoplasmic side of the bridge (). To determine whether the N and C termini of Sfi1p are distant from each other, with the N terminus associated with the edge of the central plaque, a yeast strain containing Sfi1p labeled with GFP at the N terminus was stained with anti-GFP (). To quantify these observations, these data, together with the earlier results ( ), were plotted as the presence or absence of silver deposition for 10-nm sectors along the SPBs and bridge (note that this would broaden the actual distribution, as it does not allow for the volume of deposition). The label itself before silver deposition, rabbit IgG with Fab anti–rabbit-Nanogold, would span ∼20 nm (). The data () show a different distribution for the GFP on the N terminus compared with the C terminus. The N-terminal label is always close to the junction between the SPB and the proximal end of the half-bridge or bridge and in paired SPBs shows two separate sites of localization close to each SPB. In contrast, the C-terminal label, as found previously (), is either close to the distal end of the half-bridge or close to the center of the bridge in paired SPBs (). The N-terminal label also showed another class of labeling for single SPBs that appeared to have full bridges (, bottom). These are either SPBs at a very early stage of duplication, when daughter SPB components start to assemble (), or satellite-bearing SPBs or paired SPBs where it was not possible to locate a satellite or second SPB in the serial sections. Here, the mother SPB often had reduced accessibility to the antibody, hence the difference in peak heights at the N and C termini. Again, these SPBs show a bimodal distribution of stain, peaking close to the edge of the mother SPB and the expected position of new SPB assembly. The staining with both labels was always cytoplasmic. Our interpretation of these data is that the N terminus of Sfi1p lies close to the edge of the central plaque and the filamentous Sfi1p–centrin repeats span the length of the half-bridge, ending with the C terminus of Sfi1p at the distal end of the half-bridge. During duplication, the doubling in the length of the bridge takes place by the end-to-end addition of another molecule of Sfi1p by association of their C termini, and the N terminus of this second Sfi1p molecule is eventually associated with the edge of the central plaque of the daughter SPB. Earlier work showed that, in contrast to the binding of yeast centrin to Kar1p (), Ca did not appear to be necessary for the binding between centrin and Sfi1p as assayed by pull downs (), and this was confirmed here, as all complexes were prepared in the presence of EGTA. Our structures provide a clear reason for this: Ca does not appear to change the conformation of centrin when bound to Sfi1p. Very recently, the structure of the N-terminal domain of centrin in the presence of Ca was solved by nuclear magnetic resonance (). This has a different conformation from our five structures (the rmsd against the N-terminal domain of centrin A is 2.6 Å). Probably the constraints caused by binding Sfi1p and the centrin–centrin interactions prevent the N-terminal domain taking up such a conformation, or this may reflect the different members of the centrin family (the centrin is a centrin 2, whereas yeast centrin is a centrin 3). It would be very interesting to examine the recently discovered centrin 3 (). The lack of a significant effect of Ca on the yeast centrin–Sfi1p interaction may reflect the fact that there is so far little evidence for a role of Ca in the yeast cell cycle (). The only increase in cytosolic Ca concentration that has been detected in budding yeast under laboratory conditions is after mating pheromone treatment (), and indeed yeast calmodulin is able to perform all its essential functions in the absence of Ca binding (). In higher eukaryotes, Ca transients have been detected during the cell cycle (), and we found that stabilization of several human Sfi1–centrin 3 complexes containing three repeats required the presence of Ca to remain intact on ion exchange chromatography (unpublished data). Thus, Ca may play more of a role in Sfi1–centrin interactions in higher eukaryotes. An extensive study of yeast centrin has been performed to identify mutations that cause a ts phenotype or lethality () and thus may destabilize the structure or an interaction. A total of 40 mutants were isolated in this screen and in an earlier study (). For 33 of these, we can explain their effect by either altering a particularly conserved position in the EF-hand () or changing a core residue in centrin or a residue involved in the interaction between centrin and Sfi1p (E97G, A101T, F105L or Y, and C158F). Of the remainder, four (K112E, D144N, E148A, and E148Q) are within the EF-hands and should be acceptable substitutions (), though they may perturb the relationships between the adjacent helices within an EF-hand. One mutation (R73W) places a hydrophobic tryptophan on the surface, and the final two (K91R and P94S) are in the linker region. P94 is conserved in all centrin 3 sequences (), whereas in calmodulins and other centrins it is replaced by S or T. With S or T in place of P94, helices IV and V can be continuous, as they are in Ca-calmodulin. The presence of the helix-breaker P94 between these two helices ensures they are always separate and may give additional flexibility to the linker. We cannot explain structurally why K91R gives a ts phenotype, as this is a surface residue with no interactions; however, this is one of the substitutions that causes a large decrease in Kic1p kinase activity () and thus may perturb the binding between centrin and Kic1p (). Perhaps the most similar structure to the Sfi1p–centrin filament are the IQ repeats in myosin, which bind either calmodulin or myosin light chains, sometimes in the absence of Ca (). These might be a lever arm that amplifies the movement of the myosin heavy chain along the actin filament. The structures of several of these have been solved and, like Sfi1p, all show the IQ repeats as an α helix but also with some distortions. In scallop myosin, the helix is bent between the two light chains, and toward the C terminus it is distorted into a right angle (). The light chains are less extended along the helix than centrin, they stabilize ∼20 residues, compared with 25–33 for centrin, and are in the opposite orientation to centrin. Both light chains make direct interactions with each other, but these are different from the centrin–centrin interactions found here. Myosin V has up to six continuous IQ repeats, and the structure of two of these has recently been solved. The structure is similar to that of scallop myosin but shows no direct interactions between the light chains (). These structures and the two in this paper show that by imposition of kinks, bends, and bulges in an α helix, EF-hand proteins build a filament using a long α helix as a scaffold. It is not yet clear what the particular properties of these different types of filament are, though this type of arrangement is an attractive model (; ) for Ca-dependent contractile organelles such as the spasmoneme () and centrin-containing striated flagellar roots (). The immuno-EM staining of Sfi1p at both the N terminus () and C terminus () shows only cytoplasmic staining, so it seems likely that Sfi1p, which has no transmembrane domain, either is positioned just above the densely stained outer nuclear membrane of the bridge or is a constituent of the cytoplasmic half-bridge outer layer (). This location suggests that a potential role in the elasticity of the bridge during assembly of the daughter SPB (; ) is less likely and, indeed, if Sfi1p is part of the outer layer, then its length would not change during SPB duplication (), and this would be more consistent with the structural work presented here. The cytoplasmic part of the bridge is a flat rectangular structure (see plate 5 in ), and our immuno-EM results suggest that Sfi1p is positioned across this structure with the N-terminal domain associated with the side of the SPB, the filamentous Sfi1 repeats and centrin laid in rows across the rectangle with the C-terminal domain at the distal end of the half-bridge (). The bridge, which is twice the length, would have the Sfi1p C-terminal domains associated, thus connecting the two SPBs. The flat shape of the bridge suggests that there is only one or very few layers of Sfi1p–centrin filaments. If this is the case, our crystal structures suggest that there are unlikely to be extensive similar side-to-side interactions between the centrin-coated Sfi1p filaments because of the 65° twist in the centrin positions along the Sfi1p α helix (). This arrangement would resemble a Venetian blind with the slats held together at wide intervals, and the positions of the intervals would depend on the twist between each centrin. This relative lack of side-to-side interactions could explain how both microtubules and Sfi1p filaments are accommodated at the bridge at the same time. Microtubule assembly is initiated at the bridge during G/S () and involves Kar1p, a transmembrane protein localized to the cytoplasmic side of the bridge (), which binds the γ-tubulin–containing Tub4p complex via Spc72p () and initiates microtubule assembly. Thus, these microtubules are anchored close to the outer nuclear membrane and might have to pass through cytoplasmic bridge components. If the Sfi1p filaments were organized like a venetian blind with widely spaced linkages between the filaments, then with some adjustments in filament position, microtubules could be accommodated between the filaments. The suggested arrangement of Sfi1p on the half-bridge and bridge () extends the earlier model for SPB duplication (), which proposed that the bridge had the property of binding cytoplasmic SPB components at either end. We extend this and propose that it is the N terminus of Sfi1p that either directly or indirectly binds the SPB components. An early step in SPB duplication would be the end-to-end association of Sfi1p at the C terminus, which might have to be activated at the appropriate point in the cell cycle or bind other proteins. This association would double the length of the bridge and provide a new N terminus of Sfi1p able to bind SPB components and thus start the assembly of the new SPB. The specificity of the association of the two C termini and the binding of the N terminus to SPB components would ensure that only a single copy of the SPB is produced. After assembly, the two SPBs would separate to form a spindle by dissociation of the connection between the C termini of Sfi1p. Fragments of Sfi1p containing between 2 and 15 repeats were cloned as GST fusion proteins into the dicistronic vector pGEX-6p-2rbs (a gift from A. Musacchio, European Institute of Oncology, Milan, Italy) with yeast centrin (Cdc31p) in the non-GST site. All constructs produced soluble protein at 25°C, in contrast to expression of Sfi1p alone (), which only gave soluble GST fusion proteins in the presence of sarkosyl. These bound to glutathione beads but could not be eluted with glutathione. Complexes were isolated with glutathione beads, GST was removed, and the complex was released by PreScission cleavage. Further purification was by gel filtration on Superdex 200 in 10 mM Tris-Cl, pH 8.0, 0.15 M NaCl, 1 mM DTT, and 1 mM EGTA and then ion exchange chromatography on Q Sepharose using a 0.3–0.43 M NaCl gradient in the same buffer. Crystals of first wild type and then selenomethionine-substituted protein () were obtained by sitting drop vapor diffusion at 4°C. The reservoir solutions contained 0.2 M sodium acetate, 18% polyethylene glycol 3350 (crystal 1), or 9% isopropanol, 0.1 M MES, pH 6.2, and 0.2 M calcium acetate (crystal 2). Drops contained 5 μl protein (20 mg/ml in 20 mM Tris-Cl, pH 8.0, 1 mM DTT, and 1 mM EGTA) mixed with 5 μl of the reservoir buffer. Some electron density was present in the Ca position in EF-hand 1, 3, and 4 of both centrins in crystal 1. This is probably water and not Ca because refinement as Ca resulted in a much higher average B factor than the average residue and the Ca content of the protein and crystals is very low. This was measured by inductively coupled optical emission spectrometry and gave an EF-hand occupancy of around 1%, assuming three active EF-hands/centrin. In addition, to detect Ca ions directly in both crystals 1 and 2, diffraction data were collected from crystals at a longer wavelength (λ = 1.74 Å) for crystal 1 at Beamline ID23 ESRF (Grenoble) or in-house at CuK (λ = 1.54 Å) for crystal 2. An anomalous difference map was calculated using phases from the refined model. This showed no detectable Ca in the EF-hands or elsewhere in crystal 1 and after refinement showed Ca only at EF-hands 1, 3, and 4 in crystal 2 grown at high Ca concentration. strains and yeast vectors were used as before (). Mutants in yeast centrin, , were prepared by QuikChange mutagenesis, confirmed by sequencing, and transferred to pRS314. These replaced the wild-type gene by plasmid shuffle, and if the strains were viable, PCR and sequencing were used to check that the mutation was retained. A strain containing Sfi1p labeled with GFP at the N terminus was prepared by insertion of an NcoI site and then GFP into , integration of a single copy at the locus, and plasmid shuffle to remove the wild-type gene. Nanospray mass spectrometry was performed using modified instrument technology, involving increased pressures and reduced quadrupole frequencies to assist in the analysis of high ions (), incorporated into a Q-Star XL instrument (MDS Sciex; ). Protein samples were transferred into 0.1 M ammonium acetate, pH 7.0, using three successive Bio-spin 6 columns (Bio-Rad Laboratories). Nanospray capillaries were prepared as described previously (). Desolvation of high ions was assisted by “collisional cleaning” using argon gas in the instrument's collision cell. The observed ion series were assigned manually and were consistent with a 1:15 (major component) and 1:14 (minor component) Sfi1p–centrin complex. Higher energy collision-induced dissociation experiments, either in “normal” mass spectrometry mode or by tandem mass spectrometry, caused the ejection of a single centrin monomer from the complex, to give 1:14 and 1:13 “stripped oligomers,” supporting the assigned parent ion stoichiometry. Immuno-EM and EM shadowing were done as before (; ). In thin-section EM, bridge length was measured as the distance between the edge of the central plaques to the edge of the second central plaque or satellite or duplication plaque. Half-bridge length was measured from the edge of the central plaque to the end of the electron-dense nuclear membrane or cytoplasmic outer layer. Half-bridge and bridge length between paired SPBs was measured from log phase cells, and other measurements were from α factor–blocked or –released cells. EM photographic negatives were digitized and transferred to Photoshop 7.0 (Adobe). Table S1 shows the MAD data collection, phasing, and refinement statistics. The supplemental text gives the GenBank identifiers of the sequences used in the analysis of Sfi1 repeats, describes the methods used in the refinement of the crystal structure, and gives the interhelical angles for the closed and open N- and C-terminal domains of centrin. Online supplemental material is available at .

Proper chromosome alignment at the metaphase plate ensures the fidelity of chromosome segregation in mitosis (). This process of bipolar spindle formation and chromosome congression is mediated by spindle poles, microtubules, and kinetochores through a combination of centrosome- and chromatin-mediated pathways (; ). Through a search and capture mechanism, microtubules that nucleated from the centrosomes attach to kinetochores. In a parallel pathway, chromatin promotes the nucleation of microtubules that extend toward the centrosomes to link kinetochores to the mitotic spindle poles. To ensure the equal segregation of chromosomes during mitosis, the spindle checkpoint monitors both the attachment of microtubules to kinetochores and the tension generated across bioriented sister kinetochores (; ). Satisfaction of these two requirements abrogates signaling at the kinetochores through the checkpoint proteins Mad2 and BubR1, thereby allowing activation of the anaphase-promoting complex/cyclosome (APC/C). This ubiquitin ligase targets the anaphase inhibitor securin for degradation, leading to anaphase onset. Thus, the spindle checkpoint controls the fidelity of chromosome separation by preventing inappropriate anaphase initiation. Disruption of mitotic events such as spindle formation and checkpoint response can promote genomic instability (; ). Therefore, an understanding of mitotic regulation is central to dissecting the basic mechanisms of tumorigenesis. In a functional genomic screen, we identified hepatoma up-regulated protein (HURP) as a mitotic regulator. Upon the depletion of HURP, we found that HeLa cells at metaphase had a high frequency of unaligned chromosomes lacking microtubule attachment. Even chromosomes aligned at the metaphase plate were only under partial tension. Although Mad2 and BubR1 were recruited to kinetochores, these cells bypassed the spindle checkpoint and initiated anaphase prematurely. Biochemically, HURP is a microtubule-associated protein (MAP) that colocalizes with the mitotic spindle in a gradient that increases toward the chromosomes. HURP enhances microtubule polymerization and stabilizes the mitotic spindle by reducing the turnover rate of α/β-tubulin subunits on the spindle. We concluded that HURP promotes the efficient capture of kinetochores, timely chromosome congression, and proper interkinetochore tension by controlling the stability and dynamics of spindle microtubules. Genes with mitotic functions such as , , and have similar transcriptional expression profiles during the cell cycle, as they are all induced in G2/M (). reported that 566 genes in the human transcriptome are induced in G2 or G2/M, and we expected that a subset of these genes would have mitotic functions. In our initial analysis, we focused on 30 novel genes with the best induction in G2/M (see Materials and methods for the gene list). To refine this list of potential mitotic regulators, we applied a second genomics filter based on the fact that in tumor cells, known mitotic regulators are transcriptionally up- or down-regulated in concert, thereby resulting in stereotypical covariation profiles (). By surveying the analysis of transcriptional array data from 944 cancer tissue samples covering 22 different tumor types that were reported by , we found that 12 of the 30 aforementioned genes covaried with known mitotic regulators such as , , and . The functions of these 12 potential mitotic regulators were analyzed in two assays. First, the cellular localization of these genes was determined by expression of the respective GFP fusion protein in HeLa cells. Second, the expression of endogenous gene products was reduced by RNA interference, and spindle and DNA morphology was analyzed in mitotic cells by immunofluorescence staining. These analyses led to our identification of HURP as a potential mitotic regulator (). To determine its potential role in mitosis, we ectopically expressed GFP-HURP in HeLa cells and found that although GFP-HURP was diffusely cytoplasmic during interphase, it localized predominantly to the chromatin-proximal regions of the mitotic spindle. Thus, HURP may regulate microtubule function specifically in mitosis (). Next, we depleted endogenous HURP from HeLa cells using pSUPER constructs and siRNAs targeting five different sequences, all of which generated the same phenotypes. siRNA-mediated knockdown reduced the protein level of HURP to <5% of that in control cells ( and not depicted) and abolished the localization pattern of endogenous HURP on the spindle (). We found that 44% of HURP-depleted metaphase cells contained at least one unaligned chromosome compared with only 3% in control cells (). HURP depletion in Hct116 cells also generated a similar phenotype, with 8.9% of HURP-depleted metaphase cells containing at least one unaligned chromosome compared with only 2.8% of control cells (Fig. S1 A, available at ). Based on confocal time-lapse microscopy of HURP-depleted HeLa cells stably expressing GFP–histone H2B (HeLa/GFP-H2B; = 41), these unaligned chromosomes at metaphase persisted for up to 60 min (, bottom; and Video 1). In contrast, control cells ( = 37) progressed from prophase to mitotic exit within 40 min. Thus, HURP is a mitotic regulator that is required for efficient chromosome congression. To determine the cellular basis for the lack of congression of unaligned chromosomes, we analyzed HeLa cells for indications of kinetochore attachment to microtubules. In control and HURP-depleted cells, the kinetochore marker CREST and spindle microtubules were juxtaposed for chromosomes aligned at the metaphase plate (). However, in HURP-depleted cells, the kinetochores of unaligned chromosomes were not attached to any microtubules. The lack of attachment to microtubules should abolish tension across sister kinetochores. We measured interkinetochore distance between pairs of Hec1 dots as an indication of the level of tension. In control cells, metaphase chromosomes under full tension had longer interkinetochore distances (1.90 ± 0.02 μm; ) compared with prometaphase chromosomes that were unattached and under no tension (0.69 ± 0.02 μm) and mitotic cells treated with 1 μg/ml nocodazole (0.75 ± 0.01 μm) or with 1 μM taxol (0.75 ± 0.01 μm). As expected, the interkinetochore distance for unaligned chromosomes in HURP-depleted cells (0.71 ± 0.04 μm) was similar to that of control prometaphase chromosomes, indicating that the unaligned chromosomes are not under any tension. Surprisingly, in HURP-depleted cells with unaligned chromosomes, the interkinetochore distance of aligned chromosomes (1.19 ± 0.02 μm) was shorter than that of control metaphase chromosomes. In fact, even in HURP-depleted cells with all chromosomes aligned at the metaphase plate, sister kinetochores still exhibited an intermediate interkinetochore distance (1.67 ± 0.03 μm), which is consistent with partial tension. This decrease in interkinetochore distance was confirmed by confocal time-lapse microscopy in HeLa cells stably expressing GFP–Cenp-A (HeLa/GFP–Cenp-A; ; and Video 2, available at ). For chromosomes aligned at the metaphase plate, the mean interkinetochore distance in control cells (1.16 ± 0.01 μm), as measured by the distance between pairs of GFP–Cenp-A dots, was longer compared with HURP-depleted cells (0.67 ± 0.01 μm). Finally, HURP depletion resulted in an irregular alignment of kinetochores along the metaphase plate, possibly as a result of a reduction in tension. Only 2% of HURP-depleted metaphase cells exhibited a pattern of two parallel tracks of kinetochores compared with 39% in control cells (). Thus, HURP is essential for the attachment of microtubules to kinetochores and is required for generating full tension to precisely align chromosomes along the metaphase plate. Because HURP-depleted cells at metaphase contained kinetochores that were unattached or under partial tension, we investigated the state of the spindle checkpoint. In control cells, the checkpoint proteins Mad2, which monitors attachment (; ), and BubR1, which monitors attachment and tension (; ), localized to kinetochores at prometaphase but disappeared at metaphase (). As expected, kinetochores on unaligned chromosomes in HURP-depleted cells exhibited high Mad2 and BubR1 levels. Furthermore, in HURP-depleted cells, BubR1 also remained on kinetochores of aligned chromosomes that were under partial tension. These data indicate that in the absence of HURP, checkpoint signaling on kinetochores is activated because of a lack of attachment and insufficient tension. Although HURP-depleted HeLa cells activated the spindle checkpoint, the mitotic population did not accumulate significantly. Mitotic indices for HURP-depleted cells were slightly increased compared with control cells when quantified by flow cytometry (control, 2.08%; HURP depleted, 2.75%; ) or by immunofluorescence staining (control, 2.8%; HURP depleted, 3.4%; ). A similar increase in mitotic index was also found by flow cytometry in HURP-depleted Hct116 cells (control, 1.96%; HURP depleted, 3.01%; Fig. S1 B). Furthermore, within the mitotic population itself, HURP-depleted cells were only slightly enriched at metaphase (). Thus, the spindle checkpoint only transiently arrests HURP-depleted cells at metaphase. To investigate the role of spindle checkpoint response in HURP-depleted cells, we examined the kinetics of mitotic progression. Control or HURP-depleted HeLa cells were treated for 14 h with 100 ng/ml nocodazole to stably arrest cells in prometaphase. In the presence of nocodazole, control and HURP-depleted cells had similar mitotic indices of 47.5 and 46.9%, respectively (). This indicates that HURP itself is not a checkpoint protein required for the establishment or maintenance of the spindle checkpoint. Upon release from nocodazole arrest, the fraction of cells progressing beyond prometaphase was lower in HURP-depleted samples compared with control samples at both 1 h (HURP depleted, 24.9%; control, 44.0%) and 2 h after release (HURP depleted, 38.4%; control, 61.6%; ). We concluded that the spindle checkpoint in HURP-depleted cells delays mitotic progression but does not maintain a stable mitotic arrest. To determine the defects in HURP-depleted cells that delayed mitotic progression, we analyzed the kinetics of mitotic progression by time-lapse imaging using HeLa/GFP-H2B cells. Among the cells that progressed through mitosis, we found that the duration from nuclear envelope breakdown (NEB) to anaphase was substantially longer in HURP-depleted cells (59.1 ± 3.0 min) compared with control cells (42.8 ± 1.1 min; and Video 3, available at ). To further define the mitotic stages that were delayed in HURP-depleted cells, we characterized the duration between four mitotic transitions: NEB; an unaligned state in which the metaphase plate and bipolar spindle had formed but before complete chromosome congression; complete chromosome congression at metaphase; and anaphase onset. The duration from NEB to the unaligned state was only slightly increased in HURP-depleted cells (9.5 ± 0.2 min) compared with control cells (8.2 ± 0.1 min), indicating that the kinetics of bipolar spindle formation and of the initial capture of chromosomes were only slightly affected by HURP. On the other hand, the duration from the unaligned state to metaphase was substantially longer in HURP-depleted cells (15.0 ± 1.0 min) compared with control cells (8.1 + 0.4 min), suggesting that the capture of the remaining unaligned chromosomes was much less efficient in the absence of HURP. Finally, the duration from metaphase to anaphase onset was longer in HURP-depleted cells (34.7 ± 2.9 min) compared with that in control cells (26.4 ± 1.0 min), indicating that the metaphase to anaphase transition was also delayed in the absence of HURP, likely because of insufficient tension. Thus, to varying degrees, HURP is required for timely bipolar spindle formation, chromosome congression, and anaphase onset. To investigate the fate of HURP-depleted cells in greater detail, we tracked their mitotic progression by confocal time-lapse microscopy in HeLa/GFP-H2B cells. Surprisingly, despite a delay during metaphase, 18 of 32 HURP-depleted cells containing one to two unaligned chromosomes initiated anaphase without complete chromosome congression, resulting in an unequal segregation of chromosomes ( and Video 4, available at ). To determine whether this checkpoint bypass phenotype is an intrinsic property of HeLa cells or a result of HURP depletion, we partially knocked down ch-TOG () using siRNAs targeted to four different sequences, all of which gave the same phenotypes. Although the complete depletion of ch-TOG generated the previously reported phenotype of multipolar spindles (), the partial depletion of ch-TOG using a limiting amount (1%) of specific siRNA generated only one to eight unaligned chromosomes in metaphase cells (). Among these ch-TOG–depleted HeLa/GFP-H2B metaphase cells analyzed during the 90-min time lapse, we observed that 6 of 30 metaphase cells containing one to four unaligned chromosomes initiated anaphase without complete chromosome congression ( and Video 4). Unaligned chromosomes at metaphase were also generated in wild-type HeLa/GFP-H2B cells by treatment with a very low concentration of nocodazole (10 ng/ml) for 1 h, which neither depolymerized microtubules nor stably arrested HeLa cells at metaphase. Under these conditions, 3 of 20 cells that contained one to four unaligned chromosomes initiated anaphase without complete chromosome congression within the duration of the time lapse ( and Video 4). Thus, HeLa cells have the intrinsic ability to initiate anaphase in the presence of a few unaligned chromosomes, which is a potential mechanism for genomic instability. Although only 44% of HURP-depleted cells contained unaligned chromosomes at metaphase (), all HURP-depleted cells exhibited a decrease in tension across sister kinetochores (). We measured the interkinetochore distance as an indication of tension at the metaphase to anaphase transition by confocal time-lapse microscopy of HeLa/GFP–Cenp-A cells. In control cells, the mean interkinetochore distance at anaphase onset was 1.7 ± 0.1 μm ( and Video 5, available at ). However, HURP-depleted cells with all chromosomes aligned at the metaphase plate initiated chromosome segregation at a shorter mean interkinetochore distance of 0.9 ± 0.1 μm, indicating anaphase onset with only partial tension. Because metaphase HeLa cells initiated anaphase in the presence of unaligned chromosomes or of sister kinetochores under partial tension, HeLa cells have an inherently less sensitive response to the spindle checkpoint. To determine whether anaphase onset in HURP-depleted cells is initiated through APC/C activation, we assayed for the degradation of the APC/C substrates securin and cyclin B. By quantitative analysis of cytoplasmic fluorescence intensities, we found that the levels of these proteins were lower in anaphase compared with metaphase in both control and HURP-depleted cells ( = 10; ). The anaphase markers Aurora B and Cenp-E were also assayed to determine whether other anaphase events occurred normally. In control and HURP-depleted cells, both proteins correctly localized to the centromeres/kinetochores at metaphase and redistributed to the central spindle at anaphase (). Thus, HURP-depleted cells initiated global anaphase events through the activation of APC/C even in the presence of unaligned chromosomes and kinetochores under partial tension. These observations then raised questions concerning the status of the spindle checkpoint and the mechanism of its bypass at the onset of anaphase. Through extensive immunofluorescence analysis, we observed Mad2 staining on one to three kinetochores per cell in a subset of HURP-depleted anaphase A cells (), indicating that a subset of anaphase chromosomes were unattached in these cells. Because the signaling pathway up to the kinetochore association of Mad2 remained intact, HURP-depleted cells bypassed or adapted to the spindle checkpoint at a point downstream of this signal. We next determined the localization of endogenous HURP across the cell cycle. Although HURP did not associate with interphase microtubules, it did colocalize with spindle microtubules during mitosis and cytokinesis (). Specifically, HURP is localized in a gradient along the pole-to-pole axis with higher concentrations at chromatin-proximal regions that decline toward the poles (). This localization pattern indicates that HURP is actively regulated through its association with microtubules during mitosis and its affinity for chromatin-proximal regions of the spindle. Consistent with its transcriptional induction profile () and its function in mitosis, we found that protein levels of HURP accumulated in G2 and peaked in mitosis at a stage slightly earlier than that of the cyclin B peak (). HURP was also down-regulated as cells exited from mitosis, which is consistent with a previous study (). Given its colocalization with the mitotic spindle, we determined whether HURP is a MAP. Purified recombinant GST-HURP was incubated with polymerized microtubules that were stabilized with taxol, and, upon ultracentrifugation through a glycerol cushion, HURP was found to copellet with microtubules (). We also tested the effect of HURP on microtubule polymerization in vitro. The addition of recombinant purified HURP to purified α/β-tubulin near its critical concentration increased the total yield of microtubule polymer pelleted (; odd lanes vs. even lanes for microtubule yield), which is an effect similar to the one observed with the addition of taxol. Thus, HURP directly binds to and stabilizes microtubules and promotes their polymerization. Next, we investigated the in vivo function of HURP on the mitotic spindle by assaying for morphological defects in HURP-depleted metaphase cells. When analyzed for total β-tubulin immunofluorescence, the microtubule mass on the spindle was reduced by threefold in HURP-depleted cells (). In addition, the interpolar distance, which was marked by γ-tubulin signals, was reduced from 8.84 ± 0.30 to 7.41 ± 0.27 μm in the absence of HURP (). Thus, HURP is a mitotic MAP that increases the amount and length of spindle microtubules by enhancing microtubule stability and polymerization. Given the biochemical activity of HURP as a microtubule-stabilizing factor, we investigated its effects on the dynamics of the mitotic spindle. When treated with 1 μg/ml nocodazole for 10 min, all spindle microtubules were depolymerized in both control and HURP-depleted mitotic cells (, left). However, when treated with 100 ng/ml nocodazole, disruption of the mitotic spindle was greater in HURP-depleted cells compared with control cells (, right). To corroborate this result, we assayed the stability of spindle microtubules in HeLa cells transiently expressing GFP-HURP approximately five times above the endogenous level (not depicted). At 10 and 50 μg/ml nocodazole, all spindle microtubules were disrupted in control mitotic cells, but cells overexpressing GFP-HURP retained stable spindle microtubules (). Thus, increasing the expression of HURP results in a greater stabilization of spindle microtubules in vivo. To investigate the role of HURP in microtubule polymerization, control and HURP-depleted cells were treated with 1 μg/ml nocodazole for 10 min and released into fresh media. 2 min after release, control cells formed short microtubules both at the centrosomes as well as at multiple foci within the chromatin (). Interestingly, HURP colocalized with β-tubulin foci on both the centrosomes and the chromatin. In contrast, HURP-depleted cells only showed weak β-tubulin staining at the centrosomes, and no microtubule foci were found within the chromatin. 30 min after release, control metaphase cells had reformed a bipolar spindle, whereas HURP-depleted mitotic cells showed only a minimal recovery of microtubules and no bipolar spindles (). Thus, HURP enhances the de novo polymerization of microtubules from both centrosomes and chromatin to reform the bipolar spindle. To further define the role of HURP in microtubule dynamics, the kinetics of microtubule depolymerization and repolymerization were monitored by confocal time- lapse microscopy on HeLa cells transiently expressing GFP– α-tubulin. When the depolymerization phase was monitored in the presence of 1 μg/ml nocodazole, control metaphase cells maintained a bipolar mitotic spindle structure for up to 80 s, whereas the spindle in HURP-depleted metaphase cells was disrupted by 40 s ( and Video 6, available at ). To assay for effects on microtubule polymerization, control and HURP-depleted metaphase cells were treated with 1 μg/ml nocodazole for 5 min to depolymerize all microtubules and were then released into fresh media. Control cells rapidly repolymerized microtubules within 30 s and formed multiple microtubule foci within 1 min ( and Video 7). However, this recovery was delayed in HURP-depleted cells until 4 min after release, with very few weak microtubule foci even at 5 min after release ( and Video 7). We also monitored the kinetics of bipolar spindle formation. Control cells reformed a bipolar spindle within 20 min, but this was delayed to 50 min after release in HURP-depleted cells ( and Video 8). Collectively, these data indicate that HURP increases the stability of the mitotic spindle, enhances the de novo polymerization of microtubules, and promotes timely formation of the bipolar spindle. Because HURP stabilizes the mitotic spindle, we measured the turnover rate of α/β-tubulin subunits on the spindles of HeLa cells transiently expressing GFP–α-tubulin. In these fluorescence loss in photobleaching (FLIP) experiments, cytoplasmic GFP–α-tubulin was photobleached continuously while time-lapse images recorded the decrease in fluorescence intensity in the metaphase spindle (Video 9, available at ). In control cells, the half-life of GFP–α-tubulin on the spindle was 104.47 s (, middle). In HURP-depleted cells, the half-life decreased to 83.81 s (, left); however, the half-life increased to 129.93 s in cells overexpressing RFP-HURP (, right). Thus, the presence of HURP decreases the turnover rate of α/β-tubulin subunits on the spindle, thereby stabilizing the microtubules. We have developed a genomic screen for novel mitotic regulators based on a two-filter analysis of gene expression. The first filter selects for novel G2/M up-regulated transcripts that may function in mitosis. The second filter searches for genes that covary with a set of known mitotic regulators that define a core module essential for cell cycle progression. A combination of these two filters selects for candidate mitotic regulators involved in tumorigenesis, which, when coupled to functional assays, allows us to effectively identify novel mitotic regulators. Our two-filter strategy can also be modified to investigate other physiological processes such as the DNA damage response or apoptosis. Through this screen, we identified HURP as a gene that is up-regulated during G2/M () and that covaried with other mitotic regulators across many tumor tissues (). Compared with controls, spindle microtubules lacking HURP were less stable, polymerized more slowly, and exhibited higher turnover rates. Functionally, this activity is required for the efficient search and capture of kinetochores, timely congression of chromosomes, and generation of proper tension across sister kinetochores. Specifically, HURP-depleted cells had both unattached kinetochores on unaligned chromosomes and kinetochores under partial tension on aligned chromosomes. These defects activated the spindle checkpoint and transiently arrested cells at mitosis. Interestingly, HURP-depleted HeLa cells subsequently bypassed the checkpoint arrest and initiated anaphase without resolution of these deficiencies, leading to the missegregation of their chromosomes. performed microarray analysis on genome-wide gene expression across the cell cycle in HeLa cells and identified 566 genes that were transcriptionally induced in G2 or G2/M. analyzed gene coexpression profiles among 1,975 published microarrays derived from 22 different types of tumors. Based on statistical analyses of coexpression profiling, they organized human genes into different functional modules, each of which corresponds to a set of genes that act in concert to carry out specific physiological functions. Out of 577 functional modules identified, two modules function in cell cycle regulation and consist of genes that covary with known cell cycle regulators. Our current analyses were based on these two previous studies. Among 566 G2- or G2/M-induced genes reported by , we initially analyzed the following 30 novel genes that have the best induction profile at G2/M: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , and . We found that the first 12 of the above listed 30 genes were also present in the cell cycle modules reported by . Therefore, we focused our functional analysis on these 12 genes and found that several of them, including , , , and , were associated with microtubules in vivo during mitosis. At the functional level, we found that is required for cell abscission at the terminal stage of cytokinesis and reported that HURP is required for the efficient capture of kinetochores and congression of chromosomes by the mitotic spindle during mitosis. The full-length HURP cDNA was subcloned into a modified version of the pFastBac vector (Invitrogen) containing an NH-terminal GST tag, expressed in Sf9 cells, and purified by glutathione–Sepharose 4B (GE Healthcare). GFP-HURP and RFP-HURP were subcloned into a modified version of pCS2+ containing an NH-terminal GFP or mRFP tag. GFP–α-tubulin was a gift from A. Barth (W. James Nelson Laboratory, Stanford University, Stanford, CA). HURP fragments (aa 1–280, 81–625, and 626–846) were subcloned into the pGEX vector (GE Healthcare), expressed in , purified by glutathione–Sepharose 4B, and used to immunize rabbits for the production of antisera. Antibodies were immunopurified using the respective immunogens. The following antibodies were obtained from commercial sources: anti-GFP clone 3E6 (Invitrogen); anti–β-tubulin ascites clone 2–28-33, anti–α-tubulin clone DM1α, and anti–γ-tubulin clone GTU-88 (Sigma-Aldrich); sheep anti–β-tubulin (Cytoskeleton, Inc.); CREST (Antibodies, Inc.); anti-Hec1 (GeneTex, Inc.); anti–Aurora B (BD Biosciences); antiphosphohistone H3 (Upstate Biotechnology); and anti–p38-MAPK, anti-Hps70, anti–cyclin B, and anti–Cenp-E (Santa Cruz Biotechnology, Inc.). Rabbit antibodies against Mad2 and BubR1 were described previously (). The anti–ch-TOG antibody was provided by L. Cassimeris (Lehigh University, Bethlehem, PA). The HeLa/GFP–Cenp-A cell line was a gift from J.-M. Peters (Research Institute of Molecular Pathology, Vienna, Austria). Two DNA-based pSUPER constructs targeting HURP were transfected into HeLa cells using LipofectAMINE 2000 (Invitrogen). The target sequences for HURP are 5′-GCAATGAGAGAGAGAATTA-3′ and 5′-AGACTAAGATTGATAACGA-3′. The pSUPER empty vector () was used as a negative control. siRNAs targeting HURP or ch-TOG were transfected into HeLa or Hct116 cells using Dharmafect 1 (Dharmacon). siRNAs for HURP are 5′-GGTGGCAAGTCAATAATAA-3′, 5′-AGACTAAGATTGATAACGA-3′, and 5′-CGAAATAGACACTTTGGTT-3′. siRNAs for ch-TOG are 5′-GGAAATAGCTGTTCACATA-3′, 5′-GAAGAAACCTCAAGTGGTA-3′, 5′-GGCCAAAGCTCCAGGATTA-3′, and 5′-CAAGAAACCTGGATGGAAA-3′. siCONTROL Non-Targeting siRNA #2 (Dharmacon) was used as a negative control. Partial knockdown of ch-TOG was performed by using 1 nM of specific siRNA with 99 nM of control siRNA. For the copelleting assay, recombinant HURP protein at a final concentration of 2 μM was added to the reaction mix containing 2 mM GTP, 1× protease inhibitors, 20 μM taxol, and taxol-stabilized microtubules in 1× BRB80 buffer (80 mM Pipes, pH 6.8, 1 mM MgCl, and 1 mM EGTA). The reaction was incubated at 30°C for 30 min and pelleted through a 40% glycerol cushion containing 20 μM taxol and 1× protease inhibitors in BRB80 at 100,000 for 20 min at 30°C. Pellets were washed three times with 1× BRB80 and analyzed by Western blotting. For the polymerization assay, recombinant HURP protein at a final concentration of 8 μM was added to the reaction mix containing 10 μM of purified α/β-tubulin, 2 mM GTP, and 2 mM DTT. The reaction was warmed to 37°C for 2 min, and taxol was added incrementally during the 25-min polymerization reaction. The reaction was pelleted through a 40% glycerol cushion containing 20 μM taxol in BRB80 at 100,000 for 20 min at 30°C. Pellets were washed three times with 1× BRB80 and analyzed by Western blotting. Fig. S1 shows that HURP depletion in Hct116 cells generates unaligned chromosomes and an increase in mitotic index compared with control cells. Videos 1 and 3 show control and HURP-depleted HeLa/GFP-H2B cells in mitosis. Video 2 shows control and HURP-depleted HeLa/GFP–Cenp-A cells at metaphase. Video 4 shows HURP-depleted, ch-TOG–depleted, and nocodazole-treated HeLa/GFP-H2B cells initiating anaphase in the presence of unaligned chromosomes. Video 5 shows control and HURP-depleted HeLa/GFP–Cenp-A cells initiating anaphase. Videos 6–8 show control and HURP-depleted HeLa cells transiently expressing GFP–α-tubulin in metaphase that were treated with 1 μg/ml nocodazole (Video 6) and then released (Videos 7 and 8). Video 9 shows FLIP analyses of a control HeLa cell, a HURP-depleted HeLa cell, and a HeLa cell expressing RFP-HURP (all with transient GFP–α-tubulin expression) at metaphase. Online supplemental material is available at .

Cell survival depends on the accurate transmission of the genetic material to progeny. Coordinating chromosome behavior with the cell cycle machinery guarantees that the products of cell division are two genetically identical cells. Chromosomes are replicated to create two sister chromatids held together by topological and protein-mediated linkages. At the onset of mitosis, chromosomes compact into discrete bodies, converting the chromatids into rod-shaped structures short enough to segregate away from each other. At anaphase, the protein and topological connections between sisters resolve, allowing their segregation from each other to opposite poles of the mitotic spindle. Cohesin is responsible for the protein-mediated linkages. During mitosis, cohesin's cleavage allows separation of sister chromatids (). Although this is the case for most of the genome, the repetitive ribosomal gene cluster also requires the activity of the Cdc14 phosphatase for segregation (; ; ; ; ). Cdc14 is required for rDNA segregation because it is necessary for the localization of condensin to rDNA (; ), a protein complex required for chromosome condensation and segregation (; ; ; ; ; ). However, Cdc14 is better known for its multiple roles during mitotic exit (). Cdc14 is itself regulated by an inhibitory protein (Net1) that keeps it bound to nucleolar chromatin for the entire cycle except for anaphase (; ), when the Cdc14 early anaphase release (FEAR) network and mitotic exit network (MEN) promote its release, thus allowing Cdc14 to reach its targets (; ; ; ; ). Because of these roles, temperature-sensitive mutants of Cdc14 arrest in late anaphase as binucleated cells with unseparated and decondensed rDNA (; ; ; ; ). The reason rDNA requires additional segregation mechanisms, dependent on Cdc14, is presently unclear. The locus differs from the majority of the genome in several aspects: it is highly repetitive, which increases chromosome size and the potential to undergo recombination; it replicates unidirectionally as a result of the presence of a replication barrier at the 3′ end of each 35S ribosomal RNA (rRNA) gene (; ); it is highly transcribed by dedicated polymerases (RNA polymerase I and III), accounting for 60% of all cellular transcription; and (iv) it is repressed for RNA polymerase II transcription (; ). Any or all of these differences could in principle impose segregation constraints in rDNA regions. We have investigated the reason behind the additional segregation requirements of rDNA. We show the length of the array and the transcriptional hyperactivity of the rRNA genes it contains to be the factors that differentiate its segregation from the rest of the genome. We demonstrate that shortening the array or inactivating RNA polymerase I eliminates the segregation defects of mutants. In addition to Cdc14, we uncover a second pathway designed to prevent linkages between rDNA on sister chromatids dependent on the replication fork barrier (RFB) gene . The function of Cdc14 in rDNA disjunction is probably unrelated to its role in inactivating Cdks, as several mitotic exit mutants can segregate rDNA despite being unable to lower Cdk activity (; ). However, overexpression of the Cdk inhibitor not only forces mutant cells out of mitosis but also allows their growth on solid media (; ). To resolve this paradox, we tested whether rDNA segregates correctly when cells are forced out of mitosis without Cdc14. To this aim, we analyzed the segregation of a chromosome tag inserted in the distal flank of rDNA ( tags) in cells expressing from the promoter. Inactivation of Cdc14 through temperature elevation causes arrest at telophase, whereas addition of galactose to these cells induced mitotic exit, as judged by the growth of a new bud. Three different categories were observed, with respect to the segregation of tags, in cells that had entered a new cycle (): unresolved tags (sister chromatids failed to separate), resolved but missegregated tags (separated sisters found in the same nuclear mass), and resolved and segregated tags (sisters found in different nuclear masses). A large proportion of cells showed unresolved tags, indicating rDNA nondisjunction after mitotic exit (). Therefore, the function of Cdc14 in rDNA segregation is independent from its role to drive mitotic exit. The nondisjunction of tags in cells () is intriguing because these cells have been previously reported to form colonies on solid media containing galactose at 37°C (; ). To revisit this, we plated cells on galactose at 37°C (). Consistent with previous studies, colonies formed after several days (; ); however, the amount of colonies corresponded to 1% of the total number of cells (). Therefore, the formation of survivor colonies appears to be a selection process, instead of allelic suppression. Survivor colonies remained able to grow at 37°C in galactose after being passed for 40 generations in glucose-containing media at 23°C (). The segregation of rDNA in survivor cells was significantly improved (); however, these cells were still unable to undergo cytokinesis and consequently grew as chains in culture (). These observations show that Cdc14 has at least three independent roles during mitotic exit, namely, Cdk inactivation, nucleolar segregation, and cytokinesis, the former two being the essential functions for cell viability. Our results demonstrate that both nucleolar segregation and mitotic exit are the essential functions of Cdc14. We reasoned that the appearance of survivors might be related to changes that affect the nucleolar segregation function of Cdc14. The frequency of survivors is too high (1%) to be caused by spontaneous gene mutations. Instead, survival is more likely to be associated to changes in rDNA structure that alleviate segregation defects. Compaction of rDNA has been shown to occur during anaphase, and it is required for segregation (; ). Recently, spontaneous large deletions in the rDNA have been shown to occur in ∼1% of cells (). A large size reduction in rDNA would simulate compaction and could influence segregation. To test this possibility, we compared the size of chromosome XII in survivors to that of the original strain by pulsed-field gel electrophoresis (PFGE). The chromosome XII size in all survivors was reduced compared with the original strain (). No translocations were detected (unpublished data), suggesting that size reduction was associated with rDNA loss in the chromosome. Changes in rDNA array size can also occur through the formation of extrachromosomal ribosomal circles (ERCs; ; ). However, we did not detect an increased number of ERCs in the survivors (). Furthermore, the lack of rDNA segregation in mutants is not affected by the presence of multicopy plasmids carrying rDNA (Fig. S1, available at ). We conclude that chromosome size reduction in the survivors is caused by a loss in the total rDNA copy number in the cell. The reduction of the rDNA array size is therefore a shared phenotype amongst all survivors. However, it is still possible that size reduction is not a requirement for the survival but an indirect effect of the selection that –blocked cells undergo when forced out of mitosis. To distinguish between these two possibilities, we tested whether fixing the size of the rDNA array in the original strain would prevent the appearance of survivors. Changes in rDNA copy number require the gene bound to the RFB site on rDNA (). In cells, the rDNA array size is maintained without change in copy number (). Deletion of in cells abolished the appearance of survivors in galactose media at 37°C (), suggesting that Fob1 is required for survival. However, Fob1 is an rDNA binding protein with roles that contribute to rDNA segregation ( and see ); therefore, it is possible that Fob1 is necessary for survival for reasons other than to mediate array size change. To evaluate this, we investigated whether survival requires the recombination machinery because the role of Fob1 in rDNA array expansion/contraction also involves mechanisms dependent on recombination (; ; ). Like Fob1, deletion of , an essential protein for recombination, in cells prevented the appearance of survivors (). Interestingly, Rad52 is only required at the time of selection, as deletion of in survivor strains did not affect their ability to grow in galactose media at 37°C (). These results demonstrate that a change in rDNA array size is important for the survival of cells and that such changes are mediated by recombination events. Cdc14's role in rDNA segregation is at least in part to target condensin to rDNA regions (; ), thus promoting compaction of this chromosome, which is an important feature of its segregation (). Reduction of rDNA copy number in survivor cells shortens chromosome XII, and this might be sufficient to circumvent the need for compaction and, thus, Cdc14's role in the process. To test this model, we investigated whether shortening rDNA arrays would be sufficient to bypass the role of Cdc14 in rDNA segregation. We used two strains with different rDNA array sizes, a short array of 25 units () or a long array containing 190 () copies. Both strains also contained a chromosome tag in the distal flank of rDNA () and carried a deletion to prevent any further changes in rDNA size. Surprisingly, we found no differences with respect to segregation between the two strains (). However, we noticed a genetic interaction between and genes at permissive conditions (), raising the possibility that Fob1 has additional roles in rDNA segregation that are independent of rDNA size ( and see ). To address this, we expressed from the promoter during the last few cell cycles in the and strains before inactivating Cdc14. Although >50% of cells were able to segregate in the strain, only 5% segregated in the strain when Fob1 was present (). The results demonstrate that reduction in rDNA length improves rDNA segregation in the absence of Cdc14 function. Our results demonstrate that deletion of Fob1 in a mutant background impedes rDNA segregation irrespective of array size (), suggesting that this protein has a direct role in rDNA segregation. Strains containing the normal number of units (100–200) already show low levels of segregation in the arrest (), thus making it difficult to quantify the effect of Fob1 in cells arrested by inactivation of Cdc14. To investigate the contribution of Fob1 to segregation, we used an alternative growth regimen. First, we blocked cells in anaphase (by temperature) and then returned them to permissive conditions () to allow mitotic exit. We scored rDNA segregation during mitotic exit (). We used different tags along chromosome XII to compare the segregation between and cells (). Tags in the proximal side of rDNA ( and ) were already resolved in 70–80% of cells arrested in the block before release () and showed no differences with respect to segregation (with >80% of cells segregated 150 min after release), independent of whether Fob1 was present (). In contrast, the segregation of tags in the distal side of rDNA ( and ) reached a maximum of ∼50% when Fob1 was present but dropped to <5% in cells (). These results show that Fob1 plays an active role in the segregation of rDNA distal regions in addition to that of Cdc14. These experiments also revealed several interesting observations. It seems that when a culture goes through anaphase without Cdc14, a large proportion of cells show segregation defects for the distal tags even when Cdc14 is added back (; and segregation in ). We also noted differences between the and tags in cells. Despite the fact that neither nor tags segregated, resolved in 45% of cells (localized to same nucleus) with a mean distance of 1–2 μm (), whereas tags did not resolve from each other (). Deletion of Fob1 negatively affects rDNA segregation in a mutant background (), suggesting additive effects for both proteins. However, no segregation phenotypes have been previously described for the single mutant. Next, we tested whether affects rDNA segregation in the presence of Cdc14. We could not detect missegregation of chromosome tags in cells (unpublished data); however, the resolution of tags in cells occurred at longer spindle lengths (), suggesting that cells suffered segregation delays. One possibility is that Cdc14 activity is sufficient to mask Fob1's segregation role. To test this, we investigated whether Fob1 interacts with downstream targets of Cdc14. Condensin and Top2 activities in rDNA during anaphase depend on Cdc14 (; ; ). Interestingly, shows additive growth defects with temperature-sensitive alleles of the condensin subunit , , as well as , (Fig. S2, available at ). In addition, we investigated the targeting of condensin and Top2 in cells by chromatin spreads. We did not detect any differences for condensin between wild-type and samples (unpublished data). However, Top2 was present in bright nucleolar foci only in cells (). Overexpression of in cells (blocked in a –mediated arrest) induced segregation of rDNA distal tags in >75% of cells (Fig. S3). These results show that the origins of the disjunction defects caused by and Cdc14 inactivation are similar. Our findings suggest that condensin activation and its regulation of Top2 recruitment () in a Cdc14-dependent manner is likely to resolve problems caused by the absence of Fob1, hence masking its contribution to rDNA segregation in strains. Our results demonstrate that shortening the rDNA array significantly reduces the need for Cdc14 activity to achieve segregation (). However, a proportion of mutant cells with short rDNA arrays still failed to segregate correctly (), raising the possibility that additional factors (besides rDNA size) contribute to nondisjunction in mutants. rDNA differs from the majority of the genome in several aspects, including its potential to undergo recombination (; ; ), its unidirectional mode of replication (; ), and the fact that, despite being silenced for RNA polymerase II transcription (; ), it is highly transcribed by RNA polymerase I. Next, we tested whether any of these peculiarities impose the segregation constraints in rDNA that require Cdc14 and Fob1 activities. First, we considered recombination to be the possible source of nondisjunction because, conceptually, an increased level of recombination between rRNA genes or the inability to remove recombination intermediates could interfere with segregation. However, recombination is unlikely to be the origin of nondisjunction because Fob1 is necessary for rDNA recombination (; ; ), and we predict that loss of recombination structures would promote segregation and not reduce it as we observed in the experiment (). Nevertheless, we tested the possibility in a more direct way by deleting in the strain and analyzing rDNA segregation in the resulting strain. The resolution and segregation of and tags in cells were not affected by (Fig. S4 A, available at ), confirming that recombination does not contribute to the rDNA nondisjunction phenotype in the absence of Cdc14. Moreover, the fact that deletion of did not worsen segregation as we see in allowed us to conclude that the phenotype associated to this double mutant is not due to recombination. Transcriptional silencing in the rDNA gene cluster acts on RNA polymerase II–transcribed genes (; ). Silencing on rDNA requires the silencer protein Sir2 as part of the protein complex called RENT (regulator of nucleolar silencing and telophase exit; ). RENT recruitment to rDNA depends on Fob1 (). Deletion of does not improve the segregation defect in mutants released from metaphase (). However, it is not known whether worsens segregation as observed for mutants (). To test this possibility, we investigated segregation in cells at the block. Segregation of tags in cells was comparable to that in (Fig. S4 B). These results confirm that RNA polymerase II–silent chromatin does not interfere with the segregation of nucleolar regions in the and mutants. Our results have revealed a function for Fob1 in nucleolar segregation (). Recent work has shown that Fob1 also plays a role regulating the timely activation of Cdc14 (); thus, one possibility is that these two roles are related. Inactivation of prematurely releases Cdc14, whereas overexpression causes a delay (). Because the mutant protein Cdc14-1 is rapidly delocalized from the nucleolus at 37°C (), it is possible that segregation after –block release () requires passage of the reactivated Cdc14 protein through the nucleolus. If this were the case, could potentially interfere with Cdc14 reactivation and consequently worsen segregation in our experiments. To test this possibility, we analyzed the localization of reactivated Cdc14-1 protein fused to GFP () during the release from a block (Fig. S4 C). Cdc14 was not observed in the nucleolus until 60–70 min after release (Fig. S4 C), a time when segregation has already reached its maximum levels (). Therefore, Cdc14 reactivation does involve passage through the nucleolus before segregation and, hence, Fob1 roles in segregation and Cdc14 activation are independent. Fob1 is also required for replication fork pausing in the RFB site at the 3′ end of the 35S rRNA gene (). This fork barrier is thought to prevent collisions between the replication and transcription machineries (; ; ), thus forcing replication and transcription to occur codirectionally. This function might be important because, at least in plasmids, opposing replication and transcription can generate topological problems (). Therefore, it is possible that in the absence of Fob1 a high level of collisions between transcription and replication impede mitotic disjunction of rDNA. To test this hypothesis, we investigated whether inactivation of Tof1 in cells also emulated the rDNA segregation defects of cells, as Tof1 is also required for fork arrest at the RFB site (; ; ). The levels of tag segregation in cells are comparable to those in mutants (). We thus conclude that the lack of RFB activity in cells is not the cause of its segregation defects. A major difference between rDNA and the rest of the genome is in respect to its transcriptional activity. Despite being silenced for RNA polymerase II transcription (; ), rDNA is also highly transcribed by RNA polymerase I. In higher eukaryotes, a reduction in rRNA transcriptional activity occurs during mitosis, but this is not the case in budding yeast, where rRNA transcription continues through this cell cycle stage (). It is possible that continuous transcription during mitosis requires specialized mechanisms to ensure segregation, perhaps dependent on Cdc14 and Fob1 activities. To test this possibility, we investigated rDNA segregation in mutants where polymerase I transcription of 35S rRNA was turned off. We deleted , an essential gene encoding the second largest subunit (A135) of the yeast RNA polymerase I complex in the strain. The resulting cells are able to grow because they carry a multicopy plasmid with a 35S rRNA gene driven by the RNA polymerase II () promoter (). Cells were released from G1 at 37°C to inactivate Cdc14, and the segregation of tags was scored in binucleated cells arrested in the block. Correct chromosome segregation for both tags was observed in a high proportion (>80%) of cells (). Next, we asked whether also suppressed the segregation defects in the mutant. cells (). cells were able to segregate rDNA regions correctly (), suggesting that bypasses both Cdc14 and Fob1 segregation functions. These results demonstrate that the transcription of rRNA genes imposes segregation constraints in rDNA that require Cdc14 activity for resolution. In addition, the data show that the presence of Fob1 also plays a role in reducing the levels of linkages in the rDNA that need to be resolved by Cdc14. Thus, we identify polymerase I transcription as a novel means of establishing linkages between chromosomes. To ensure genomic stability through generations, cells need to hold sister chromatids together until metaphase and then remove all the physical connections between them in anaphase. It has long been known that the nucleolus requires Cdc14 to segregate (). However, the reason for this specific requirement was unknown. Here, we have shown that rDNA requires Cdc14 for segregation partly because of its physical length but most importantly because a fraction of rRNA genes are transcribed at very high rates. We show that the rDNA segregation function of Cdc14 can be bypassed through genetic rearrangements that involve a gross reduction in the number of rDNA copies, thus reducing chromosome size. We also demonstrate that besides rDNA size, the transcription of rRNA genes by RNA polymerase I cells generates linkages between sister chromatids that prevent segregation in the absence of Cdc14. In addition, our study shows that Fob1 has a novel function in rDNA segregation independent from that of Cdc14. Thus, our data not only provide an insight into the mechanisms that give rise to constraints on mitotic rDNA sister chromatid disjunction (i.e., rRNA transcription) but also reveal the presence of two pathways to deal with these problems, one dependent on Fob1 and the second requiring Cdc14. These findings thus explain the reason behind the segregation phenotypes observed in mutants (). All yeast strains used were S228C background, except for strains and the strains bearing a fixed number of rDNA units (25 or 190 copies) that were W303 (a gift from F. Cross, The Rockefeller University, New York, NY). Chromosome tags have been described elsewhere (). COOH-terminal epitope tagging with GFP and gene deletions, including , were performed using PCR allele-replacement methods. The allele was transferred between strains also using PCR allele-replacement strategy where a 9myc epitope and a selective marker () are tagged to the COOH terminus. Western blotting and thermo-sensitivity assays were used to confirm transformations. Plasmid pNOY103 () was a gift from M. Nomura (University of California, Irvine, Irvine, CA) and K. Kobayashi (National Institute for Basic Biology, Tokyo, Japan). Plasmid for the overexpression of Cdc14 was a gift from A. Amon (Massachusetts Institute of Technology, Cambridge, MA). The allele was a gift from M. Sullivan (Cancer Research UK, London, UK). Relevant genotypes of strains used in this study are shown in Table S1 (available at ). To arrest cells in G1, we used . Cells were treated with 50 ng/ml α-factor for 3 h at 25°C. To release cells from the block, we transferred them to fresh media plus pronase E (0.1 mg/ml). For releases at nonpermissive temperatures, we exposed cells to 37°C for 30 min before their transfer to fresh media (also at 37°C). To release from a block, G1-released cells were incubated at 37°C for 150 min before shifting them back to 25°C to reactivate Cdc14. G2/M arrest in was obtained by adding 15 μg/ml nocodazole to the media and incubating for 3 h. For the experiments in (lack of RNA polymerase I transcription), parental strains ( without the pGAL-35S plasmid) and strains bearing were grown in YPgal at 25°C until log phase (up to 3 d for strains). Strains with the pGAL-35S plasmid were grown in YPgal only for 9 h after a first overnight growth in SC-galactose-ura). strains were arrested in α-factor for 3 h and released into 37°C for 3 h (OD600 doubling time in YPgal ∼3 h). RNA polymerase I–deficient strains were arrested in α-factor for 6 h and released into 37°C for 7 h (OD600 doubling time in YPgal ∼7 h). The α-factor block arrests >98% of the cells in G1. About 50% of the RNA polymerase I–deficient cells enter a new cell cycle after the G1 release. Only cells clearly in anaphase (stretched nucleus across the neck or binucleated) were counted. PFGE to see chromosome XII was performed in a 0.8% agarose gel in 0.5× TBE buffer run for 68 h at 6 V/cm with an initial switching time of 60 s, a final of 120 s, and an angle of 120°. ERC analysis was performed as described by . The total running time was doubled to 48 h, and a long agarose gel was used. Yeast cells with GFP-tagged proteins were analyzed by fluorescence microscopy after DAPI staining. Series of z focal plane images were collected on a microscope (IRB; Leica) using a digital camera (C4742-95; Hamamatsu) and OpenLab software (Improvision). A tuneable light source (Polychrome IV) with a Xenon lamp was used. Images in different z axis planes were flattened into a 2D projection and processed in OpenLab. DNA was stained using DAPI (Invitrogen) at a final concentration of 1 μg/ml after short treatment of the cells with 1% Triton X-100. Imaging was done in antifade/DAPI medium (Invitrogen) at room temperature. Micrographs were taken with either 63×/1.4 or 100×/1.35 lenses. Fig. S1 shows that multicopy plasmids bearing the rDNA unit cannot rescue the chromosome XII segregation impairment in mutants. Fig. S2 shows that condensin and topoisomerase II mutants show synergistic genetic interactions with . Fig. S3 demonstrates overexpression of Cdc14 rescues and rDNA segregation defects. Fig. S4 shows that Fob1 function in chromosome XII segregation does not act through its role in rDNA recombination, FEAR network, or rDNA silencing. Table S1 shows relevant genotypes of strains used in this study. Online supplemental material is available at .

Nebulin is a giant modular sarcomeric protein (500–900 kD) in skeletal muscle, where it comprises ∼3% of total myofibrillar protein (). A critical role for nebulin in skeletal muscle function is demonstrated by human mutations in , which are causative for nemaline myopathy, a neuromuscular disorder characterized by muscle weakness and the presence of rodlike nemaline bodies in the muscle fibers that contain abnormally arranged Z-line and I-band proteins (, ). Nebulin has been proposed to be important for multiple aspects of striated muscle form and function (). However, its exact role in vivo remains elusive. Nebulin is encoded by a single gene, and its molecular weight ranges from 500 to 900 kD because of extensive isoform diversity in different muscle types, species, developmental stages, and in disease (; ; ). Single molecules of nebulin are associated with thin filaments in skeletal muscle and span the entire length of the thin filament with the COOH terminus anchored in the Z-line and the NH terminus extending to the pointed end of the thin filaments (; ; ). The molecular size of nebulin correlates with variations in thin filament lengths in different muscle types, suggesting that nebulin may act as a molecular ruler to regulate thin filament length in skeletal muscle (; ). According to the cross-bridge theory of muscle contraction (), the amount of force that a muscle can exert at different sarcomere lengths is determined by the amount of overlap between thin and thick filaments, which is dependent on thin filament length and the contractile state (; ). In particular, thin filament lengths are fine tuned in different vertebrate muscles in vivo to overlap with thick filaments by an amount characteristic for each muscle such that the amount of force generated is suited to the physiological requirements for that muscle (). Importantly, muscles working beyond their optimal length range, causing an incomplete overlap of thick and thin filaments, results in instability and damage (). Nebulin might also play a role in myofibrillogenesis, where it appears to participate in the early assembly of precursor I-Z-I bodies, and assembles in a striated pattern before thin filaments attain their mature length (; ; ). This hypothesis is supported by a recent study showing the failure of myofibril assembly in cultured myotubes upon the knockdown of nebulin for 5 d (). In addition, nebulin has been proposed to play a role in the regulation of muscle contraction by modulating actomyosin ATPase activity in a Ca-calmodulin–dependent manner (, ). Interactions of nebulin with diverse sarcomere-associated proteins suggest potential roles in sarcomeric architecture, signaling, and force transduction. Nebulin is composed mainly of ∼35 amino acid modules (M1–185), which are further organized into super-repeats of seven modules within the central region of nebulin (M9–162). Each nebulin module interacts with a single actin monomer, and each super-repeat associates with each tropomyosin–troponin regulatory complex along the length of the thin filament (; ; ). Within the COOH-terminal region of nebulin, modules M163–170 interact with the intermediate filament desmin in the periphery of the Z-line, suggesting that nebulin might play a role in the lateral registration of sarcomeres and in force transmission (). Nebulin's extreme COOH-terminal end contains a unique serine-rich domain with several potential phosphorylation sites and a Src homology 3 (SH3) domain, suggesting that nebulin might be involved in signaling pathways at the Z-line. The SH3 domain interacts with myopalladin, which, in turn, binds to α-actinin, thereby tethering nebulin at the Z-line (; ). In addition, recent evidence suggests that nebulin's SH3 domain might also bind to the springlike PEVK domain in the I-band region of titin (). In nebulin's NH-terminal region, modules M1–3 bind to the thin filament pointed end–capping protein tropomodulin (), which is critical for maintaining thin filament length at the pointed ends (; ). The interaction between nebulin and tropomodulin as well as a recent RNA interference study support the proposed function of nebulin as a regulator of thin filament length (). Nebulin was long thought to be absent from cardiac muscle, where the homologous but smaller protein nebulette is expressed (). However, recent studies have shown that is expressed in cardiac muscle and is localized in a layout identical to that in skeletal muscle, although at lower levels (; ; ; ). In addition, an RNA interference study in cardiomyocytes has suggested that nebulin is involved in thin filament length regulation in both cardiac and skeletal muscle (). Based on this study, two distinct models have been proposed to explain how nebulin functions to regulate thin filament lengths in the heart (; ). As discussed above, multiple roles have been suggested for nebulin, including the regulation of thin filament length, myofibrillogenesis, signal transduction, regulation of muscle contraction, and myofibrillar force generation. However, requirements for nebulin in vivo have not yet been addressed. To study the functional role of nebulin in vivo, we generated nebulin-deficient mice by using a Cre knockin approach. This strategy also allowed us to study the endogenous expression pattern of by crossing these mice with Rosa26 lineage reporter mice (). Our lineage studies revealed that is expressed in all skeletal muscle myocytes. expression was also identified in the heart. However, was shown to be expressed mainly in the atria, where it was expressed heterogeneously in ∼50% of atrial cardiomyocytes, whereas it was expressed only in a minor percentage of ventricular cardiomyocytes. These results suggest that nebulin may have a more critical role in skeletal muscle than in cardiac muscle. Nebulin-deficient mice die within 8–11 d after birth, with symptoms including decreased milk intake and muscle weakness. Transmission EM (TEM) and immunostaining analyses demonstrated that nebulin is not important for the normal assembly of sarcomeres. However, nebulin-deficient mice had skeletal muscle thin filaments that were decreased in length by up to 25% compared with wild type. Also, thin filaments were uniform in length both within and between muscle types. This is in contrast to a recent study () in cultured cardiomyocytes in which the RNA interference knockdown of nebulin resulted in a 30% increase in thin filament lengths. Our analyses further indicate a critical role for nebulin in the maintenance of sarcomere structure and demonstrate a dramatic reduction in force production by nebulin-deficient skeletal muscle. To study the function and expression pattern of in vivo, we generated -deficient mice by gene targeting. Exon 1 was deleted and replaced by recombinase cDNA as well as the neomycin resistance gene flanked by frt sites (). After electroporation of the targeting vector into R1 embryonic stem (ES) cells, one clone was identified that had undergone homologous recombination (). The clone was injected into blastocysts from C57/B6 mice and gave rise to chimera mice that were then bred with Black Swiss mice to generate germ line–transmitted heterozygous mice. These mice were subsequently mated to generate homozygous mice. To verify that homozygous knockout mice were null mutants, we performed Southern blot analysis on DNA isolated from and mice () as well as Western blot analyses for nebulin protein using a polyclonal antibody raised against domain M161–165. As shown in , no nebulin protein was detected in mice. In addition to the high molecular mass band at nebulin's expected size, lower molecular mass bands of ∼130 and 60 kD were detected that were also absent in mice. This could either be alternative splice isoforms of nebulin or degradation products from the lysate preparation, which has also been seen in other studies using various antibodies (; ). We also performed RT-PCR analysis on RNA isolated from gastrocnemius muscle using primers located in exons 2, 3, 98–102, 163, and 164 (mouse has 165 exons; ), which suggested that the entire nebulin was successfully knocked out (). mice were viable and indistinguishable from their wild-type littermates. To determine the expression pattern of endogenous in vivo, mice were crossed with the Rosa26 lineage reporter mouse (). In the progeny of this cross, Cre-mediated excision at the Rosa26 locus results in the expression of the β-galactosidase (lacZ) gene under the control of the ubiquitously expressed Rosa26 locus, thus allowing lineage analysis of nebulin-expressing cells. Various tissues from 10-wk-old Rosa26 mice were fixed and stained with X-galactosidase. As expected, strong β-galactosidase activity was detected homogenously in all skeletal muscle types (). In addition, β-galactosidase expression was detected in the heart, aorta, and liver (). No β-galactosidase activity was detected in other tissues. Because the presence of in the heart has only recently been reported (; ; ; ) and has not yet been thoroughly characterized, we further examined its expression pattern in the heart. X-galactosidase staining of frozen sections from the heart revealed that is expressed mainly in the left and right atria (∼50% of cardiomyocytes), whereas in the ventricles, is expressed only in a small percentage of cardiomyocytes (). mice were born in Mendelian ratios with a similar body weight compared with their wild-type littermates. mice were able to breathe and move their legs and were indistinguishable from their wild-type littermates except that mice had little or no detectable milk in their stomachs, presumably reflecting an inability to suckle. Consistent with this observation, mice barely increased in weight after birth and exhibited minimal subcutaneous fat. mice died ∼8–11 d after birth with a weight ∼25% of that of their wild-type littermates, which was most likely the result of decreased milk intake caused by muscle weakness (). Because mice were able to survive up until 11 d, we assumed that they were able to drink some milk. Therefore, we tested whether mice would be able to survive longer in the absence of competition from wild-type littermates by removing wild-type littermates after birth. However, this had no effect on the size or lifespan of the mice. Histological analysis by hematoxylin and eosin staining of frozen sections of skeletal and heart muscle from postnatal day 1 mice showed no obvious differences between and wild-type mice (unpublished data). No fibrosis was detected by trichrome staining of either skeletal or heart muscle from mice (unpublished data). We also analyzed myosin heavy chain composition in several distinct skeletal muscles by gel electrophoretic analyses and observed no differences between mutant and wild-type mice (unpublished data). Both had prominent neonatal and embryonic myosin heavy chain bands that were consistent with previous findings (). To determine the structure of skeletal muscle in more detail, TEM analysis was performed on tibialis anterior (TA) muscle from 1-d-old and littermate control mice. Postnatal day 1 was chosen because and wild-type littermate mice were still similar in size at this stage, thus facilitating a valid comparison. To allow us to study the ultrastructure of TA muscle at different sarcomere lengths, legs were fixed with the knee joint at 90° and with the ankle at 90° (neutral) or 180° (fully plantarflexed), as plantarflexion of the ankle joint causes stretching of the TA muscle. At resting length, muscles from mice had relatively normal sarcomeric structure with wide myofibrils, distinct A-bands, and narrow, uniformly spaced Z-lines (), suggesting that sarcomere assembly and organization were preserved in the absence of nebulin. However, the misalignment of myofibrils was often observed. In moderately stretched muscle, myofibrillar misalignment was more pronounced (), and myofibril splitting frequently appeared. In addition, fragmented Z-lines were apparent, whereas M-lines appeared unaffected. Similar results were obtained by TEM of muscle tissue from mice at postnatal day 6 (unpublished data). To determine whether the observed ultrastructural abnormalities in muscle resulted from abnormal sarcomere assembly or an inability to maintain myofibrillar integrity during muscle contraction, we compared embryonic and postnatal diaphragm muscle in mice with littermate controls. Diaphragm muscle remains inactive during embryogenesis, thus allowing a determination of whether nebulin is required for myofibrillogenesis before the onset of contraction. After birth, the diaphragm is continuously contracting and, therefore, is useful for studying the requirement for nebulin in actively contracting muscle. At embryonic day 18.5, diaphragm muscle had well-aligned sarcomeres that were virtually indistinguishable from those of wild-type muscle (). At postnatal day 1, the misalignment of myofibrils and moderate thickening of Z-lines were observed in diaphragm muscle from mice (). At postnatal day 9, diaphragm muscle exhibited a severe disruption of myofibrillar structure, including a lack of well-defined A- and I-bands (). Z-lines exhibited various types of morphologies. Some Z-lines were extremely thick, rounded, and highly electron dense, resembling nemaline rod bodies (; , ), whereas others were punctate and appeared fragmented (). Moreover, some fibers exhibited a complete lack of organization with Z-lines that appeared to be dissolving (). Another interesting observation was the abnormal accumulation of mitochondria within myofibers (). These observations suggest that nebulin is not required for myofibrillogenesis but is important for the maintenance of myofibrillar integrity during muscle contraction. To determine how the ablation of nebulin affected other cytoskeletal proteins, we stained TA and extensor digitorum longus (EDL) muscles from 1-d-old mice and littermate controls for α-actinin 2, α-myosin, actin (phalloidin), tropomodulin, tropomyosin, desmin, and palladin (), many of which have been shown to interact with nebulin. Although sarcomeric structure was less well organized compared with wild-type control mice, no obvious differences in the localization of any of these proteins were found. We also assessed the localization of these proteins at postnatal day 8 before the death of mice and found no significant changes (unpublished data). Thus, nebulin is not critical for normal sarcomere assembly or the localization of several sarcomeric proteins. Because nebulin has been proposed to play a role in the regulation of thin filament length, we examined thin filament lengths in mice. In mice, tropomodulin was localized in its typical striated pattern at the pointed ends of the thin filaments and was expressed in similar amounts compared with wild type as assessed by Western blot analysis (unpublished data). Thus, we measured thin filament lengths either by double immunostaining for α-actinin and tropomodulin, which stain Z-lines and thin filament pointed ends, respectively, or by fluorescently labeled phalloidin to directly localize thin filaments. Stained samples were analyzed by distributed deconvolution, which determines the location, brightness, and dimensions of each thin filament array (I-Z-I body) along a myofibril at subpixel resolution and is independent of sarcomere length differences (). Using distributed convolution on α-actinin– and tropomodulin-stained muscle sections, we measured thin filament lengths in four different muscles from wild-type and mice: gastrocnemius, TA, vastus lateralis (VL), and EDL (). In addition, our results were independently verified by phalloidin staining for TA muscle. A total of three wild-type and three mice were used for the analysis. Examples of the analysis on TA and gastrocnemius muscle are shown in and as supplemental data (Fig. S1, available at ), respectively. As shown in , the absence of nebulin resulted in a significant shortening of thin filament lengths by 0.07–0.28 μm in all four muscle types examined. Intriguingly, the absence of nebulin resulted in similar thin filament lengths of ∼0.95–1.07 μm in all four muscles (), whereas in wild-type muscle, thin filaments lengths were 1.27 ± 0.06, 1.29 ± 0.05, 1.29 ± 0.05, and 1.16 ± 0.06 μm in gastrocnemius, VL, TA, and EDL muscle, respectively. Thus, gastrocnemius, VL, and TA thin filaments were shortened the most (0.28, 0.25, and 0.21 μm, respectively), whereas EDL thin filaments, which were the shortest in wild-type mice, only shortened by 0.07 μm in the mice. To determine the functional effects of the ultrastructural changes in the mice, we measured force production. At postnatal day 1, TA muscle in mice were grossly indistinguishable from wild-type littermates (, inset). However, the neonatal tendon was translucent and mechanically fragile, making it difficult to test contractile force in mice using traditional techniques. As a result, isometric contractile testing of the isolated TA bone–tendon–muscle–tendon–bone complex was performed using high resolution force transducers to yield a signal to noise ratio of >20 (). A total of 16 TA muscles from wild-type and 9 from mice were used for the analysis. Electrophysiological threshold contractile kinetics were similar between and wild-type mice, which suggested that excitation/contraction coupling was not significantly altered in mice (unpublished data). However, the magnitude of the stress generated by mice was <50% of the wild-type value (). Overall stress generated by neonatal muscle was 5–10% that of adult mammalian muscle, probably because of the fact that the myofibrils were less dense (). It is important to note that the low stress was not simply because TA muscles were smaller since there was no significant difference between the physiological cross-sectional area (PCSA; ) of TA muscles. Thus, the force transmission efficiency of the contractile complex of the mouse was critically compromised even though the contractile machinery appeared to be of normal morphology () and fiber size between the two genotypes was similar (not depicted). xref #text genomic DNA was isolated from a 129SVJ mouse genomic library (Stratagene) and was used to construct a -targeting construct for the replacement of exon 1 by the recombinase cDNA and the resistance gene. The construct was generated in the pBluescript II KS+ vector, and the 5′ arm of homology consisted of a 4.096-kb NotI–SalI fragment upstream of exon 1 fused in frame with followed by flanked by sites. The 3′ arm of homology was a 3.736-kb SalI–XhoI fragment located downstream of exon 2 (). The targeting construct was verified by sequencing and linearized with NotI before electroporation into R1 ES cells at the Transgenic Core Facility at the University of California, San Diego. 1,000 G418-resistant ES clones were screened for homologous recombination by Southern blot analysis as described in the next section. Genomic DNA was extracted from G418-resistant ES cell clones and mouse tails as previously described (). ES cell DNA was digested with SacI and analyzed by Southern blot analysis. A 507-bp fragment corresponding to 2,676–3,182 kb of the target vector (indicated in ) was generated by PCR using mouse genomic DNA and specific primers (forward, CGTGTGAGGATTGGAGGTTT; reverse, AGTGCATCACAGGGGTAAGG). The PCR product was subsequently radiolabeled using α-[P]dATP by random priming (Invitrogen). DNA blots were hybridized with the radiolabeled probe and visualized by autoradiography. The wild-type allele is represented by a band of 8.977 kb, whereas a band of 6.982 kb represents the correctly targeted mutant allele. One independent homologous recombinant ES clone was microinjected into blastocysts from C57/B6 mice at the Transgenic Core Facility at the University of California, San Diego. Male chimeras were inbred with female Black Swiss mice to generate germ line–transmitted heterozygous mice (). mice were subsequently intercrossed to generate mice, which were homozygous null mutant mice (). Offspring from intercrosses were genotyped by PCR analysis using mouse tail DNA and wild-type (forward, ATGGCATATGGAAAGTTTGTAGGT; reverse, AACATGAAACATGCCTTCTTTGTA) and mutant allele-specific primers (forward, GTTCGCAAGAACCTGATGCACA; reverse, CTAGAGCCTGTTTTGCACGTTC). Total protein extracts were prepared () from postnatal day 1 skeletal muscle and separated on a 3–8% NuPAGE Tris-acetate gel (Invitrogen) and subjected to Western blot analysis using polyclonal antibodies against nebulin domain M161–165 (1:50; provided by S. Labeit, Universitätsklinikum Mannheim, Mannheim, Germany) according to the manufacturer's instructions. Glyceraldehyde-3-phosphate dehydrogenase antibodies were used for normalization (1:1,000; Sigma-Aldrich). Total RNA was isolated from skeletal muscle and heart tissue from 1-d-old neonatal mice by using TRIzol reagent (Invitrogen). First-strand cDNA synthesis was performed with the random primer and Superscript kit (Invitrogen). The cDNA was used as a PCR template to perform PCR by standard protocols. Specific primers for exons 2 and 3 (forward, TCTGTTTACAGTACTACACAGAGGA; reverse, ACAATGGTGGCGACATAATGAACAAG), 98–102 (forward, CAAAAATTGCCTATGACCTTCAGAGTGACA; reverse, TCATCCAGGGTGTAGCCATAGGCCTTGGTG), and 163 and 164 (forward, GAAGCTGCGGACCAGCGCCTTTCCAC; reverse, ACCCCGGCAGTAGACGGGTGTGATGGG) were used. Fate mapping of -expressing cells was assessed by crossing heterozygous mice with the reporter Rosa26 mouse line to generate double heterozygous Rosa26 mice in which lacZ is expressed from the Rosa26 promoter after Cre-mediated recombination (). Tissue was stained for β-galactosidase activity according to the standard procedure. Tissue was fixed for 30 min at room temperature in 4% PFA in PBS and washed in PBS. Subsequently, tissue was stained overnight at 37°C using the X-galactosidase staining solution (5 mM potassium ferricyanide, 5 mM potassium ferrocyanide, 2 mM MgCl2, and 0.4% X-galactosidase in PBS) and rinsed twice in PBS. For frozen sections, tissue was frozen in optimal cutting temperature after fixation and sectioned at a thickness of 10 μm. Sections were incubated overnight at 37°C in X-galactosidase staining solution and counterstained with eosin followed by dehydration and mounting in permount. Cardiomyocytes were isolated as described previously (), fixed, and stained in X-galactosidase staining solution before mounting in permount. Tissue was analyzed and photographed on a dissecting microscope (SV-6; Carl Zeiss MicroImaging, Inc.) with a 35-mm camera (C-mount; Nikon). For TEM on TA muscle, hind limbs from day 1 neonatal mice were transected midfemur and pinned to cork with the knee joint fixed at 90° and the ankle joint fixed either at 90° (neutral) or 180° (fully plantarflexed), which results in stretching of the TA muscle. After fixing overnight in 2% PFA and 2% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4, the TA muscle was dissected out and cut into smaller pieces. Diaphragm muscle was dissected out and cut into smaller pieces before subsequent fixation overnight. Muscle tissue was postfixed and stained for 2 h in 1% osmium tetroxide and 1% potassium ferrocyanide followed by 1 h in 1% uranyl acetate. After dehydration in a series of ethanol and acetone, muscle tissue was embedded in Durcupan resin (EMD). Ultrathin sections (60–70 nm) were stained with lead citrate, and electron micrographs were recorded by using an electron microscope (1200EX; JEOL) operated at 80 kV. Hind limbs were fixed overnight in 4% PFA in either an unstretched or stretched position as described in the previous section. TA, EDL, gastrocnemius, and VL muscle were dissected out and incubated in 10, 15, and 30% sucrose in PBS before freezing in optimal cutting temperature. 10-μm longitudinal frozen sections were permeabilized and blocked in a solution containing 1% normal goat serum, 0.3% Triton X-100, 50 μm glycine, and 1% cold water fish gelatin (Sigma-Aldrich) in 1× PBS for 30 min followed by incubation overnight at 4°C in a humidified chamber with various antibodies in wash buffer (0.01% Triton X-100, 5 μm glycine, and 0.1% fish gelatin in PBS). Phalloidin (1:100; Sigma-Aldrich) as well as the following antibodies were used: sarcomeric α-actinin antibody EA-53 (1:1,000; Sigma-Aldrich), nebulin M161–165 (1:50; provided by S. Labeit), α-myosin F59 (1:50; Developmental Studies Hybridoma Bank), tropomodulin (polyclonal 1:50 and monoclonal E-Tmod 204; 1:100; ; provided by M. Sussman [San Diego State University, San Diego, CA] and A. Sung [University of California, San Diego, La Jolla, CA], respectively), tropomyosin CH1 (1:50; Sigma-Aldrich), desmin D33 (1:25; Sigma-Aldrich), and palladin (1:50; provided by C. Otey, University of North Carolina, Chapel Hill, NC). After rinsing in wash buffer, sections were incubated at room temperature for 4 h with fluorescently labeled secondary antibodies (goat anti–mouse FITC, goat anti–rabbit FITC, or goat anti–mouse RedX antibody; Sigma-Aldrich) at a final dilution of 1:100 in wash buffer. Slides were rinsed in wash buffer, dried, and mounted in Gelvatol. Confocal microscopy was performed using a confocal microscope (Radiance 2000; Bio-Rad) with a 60× plan- Apochromat NA 1.4 objective (Carl Zeiss MicroImaging, Inc.). Individual images (1,024 × 1,024) were converted to tiff format and merged as pseudocolor RGB images using Imaris (Bitplane AG). Thin filament length analysis was performed by distributed deconvolution using a custom plugin written for ImageJ (National Institutes of Health [NIH]; ). This method determines the best fit of a probe-specific intensity (model) distribution to 1D myofibril fluorescence intensity profiles (line scans) by estimating the spread of light along the line scan and the positions and intensities of each probe with adjustable intensity distribution models for the entire thin filament array (I-Z-I body; ). Image regions containing at least three stretched or hypercontracted sarcomeres were identified based on the appearance of tropomodulin doublets or phalloidin gaps (H zones). Regions were not analyzed if tropomodulin doublets or phalloidin gaps were not visible. Background-corrected line scans were calculated by averaging the intensity across the width of the myofibril for each point along the selected myofibril length by subtracting the minimum intensity above and/or below the myofibril from the mean intensity across the myofibril width. Line scans were analyzed by distributed deconvolution using model distributions for α-actinin, phalloidin, and tropomodulin as described in . Thin filament length was defined as half the distance between tropomodulin peaks or half the width of the phalloidin bands. Based on Z-line positions in muscles costained with α-actinin and tropomodulin and corrected for possible image registration, we estimated the positional error associated with the method to ∼50 nm. For example, for gastrocnemius muscles from mice, there was an estimated error of 66 nm in sarcomere lengths and of 47 nm in Z-line positions, which could not result from image misregistration. To provide the best estimate of muscle mechanical properties that did not reflect the progressive deterioration of the pups from days 1–11 (), mechanical experiments were performed on 1-d-old pups. Animals were killed by decapitation, and hind limbs were transected at the midfemur, immediately placed into a mammalian Ringer's solution (137 mM NaCl, 5 mM KCl, 24 mM NaHCO, 1 mM NaHPO, 2 mM CaCl, 1 mM MgSO, 11 mM glucose, and 10 mg/L curare), and kept on ice until tested. Pilot experiments revealed that the neonatal tendon was extraordinarily fragile and, thus, could not be secured directly to the testing apparatus. Therefore, we tested the muscle–tendon–bone unit associated with the TA muscle. Distal to the TA muscle, the tarsal bone was secured to the lever arm of a servomotor (305B; Aurora Scientific). Proximal to the TA muscle, an anodized stainless steel minutien pin (0.2-mm diameter; Fine Science Tools) was bent to a 90° angle, and the sharp end was driven down the shaft of the femur until it protruded from the femoral condyle. The blunt half was then secured with a set screw to an XYZ translator (Newport Corporation) and adjusted until both the ankle and knee joints were at 90° and the tibia was perpendicular to the foot. Once the leg was fixed in this position, the plantar flexors were severed at the Achilles tendon and dissected away. The length of the TA muscle was measured using a stereomicroscope (Leica MZ16; McBain Instruments) fitted with an eyepiece crosshair reticle, translating the chamber under the field of view from origin to insertion of the muscle using a digital micrometer (350-712-30; Mitutoyo), and taking the mean of three measurements. Data acquisition was performed with a custom-written LabVIEW program (National Instruments) to trigger the 6-bp stimulator (Pulsar; FHC) and record force from the servomotor using a data acquisition board (PCI-6040; National Instruments) sampling at 4,000 Hz. Maximum isometric tension was measured by stimulating dorsiflexors via platinum plate electrodes with a bipolar 400-ms train of 0.3-ms pulses delivered at 100 Hz. Two measurements were taken 2 min apart and averaged. After mechanical testing, the legs were removed from the chamber, and the TA and EDL muscles were dissected and weighed. The force-time records were then analyzed by a computer algorithm written in MatLab (The MathWorks) to calculate isometric force. To determine the specific force generated by each muscle, raw force was normalized by PCSA, the only anatomical value, which has been shown to be proportional to muscle force generation (). PCSA takes into account both fiber length and fiber pennation angle to yield a value that represents the total cross-sectional area of muscle fibers. This value permits comparison between the intrinsic contractile properties of muscles of different sizes. Because mice are slightly smaller compared with wild-type littermates, we did not want the small size of the muscle to dominate the analysis. Thus, PCSA was calculated for each muscle using the following equation:where L is fiber length, θ is the fiber pennation angle, and ρ is muscle density. L was determined for each specimen using the TA L ratio of 0.65 (). Muscle-specific tension was then calculated as muscle force/PCSA. Because normal mammalian muscle generates a specific tension of ∼250 kPa, this method also allowed us to compare these neonatal muscles with their adult counterparts. Fig. S1 shows an example of the thin filament length analysis in gastrocnemius muscle using distributed convolution. Online supplemental material is available at .

Human cells encode ∼70 Rab GTPases that are localized to distinct membrane-bound compartments (, ; ). These proteins regulate transport vesicle formation, motility, docking, and fusion via interaction with so-called effector proteins that bind with preference to Rabs in their GTP-bound conformations (). The number of identified Rab effector proteins is growing steadily, yet little is known about how Rabs are localized correctly within cells (). COOH-terminal prenylation contributes to the stable membrane association of Rab proteins. Rabs also interact with numerous effectors to form microdomains on organelle surfaces (). For example, Rab5 binds to early endosome antigen-1 (EEA1), which binds to early endosomes via Rab5 and also by binding to phosphatidylinositol-3-phosphate. Rab5 recruits the kinase that generates this lipid, thereby catalyzing the generation of a membrane microdomain. The Rab9 GTPase recruits a cytosolic protein, tail-interacting protein of 47 kD (TIP47), which binds both to Rab9 and to the cytoplasmic domains of two mannose 6-phosphate receptors (MPRs; ). In both cases, combinatorial recognition of a Rab and a membrane constituent enhances the selective recruitment of a cytosolic effector protein. Thus, there are many examples of Rabs that serve as determinants of effector–membrane binding. But how are Rabs themselves localized? Prenylated Rabs are delivered to membranes by a protein named GDP dissociation inhibitor (GDI; ). Complexes of Rabs bound to GDI bear all of the information needed to accomplish their specific membrane delivery (; ). Proteins called GDI displacement factors (GDFs) may facilitate Rab recruitment; e.g., Yip3 protein was recently shown to be able to release Rab9 from GDI and lead to its membrane association (). But Yip3 is not a Rab receptor, and it acted catalytically to permit Rab9 to associate with membranes. Yip3 catalyzes the release of endocytic, but not exocytic, Rabs from GDI (). Thus, although GDFs likely contribute to Rab delivery, we know little about how they distinguish between Rab types or the breadth of their substrate recognition. Their partial specificity cannot by itself, explain the sequestration of this category of Rabs into early endosomes, recycling endosomes, or late endosomes. Moreover, steady-state localization of Rab proteins is likely to include interactions of Rabs with other constituents, after they are delivered to a membrane surface. Early work on Rab localization suggested that COOH-terminal Rab hypervariable domains specified their distinct localizations (; ; ). But more recent analyses suggest a more complex scenario. found that several Rabs were correctly localized, despite bearing significant alterations in their hypervariable domains. As described herein, we obtained similar findings for a different set of Rab chimeras, and we used this as a starting point to explore the mechanisms of Rab localization. In this study, we present evidence that certain effectors may play a special role in Rab9 localization. Hints of this came from a recent study in which the Rab9 effector TIP47 was depleted from cells using RNAi (). Loss of most of the predominantly cytosolic TIP47 protein led to the destabilization of Rab9; its half-life decreased from 32 to 8 h. This was unexpected, because we think of prenylated Rabs as independent entities residing on organelle surfaces or as a complex with GDI in the cytosol. We show that TIP47 is a “key” effector, in that it controls Rab9 stability (), as well as its steady-state localization. In addition, TIP47 can compete with Rab1 and Rab5 effectors to relocalize Rab1 and Rab5 chimeras to late endosomes. We generated and purified hypervariable domain chimeras based on late endosome–localized Rab9, early endosome–localized Rab5, and Golgi-localized Rab1 (). The hypervariable domain junction was selected based on sequence alignments (; ) and on the three-dimensional structures of Rab3A (), Rab5C (), and Rab9 (). Junctions were placed near the end of helix 5 at a dibasic repeat that is conserved in Rab9, Rab5, and Rab1 sequences (, blue residues). Most of the hypervariable domain is unstructured and extends beyond the GTPase fold (and is not included in the structures shown). By splicing sequences at or near the end of the structured portion, we hoped to avoid significant alterations in Rab structure. When expressed as GFP fusions in cells, the chimeras were efficiently prenylated (). In all cases, prenylation efficiency was ∼50%, as determined by comparative analysis of membrane and cytosolic fractions (). Membrane-associated forms migrated at a different mobility upon SDS-PAGE, which is consistent with prenylation. This suggests that the chimeras were folded well enough to be recognized by Rab prenyltransferase. In addition, all of the purified chimeras were highly active for nucleotide binding and, in most cases, effector binding ( and and ), which are both important tests of proper protein folding. shows the ability of the purified GST chimeras to bind radiolabeled GTP in a nucleotide exchange reaction. On average, these chimeras were ∼75% active in terms of their abilities to release bound GDP and to bind added GTP. Although there was slight variability between each purified Rab protein preparation, the 20% maximum differences observed could not account for the differences in effector binding shown in and . shows the ability of each of the chimeras to bind to the Rab9 effectors TIP47 (; ) and p40 (). Pure, nonepitope-tagged Rab proteins were prebound to [S]GTPγS; unbound nucleotide was removed by gel filtration, and the active proteins were incubated with His-tagged effectors, and then immobilized on an Ni-NTA resin. As expected, Rab9 bound specifically to TIP47 and p40; Rab1 and Rab5 showed very low binding ability (). The Rab9 hypervariable domain was required both for TIP47 and p40 binding, as neither Rab9/1 () nor Rab9/5 () bound significantly to either effector (). Surprisingly, Rab5/9 and Rab1/9 bound well to Rab9 effectors compared with the parental Rabs 1 and 5. Therefore, the Rab9 hypervariable domain is not only required for binding TIP47 and p40, but it is also sufficient for Rab9 effector binding (especially to TIP47) within the context of a Rab GTPase structure. Although the hypervariable domain was sufficient for TIP47 binding to Rab5/9 and Rab1/9 chimeras, binding also required the presence of a GTPase domain. This was clear from experiments in which we tested the nucleotide dependence of the interaction (). For both Rab5/9 and Rab1/9, interaction with TIP47 showed the same nucleotide dependence of binding that is seen with Rab9 (), with a preference for the GTP-bound state. Thus, interaction with TIP47 requires more than just the Rab9 hypervariable domain, and likely involves the switch domains of Rab5/9 and Rab1/9, as these domains are the only parts of a Rab that change conformation between GTP- and GDP-bound states. In addition, Rab recognition by p40 appears to require more significant interaction with nonhypervariable domain sequences than Rab protein recognition by TIP47. The same Rab proteins were next tested for binding to the Rab5 effectors rabaptin-5 () and EEA1 (; ), and to the Rab1 effectors GM130, golgin-84, and p115 (; ; ; ). In contrast to what was observed for Rab9 effector interactions, Rab5 binding to two of its effectors did not require Rab5 hypervariable domain sequences. As shown in , GST-Rab5, but not GST-Rab9, bound to the Rab5 effectors EEA1 and rabaptin-5 from bovine brain cytosol. GST-Rab5/9 was fully functional for binding both proteins, whereas active GST-Rab9/5 did not bind either effector. These data confirm that the Rab5 hypervariable domain is neither required nor sufficient for EEA1 () or rabaptin-5 interaction. The published structure of COOH-terminal–truncated (and hypervariable domain–truncated) Rab5 bound to the coiled coil region of rabaptin-5 () is consistent with the hypervariable domain independence of rabaptin-5 binding shown in this study. Rab1 effector binding was studied using GST-Rab proteins and detergent-extracted rat liver Golgi membranes as an effector protein source (). As with the Rab5 effectors rabaptin-5 and EEA1, GM130 did not require the Rab1 hypervariable domain for binding, as it bound well to both GST-Rab1 and GST-Rab1/9 (). In contrast, the Rab1 hypervariable domain was more important for binding to both golgin-84 and p115 (). Rab1/9 failed to bind p115, and showed weaker binding to golgin-84. Although the Rab9 hypervariable domain was sufficient for TIP47 and p40 interaction (), the Rab1 hypervariable domain was necessary, but not sufficient, for golgin-84 and p115 interaction (). Therefore, these two Rab1 effectors also require nonhypervariable domain determinants for Rab recognition. The differences in the Rab-binding profiles of GM130, golgin-84, and p115 show that these Rab1 effectors rely, to different extents, on distinct binding determinants in the Rab1 structure. We determined the intracellular localizations of each of the chimeras as GFP-fusions in mammalian cells; the results are summarized in .Rab5/9 was clearly present in early endosomes, as judged by colocalization with EEA1 (Fig. S2, available at ). Very little Rab5 or Rab5/9 was present in late endosomes (Fig. S2). For the chimeras studied, the Rab5 hypervariable domain was dispensable for both effector binding () and for early endosome localization (; Fig. S2). Rab5/9 and Rab5/7 (not shown) were present on numerous peripheral early endosomes, many located near the plasma membrane (; Fig. S1, available at ). Quantitation of Rab5/9 colocalization with endogenous EEA1, which is a marker for early endosomes, confirmed that Rab5/9 has a localization profile identical to that of Rab5. 71% of EEA1-positive structures contained exogenous Rab5/9; in a parallel experiment, 74% of EEA1 structures contained exogenous Rab5 (). In contrast, only 25% of EEA1-positive structures contained exogenous Rab9 (). Rab5/9 and Rab5 colocalized significantly when coexpressed; 86% of CFP-Rab5/9–positive structures (1,384/1,618) were also positive for YFP-Rab5. Given 85% colocalization between coexpressed CFP- and YFP-Rab5 in control experiments (), we conclude that Rab5/9 shows identical localization to Rab5. Finally, Rab5/9 was not found on late endosomes any more than Rab5 was, as both Rabs showed low colocalization (11 and 12%, respectively) with endocytosed anti–cation-independent (CI) MPR (CI-MPR) IgG that had been chased into late endosomes (). These data demonstrate that the Rab5 hypervariable domain is dispensable for the early endosome localization of Rab5/9. A lack of a requirement for certain hypervariable domain sequences has also been reported by . Together, these data confirm that Rab targeting is more complex than originally believed. In contrast to Rab5, the Rab9 hypervariable domain was required for late endosome localization. Replacement of the Rab9 hypervariable domain in Rab9/5 and Rab9/1 chimeras disrupted late endosome targeting and relocalized both proteins to the Golgi (). Rab9/5 and Rab9/1 were clearly Golgi localized, as determined by their colocalization with p115, but not with endocytosed anti–CI-MPR antibody, in late endosomes (Fig. S1). The Golgi localization of Rab9/5 was not expected if hypervariable domain sequences were key; the protein should have been present on early endosomes. Although it has been proposed that the ER and Golgi may be sites for Rab mislocalization (), we favor an alternative possibility. This laboratory has recently identified two new Rab9 effectors that are localized at the TGN. It is possible that one of these proteins interacts with Rab9/5 preferentially and retains the protein at that site. The Golgi localization of Rab9/1 may also be attributable to its hypervariable domain, as at least one Rab1-binding protein recognizes Rab1 hypervariable domain sequences (). Rab1 contains effector binding information in nonhypervariable domain sequences (), and accordingly, Rab1/9 localized to the Golgi ( and Fig. S3, available at ). Although we cannot yet fully explain the localizations of Rab9/5 and Rab9/1, we conclude that the Rab9 hypervariable domain is required for proper late endosome localization of the Rab9 GTPase. We have presented a test of the ability of Rab effectors to direct the localization of Rab proteins to distinct membrane compartments. Such an analysis was only possible because of our serendipitous generation of two Rab chimeras that were able to bind two distinct classes of Rab effector proteins. A Rab5 protein containing the Rab9 hypervariable domain (Rab5/9) bound well to the Rab5 effectors rabaptin-5 and EEA1, and also showed significant binding to the Rab9 effectors TIP47 and p40. Similarly, a chimera comprised of Rab1 linked to the Rab9 hypervariable domain bound to the Rab1 effectors GM130 and golgin-84. This Rab1/9 hybrid also bound strongly to the Rab9 effector TIP47 and somewhat less strongly to another Rab9 effector, p40. Upon expression in cultured cells, the chimeric Rabs localized, together with their Rab backbone parental counterparts, on early endosomes and the Golgi complex, respectively. Yet, when the Rab9 effector TIP47 was coexpressed, the Rab5/9 and Rab1/9 chimeras moved to late endosomes, together with endogenous Rab9. These data demonstrate that effector binding can relocalize a Rab from one membrane compartment to another. Only TIP47 had the capacity to relocalize the Rabs; another Rab9 effector (p40) did not. Thus, certain Rab effectors can play a dominant role in Rab9 localization. Our analyses confirmed the importance of Rab9 hypervariable domain sequences for interaction with the effector proteins TIP47 and p40; in both cases, this domain was necessary and sufficient for Rab–effector interaction, within the context of a Rab GTPase. A recent study has shown a role for hypervariable domain sequences in a Rab–effector interaction; Rab7 COOH-terminal residues are required for binding to the Rab7 effector RILP (). In addition, polo-like kinase, which is a Rab1-binding protein, interacts with the phosphorylated Rab1 hypervariable domain (). In the structure of Rab3 bound to Rabphilin-3A, residues in α helix 5 that are part of the hypervariable domain contribute to the Rab-binding interface (). In contrast, two Rab5 effectors and two out of three Rab1 effectors that were tested failed to require the presence of hypervariable domain sequences for binding. Indeed, in the case of EEA1, biochemical experiments ruled out a significant role for the Rab5 hypervariable domain in EEA1 binding (). In addition, the crystal structure of Rab5 missing its hypervariable domain with a fragment of rabaptin-5 suggested that the hypervariable domain would not contribute to this Rab–effector pair (). Rab5 hypervariable domain sequences may contribute to effector interactions not tested in this study, but they were not required for the localization of Rab5/9. Thus, for Rab9, hypervariable domain interactions with TIP47 do direct localization, which is consistent with the original proposals of and . For the Rab5/9 and Rab1/9 chimeras, nonhypervariable domain–dependent interactions were sufficient for their initial cellular localizations, which were early endosomes and the Golgi complex, respectively. Our working model for how effectors contribute to Rab localization is shown in . Rabs are delivered to membranes by GDI in their GDP-bound forms (; ). Release of Rabs from GDI may be catalyzed by GDFs (). If nucleotide exchange occurs, the Rab will be able to bind to effectors. This may involve recruitment of cytosolic effectors (, option A) or binding to effectors that are already membrane associated (, option B). Effector binding often stabilizes a Rab in its GTP-bound form, enhancing both Rab–membrane and –effector association. Rab guanine nucleotide exchange factors (GEFs) associate with Rab effectors (), thus, activation of Rabs occurs in the vicinity of specific Rab-binding partners. If a Rab is delivered to the wrong compartment, it will not be a substrate for nucleotide exchange, and GDI can remove it. In this manner, a microdomain can form, containing active Rabs bound to an array of effector proteins. Importantly, the overall specificity of Rab localization is contributed to by multiple components, such as GDFs, the proximity of Rab-specific GEFs, and subsequent effector interactions. For every Rab, there will be numerous potential effector interactions that will depend upon the relative affinity and abundance of each effector. These multiple interactions will contribute to the steady-state localization of a Rab that we score by immunofluorescence microscopy. What is the mechanism by which TIP47 stabilizes and localizes Rab9? TIP47 is a predominantly cytosolic protein () that interacts with Rab9 and the cytosolic domains of the two MPRs via two distinct binding sites (; ; ; ; ; ). TIP47 binding to Rab9 enhances its affinity for MPR cytoplasmic domains (). Cytosolic TIP47 occurs as a homohexamer or larger oligomer, and may further oligomerize when it is membrane associated (). Thus, a TIP47 hexamer has the capacity to interact with multiple Rab9 molecules and multiple MPRs. This combinatorial requirement for both a Rab and MPR cytoplasmic domains enhances the specificity of TIP47 membrane association; the protein binds preferentially to membranes containing both Rab9 and MPRs. Once TIP47 has found a membrane containing both Rab9 and MPRs, it will bind there and drive transport from late endosomes to the TGN (; ). If a hybrid Rab is nearby, either as a consequence of GDI-mediated delivery or because of membrane trafficking events, it has the potential to become part of a TIP47–Rab9 microdomain. All Rabs are in equilibrium with their own effectors, and TIP47-mediated localization of the chimeras was only observed when it could predominate in relation to Rab1 and Rab5 effectors by overexpression in cells. Our ability to relocalize Rab5/9 from early to late endosomes and Rab1/9 from the Golgi to late endosomes is most easily explained by a binding competition between Rab5, Rab1, and Rab9 effectors. NH-terminal His6-tagged p40 (), His6-tagged TIP47 (152–434; ), and Rab9-CLLL () were purified. Human Rab5a, Rab1a, and Rab9/5, Rab5/9, Rab9/1, and Rab1/9 were cloned into pET14b (Novagen), for expression in as unprenylated Rabs, or into pGEX-4T1 (GE Healthcare). Rab9-CLLL was also cloned into pGEX-4T1. Untagged or GST-tagged Rab1a, Rab5, Rab1/9, Rab9/1, Rab5/9, Rab9/5, and Rab9-CLLL were expressed in BL21(DE3) RIL (Stratagene) or Rosetta (Invitrogen) cells and induced with 0.5 mM IPTG for 3 h at 37°C (Rab5 and Rab9/1) and 16 h at 25°C (Rab1a, Rab1/9, and Rab5/9, GST-Rab9CLLL, GST-Rab1a, GST-Rab1/9, and GST-Rab9/1), and with 1 mM IPTG for 4 h at 30°C (Rab9/5). Cells (6 L) were resuspended in 40 ml of 50 mM Tris-HCl, pH 8.0, 8 mM MgCl, 2 mM EDTA, 0.5 mM DTT, 10 μM GDP, and protease inhibitor cocktail (Roche; Rab1a, Rab5, Rab9/1, Rab1/9, GST-Rab1a, GST-Rab1/9, GST-Rab9CLLL, and GST-Rab9/1), or 50 mM MES, pH 6.5, 8 mM MgCl, 2 mM EDTA, 0.5 mM DTT, 10 μM GDP, protease inhibitor cocktail (Rab9/5 and Rab5/9). Cells were lysed using a French press, and lysates were centrifuged at 15,000 rpm for 20 min (JA 20 rotor; Beckman Coulter). Homogenates were centrifuged at 55,000 rpm for 30 min (70Ti rotor; Beckman Coulter). Supernatants were diluted 10-fold with lysis buffer and loaded onto a 25-ml Q–Sepharose Fast Flow column (GE Healthcare; Rab1a, Rab5, Rab1/9, Rab9/1, Rab9/5, GST-Rab1a, GST-Rab1/9, GST-Rab9CLLL, and GST-Rab9/1), or a SP–Sepharose Fast Flow column (GE Healthcare; Rab5/9). With Rab5, Rab9/5, Rab5/9, Rab1/9, and Rab9/1, proteins were eluted with 0–400 mM NaCl in lysis buffer. With Rab5, Rab1/9, and Rab9/5, fractions were precipitated with 60% (Rab5) or 40% (Rab1/9, Rab9/5, GST-Rab1a, GST-Rab1/9, GST-Rab9CLLL, and GST-Rab9/1) (NH)SO4, resuspended in buffer A (20 mM Hepes, pH 7.4, 200 mM NaCl, 1 mM MgCl, 0.5 mM DTT, and 10 μM GDP) and loaded onto a Superdex 200 column (GE Healthcare). Pools were brought to 10% (vol/vol) glycerol and stored at −80°C. With Rab9/1 and Rab5/9, fractions were pooled and 4 ml loaded onto a Superdex 200 column. With Rab1a, the Q–Sepharose flow-through was subjected to 60% (NH)SO4 precipitation, and the precipitate was resuspended in buffer A. GST-Rab5, GST-Rab5/9, and GST-Rab9/5 were expressed and purified on glutathione–Sepharose beads (GE Healthcare), as recommended by the manufacturer. Rab nucleotide-binding activity was performed as previously described (), except for nucleotide exchange; Rabs (200 nM) were incubated in 50 mM Hepes, pH 7.5, 150 mM KCl, 2 mM EDTA, 1 mM MgCl, 0.1% BSA, and 2 μM [S]GTPγS (4 nCi/μl) for 45 min at 25°C in 50 μl of extract. Rab proteins (2.6 μM) were preloaded with 3 μM [S]GTPγS or [H]GDP (4.5 nCi/μl) in 2.2 ml of extract containing 50 mM Hepes, pH 7.4, 150 mM KCl, 15 mM imidazole, 4.5 mM EDTA, 5 mM MgCl, 0.5 mM DTT, and 100 μg/ml BSA. Rab-GTPγS was separated from free nucleotide by gel filtration (G-25). TIP47-binding reactions (450 μl) were in 50 mM Hepes, pH 7.4, 150 mM KCl, 15 mM imidazole, 5 mM MgCl, and 100 μg/ml BSA. Rab–[S]GTPγS complex (571 nM) was incubated with TIP47 (2.5 μM) at room temperature for 1.5 h. TIP47-bound Rabs were recovered on Ni-NTA resin (QIAGEN; 50 μl of a 50% slurry) and counted. Rab–[S]GTPγS complexes (571 nM) were incubated with 2.5 μM p40 at room temperature for 1.5 h in 50 mM Hepes, pH 7.4, 150 mM KCl, 25 mM imidazole, 5 mM MgCl, and 100 μg/ml BSA. p40-bound GTPases were recovered on Ni-NTA resin (QIAGEN; 50 μl of a 50% slurry) and counted. GST-Rab fusion proteins (3.0 μM) were incubated in binding buffer (20 mM Hepes, pH 7.5, 100 mM NaCl, 4.5 mM MgCl, 5 mM EDTA, 300 μM GTPγS, and 0.2 mM DTT) for 1.5 h at room temperature in 200 μl of extract. Bovine brain cytosol (40% (NH)SO4 precipitate; 200 μl of ∼15 mg/ml) in binding buffer (+1 mM PMSF, 5 mM MgCl and protease inhibitor cocktail, but no EDTA) was added at 20°C for 1 h. Glutathione–Sepharose (GE Healthcare; 10% slurry in binding buffer) was added for 20 min on a rotator. Beads were washed three times with 1 ml of wash buffer (20 mM Hepes, pH 7.5, 100 mM NaCl, 5 mM MgCl, and 0.2 mM DTT). Proteins were eluted with SDS-PAGE loading buffer and analyzed by immunoblot with anti-EEA1 and anti–rabaptin-5. Binding of Rab1 effectors from rat liver Golgi extracts to GST-Rabs was measured as previously described (), with minor modifications. In brief, 10 nmol (∼0.5 mg) of each GST-Rab was loaded on glutathione–Sepharose. Bead-bound GST-Rab1 and GST-Rab1/9 were loaded with GTPγS (), whereas GST-Rab9CLLL and GST-Rab9/1 were loaded with GTPγS in 20 mM Hepes, pH 7.4, 150 mM KCl, 5 mM EDTA, 4.5 mM MgCl, 1 mM DTT, and 100 μM GTPγS for 1.5 h at room temperature, followed by a 30-min incubation in 20 mM Hepes, pH 7.4, 100 mM NaCl, 5 mM MgCl, and 100 μM GTPγS at room temperature. Golgi extracts were made by diluting rat liver Golgi membranes to 0.5 mg/ml with 20 mM Hepes, pH 7.4, 100 mM KCl, and 5 mM MgCl, and adding 1% Triton X-100, 1 mM DTT, protease inhibitor cocktail (Roche), 1 mg/ml soybean trypsin inhibitor, and MgCl to a final concentration of 8 mM. After incubation on ice for 30 min, extracts were spun at 16,000 for 20 min. 1 mM GTPγS was added to the supernatant. Extract (200 μl) was incubated with GST-Rab–loaded beads for 1 h at 4°C. Beads were washed, eluted, and analyzed by SDS-PAGE and Western blot, as previously described (). cDNAs for canine Rab9a and human Rab5a, Rab1a, and Rab9/5, Rab5/9, Rab9/1, and Rab1/9 were cloned into pECFP-C1, pEYFP-C1, and pEGFP-C3 (CLONTECH Laboratories, Inc.) for expression. For fixed cell indirect immunofluorescence, HeLa cells were grown and transfected on glass coverslips. Cells were transfected (Fugene 6; Roche); 16–24 h after transfection, cells were fixed, permeabilized, and stained (), and then mounted in ProFade mounting medium (Invitrogen). Micrographs in and and Figs. S1–S4 were acquired using a deconvolution microscopy system (Spectris; Applied Precision, LLC) with an inverted epifluorescence microscope (IX70; Olympus), a PlanApo 60×, 1.40 NA, oil immersion objective (Olympus), a charge-coupled device (CCD) camera (CoolSNAP HQ; Roper Scientific), and acquisition and deconvolution software (DeltaVision; Applied Precision, LLC). Micrographs in were acquired using a microscope (Axioplan2; Carl Zeiss MicroImaging, Inc.) fitted with 63×, 1.30 NA, and 100×, 1.3 NA, Plan Neofluar objective lenses, a CCD camera (AxioCamHRc; Carl Zeiss MicroImaging, Inc.), and controlled by Axiovision 4.2 software (Carl Zeiss MicroImaging, Inc.). Pictures were analyzed using ImageJ (National Institutes of Health) and Photoshop (Adobe) software. For late endosome labeling, cells were incubated with 30 μg/ml of mouse anti–CI-MPR (2G11) IgG with a 20-min pulse and a 40-min chase in complete medium. Endogenous organelle markers used were as follows: anti-p115 (Golgi; gift from G. Waters, Merck Research Laboratories, Rahway, NJ), anti-TGN46 (TGN; Serotec), anti-EEA1 and anti-rabaptin-5 (early endosomes; BD Biosciences), and CI-MPR (primarily late endosomes; ). Rabbit anti-GFP and secondary antibodies derivatized with Alexa Fluor dyes (Invitrogen) were used. The extent of colocalization between endosomal Rabs and organelle markers was quantitated by generating contour maps of each image by overlaying the Rab and marker channels. The total number of vesicles and the number showing overlap between both markers were tallied. All fixed cell microscopy was performed at room temperature. For live cell microscopy, BS-C-1 cells on glass coverslips were observed 16–20 h after transfection at 37°C on a temperature-controlled stage of an inverted microscope (Diaphot-300; Nikon) equipped with a cooled CCD camera (NDE/CCD; Princeton Instruments), Plan Apo 60×, 1.40 NA, and 100×, 1.40 NA, oil objectives, and using MetaMorph software (Molecular Devices); images were analyzed using Photoshop (Adobe). Dual CFP/YFP filters were obtained from Chroma Technology Corp. CFP- or YFP-tagged wild-type Rabs or CFP-Golgi (CLONTECH Laboratories, Inc.) were used as organelle markers. HeLa cells on 100-mm dishes were transfected for expression of Rab fusion proteins with GFP, CFP, or YFP. After 23 h, cells were washed three times with 10 ml of phosphate buffered saline, and then two times with 10 ml ice-cold HM buffer (25 mM Hepes, pH 7.4, and 2.5 mM Mg(OAc)). Cells were swollen on ice for 5 min, washed gently with 5 ml SEAT buffer (10 mM ethanolamine, 10 mM acetic acid, 1 mM EDTA, and 0.25 M sucrose) leaving ∼0.5 ml of buffer, and then scraped off of the plate with a rubber policeman and transferred to microfuge tubes. Protease inhibitor cocktail (Complete EDTA-Free; Roche) and 0.1 M PMSF were added to each sample. Lysates were passed through a 27-gauge needle 20 times and spun for 20 min at 341,000 and 4°C in a rotor (TLA-100.1; Beckman Coulter). Pellets and supernatants were analyzed by SDS-PAGE and an anti-GFP Western blot. Rab nucleotide–binding activity was performed according to , with minor changes. Rabs (200 nM) were incubated in 50 mM Hepes, pH 7.5, 150 mM KCl, 2 mM EDTA, 1 mM MgCl, 0.1% BSA, and 2 μM [S]GTPγS (4 nCi/μl) for 45 min at 25°C in 50 μl of extract. Reactions were in triplicate; 40-μl samples were removed, diluted with 3 ml of ice-cold wash buffer (20 mM Tris-HCl, pH 8.0, 100 mM NaCl, and 25 mM MgCl) and passed through 24-mm HA filters. Filters were washed three times with 3 ml of wash buffer, dried, and counted in scintillation fluid. Fig. S1 shows deconvolution microscopic localizations of Rab9/5 and Rab9/1 in fixed HeLa cells. Fig. S2 shows deconvolution microscopic localizations of Rab5/9 and Rab5 in fixed HeLa cells. Fig. S3 shows deconvolution microscopic localizations of Rab1/9 and Rab1 in fixed HeLa cells. Fig. S4 shows p40 is not a key effector for Rab9 localization. Online supplemental material is available at .

The majority of cell types in multicellular organisms are polarized and face two different environments. For example, epithelial cells face the outside world or lumen of an organ on one side, and the interstitial environment and basement membrane on the other. These cells exhibit functional and structural asymmetry in their apical and basolateral plasma membranes that is essential to their function. Epithelial cell polarity depends on the accurate targeting of apical and basolateral plasma membrane proteins (; ). Targeting information is usually present in the cargo proteins themselves. These targeting signals are thought to be recognized in the TGN or endosomes, which leads to the sorting of cargo proteins into transport vesicles destined for the apical or basolateral plasma membrane. Like most intracellular membrane fusion events, vesicle fusion with the apical and basolateral plasma membranes is mediated by the SNARE machinery (; ). Mammals express >30 different members of the SNARE superfamily, each one of them associated with a particular fusion event (; ; ). SNAREs are characterized by one or two conserved SNARE motifs of ∼60 amino acids (, ) that mediate the SNARE–SNARE interactions. SNARE complexes contain at least one member of the syntaxin family, in addition to two or three other cognate SNAREs. SNARE pairing was initially proposed to contribute to the overall specificity of membrane trafficking pathways (). Using in vitro–reconstituted fusion assays, it has been demonstrated that only matching combinations of cognate SNAREs lead to successful membrane fusion (; ). An important question is whether SNAREs, indeed, contribute to specificity of trafficking in living cells. Epithelial cells generally contain at least two different plasma membrane syntaxins. Syntaxins 3 and 4 localize to the apical and basolateral plasma membranes, respectively, in virtually all epithelial cell types investigated to date. This includes the MDCK cell line (), all epithelial cell types along the renal tubule in vivo (), and a range of epithelial cells from other tissues (). Syntaxin 3 is involved in biosynthetic trafficking from the TGN to the apical plasma membrane and in apical recycling (). Syntaxin 4 functions in trafficking from the TGN to the basolateral plasma membrane (). The high degree of conservation of the polarity of syntaxin 3 and 4 suggests that their function and proper localization may play an important role in epithelial polarization. The clear distinction between apical and basolateral trafficking pathways makes epithelial cells a good system in which to test the central prediction of the SNARE hypothesis on their contribution to the overall specificity of trafficking pathways. For example, one would predict that mislocalization of the apical syntaxin 3 to the basolateral domain would allow the inappropriate fusion of apical transport vesicles with that domain and reduce the fidelity of polarized trafficking. This is supported by our previous results; disruption of microtubules leads to partial mislocalization of syntaxin 3 to the basal membrane. Under these conditions, post-Golgi transport vesicles carrying apical cargo are able to fuse with the basal membrane (). Although these results were consistent with the idea that syntaxin 3 must be restricted to the apical membrane to achieve maximal fidelity of apical cargo transport, it could not be excluded that the observed cargo mistargeting was an indirect effect of microtubule disruption. To more fully test the contribution of apically localized syntaxin 3 to the fidelity of polarized trafficking, we have now investigated the mechanism of apical targeting of syntaxin 3. We report that syntaxin 3 contains a necessary and sufficient apical targeting signal in its NH-terminal helical domain and that disruption of this signal leads to the random localization of syntaxin 3 at the apical and basolateral domain. Expression of mislocalized syntaxin 3 results in mistargeting of apical cargo proteins and in the overall disruption of epithelial cell polarity. These results indicate that proper SNARE pairing, indeed, contributes to the specificity of membrane trafficking pathways in vivo. Furthermore, these results show that epithelial cell polarity is dependent not only on the function of syntaxin 3 but also on its polarity. At steady state, syntaxin 3 is highly enriched at the apical plasma membrane domain of MDCK cells (). To test whether newly synthesized syntaxin 3 is sorted in the biosynthetic pathway and directly delivered to the apical membrane, we established an assay based on pulse-chase metabolic labeling and detection of syntaxin 3 at the surface. Syntaxin 3 lacks an extracytoplasmic domain. To enable the detection of surface-delivered syntaxin 3 in intact cells, we engineered a fusion protein containing two COOH-terminal myc epitope tags (). We have previously shown that the added epitope tags are accessible to binding by anti-myc antibody in intact cells and do not interfere with the apical targeting of wild-type syntaxin 3 (). MDCK cells that were stably transfected with this syntaxin 3 fusion protein were cultured on permeable filter supports to establish polarized monolayers. Cells were pulse-labeled with [S]methionine, and newly synthesized syntaxin 3 was chased to the surface for different periods of time in the presence of anti-myc antibody in either the apical or basolateral media compartment to capture surface-delivered syntaxin 3. Successive immunoprecipitation of antibody-tagged and untagged radiolabeled syntaxin 3 allows quantitation of the kinetics of surface delivery (see Materials and methods). As shown in , although syntaxin 3 delivery is primarily apical, a significant fraction is also initially targeted to the basolateral domain. This suggests that syntaxin 3 undergoes sorting both during the initial delivery and at a later step, presumably after endocytosis from the basolateral membrane. Most apical targeting signals have been identified within extracytoplasmic domains of membrane proteins. Because syntaxin 3 does not contain an extracytoplasmic domain, its apical targeting must be specified by a determinant within the cytoplasmic or transmembrane domains. To identify an apical targeting signal of syntaxin 3, we generated mutants with successively deleted domains. Structural domains of syntaxin 3 were identified by sequence alignment with the highly homologous syntaxin 1 whose structure has been previously reported (; ; ; ). Five domains are identified () as follows: an unfolded NH-terminal domain is followed by a bundle of three α helices (H), an unfolded linker domain, the SNARE domain, and the COOH-terminal transmembrane domain. Deletion of the NH-terminal unfolded domain (Syn3-Δ27) has no effect on polarized targeting (). However, deletion of the H domain (Syn3-Δ146), and any further deletion, results in loss of apical-specific targeting and random localization at the apical and basolateral domain (), indicating that the H domain contains a necessary apical targeting signal. Fusing the H domain directly to the transmembrane domain and omitting all other regions of syntaxin 3 (Syn3-27-146+TM) restores specific apical targeting (). These results indicate that the H domain of syntaxin 3 contains a necessary and sufficient apical targeting signal. To further locate this signal, we generated additional deletion mutants. The region of the syntaxin 3 gene encoding the H domain contains four exon boundaries. Because exons often encode structural or functional domains, we designed deletion mutants according to their boundaries (). Deletion of the NH-terminal 38 residues (Syn3-Δ38) and any further deletion prevents specific apical targeting (), indicating that the region between residues 27–38 is critical. Comparison of the primary structures of the four closely related plasma membrane syntaxins (1–4) revealed that this region contains a six-residue sequence (FMDEFF) that is conserved between syntaxins 1–3, but differs in syntaxin 4 (). The syntaxins 1–3 are known to target to the apical plasma membrane domain in polarized epithelial cells, whereas syntaxin 4 is strictly basolateral (; , ; ; ). To test whether the FMDEFF motif is critical for apical targeting, we mutated each residue individually to an alanine. Three mutations (Syn3-F31A, D33A, and E34A) result in the loss of specific apical targeting, whereas the three others (Syn3-M32A, F35A, and F36A) behave like wild type (). This result indicates that the apical targeting signal of syntaxin 3 is centered around the first four residues (FMDE) of this conserved motif and that the residues F31, D33, and E34 play a critical role. In neurons, syntaxin 1 binds to SNAP-25 to form a functional t-SNARE that can interact with the v-SNARE on synaptic vesicles. The interaction between syntaxin 1 and SNAP-25 depends solely on the SNARE domains of these proteins (). It has previously been reported that syntaxin 1 and SNAP-25 may be targeted to their final destination together in a complex, but this has been controversial (). SNAP-23 is a nonneuronal isoform of SNAP-25, binds to both syntaxin 3 and 4 (), and localizes to both the apical and basolateral plasma membrane in MDCK cells (). To test whether mutagenesis of the apical targeting signal affects binding to SNAP-23, wild-type syntaxin 3, Syn3-Δ38, and the six alanine point mutations were expressed in MDCK cells, which were immunoprecipitated using an anti-myc epitope antibody, and the binding to SNAP-23 was analyzed by immunoblotting. As shown in , wild-type syntaxin 3 and all mutants bind to SNAP-23 to a similar degree. This indicates that the loss of specific apical targeting in these mutants is not caused by an inability to bind to SNAP-23. Members of the SM protein family regulate syntaxin function (). In the case of syntaxin 1, the SM protein Munc18a has been shown to bind to a conformation in which the H domain is tightly bound to the SNARE domain. Munc18a binding is thought to stabilize this closed conformation of syntaxin 1 and prevent interactions with other SNAREs. Munc18b is a nonneuronal homologue that specifically binds to syntaxin 3 in a region that includes its H domain (). Therefore, we tested whether binding to Munc18b may be required for the apical targeting of syntaxin 3. Wild-type syntaxin 3, Syn3-Δ38, and the six alanine point mutations were again expressed in MDCK cells, and Munc18b binding was analyzed by immunoprecipitation and immunoblotting. Wild-type syntaxin 3 coprecipitates with Munc18b, but Syn3-Δ38 does not (). Four of the point mutants (Syn3-F31A, D33A, E34A, and F35A) are able to bind to Munc18b, whereas two of the mutants lost binding activity (Syn3-M32A and F36A). However, the ability to bind to Munc18b does not correspond to the ability of the syntaxin 3 mutants to be correctly apically targeted. For example, Syn3-F31A, which is mislocalized, is still able to bind to Munc18b. In contrast, Syn3-F36A is properly apically localized, but has lost its ability to bind to Munc18b. This result indicates that apical targeting of syntaxin 3 is independent of its binding to Munc18b. Next, we tested whether the localization of Munc18b in turn may be determined by the localization of syntaxin 3. Munc18b localizes to the apical plasma membrane of renal epithelial cells (). Because our available antibodies did not allow us to reliably detect the localization of endogenous Munc18b in MDCK cells, we transfected cells with epitope-tagged Munc18b. As shown in , Munc18b expressed alone exhibited a cytoplasmic distribution. However, cotransfection with wild-type syntaxin 3 resulted in colocalization of Munc18b with syntaxin 3 at the apical plasma membrane. In contrast, coexpression with the mistargeted syn3-E34A mutant resulted in membrane association of Munc18b in a nonpolarized manner. Altogether, these results indicate that both membrane-anchoring and the proper polarized localization of Munc18b depend on syntaxin 3. It has previously been reported that a fraction of syntaxin 3 can be recovered in detergent-insoluble membranes, and it was proposed that raft-association may play a role in apical targeting of syntaxin 3 (). We tested whether our syntaxin 3 mutants may fail to be properly targeted apically because of defective raft association. MDCK cells stably expressing wild-type syntaxin 3, syn3-Δ38, or wild-type syntaxin 4 as a control were subjected to detergent extraction and floatation gradient centrifugation, as previously described (). Caveolin-1 served as a raft-associated positive control and calnexin served as a nonraft control. As shown in , a large fraction of caveolin-1, but not calnexin, can be recovered in fraction 7 of the sucrose gradient. A smaller fraction of wild-type syntaxin 3 also partitions in this raft fraction, whereas syntaxin 4 does not. The syn3-Δ38 mutant partitions in the raft fractions to a similar extent as wild-type syntaxin 3. This result indicates that raft partitioning is not affected by deletion of the apical targeting signal of syntaxin 3. Therefore, although raft partitioning may be necessary for apical targeting of syntaxin 3, it is not sufficient. 3D structures of apical or basolateral targeting signals have not yet been clearly elucidated. The four-residue FMDE motif that we have identified as the apical targeting signal of syntaxin 3 is 100% conserved in syntaxin 1 (). The structure of the H domain containing this motif has been reported for syntaxin 1 (; ). Assuming that the same motif is used for apical targeting of syntaxin 1 in epithelial cells, this would therefore be the first known 3D protein structure containing a signal involved in polarized targeting. This allowed us to generate a model for the H domain of syntaxin 3 using the syntaxin 1 structure as a template. As shown in , the side chains of all three critical residues, which affect localization (F31, D33, and E34), are exposed on the surface of the protein. This suggests that these residues may be contacted directly by a targeting factor recognizing this signal. The side chains of D33 and E34 are completely exposed, and one face of the F31 side chain is exposed. The other side of F31 faces the interior of the three-helical bundle formed by helices a–c, and potentially engages in a weak interaction with a methylene group of R116 on helix c. All three residues of the FMDEFF motif, whose mutation has no effect on apical targeting (M32, F35, and F36), engage in hydrophobic interactions with side chains of the opposing helices b or c (). These residues may, thus, help to stabilize the three-helical bundle, but would be unlikely to interact with a putative apical sorting adaptor, which is consistent with our targeting results. The crystal structure of the syntaxin 1–Munc18a complex has also been reported () and also contains the conserved FMDEFF motif. Therefore, we generated a model of the syntaxin 3–Munc18b complex, based on this crystal structure (). In the syntaxin 1A-Munc18a structure, the first turn of the “a” α-helix of syntaxin 1A is partially unwound, relative to the uncomplexed structure. We have modeled the syntaxin 3–Munc18b complex accordingly. In this model, F31 contacts S70 and L71 of Munc18b. It is therefore unlikely that F31, which is critical for apical targeting of syntaxin 3, would be accessible to a putative apical sorting adaptor if syntaxin 3 is in a complex with Munc18b. This is consistent with our data (), indicating that syntaxin 3 targeting is independent of Munc18b. F36 interacts with W28 of Munc18b (). This extensive hydrophobic contact was also noted in a syntaxin 3–Munc18b model by , and F36 is conserved in syntaxins 1–4. Our results verify that this contact is required for the association of syntaxin 3 and Munc18b (). Our results also showed that mutation of M32 substantially reduces the syntaxin 3–Munc18b interaction (). Our model does not suggest a direct basis for this effect because M32 is not located within contact distance of Munc18. However, the side chains of M32, F36, and F31 pack tightly together into a hydrophobic bundle. Thus, it is possible that M32 acts as a buttress for the side chains of F36 and F31, indirectly stabilizing their interactions with Munc18b. The three mistargeted point mutants (F31A, D33A, and E43A) of syntaxin 3 are likely to be fully functional because their SNARE domain is unaffected, and we observed normal binding to SNAP-23 and Munc18b. This allowed us to test the central aspect of the SNARE hypothesis, which is that SNARE pairing contributes to the specificity of vesicle-trafficking pathways. If correct, then the purposeful mistargeting of a t-SNARE to an aberrant location should make that location fusion-competent for transport vesicles carrying cargo intended for the original location of this t-SNARE. We investigated a cargo protein (p75-GFP) whose apical trafficking has previously been shown to depend on syntaxin 3 (). It was also shown that in polarized MDCK cells, apical post-Golgi vesicles carrying p75-GFP can reach the basolateral plasma membrane, presumably because of the infidelity of prior targeting mechanisms, but are unable to fuse there (). p75-GFP, which is transiently expressed in MDCK cells, targets to the apical plasma membrane (). As expected, cotransfection with wild-type syntaxin 3 does not change the apical polarity of p75-GFP. However, expression of mistargeted syntaxin mutants (F31A or E34A) resulted in partial mistargeting of p75-GFP to the basolateral plasma membrane (). In contrast, expression of mistargeted syntaxin 3 had no effect on the localization of the basolateral protein p58 (). We next asked whether the mistargeting of apical membrane proteins, which was induced by the expression of mistargeted syntaxin 3, would affect the cells' overall ability to acquire a polarized phenotype. The kinetics of the formation of tight junctions has frequently been used as a measure of the ability of epithelial cells to polarize. For example, disruption of “polarity proteins” such as PATJ, Par-1, and Par-6 in MDCK cells does not result in the complete inability to ultimately form a polarized monolayer, but, rather, causes a kinetic delay (; ; ). Therefore, we tested whether expression of mislocalized syntaxin 3 mutants would affect overall epithelial polarity in a similar fashion. We first cultured parental MDCK cells or cells stably transfected with syn3-E34A on permeable filters at high density for 2 d. Syntaxin expression was induced with doxycycline for 8 h, and cells were subjected to calcium-deficient medium for 15 h, which results in the opening of tight junctions. At time zero, cells were switched back to high calcium medium, and the reestablishment of tight junctions was monitored by measuring the transepithelial electrical resistance (TEER). As shown in , expression of syn3-E34A caused a significant delay of ∼10 h in the characteristic peak of TEER indicative of tight junction reformation. This delay is similar to the effects observed with the disruption of polarity proteins such as PATJ and Par-6 (; ). We also monitored tight junction reformation by immunofluorescence microscopy at different time points after calcium switch. As shown in , tight junctions are only incompletely formed in cells expressing syn3-E34A at 6 h after calcium switch, a time point at which control cells already exhibit extensive, circumferential immunostaining for the tight junction protein ZO-1. This effect of delaying the formation of tight junctions is similar to the effect observed by inhibition of expression of the tight junction protein ZO-1 by RNAi (). These results suggest that syntaxin 3–dependent apical targeting pathways are involved in the polarization events necessary for tight junction formation. Whereas disruption of proteins important for epithelial polarity often only results in a delay in polarization in a 2D culture system, as described above, MDCK cells are more sensitive when cultured in 3D collagen gels (; ; ). Therefore, we cultured MDCK cells in collagen gels for 7–9 d under conditions where they form spherical cysts in which the apical membrane faces a single lumen. Expression of wild-type syntaxin 3 did not interfere with the development of cysts (). In contrast, expression of syn3-E34A results in the inability to form organized cysts (). Instead, the cells formed tumor-like structures consisting of disorganized cells that were apparently unable to properly polarize and form a central lumen. Tight junctions were barely detectable or absent in these structures. This indicates that appropriately polarized syntaxin 3 plays a critical role in apicobasolateral epithelial polarization. We have identified a region centered around a conserved motif at the beginning of the H domain of syntaxin 3 as a necessary and sufficient apical targeting signal. In contrast to basolateral targeting signals, the structure and function of apical targeting signals are not well understood. Basolateral targeting signals are typically found in cytoplasmic domains of integral membrane proteins, and some of these signals are thought to be recognized by clathrin adapters at the level of the Golgi apparatus and/or endosomes. In contrast, most known apical targeting signals do not reside in cytoplasmic domains of membrane proteins. Glycosylphosphatidylinositol anchors and lumenal glycosylation sites have been shown to confer apical targeting information on some proteins. Syntaxin 3 is neither glycosylphosphatidylinositol anchored nor does it possess a lumenal domain. Raft association mediated by transmembrane domains, and possibly by adjacent regions, has been implicated in apical targeting of other membrane proteins. Our results indicate that neither raft-association nor the transmembrane domain of syntaxin 3 are involved in apical targeting. Only recently, apical targeting signals in cytoplasmic domains of a handful of membrane proteins have been identified (; ; ). Interestingly, the cytoplasmic tails of both CFTR and rhodopsin can target to the apical membrane in the absence of transmembrane domains (; ). In the case of CFTR, this depends on a COOH-terminal PDZ-binding domain, suggesting a mechanism of selective retention at the apical plasma membrane. Apical targeting of the GABA transporter GAT-3 () also depends on a COOH-terminal PDZ-binding motif, suggesting a common mechanism. However, this mechanism is clearly different from the apical targeting of syntaxin 3, which does not possess a PDZ-binding domain. Several other diverse cytoplasmic apical targeting signals have been identified in polytopic membrane proteins, but this has not yet led to the identification of a possible common mechanism. Only one potential secondary structure has been reported for the apical targeting signal of a bile acid transporter (). Based on NMR analysis of a synthetic peptide, this has revealed a possible β-turn conformation of a four-residue sequence. Fortuitously, the apical targeting signal that we identified in syntaxin 3 falls in a region that is identical to that of syntaxin 1, whose crystal structure has been solved both for the free protein and for a complex with Munc18. This has allowed us to obtain the first structural model of any polarized targeting signal in the context of the bulk of the protein. The three critical residues that we have identified are all exposed on the surface of a triple-helix structure () and should be accessible for interaction with a putative apical sorting adaptor. Altogether, the targeting motif of syntaxin 3 appears to differ from all other known polarized targeting signals and its characterization may aid in the identification of the machinery required for its recognition. Our results indicate that Munc18b is not involved in the apical targeting of syntaxin 3, even though it binds to a region that overlaps with the identified apical targeting signal. The targeting phenotype of our Ala mutants does not correlate with their ability to bind to Munc18b (). Furthermore, structural modeling suggests that access to the apical targeting signal would be partially blocked in the syntaxin 3–Munc18b complex (). Therefore, we suggest that syntaxin 3 and Munc18b are not targeted together as a complex. This conclusion is consistent with the recent finding that the synaptic targeting of syntaxin 1 is not affected in Munc18a-null animals (). Our experiments () indicate that both membrane-association and apical polarity of Munc18b depend on syntaxin 3 and that it does not contain any polarized targeting information in itself. Interestingly, the FMDE motif of syntaxin 3 overlaps with the predicted binding site on syntaxin 1 (FMDEFFEQVE) of botulinum neurotoxin C (). This toxin inactivates syntaxin 1 by proteolytic cleavage. Syntaxin 3 is also subject to botulinum neurotoxin C cleavage (), suggesting that the same region is recognized. Therefore, we suggest that bacterial neurotoxins, to specifically recognize SNARE proteins, evolved to exploit the exposed domains in SNAREs, which were originally meant for the binding of adaptor proteins that were essential for their subcellular targeting. Another example may be VAMP2, which is recognized by botulinum neurotoxin D in the same region (); it was shown to be required for targeting to synaptic vesicles and endocytosis (), although this coincidence was not recognized. Based on the few cases in which targeting signals of other syntaxins have been identified, one can conclude that there is not a single conserved region that generally contains the signals. The Golgi-targeting signal of syntaxin 5 is contained within its SNARE domain. This domain targets to the Golgi, even in the absence of the transmembrane domain (). In contrast, a longer splice isoform of syntaxin 5 contains an NH-terminal ER retrieval signal (). The localization of syntaxin 6 to the TGN also depends on its SNARE domain, but there is an additional tyrosine-based signal in the middle of the molecule that may act as an internalization signal to facilitate the retrieval of syntaxin 6 back to the TGN (). Finally, the retention in the ER membrane of the yeast syntaxin Ufe1p depends only on the length, but not the sequence, of its transmembrane domain (). The region containing the apical targeting signal in syntaxin 3 has not previously been implicated in the targeting of a syntaxin. However, given that the critical FMDE motif is also conserved in syntaxins 1 and 2 suggests that it may also be used in apical targeting of these syntaxins. Syntaxin 2 has been shown to target to the apical membrane of pancreatic acinar cells (). In the kidney, syntaxin 2 localizes to the apical domain of medullary collecting duct cells, but to the basolateral domain of cortical-collecting duct principal cells (). Furthermore, syntaxin 2 localizes to the basolateral domain of retinal pigment epithelial cells (). This suggests that syntaxin 2 may contain a competing basolateral targeting signal that is recognized in a cell type–dependent fashion. MDCK cells target syntaxin 2 to both the apical and basolateral domain (; ), which may suggest that they can recognize both signals. Whether the FMDE-motif of syntaxin 1 is involved in neuronal targeting remains to be determined. It is now widely accepted that SNAREs are intimately involved in the mechanism of fusion. The question of specificity, however, had been controversial since it was found that SNAREs in solution can bind promiscuously (). Subsequent results from in vitro reconstituted fusion assays with artificial liposomes established that membrane-anchored SNAREs allow only fusion of cognate SNARE complexes (). Our results provide evidence that SNARE pairing also contributes to the overall specificity of trafficking pathways in intact cells. Our results are consistent with a model in which the mislocalization of functional syntaxin 3 to the “incorrect” basolateral membrane makes this membrane permissive for fusion of apical post-Golgi vesicles and leads to the incorrect basolateral delivery of apical proteins. As previously shown by time-lapse imaging of post-Golgi transport vesicles in polarized MDCK cells, the fidelity of targeting of p75-GFP is not absolute, and vesicles carrying p75-GFP can reach the basolateral membrane, but are unable to fuse there (). Our results suggest that the expression of mistargeted mutants of syntaxin 3 renders the basolateral membrane fusion competent for such vesicles, which results in the accumulation of p75-GFP at the basolateral domain (). This strongly supports the notion that SNARE pairing contributes to the overall specificity of membrane trafficking pathways in vivo and suggests that SNARE-mediated membrane fusion acts as a final proofreading mechanism to allow the fusion of “correct” vesicles and deny the fusion of incorrect vesicles with a given target membrane. The effect of mistargeting of syntaxin 3 to the basolateral domain strikingly resembles the defects of apicobasolateral polarity caused by the disruption of so-called polarity proteins. Proteins such as PATJ, Par-1, and Par-6 have been shown to be important for epithelial polarity (; ; ). Their inactivation, usually by siRNA, typically results in kinetic delays in tight junction formation in MDCK cells cultured on permeable filters. For unknown reasons, cell polarity is more severely affected when MDCK cells are cultured in 3D collagen. In the case of syntaxin 3, we find that merely disrupting its specific apical targeting results in a dominant effect that causes polarity defects very similar to those caused by inactivating PATJ, Par-1, Par-6, and other polarity proteins. This indicates that not just the function of syntaxin 3 but also its apical-specific localization is essential for epithelial polarity. MDCK clone #11 cells were used for all experiments. These cells were made from MDCK strain II cells by stable transfection with the tetracycline repressor (Invitrogen), cloning, and extensive characterization of tetracycline inducibility and epithelial polarity parameters. These cells were used for subsequent stable transfections using pcDNA4-TO plasmids (Invitrogen) for tetracycline-inducible expression of proteins of interest. Cells were maintained in MEM containing 5% FBS and penicillin/streptomycin at 37°C and 5% CO. For transgene induction, the cells were induced with 50 ng/ml of the tetracycline analogue doxycycline for at least 16 h. For microscopy studies with polarized syntaxin 3 mutants, the cells were grown on polycarbonate filters (12-mm diam; 0.4 μM pore size; Costar Corning) for at least 48 h. For culture in 3D collagen gels, MDCK cells were seeded from 0.5 to 10 cells/ml in 80% collagen (Vitrogen) and 20% MEM containing 0.02 M Hepes, pH 7.4, and 0.02 M NaHCO on either 16-well chambered slides (Lab-Tek; Nunc) or on 0.2 μm membrane inserts (Anapore; Nunc). The filters were kept at 37°C for 30 min to solidify the collagen, after which media containing 5% FBS and penicillin/streptomycin was added. Gene expression was induced by adding doxycycline after 2 d of seeding, and the cultures were continued for a total of 7–10 d. All expression constructs are based on human syntaxin 3, using a modified pcDNA4-TO expression vector for the addition of two COOH-terminal myc epitope tags and one hexa-histidine tag. Deletion mutants were made by PCR. Point mutants were generated using complementary sense and antisense primers containing the desired mutation in the middle of the primers. PCR products were digested with the enzyme DpnI before cloning into the expression vector. All inserts were confirmed by sequencing. An assay for the quantitation of the kinetics of surface delivery of newly synthesized syntaxin 3 was established by modification of a protocol for measuring surface delivery of the polymeric immunoglobulin receptor in MDCK cells (). In brief, MDCK cells that stably express myc-tagged syntaxin 3 were cultured on transwell filters for 3 d. After 12 h of induction with doxycycline for the expression of syntaxin 3, cells were starved for 30 min in methionine-deficient media (DMEM; Invitrogen). After starvation, cells were metabolically labeled for 15 min with [S]methionine (GE Healthcare), followed by a chase with unlabeled methionine for different time intervals. Anti-myc antibody was present throughout the chase, in either the apical or basolateral media compartment. Antibody binding was allowed to proceed for 60 min on ice, after which excess antibody was washed away. Cells were lysed in a buffer containing Triton X-100 with the addition of MDCK cell lysates containing an excess of unlabeled myc-tagged syntaxin. Antibody-tagged syntaxin molecules that had been exposed to the surface were precipitated with protein G–Sepharose. The remaining syntaxin molecules that had not reached the surface were subsequently immunoprecipitated with fresh antibody and protein G–Sepharose. Immunoprecipitates were separated by SDS-PAGE, gels were dried, and radioactive bands were imaged using a Molecular Imager FX (Bio-Rad Laboratories). Images were quantitatively analyzed using Quantity One analyzing software (Bio-Rad Laboratories). MDCK cells were transiently transfected with myc-tagged syntaxin 3 plasmids. After 24 h, cells were lysed in buffer containing Triton X-100, and syntaxin 3 was immunoprecipitated with cross-linked 9E10 antibody. Syntaxin 3 was detected by Western blot using an affinity-purified polyclonal antibody made against a GST fusion protein with human syntaxin 3. Munc18b was detected by a polyclonal antibody (Affinity BioReagents). A polyclonal antibody against a COOH-terminal peptide of SNAP-23 has been previously described (). Homology models of syntaxin 3 and a syntaxin 3–Munc18b complex were constructed using structures of syntaxin 1A (PDB:1EZ3; ) and a syntaxin 1–Munc18a complex (PDB:1DN1; ref 2) as templates. Models were constructed and optimized using the Swiss-Model website () in project mode. Structures were minimized in the SwissPBDViewer program (), and side chains of residues making obvious clashes were adjusted using rotamers from an extended rotamer library () in the program O (). Figures were generated using PyMOL (Delano, W.L.; ).

All developmental and physiological functions performed by epithelia depend on the polarized targeting of the plasma membrane and secreted proteins to either the apical or basolateral plasma membranes (). Cargo proteins sorted in the Golgi apparatus and the endosomal system through sets of basolateral- and apical-specific sorting determinants are transported to the plasma membrane following partially different routes (; ). Although basolateral secretion has been fairly well characterized, the mechanisms involved in apical trafficking remain poorly defined (). Basolateral sorting signals usually correspond to tyrosine or dileucine residues found in the COOH terminus of proteins. They are recognized by basolateral-specific adaptor complexes (; ), such as AP-1B in epithelial cells (). Before membrane fusion and SNARE action, vesicles are thought to be tethered to the basolateral membrane by the exocyst complex (), which was initially identified in yeast (). In metazoans, the exocyst is required for basolateral delivery of the LDL receptor in MDCK cells (; ), of E-cadherin in the notum (), and for Rhodopsin1 transport in photoreceptor cells (). Recent results suggest that AP-1B and the exocyst act primarily in recycling endosomes (; ; ; ; ), which underlines the central role of this organelle in sorting processes. Indeed, recycling endosomes may be compartmentalized into apical- and basolateral-related domains, or even divided into distinct organelles, suggesting that they could also play a critical role in apical trafficking (; ). Aside from this possible role of recycling endosomes, all other aspects of sorting along the basolateral and apical routes seem to differ. Apical signals are more diverse and often correspond to posttranslational adducts, such as lipids or glycans (; ). For instance, the Hedgehog morphogen is secreted apically upon cholesterol addition, but basolaterally otherwise (). No specific apical cytosolic complex, akin to AP-1B or the exocyst, has been identified so far. Instead, protein clustering, possibly through lipid rafts, is thought to mediate the sorting and transport of apical cargoes (; ). In particular, glycosyl phosphatidylinositol–linked proteins appear to form oligomers that are directly targeted to the apical membrane (, ; ). Several proteins have been proposed to play an active role in apical protein clustering, raft formation, and/or apical delivery, such as caveolins (), annexin 13b (), and the tetraspan protein VIP17/MAL (; ). However, their mechanistic roles have not been fully elucidated, or their implication has been questioned (). In addition to the limited understanding of apical secretion at the molecular level, it is not clear whether the terminal fusion process involves small vesicles, such as those defined at synapses, or larger organelles, such as secretory lysosomes (). Hence, despite the many critical findings originating from tissue culture cells (), investigations with other systems and other cargo proteins could help to elucidate the mechanisms involved in apical exocytosis. , which has contributed to decipher the mechanisms controlling vesicular trafficking (), provides such an in vivo model. We have chosen to analyze apical secretion of cuticle proteins by the epidermis. The cuticle includes glycosylated collagens, glycosyl phosphatidylinositol–linked cuticlins, and lipid-modified Hedgehog-related peptides (; ; ). We previously suggested that the gene is required for cuticle secretion (). The CHE-14 protein is the orthologue of Dispatched, which participates in apical targeting of cholesterol-modified Hedgehog (; ). While searching for alleles (), we uncovered several additional mutations inducing –like phenotypes and reasoned that they might identify new components of the apical trafficking pathway. Two such mutations, and , proved to be small deletions behaving as genetic-null alleles of the gene (unpublished data). The gene encodes one of the four “a” subunits of the V0 sector of the vacuolar H-ATPase (V-ATPase), and is required for development beyond the L2 larval stage (; ). The V-ATPase is a multisubunit protein complex consisting of two subcomplexes called the V0 and V1 sectors (). The cytosolic V1 sector hydrolyses ATP and provides the energy to pump protons through the transmembrane proteolipid pore formed by the V0 sector (). The V-ATPase is the main proton pump establishing a pH gradient in the secretory and endocytic pathways. It generates a proton-motive force that is essential to load synaptic vesicles with neurotransmitters before secretion (). The V-ATPase is also found at the apical plasma membrane of polarized cells, where it is essential for osmoregulation in animal excretory systems (). More recently, biochemical and genetic data suggested that the V0 sector can play a role independently from the V1 sector. In , vacuoles deficient for the “a” subunit Vph1p do not fuse efficiently (; ). In , neurons lacking the “a” subunit Vha100 accumulate vesicles in synaptic terminals (). In both cases, the defects were independent of the proton gradient and placed downstream of SNARE function (; ; ). By further dissecting the role of using targeted mutagenesis, and by comparing phenotypes resulting from the inactivation of V1 or V0 subunits, we uncover a specific role for the V0 sector in mediating secretion to the apical membrane. In particular, we show that the V0 sector is required for apical secretion of Hedgehog-related peptides through a multivesicular compartment able to release exosomes. To determine the distribution and subcellular localization of VHA-5, we raised polyclonal antibodies against its cytoplasmic NH terminus. In addition, we generated a COOH-terminal VHA-5∷GFP fusion, which rescued the larval lethality caused by the deletion (). The VHA-5 antiserum recognized a 105-kD protein in wild-type extracts (, lane a). To prove its specificity, we examined extracts from homozygous animals carrying the rescuing VHA-5∷GFP construct. The VHA-5 antiserum failed to detect the ∼105-kD band in these extracts, but detected an ∼135-kD band (, lane b). These results are consistent with being a small deletion associated with a frameshift ( and not depicted) and with the presence of 257 additional residues in the GFP-fusion protein. We conclude that the VHA-5 antiserum is specific and that is a molecular null mutation. In agreement with previously published observations (; ), we found that VHA-5 is expressed in the H-shaped excretory cell corresponding to the kidney-like organ (). It is also expressed in the main epidermal syncytium (), which had previously been overlooked. The excretory cell extends long processes called excretory canals where osmoregulation takes place (), whereas the epidermis controls body length and apical cuticle secretion (). VHA-5 colocalized apically with the V1 subunit VHA-8 in both tissues (; note that VHA-5 is not expressed in the lateral epidermis). VHA-5 was localized at the level of apical membrane stacks by immunogold staining (). Consistent with VHA-5 distribution and a role of the V-ATPase in osmoregulation (), larvae filled with fluid and died at the L1 stage (unpublished data), which corresponds to the phenotype observed after laser ablation of the excretory cell (). In addition, L1 larvae had a severe malformation of the lateral cuticular specializations known as alae ( and ), which are primarily synthesized by the lateral seam cells. Although VHA-5 is not expressed in these cells, the main epidermal syncytium also contributes to their morphogenesis (). Because VHA-5 is transmembraneous and not cuticular, the simplest interpretation for this phenotype is that mutations compromise the secretion of proteins needed for alae formation. As outlined in the previous section, the V0 sector may fulfill two distinct functions; either working together with the V1 sector to mediate proton pumping or working alone, as in yeast and neurons, to mediate membrane fusion. To determine which of these functions could account for the cuticle secretion defect observed in larvae, we examined the role of other V-ATPase subunits in cuticle formation using the RNAi approach. If improper proton pumping is responsible for the aforementioned cuticle defects, RNAi knockdown of either V0 or V1 subunits should result in similar cuticular defects. Conversely if the loss of a V0-specific function accounts for the cuticular phenotype, only RNAi knockdown of V0 subunits should phenocopy cuticle defects. We chose two V1 subunits (VHA-8 and -13) and one V0 subunit (VHA-4) encoded by single-copy genes, which were, thus, expected to be ubiquitously expressed and to display RNAi phenotypes of comparable severity. In addition, we tested RNAi against the three genes encoding the V0 “c” subunit (, , and ), which are >78% identical at the nucleotide level. We found that the RNAi phenotype of was the strongest and was directly comparable to that of , and (), presumably because it reflects knockdown of all three paralogs. RNAi against these V1 or V0 subunit genes led to 100% lethality in the progeny of treated animals (‘, bottom bars). It is likely that most embryos died because of a defect in yolk endocytosis, which is known to be sensitive to proton pumping (). In agreement, we found that yolk vitellogenin-GFP accumulated in the pseudocoelom of RNAi-treated animals rather than in oocytes and embryos (Fig. S1, available at ). Despite the strong lethality induced by loss of V-ATPase function, a few L1 larvae hatched and, invariably, died filled with fluid before the L2 larval stage (’‘), as observed for larvae. Strikingly, we observed before their death that L1 hatchlings displayed severe alae defects when RNAi targeted the V0 subunits or , but had wild-type alae after knockdown of the V1 subunits or (, B and B’). Consistent with the phenotype of –null mutants, RNAi against also affected alae formation (, B and B‘–C), although lethality was weaker because VHA-5 is not ubiquitously expressed like VHA-4. One trivial explanation for the persistence of normal alae after V1 subunit knockdown could be that RNAi was less efficient than for V0 subunits. It is unlikely, as the lethality rates and the larval osmoregulation defects observed after V1 and V0 subunit knockdown were comparable (, B’ and B‘’), hinting that both RNAi were equally effective. To support this idea, we submitted a transgenic strain to RNAi and verified that it induced a drastic decrease of VHA-8∷GFP fluorescence (). We conclude that the V0 sector is required independently from the V1 sector for apical secretion of some cuticle components. If the V0 sector has two distinct functions, it should be possible to recover alleles that affect either its V0-specific secretion function or its V0+V1 proton-pump function. The distribution and the aforementioned phenotypes indicate that reducing V0-specific function should affect cuticle secretion, whereas impairing proton pumping should affect the excretory canal responsible for osmoregulation. To identify such mutations, we used a plasmid rescue strategy, whereby we generated mutations by using PCR on a rescuing construct, introducing them into animals, and recovering live homozygous animals whenever possible (). We modified charged or large hydrophobic residues, as well as residues previously mutated in the yeast Vph1p (, ). We generated 56 mutations (, B and C; and Fig. S2, available at ); 42 had no obvious phenotype by differential interference contrast (DIC) microscopy (, stars), and eight failed to rescue, indicating that those residues are essential for VHA-5 function (, white boxes). More interestingly, six substitutions rescued the –induced lethality and affected the cuticle, the excretory canal, or both. These six mutations defined three classes, which we will call “cuticle mutations” (L786S, E830Q, and V844F), “canal mutations” (W190A and R191A), or “mixed mutations” (W327A). First, animals carrying cuticle or mixed mutations were significantly shorter and dumpier than wild-type animals or animals carrying canal mutations (). This phenotype is frequently observed for mutations affecting cuticle components (). Western blot analysis using the VHA-5 antiserum detected similar amounts of mutant VHA-5∷GFP proteins (), implying that expression level differences do not explain phenotypic differences. Second, scanning electron microscopy (SEM) showed that adult alae were strongly affected in the former, but not the latter, animals ( and Fig. S3 A, available at ). Third, confocal microscopy using the mutant VHA-5∷GFP as a marker and transmission electron microscopy (TEM) revealed that the excretory canal of animals with canal or mixed mutations, but not with cuticle mutations, was abnormal (, D and E; and Fig. S3, A and B). Their excretory canals had an increased section, often with multiple lumens, and 3–10 abnormal whorls per canal (see in [D and E]). Strikingly, we observed similar phenotypes by knocking down V1 or V0 subunits by RNAi from L3 larval stage until adulthood (). Thus, we infer that the defects induced by canal mutations reflect an impairment of V0+V1 proton pumping, and that they are caused by loss-of-function rather than by gain-of-function mutations. Expansion of the excretory canal in these mutants might help to compensate for the decrease in proton-pumping efficiency. Surprisingly, animals with canal mutations did not show any proton-pumping defect in the epidermis. Possibly, proton pumping is preserved in this tissue because the “a” subunit VHA-7, which is expressed in the epidermis but not in the excretory cell (; ), compensates for the mutated VHA-5. In contrast, we infer that the cuticle defects induced by cuticle mutations reflect an impairment of the V0-specific secretion function, which would not be compensated by other “a” subunits, probably because they are not endowed with this specific function. The existence of various “a” subunits with possibly different functions in the epidermis is reminiscent of the difference observed in yeast between Vph1p and Stv1p (). Lastly, we suggest that the mixed mutation W327A affects both V0+V1 and V0-specific functions. We note that cuticle mutations are located in the last transmembrane domain or in the COOH-terminal luminal tail, whereas canal mutations are in the NH-terminal cytoplasmic part, which is more likely to interact with the V1 sector (). We conclude that the V0-specific and V0+V1 functions of VHA-5 are genetically separable. If indeed the V0 sector is involved in secretion, cuticle mutants should accumulate secretory organelles. At low magnification, TEM through the epidermis showed that animals carrying cuticle or mixed mutations contained significantly more and larger dense organelles than wild-type adults or animals carrying canal mutations ( and Fig. S3 C). At higher magnification these organelles appeared as multivesicular bodies (MVBs; ). MVBs are endosome-derived organelles containing 30–90 nm vesicles, which grow from early and recycling endosomes or from the trans-Golgi network and evolve into lysosomes or into secretory organelles (). Hence, MVB accumulation may reflect either an endocytic/degradation pathway or a secretory defect. To distinguish between these two possibilities, we examined whether mutants had normal lysosomes. In addition, we compared defects to those induced by strong mutations in , , and , which are three essential genes acting at different steps along the endocytic route (; ; ). The rationale for this comparison is that if cuticle mutations affect endocytosis, then , , or mutations should induce cuticle phenotypes comparable to those of mutants. As found in other systems (), we could observe in all mutants intermediate late endosome–lysosome compartments corresponding to enlarged MVBs with multilamellar structures (called hybrid-MVBs in the next section) and normal lysosomes, suggesting that the endocytic/degradation pathway was not qualitatively affected (). In contrast, mutants accumulated large MVBs in their epidermis that did not evolve into lamellar structures (). Furthermore, , , and mutants had normal alae, unlike cuticle mutants (). We conclude that cuticle mutations are unlikely to affect degradation, and, rather, they affect a secretory pathway. To understand the relevance of MVB accumulation in mutants, we reinvestigated secretion in the wild-type epidermis. In hematopoietic cells, some MVBs release their vesicle content into the extracellular space and, thus, play a role in exocytosis, in addition to their well defined role in the endosomal pathway (). The vesicles released by fusion of MVBs with the plasma membrane were originally called exosomes in antigen-presenting cells. In support of the notion that secretion in the epidermis involves exosomes, we observed small light MVBs containing 50–100-nm vesicles just beneath the apical plasma membrane. Moreover, we occasionally saw vesicles immediately above the plasma membrane in the inner cuticular layer, strongly suggesting that a MVB had released its intralumenal vesicles (). These MVBs were always found in the vicinity of epidermal apical membrane stacks, a structure whose role is so far unknown (). In contrast, we rarely observed similar MVBs adjacent to the plasma membrane in cuticle mutants, or they were darker (). These data suggest that the MVB-limiting membrane can fuse with the apical membrane to release exosomes in the cuticle, and show that this process is impaired in cuticle mutants. It raises the possibility that the V0 sector is critical for MVB fusion with the apical membrane during exosome release. An important expectation of the cuticle defects described so far is that we should be able to identify cuticular proteins whose secretion depends on VHA-5 activity. Cuticle proteins include collagens and Hedgehog-related peptides (; ). We found that the collagen DPY-7 was efficiently secreted in –null animals, in animals carrying cuticle mutations, as well as in mutants (Fig. S4, available at ). We turned our attention onto Hedgehog-related peptides, which appeared as good candidates for three reasons. First, alae defects partially resemble those observed in mutants. Second, CHE-14 is homologous to Dispatched, which is a protein required for Hedgehog release (; ). Third, despite the absence of a Hedgehog homologue in , its genome contains several Hedgehog-related peptides required to generate a normal cuticle, although their precise roles remain unknown (; ). We tagged with GFP the secreted domain of the Hedgehog-related peptides WRT-2 and -8 ( and Fig. S3 E), which are expressed in the epidermis (). We found that animals bearing cuticle mutations, but not canal mutations, accumulated VHA-5∷RFP and WRT-2∷GFP or -8∷GFP in discrete entities in their epidermis (, B and C; and Fig. S3, D and F). These entities most likely correspond to the dense and hybrid MVBs () because VHA-5∷RFP also colocalized () with the MVB marker VPS-27∷GFP (). Moreover, both VHA-5 antiserum and a GFP antiserum targeting WRT-2∷GFP decorated the MVBs of cuticle mutants ( and Fig. S5 B, available at ). Last, in wild-type nontransgenic animals, in addition to membrane stacks (), VHA-5 was found at the MVB-limiting membrane, in intralumenal vesicles, and in the cuticle ( and Fig. S5 A), suggesting that it could act at different steps in the secretion of vesicle (see Discussion). Importantly, the VHA-5 protein with the substitutions L786S () or E830Q (not depicted) could reach the plasma membrane in heterozygous animals, which strongly suggests that their intracellular retention in homozygous animals is caused by the loss of a trafficking function rather than by misfolding. Consistently, the WRT-2/8 proteins were not retained intracellularly either in heterozygous animals, despite the presence of the L786S (), or in E830Q transgenes (not depicted). These results indicate that the V0 sector plays a key role in a specific apical secretion pathway that is taken on by Hedgehog-related proteins, but not by collagens. Whereas basolateral secretion is known to depend on the activity of specific complexes (AP-1B and the exocyst), no such complex has been implicated in apical secretion. In addition, although recycling endosomes appear to play a central role in basolateral secretion, their importance in apical secretion is still under active investigation. Our characterization of mutations affecting the V-ATPase “a” subunit VHA-5 sheds new light on the apical biosynthetic secretory pathway. We could observe the fusion of MVBs with the apical plasma membrane in wild-type animals, and the subsequent release of exosomes. In contrast, we found that some VHA-5 mutations accumulate MVBs in their epidermis and prevent the secretion of Hedgehog-related proteins. Thus, we propose a model whereby apical secretion of Hedgehog-related proteins involves their incorporation into the intralumenal vesicles of MVBs, and their subsequent release when MVBs fuse with the apical plasma membrane (). Furthermore, we suggest that the V0 sector of the V-ATPase plays a key role in this process. We can envision two scenarios for the role of the V0 sector. First, mutations affecting cuticle formation could decrease V-ATPase proton pumping along the biosynthetic secretory route to indirectly impair secretion. Consistent with this possibility, mutations in Vph1p (L746S, E789Q, and V803F) corresponding to the cuticle mutations (L786S, E830Q, and V844F, respectively) strongly reduce, but do not eliminate, proton pumping (, ). We think, however, that this possibility is unlikely because in yeast proton uptake is not limiting for fusion of Vph1p-defective vacuoles (). Alternatively, these mutations could uncover a direct role of the V0 sector in apical exocytosis, independently of the V1 sector. Four observations strongly support this notion. First, inhibition of V1 subunits did not affect cuticle formation, although it strongly impaired osmoregulation and endocytosis. Second, we obtained specific V0 mutations inducing a strong cuticular phenotype without any apparent pumping-related defect in the excretory cell. Third, these mutations selectively affected the secretion of Hedgehog-related proteins, but not collagens. Last, strong mutations in well characterized genes blocking different steps of fluid-phase endocytosis did not affect cuticle structure. The specific accumulation of dense and hybrid MVBs, but not of light MVBs, in cuticle mutants, as well as the detection of significant amounts of WRT-2∷GFP within these MVBs, suggests that VHA-5 is neither required to form MVBs nor to load them with cargo proteins. Instead, the presence of VHA-5 at the MVB-limiting and apical membranes suggests that the V0 sector could drive the fusion of MVBs with the apical membrane via the formation of V0 sectors transcomplexes between both membranes (), as suggested in yeast vacuole fusion and at the synapse (; ; ). In vertebrates, 300-nm procollagen-I rod bundles assemble in the ER and travel through the Golgi lumen (). Assuming that worm collagens are secreted this way, they would not fit into exosomes, suggesting that there are at least two distinct secretion pathways in the epidermis, one involving the V0 sector and another followed by collagens. What could explain a common requirement for the V0 sector during the apical exocytosis, yeast vacuole fusion, and synaptic transmission? The prevailing view is that a SNAREpin complex initiates membrane fusion once a vesicle has been docked to a proper membrane (; ). Although the V0 sector is thought to act downstream of SNAREs in yeast and (; ; ), we cannot exclude that it also acts in parallel to SNAREs, at least in , to dock MVBs. Another possibility is that V0 transcomplexes initiate the formation of a protein pore, as initially suggested in yeast (). On the other hand, expansion of the fusion pore is considered as the limiting step in membrane fusion, and might require additional catalysts in vivo (). Such a role could be fulfilled by the V0 sector, either to overcome constraints caused by the big size and/or the specific lipoproteic content of epidermal MVBs and yeast vacuoles, or to allow rapid synaptic transmission in neurons (). Irrespective of the precise role of the V0 sector in membrane fusion, our findings bear potentially important implications. First, morphogens such as Wingless and Hedgehog in , or Sonic-Hedgehog at the mouse node, might be secreted through a similar pathway because their secretion involves particles possibly related to exosomes (; ; ; ). A major objective will be to determine whether CHE-14 and Dispatched act in the aforementioned secretory pathway, and, if so, at which step. Second, several other cell types, such as antigen-presenting cells, reticulocytes, and some epithelial cells, can release exosomes (), which might thus also require the V0 sector for their secretion. In particular, the V0 sector might be directly associated with the transmission of some infectious diseases because viruses, such as HIV and the prion protein, can be disseminated through MVBs and the exosome-releasing machinery (; ). Likewise, the aforementioned secretory pathway could be involved in innate immunity because expression of the Hedgehog-related peptide GRD-3 is induced in upon bacterial infection (). Third, our findings raise the issue of the origin of the MVBs. Interestingly, the apical recycling endosomes have been recognized to play an important role in biosynthetic secretory pathways (). Future studies should reveal whether the secretory MVBs that we described could originate from this compartment. In conclusion, our work shows that trafficking to the apical membrane of at least some lipid-modified proteins involves specific protein complexes (the V-ATPase V0 sector), much as trafficking to the basolateral membrane, and predicts a key role for MVBs in apical exocytosis. Worms were grown at 20°C (unless noted otherwise; ). The identification of and as /-null mutations will be described elsewhere (the gene affected by and was initially named ; see ). Marker and alleles used were as follows: (), (), and / (). The L1 alae phenotype of was scored by allowing adults to lay eggs for 1 h at 15°C, and then transferring embryos to 25°C after egg laying. RNAi was performed using the following bacterial clones from the Wellcome–Medical Research Council library (): , III-5A20; , II-5J16; , IV-4O13; , IV-3I08; and , V-9O06 (). To score L1 larvae, RNAi was induced in L4 larvae; to score adults, RNAi was induced in larvae at the L2–L3 molt. Feeding plasmids were retransformed into fresh HT115 (DE3) bacteria, selecting for tetracycline and ampicillin resistance. Cloning of the coding sequence with a 2.8-kb promoter upstream of the GFP coding sequence in the pPD95.75 vector (Fire kit) generated a rescuing construct. A construct was obtained by replacing the GFP with the monomeric red fluorescent protein (mRFP) coding sequence in the construct. A construct was obtained by cloning the coding sequence and a 3-kb promoter upstream of the YFP coding sequence in the pPD136.64 vector (Fire kit). To generate and constructs, we cloned and genomic DNA with their 5′ and 3′ regulatory sequences into pBSKII-derived plasmids. The GFP coding sequence was inserted in a nonconserved region of the predicted secreted peptide ( and Fig. S3 D). The construct was mutated using the QuikChange Site-Directed Mutagenesis kit (Stratagene). Each desired mutation, and the entire coding sequence of most important plasmids, was verified. Mutant plasmids were microinjected in heterozygous at 3 ng/μl, along with the marker pRF4 at 100 ng/μl, or constructs at 30 ng/μl (when relevant), and pBSKII plasmid at up to 200 ng/μl. Absence of animals in the progeny was used as a criterion for rescue. mRFP versions for the mutations W190A, R191A, W327A, L786S, E830Q, and V844F were obtained from GFP derivatives without PCR amplification and resulted in similar phenotypes. At least two independent extrachromosomal lines were initially examined for each mutation. More detailed analysis was performed on a representative line. VHA-5 polyclonal antibodies were raised in rabbits injected with a purified GST fusion protein containing VHA-5 residues I29–M302, which was obtained by cloning a fragment amplified from the cDNA (a gift from Y. Kohara, National Institute of Genetics, Mishima, Japan) into the vector pGEX-2T. Total worm extracts were solubilized in 8 M urea/2% SDS by sonication, before 8% acrylamide gel electrophoresis and Western blotting. VHA-5 antiserum was used at 1:2,000, the actin monoclonal antibody (act-2D7; Institut de Génétique et de Biologie Moléculaire et Cellulaire collection) at 1:4,000; primary antibodies were revealed with a SuperSignal kit (Pierce Chemical Co.). Immunofluorescence was performed using the VHA-5 antiserum at 1:1,000 dilution and the DPY-7 monoclonal antibody (gift from I. Johnstone, Wellcome Centre for Molecular Parasitology, Glasgow, Scotland) at a 1:50 dilution. Animals were mounted on 4% agarose pads in M9, anaesthetized with 0.2% tricaine/0.02% tetramisole in M9. For DIC imaging, we used a microscope (Axioplan; Carl Zeiss MicroImaging, Inc.) coupled to a camera (CoolSNAP; Roper Scientific) under a 100× objective (PlanApo; Leica). For , we took at least 40 pictures of adult worms per strain and used ImageJ (National Institutes of Health) to measure the distance between the rectum and the grinder. Confocal images were captured on a confocal microscope (SP2-AOBS; Leica), scanning every 122 nm for XZ sections through a 100× objective with a 2.15× zoom (, B and C; and Fig. S3 D) or a 4× zoom ( and ). Images were processed with the Tcstk software () and edited using Photoshop 7.0 (Adobe). Microscopes were in an air-conditioned room (20–21°C). L4 larvae were transferred onto fresh plates for 24 ± 2 h at 20°C before fixation. For TEM, but not for SEM, animals were sectioned and fixed for at least 24 h in 2.5% glutaraldehyde/2% paraformaldehyde/0.1 M sodium cacodylate, pH 7.2, at 4°C, and then postfixed for 5 h with 2% osmium tetroxide in the same buffer at 4°C, dehydrated in graded alcohol/water mixes, and embedded in Epon. Ultrathin 70-nm sections were contrasted with uranyl acetate and lead citrate. Sections were observed with a microscope (CM12; Philips) operating at 60 kV. Quantification of the excretory canal section area was obtained using the Metamorph software after scanning images were captured at a 17,000× magnification. Quantification of MVBs was performed on 3,600×-magnified images. Quantification of the mean area occupied by organelles () was obtained using ImageJ and dividing the total surface of each organelle subtype by the cytoplasmic surface of the hyp7 epidermis section. At least four animals per mutant strain were examined, and more than nine pictures per animal from different ultrathin sections were analyzed. For SEM, animals were postfixed for 1 h with 2% osmium tetroxide at 4°C, dehydrated, and critical point dried in hexamethyldisilazane. Fixed animals were mounted on stubs, coated with palladium, and observed through a microscope (XL20; Philips). At least 20 animals per strain were analyzed. Adult worms were frozen with a high pressure freezing apparatus (EMPACT-2; Leica) in 20% BSA/M9 medium. Cryosubstitution was conducted as in . Ultrathin sections were collected on formvar-carbon–coated copper grids and processed for immunogold labeling. Blocking was performed in PBS/glycine 150 mM, and then in PBS/1% BSA/0.1% Cold Water Fish Skin Gelatin (CWFSG; Aurion) for 30 min. Rabbit anti–VHA-5 at 1:1,000 and rabbit anti-GFP at 1:500 (ab6556; AbCam) were incubated for 1 h in PBS/0.1% CWFSG. 10 nm protein A–coupled gold beads (1:50; University Medical Center, Utrecht, Netherlands) were incubated for 1 h in PBS/0.1% CWFSG. Postfixation was achieved in 2.5% glutaraldehyde, contrasted by uranyl acetate/lead citrate. Images were acquired at 60 kV on a microscope (Morgagni; FEI) with a charge-coupled device camera (Megaview III; Soft Imaging System). Fig. S1 provides a control for , showing that RNAi against was efficient. In addition, it presents the yolk endocytosis defects induced by the loss of V-ATPase activity (yolk proteins are produced by the intestine and are essential for embryonic development); it suggests that RNAi treatment against V0 and V1 subunits was equally effective, and contributes to establish that alae differences described in are meaningful. Fig. S2 summarizes the main phenotypes observed in - animals carrying transgenes with the mutations shown in . Fig. S3 presents the excretory canal, cuticle and MVB phenotypes induced by the mutations R191A, W327A and V844A, which are discussed but not illustrated in the main text, and shows that WRT-8∷GFP accumulates in animals; it should be viewed along with , and . Fig. S4 shows that secretion of the collagen DPY-7 is not affected by or mutations. Fig. S5 provides larger pictures and controls for the immunogold experiments shown in . Online supplemental material is available at .

The tetraspanins are a family of proteins containing four transmembrane domains (TMs) linked by a small, first extracellular domain (EC1) connecting TM1 and TM2, and a large, second extracellular domain (EC2) connecting TM3 and TM4, as well as the usually short NH- and COOH-terminal cytoplasmic tails (; ; ). The tetraspanin proteins are distinguished from other membrane proteins with four TMs (such as sacospan and stargazin) by their highly conserved transmembrane helices and the unique EC2 domain with a Cys-Cys-Gly motif (). The EC2 domain contains four to eight Cys residues forming disulfide bonds (; ) that can stabilize the structure of EC2 domain. In human, the tetraspanin family consists of ∼30 members; ∼20 of them, including CD9, CD37, CD53, CD63, CD81, CD82, and CD151, are expressed in leukocytes (). Tetraspanins can also be found in , , and zebrafish, as well as in plants. Some of the tetraspanins, such as CD9 and CD81, are rather ubiquitous, whereas others, such as uroplakins (UPs) Ia and Ib (in mammalian urothelium) and peripherin/RDS (in retina), are highly tissue restricted. Tetraspanins have been implicated in many cellular functions, including cell proliferation, fusion, development, motility, and tumor growth and invasion. There is as yet no unified view of how tetraspanins perform these diverse functions. Nevertheless, several pictures have emerged from recent studies of tetraspanins. First, tetraspanins tend to associate with other membrane proteins to laterally organize structural or signaling networks, often referred to as tetraspanin webs, on cell surfaces (). Many integrins, such as α2β1, α3β1, α4β1, α5β1, and α6β1, can associate with one or more tetraspanins (; ). Tetraspanins have also been found to associate with the Ig superfamily of proteins, including CD2, CD4, CD8, CD19, MHC I, and MHC II (). The ability of tetraspanins to organize membrane networks has earned them the name “molecular facilitators” (). The tetraspanin networks in cell membranes may actually form a distinct type of tetraspanin-enriched lipid microdomains (; ) that differ from the ordinary raft lipid domains in that they are less resistant to Triton X-100 but more resistant to cholesterol depletion and elevated temperature (; , ). The second emerging picture of tetraspanin function is that these proteins may be involved in transmembrane signaling. For example, tetraspanin CD151 can serve as a transmembrane linker between integrin α3β1 and cytoplasmic phosphatidylinositol 4-kinase, playing a role in regulating cell migration (; ). In addition, CD151 can stabilize α3β1 integrin in its activated conformation; hence, CD151 may be regarded as a “conformation facilitator” (). Third, several tetraspanins can serve as pathogen receptors; e.g., tetraspanin UP Ia as the receptor for uropathogenic (; ; ) and CD81 as the receptor for hepatitis C virus (). Hence, the tetraspanins have a remarkable capability to interact with diverse partner proteins. However, the structural basis of how the tetraspanins bind to their partner proteins and how they form the tetraspanin web has not yet been elucidated. The only currently available, high-resolution crystal structure of any tetraspanin is that of the large extracellular loop (EC2) of CD81; this study indicates that EC2 is a mushroom-like structure with a head domain connected to a two-helix stalk that may anchor to the transmembrane helices (, ). Sequence analyses suggest that all tetraspanin EC2 domains share such a mushroom-like structure with a highly variable region embedded in the head domain (). The EC2 structure of CD81 has been used to model the structure of the EC2 domain of other tetraspanins (; ). Because the crystal structure of the EC2 domain of CD81 contains two EC2 fragments packed against each other to bury a hydrophobic region of the stalk, this dimerization has been assumed to occur naturally and has been used often as a paradigm for predicting tetraspanin–tetraspanin interaction. However, such an EC2–EC2 binary interaction cannot explain the formation of a tetraspanin network. The lack of structural information on the transmembrane helices of tetraspanins and on how these helices are connected to the extracellular loops has severely limited our understanding of the structure-function of tetraspanins and the formation of the tetraspanin network. UP tetraspanin complex is a crystalline tetraspanin web uniquely suitable for structural studies using EM (; ; ; ; ; ; , ). UP tetraspanin web consists of hexagonal arrays of 16-nm UP protein particles. Each of these particles contains two highly homologous (34% identical) tetraspanins, UP Ia and Ib, which pair specifically and stoichiometrically with their single-spanning transmembrane protein partners, UP II and UP IIIa, respectively (; ; ; ; ; ). The naturally occurring crystalline UP web forms rigid-looking plaques (urothelial plaques; also known as asymmetric unit membrane, or AUM) covering almost the entire apical surface of the mammalian urothelium (; ; ). These 2D membrane plaques can be isolated in milligram quantities from mammalian urothelia (; ), and they can provide intermediate resolution structural information of the UP molecules using electron crystallography (). Functionally, the urothelial plaques play an important role in the formation of one of the most efficient permeability barriers known to exist in nature, separating the urine from the cellular contents (). In addition, UP Ia has been shown to be the urothelial receptor for type 1–fimbriated (; ; ), a major causative agent of urinary tract infection (). The tethering of uropathogenic bacteria to the urothelial surface UP Ia receptor, via the lectin adhesin FimH located at the tip of bacterial fimbria, can trigger a transmembrane signaling transduction cascade leading to the urothelial engulfment of the bacteria (). Relatively little is known, however, about the mechanism of this transmembrane signal transduction (). We present here a 3D structure of the 16-nm UP tetraspanin complexes at 6-Å resolution. Our data revealed the secondary structural elements of the UP molecules and enabled us to construct a model of UP tetraspanins, showing that UP tetraspanins interact intimately with their single-spanning transmembrane protein partners through their transmembrane as well as extracellular domains. Our results have implications on the structural basis for the formation of the tetraspanin network in general (). To obtain the 3D structure of the 16-nm UP particle, we recorded, under low-dose conditions, tilted series of images of frozen-hydrated, purified mouse urothelial plaques that have a diameter of up to ∼1 μm (). This enabled us to reconstruct a 3D structure of the mouse UP particle at 6-Å resolution in the membrane plane and 12.5 Å in the vertical direction (, Fig. S1, and Table S1, available at ). The overall shape of the 16-nm particle is very similar to the structure visualized at 10-Å resolution ( [compare with ]). A top view of the 3D map showed that the 16-nm particle has a hexagonal stellate shape with six subunits (); each subunit (, blue outline) in turn consists of an inner (, arrowhead) and an outer subdomain (, arrow). The largest diameter of the particle is ∼17.5 nm. The six inner subdomains encircle an area of ∼6 nm in diameter in the center of the particle. This central region is likely to be occupied by lipids, as it cannot be penetrated by negative stains (; ). The electron density map–enclosed volume, when contoured at the 1-σ level, of one 16-nm particle is ∼6.13 × 10 Å, which can accommodate ∼650 kD of protein, consistent with the estimated total molecular mass (702 kD) based on the sequences of the four major UPs (; ). A side view of the 3D map showed that the 16-nm UP particle has a cylindrical shape of ∼12.5 nm in height (). This cylindrical shape of the particle has about the same diameter throughout its length. The density constriction at the region of the exoplasmic surface of the lipid bilayer, previously seen in a 10 Å–resolution 3D reconstruction, is less evident in the current 3D model (); whether this density constriction reflects a flexibility in this region remains to be investigated. Consistent with our previous results (), the 16-nm particle can be divided vertically into four zones (): from the top down, the joint, the trunk, the transmembrane, and the cytoplasmic. Our 3D reconstruction clearly revealed, for the first time, secondary structural elements of the 16-nm particle. This is particularly obvious for the transmembrane zone of the particle ( and ). The transmembrane zone (, TM; and , TM; ∼4 nm in height) of the 16-nm particle in the 3D reconstruction consists of only rod-shaped electron densities, which likely represent the transmembrane helices of the UP molecules. The TMs of the inner and outer subdomains, with five helices each, have almost the same shape (), consistent with our previous prediction that each subdomain is formed by a heterodimer of UPs consisting of a tetraspanin UP (Ia or Ib) and its single-spanning transmembrane UP partner (II or IIIa; , ). The distances between the centers of the nearest neighbor transmembrane helices in the middle of the transmembrane region range from ∼9.3 to 11 Å. For both the inner and outer subdomains, the electron densities of four of the transmembrane helices (, helices 1–4; see below for the assignment of helix number) are rather closely connected, whereas an additional single transmembrane helix (helix 1') is clearly less connected to the other helices ( and ). Together with our previous localization of UP Ia to the inner subdomain (), this suggests that the four closely packed transmembrane helices belong to the tetraspanin UP Ia (for the inner subdomain) or Ib (outer subdomain), whereas the somewhat detached, single transmembrane helix belongs to the single-spanning transmembrane UP II (inner subdomain) or IIIa (outer subdomain). The four transmembrane helix bundles of the UP tetraspanins are left-handed, although the cross-angles, as defined according to , between these TMs are somewhat variable ( and , top). In this regard, the four transmembrane helices of the UP tetraspanins can be grouped into two pairs, the TM1–TM2 pair and the TM3–TM4 pair (). The cross-angle between the two helices within each pair is ∼10°, which is relatively small, whereas the cross-angle between the two pairs is ∼25°, which is slightly larger than the 20° common cross-angle in transmembrane proteins in general (). The small packing cross-angles between the helices in the two pairs allow the helices within a pair to be closely associated with each other along the entire length (). The electron densities of inner and outer subdomains of the 16-nm particle can be clearly segmented according to the connectivity of the densities; i.e., closely connected densities are considered to be parts of the same molecule, into the tetraspanin region (, UP Ia and Ib [brown and yellow, respectively]) and the single-spanning UP region (, UP II and IIIa [blue and green, respectively]). The single-spanning transmembrane protein region of the inner and outer domains have roughly an inverted L-shape, with the vertical arm of the L interrupted by a region of discontinuity, most likely reflecting a flexible local structure. This inverted L is anchored by its transmembrane helix (, blue and green densities), which is packed against the four transmembrane helices of the partner tetraspanins, forming the five-helix bundle of the transmembrane zone of each subdomain. The long arm of the inverted L continues up against the cylindrical UP tetraspanins and makes a bend at the top to form the short arm of the inverted L. This short arm joins the other short arm of the inverted L from the paired subdomain within the same subunit (, blue and green densities; and ), hence the name “joint” ( and ; ). Interestingly, the joint is the only contact between the two subdomains within a subunit; the two tetraspanins, UP Ia and Ib, do not appear to have direct contacts (). This kind of loose connection between the two subdomains within a subunit suggests a flexible interaction between the inner and outer subdomains, thus providing a basis for possible structural changes of UPs upon binding by bacterial fimbria (; see Discussion). Consistent with this, the volume of the electron density of the joint goes down much faster than the rest of the particle, when contoured with increased contour levels (, bottom), suggesting that the joint is flexible. Visualization of the secondary structural elements of the 16-nm particle enabled us to construct a molecular model of the UP tetraspanins () by following the electron densities, as well as several additional considerations. First, the rod-shaped electron densities of the transmembrane helices of the two UP tetraspanins extend continuously into the extracellular region of the particle (, top; densities of UP Ia and Ib are colored brown and yellow, respectively). Second, The x-ray structure of the EC2 of the homologous tetraspanin CD81 shows that the stem of the EC2 domain of tetraspanins, which are contiguous with TM3 and TM4, are made up of two closely packed helices (). Thus, TM3 and TM4 are likely packed against each other. Third, because TM2 and TM3 are connected by a short, five-amino-acid cytoplasmic loop (; ), they must be packed closely against each other. Fourth, the shape of the EC2 from that of the x-ray structure of CD81 allows only a unique placement of EC2 of the UP tetraspanins into the electron density of the extracellular portion (TK and JT) of a subdomain. These constraints allowed us to assign the transmembrane helices of the UP tetraspanins (). They also allowed us to construct a poly-alanine atomic model of the tetraspanin UP Ia and Ib by building α-helices into the electron densities of the TMs and by placing the models of the extracellular loops based on homologous x-ray structure of the EC2 domain of CD81 (). Our model of tetraspanin UPs has an overall cylindrical shape () with a largely four-helix bundle structure capped at the top by a head domain, which consists of the disulfide-stabilized region of EC2 (). The small extracellular loop EC1 extended from TM1 and TM2 is packed against the hydrophobic part of the large loop (). This model differs from the dimeric crystal structure of the CD81 EC2, where this hydrophobic region of EC2 is packed against that of the other EC2 to form a homodimer (). Our cylindrically shaped model of UP tetraspanins is similar to a just-published theoretical 3D model of CD81 () but very different from a model of CD82 (KAI1) that has skewed (side displaced) extracellular domains (). Interestingly, our electron density shows that the TM1 and TM2 of the UP tetraspanins are packed more tightly at the exoplasmic end than the cytoplasmic end (most visible in the orientation shown by the arrowhead in ). This is consistent with a recent modeling study of another tetraspanin, CD9, by . These authors showed that TM1 and TM2 of CD9, and of tetraspanins in general, have conserved complementary heptad repeats, and their model suggested that TM1 and TM2 pack against each other left-handedly, with the extracellular ends packed more tightly (). Our assignment of the transmembrane helices of the UP tetraspanins is also consistent with the general observation that neighboring transmembrane helices often contact each other and that the antiparallel packing is preferred (). Our 3D reconstruction shows that the UP tetraspanins Ia and Ib have an overall cylindrical structure, formed by the bundled TM helices that extended into the extracellular loops and capped at the top by the disulfide-stabilized regions of EC2 (– ). This cylindrical bundle likely represents a stable protein structure. First, the four-helix bundles in the TM of UP Ia and Ib are closely packed. They have axial distances of ∼9.3 to 11 Å, comparable with the average 9.6-Å axial distance in the closely packed transmembrane helices in other membrane proteins (). They have left-handed cross-angles, which are more closely packed in general than the right-handed ones (), also suggesting stability. The first three transmembrane helices of tetraspanin family of proteins exhibit a heptad repeat of Gly residues (), which may allow the helices to interact closely by favoring van de Waals contacts with surrounding residues and forming backbone CH-amide carbonyl hydrogen bonds (; ; ). The very short, five-residue cytoplasmic loop connecting TM2 and TM3 ensures that these two TMs are closely packed together. Second, tetraspanin proteins, including UP Ia and Ib, have a highly conserved Asn in TM1 and a Glu/Gln in TM3 that may be able to form interhelical hydrogen bonds. These conserved polar/charged residues are located on the faces of the helices containing many conserved residues (unpublished data). It has been proposed that residues mediating interhelical contacts would be more evolutionarily conserved than those that face the lipids because mutations of the former are more likely to destabilize the protein (; ; ). Hence, these conserved polar/charged residues are likely involved in the interhelical interactions in tetraspanin TMs (; , ). In fact, mutation of the conserved Glu residue in TM3 of UP Ib disturbs its transmembrane structure, leading to its ER retention (Kreibich, G., personal communication). Third, the extracellular domains of UP Ia and Ib also form a closely packed structure. In our model, the small extracellular loop covers up the hydrophobic part of the large extracellular loop (), forming an integral part of the cylindrical tetraspanin structure. Although this proposed structure is different from what was suggested by the x-ray structure of the EC2 domain of CD81 (), it is consistent with the finding that the small loop is necessary for optimal surface expression of the large loop and that it may contribute to the stability of the tetraspanin structure (; ). Collectively, these data suggest that the four-helix bundle of the UP tetraspanins, and likely that of the tetraspanins in general, is a stable rod-shaped structure. With the lower half of the bundle embedded in the lipid bilayer, the tetraspanins may serve as protein pilings in the lipid sea, ideal for docking other transmembrane proteins. We have previously identified tetraspanin UP Ia as the urothelial receptor for the type 1–fimbriated bacteria by protein and bacterial overlay assays (; ) and by EM localization (). We have proposed a mechanism by which the urothelial cell senses the bacterial attachment, i.e., through the bacterium binding–induced conformational changes of the TMs of the 16-nm particle (). Our current model of the 16-nm particle, with secondary structural elements, provides some insights into the possible mechanism of this transmembrane signal transduction. Because UP Ia tetraspanin is shorter than UP II and IIIa, for the bacterial adhesin FimH molecule to bind to UP Ia (, arrow; and Fig. S2, available at ), the FimH has to reach into a crevice created by the two joints (made up of UP II and IIIa) of the neighboring subunits of the 16-nm particle. This crevice is ∼3 nm wide, just large enough to accommodate the lectin domain (∼2.5 nm in diameter and 5 nm long) of the FimH molecule (). Therefore, the FimH binding to UP Ia most likely involves additional protein–protein contacts to joints of the 16-nm particle. These broad contacts would be ideal for exerting a force at the joint region of the particle and for inducing a conformational change that can propagate to the TMs. Interestingly, the binding between FimH and the mannose moiety of UP Ia can exert a mechanical force in the range of pico-Newtons (). A mechanical force of this magnitude should be sufficiently strong to be sensed by cytoskeletal elements that interact with the cytoplasmic tails of the UPs (; ). Because the tetraspanin UP Ia is basically a closely packed four-helix bundle, likely to be relatively rigid, it is possible that conformation changes induced by the bacterial binding will lead to changes in the relative orientation (twisting), or sliding, between the tetraspanin UP Ia and its single-spanning transmembrane UP II partner. Interestingly, a “twisted ribbon” model of the 16-nm particle was proposed based only on the negative-stained structure of the extracellular portion of the 16-nm particle (). Some other members of the tetraspanin family are also known to serve as receptors, e.g., CD81 as the receptor for hepatitis C virus and, as has recently been proposed, CD9 as an alternative receptor for interleukin 16 (; ). Additional studies are needed to elucidate the mechanistic details of how these receptors transmit signals. The extracellular part of the tetraspanin UPs extends to ∼5 nm above the lipid bilayer (the height of the trunk domain in ; ). Because this height is about the same as that of an Ig or Ig-like domain, the extracellular domain of tetraspanins is ideally suited for binding to the tetraspanin-associated proteins, many of which contain an Ig or Ig-like (such as Calf-1 and -2 in integrins) extracellular domain immediately above the lipid bilayer (; ). The ability of a tetraspanin to bind its partner through both its extracellular domain and TM helices may play an important role in supporting its tall partners for signaling purpose. For example, CD151 can stabilize its partner, α3β1 integrin, in its activated configuration, as has recently been reported by . One of the key concepts in tetraspanin functions is the formation of a tetraspanin network (or web) by tetraspanins and their partner proteins (; ; ). Studies have been performed to delineate the levels and strengths of interactions in these networks (; ; ). The existing data indicate that there are three levels of interactions: the primary interaction between a tetraspanin and its partner, the secondary interaction between the primary complexes, and the tertiary interaction between these secondary complexes (; ; ). However, most of the existing studies relied on examinations of the tetraspanin complexes based on their detergent resistance, which are relatively nonspecific and insensitive. Our current structural data and molecular model of the UP tetraspanin complexes have revealed in molecular details several levels of interactions in the UP tetraspanin network (). The first level of interaction is clearly the primary interaction between UP tetraspanins Ia and Ib and their single-spanning transmembrane partners UP II and IIIa, respectively (). This interaction is very extensive, involving both TM helices and extracellular domains, and forms the primary complexes—the subdomains in the 16-nm particle ( and ). Supporting this concept are the findings that UP Ia and Ib can be readily cross-linked with UP II and IIIa, respectively, to form heterodimers UP Ia/II and Ib/IIIa () and that the formation of the UP Ia/II and Ib/IIIa heterodimers is a prerequisite for their ER exit (; ; ). There are two types of secondary interactions between the primary complexes in the UP tetraspanin network (). One is between the UP Ia/II and Ib/IIIa primary complexes (i.e., the inner and outer subdomains), via the contact of UP II and IIIa at the joint (, B [left] and C [red bars]), to form a subunit (six of which form a 16-nm particle; and ). The other secondary interaction is between the UP Ia/II complexes (the inner subdomains; , B [right] and C [red arrowheads]), via the contact between UP Ia of one primary complex and UP II of a neighboring primary complex; this secondary interaction is responsible for linking the six inner subdomains to form the inner ring of the 16-nm particle (). Strong detergents such as octyl glucoside can break up these secondary complexes but not the primary complexes (), indicating that the secondary interactions are weaker than the primary ones. Although the first type of secondary interaction (, left) within a subunit can bridge only two primary complexes, the second type of interaction (, right) can potentially bridge more than six primary complexes when involving non-UP tetraspanins. It is conceivable that these two types of secondary interactions also exist in other tetraspanin networks so that multiple primary complexes can link together to form an extensive tetraspanin web. One of the widely reported secondary interactions in tetraspanin network is the formation of homo- or heterodimer of tetraspanins; some of them can be cross-linked using short, or even “zero-length” cross-linking reagents (, ). Although we do not find direct contacts between UP tetraspanins in the structure of the 16-nm particle, the distances between UP Ia and Ib within a subunit or between the neighboring subunits are quite small. Moreover, our data do not exclude the possibility of forming homo- or heterodimer of tetraspanins in other tetraspanin networks because the cylindrical shape of the tetraspanin structure can clearly allow direct contacts between tetraspanins in both transmembrane and extracellular domains. In fact, isolated UP Ia and Ib can readily form SDS-resistant multimers (, ). The third level of interaction is between the outer subdomains of neighboring 16-nm particles (, dotted red lines). This interaction is quite weak and can be visualized only in the electron density map at very low contour levels. This is consistent with the observation that the 16-nm particle as a single unit can be separated from the urothelial plaque with a mild detergent wash (unpublished data). We have therefore observed three levels of interactions in the UP tetraspanin network, and our data provide the structural basis, at the molecular level, of these interactions. Although the details may vary in other tetraspanin networks, our model provides a framework for a better understanding of the formation of tetraspanin networks in general. Mouse urothelial plaques were isolated by sucrose density gradient and differential detergent wash as described previously (; ); the quality of the purified urothelial plaques was assessed by negative staining and EM. For cryo-EM, 5 μl of the purified plaques, adjusted to ∼0.1 mg/ml, was applied to a molybdenum grid (300 meshes) with a layer of newly prepared carbon film and transferred to a tannic acid (0.75%) solution. After the excess liquid was blotted with filter paper, the sample was quickly immersed into liquid nitrogen. The frozen sample was loaded onto a Gatan cryo-holder and transferred to an electron microscope (CM200 FEG; Philips) operated at a voltage of 200 kV. Electron micrographs were taken at a magnification of 50,000 in low-dose mode and with 0.5–1.7 μm defocus, at up to 50° tilt angles. The micrographs were screened using an optical diffractometer to select the regions with the best diffraction spots. Because of the heterogeneity of the sample, <10% of the images provided high-resolution information. Selected micrographs were digitized using a scanner (Carl Zeiss MicroImaging, Inc.) at a step size of 14 μm, which corresponds to 2.8 Å in the crystal. A total of 55 images were selected from >1,000 micrographs, and the unbending of the images was performed using MRC software suites (). The 3D density map was visualized using O () or AMIRA (Mercury Computer System, Inc.) software packages. Segmentation of the density map of the 16-nm particle was performed using AMIRA. For model building of the UP tetraspanins, transmembrane poly-alanine helices were first built into the electron density using O and then the poly-alanine model of EC1 (based on the homology region of casein kinase-1; Protein Data Bank accession no. 2CSN) and EC2 (based on the structure of CD81 large extracellular loop; Protein Data Bank accession no. 1G8Q) were manually docked into the electron density map. The modeled extracellular loops were connected to the TMs following the density map and were locally adjusted using O. Fig. S1 shows calculated diffractions and several lattice lines of 2D crystals of mouse UPs. Fig. S2 is a hypothetical model of the FimH–uroplakin interaction, illustrating that FimH has to reach into the crevice formed by two neighboring joints. Online supplemental material is available at .

Vertebrates express at least 25 chondroitin sulfate proteoglycans (CSPGs), distinguished by the primary sequence of their core protein (). Major groups include the aggrecan family of large, highly glycosylated proteoglycans present in connective tissues, the small leucine-rich proteoglycans, collagen α2IX, basement membrane proteoglycans (e.g., leprecan and perlecan), membrane bound proteoglycans (e.g., CD44, NG2, and phosphacan), and hybrid proteoglycans containing both chondroitin sulfate and heparan sulfate (syndecans). Secreted CSPGs create a hydrated matrix allowing for tissue expansion, and in cartilage they confer the ability to absorb compressive loading. In other tissues, they bind and help organize fibrillar collagens and can affect growth factor signaling pathways. Mutations affecting the core proteins or the enzymes involved in the assembly of the chondroitin chains result in developmental abnormalities in skin, cartilage, bone, tendon, eyes, brain, and the microvasculature (; ; ; ; ; ; ; ). Each chondroitin proteoglycan (CPG) consists of a protein core and one or more covalently attached chondroitin chains. Assembly occurs in a stepwise manner, starting in the endoplasmic reticulum with translation of core proteins on membrane bound ribosomes and transfer of xylose to specific serine residues. Xylosylation exhibits specificity, occurring only at serine residues that have an adjacent glycine to the COOH-terminal side and one or more flanking acidic residues usually within eight amino acids (). After xylosylation, synthesis of a tetrasaccharide primer (-GlcAβ3Galβ3Galβ4Xylβ-O-Ser) takes place in the Golgi, followed by the polymerization of the chain by the alternating addition of -acetylgalactosamine (GalNAc) and glucuronic acid (GlcA; [GlcAβ3GalNAcβ4]; ). In vertebrates, the chains undergo further modification by the addition of sulfate at C4 and C6 of GalNAc and C2 of GlcA residues (). Additionally, an epimerase can convert a subset of GlcA residues to -iduronic acid, which can be subsequently sulfated as well (dermatan sulfate). The pattern of sulfation varies in different tissues during development and in relation to age. Interestingly, very little is known about the proteoglycans present in invertebrates, even in the well-studied nematode . All of the components of the chondroitin biosynthetic machinery are completely conserved between , mice, and humans, including formation of nucleotide sugar precursors and their import into the Golgi, assembly of the linkage region, and polymerization of the chain (; ; ,; ; ,; ). However, lacks the sulfotransferases and the epimerase present in vertebrates, so the chains consist of unmodified GalNAc and GlcA residues (; ). In spite of its simplicity, genetic experiments demonstrate a crucial role for chondroitin in embryonic cell division and vulval morphogenesis (; ; ,; ,; ; ). Although all of the enzymes required for chondroitin synthesis have been identified in , no protein cores that bear the chondroitin chains have been described. To identify CPGs in , we conducted searches using BLAST (Basic Local Alignment Search Tool) with all known mammalian CSPG core protein sequences but did not identify any obvious homologues. We therefore pursued a proteomics-based approach taking advantage of the chemical properties of the long, negatively charged chondroitin chains, a tagging method to modify the serine attachment sites, and mass spectrometry. We report the discovery of nine novel CPG proteins, none of which show homology to vertebrate CSPGs. Simultaneous depletion of two of these proteins, CPG-1/CEJ-1 and CPG-2, by RNAi results in defective cytokinesis during the first embryonic cell division. This phenotype is identical to that observed when the chondroitin synthase was silenced by RNAi or by a loss-of- function mutation (; ; ; ), suggesting that these two major chondroitin-carrying proteoglycans are essential for embryonic development. Proteoglycans were solubilized from a mixed-stage worm population by a combination of sonication and extraction with guanidine hydrochloride. After dialyzing the crude extract into urea, the proteoglycans were partially purified by stepwise elution from an anion-exchange column, using 0.2 M NaCl with urea to remove contaminating proteins and 1 M NaCl without urea to elute the proteoglycans (see Materials and methods). Treatment of the sample with sodium hydroxide released the chains from their core proteins by β-elimination reaction. The liberated chondroitin chains were generally <20 kD based on their elution by gel filtration chromatography relative to shark cartilage chondroitin sulfate (; ), but they exhibited considerable heterogeneity. Treatment of the sample with chondroitinase ABC converted all of the material into nonsulfated disaccharides that eluted near the total volume of the column, confirming that the preparation contains predominantly chondroitin and very little heparan sulfate (; ). To estimate the number of CPGs, the crude proteoglycans were digested with chondroitinase ABC, separated by SDS-PAGE, and analyzed by Western blotting with the 1B5 mAb that recognizes a neoepitope generated by chondroitinase digestion (). Several CPG core proteins ranging in mass from 10 to >200 kD were detected in whole worm (lanes 2 and 3) and embryo extracts (lane 4), whereas none were detected if chondroitinase ABC was omitted (lane 1). The major species had masses of ∼118, ∼110, and ∼60 kD. Treatment of samples with PNGase F had no effect on the mobility of these bands, suggesting that none of the proteins contained Asn-linked glycans (unpublished data). also synthesizes at least three heparan sulfate–bearing proteoglycans with approximate masses of 45, 28, and 22 kD based on digestion of extracts with heparinase followed by Western blotting with 3G10 mAb, which is specific for a heparan sulfate neoepitope generated by enzyme digestion (). In vertebrates, some proteoglycans contain both chondroitin sulfate and heparan sulfate chains (e.g., syndecans 1 and 3). Simultaneous digestion of samples with chondroitinase ABC and heparin lyase II did not alter the intensity of any of the CPG bands or give rise to new ones reactive with mAb 1B5 or 3G10 (unpublished data). Thus, hybrid proteoglycans might not exist in . To identify the CPG core proteins detected by Western blotting, we searched the genome for sequences related to any of the 25 known mammalian CSPG core proteins (). No obvious homologues were found, with the exception of ORF Y47D3A.26, which resembled bamacan/SMC3 (). However, the coding sequence for this gene, as well as vertebrate bamacan, lacks a signal peptide present in all known proteoglycans, consistent with the observation that most of bamacan is found in the nucleus in a nonglycosylated form (). Thus, a new method was needed to identify the various CPGs in . describes the scheme used to identify CPGs. After the DEAE chromatography step, samples were reduced, alkylated, and digested with trypsin. Peptides containing chondroitin chains (glycopeptides) were recovered by a second round of anion-exchange chromatography, which removed much of the contaminating protein from the sample. In some experiments, we used gel filtration to separate glycopeptides of large hydrodynamic size from contaminating peptides rich in acidic amino acids, which coeluted under the low salt conditions used to prepare the samples. Next, the samples were treated with sodium hydroxide, which caused β-elimination of the chains and the formation of dehydroalanine from serine residues. The dehydroalanine residues were then reacted with DTT, which adds across the double bond (). In some experiments, the DTT-tagged peptides were further purified with a thiol column. Samples were then analyzed by mass spectrometry using multidimensional protein identification technology (MUDPIT) to identify peptides containing DTT (+167 D) or dehydroalanine (−18 D; ; ). The corresponding full-length proteins were identified by searching a proteome database, parsing the data using the following criteria: putative CPGs had to contain at least one glycosaminoglycan attachment site, defined as a modified serine residue flanked on the COOH-terminal side by glycine with one or more aspartate or glutamate residues near the site of chondroitin addition (), and the proteins had to have a hydrophobic signal peptide, which would direct the protein into the secretory pathway, where chondroitin synthesis occurs. Nine tentative CPG core proteins were identified by this procedure and designated CPG-1 through -9 (). CPG-1 (C07G2.1a) and CPG-2 (B0280.5) have predicted masses of 62 and 54 kD, respectively (). CPG-1 was identified previously as CEJ-1 based on its reactivity with an antibody to a mammalian tight junction protein (Siddiqui, S.S., personal communication). It contains five putative chondroitin attachment sites based on amino acid sequence. One of these sites was confirmed by mass spectrometry based on presence of the DTT tag. In contrast, CPG-2 has 34 potential chondroitin attachment sites (four confirmed by mass spectrometry; ). CPG-1 and -2 also have in common multiple peritrophin-A chitin binding motifs, defined by the arrangement of six cysteine residues that can form three disulfide bridges in a characteristic pattern (; ; ; ). In CPG-1, two peritrophin-A domains lie in the NH-terminal half of the protein and one toward the COOH-terminal end, whereas in CPG-2 the six peritrophin-A domains lie between sites predicted to be glycosylated (). Additionally, the COOH-terminal half of CPG-1 contains many Thr, Val, and Pro residues, resembling a segment of Muc-2, a membrane mucin present in vertebrates. CPG-3 (R06C7.4) was not originally identified by the DTT-tagging method or by the presence of dehydroalanine () but rather by 5′-RACE using PCR primers designed against the repetitive SG region of . Reevaluation of the mass spectral data revealed that peptides from CPG-3 were present in nearly all datasets (). Based on primary sequence, CPG-3 contains 15 putative chondroitin attachment sites and shows 21% sequence identity to CPG-2, mainly in the putative glycosylated region. The lack of mass spectral data for peptides that contain putative attachment sites in CPG-3 may have been caused by the failure of trypsin to gain access to and cleave the protein in heavily glycosylated regions. CPG-4 (C10F3.1) has the largest predicted mass of the CPGs at 84 kD. The majority of its 35 predicted glycosaminoglycan attachment sites reside in the COOH-terminal half of the protein, of which four sites were confirmed by DTT modification (). CPG-5 (C25A1.8, CLEC-87) and CPG-6 (K10B2.3a, CLEC-88) are highly homologous, showing 67% identity and 90% similarity. CPG-5 and -6 contain only one putative chondroitin site toward the NH terminus, which was identified in both proteins by DTT modification. Both proteins contain a C-type lectin domain in the COOH-terminal half of the protein, like the aggrecan family of vertebrate CSPGs (). This motif therefore classifies and as members of the (C-type lectin) gene class. Proteins containing C-type lectins are common in (88 members have been found thus far), but their glycan binding specificity has not been determined. CPG-7 (K09E4.6) and CPG-8 (K03B4.7a) have predicted masses of only ∼12 kD (). CPG-7 contains 11 putative chondroitin attachment sites, whereas CPG-8 contains 6 sites. In both cases, five of the sites were confirmed by mass spectrometry. Of the multiple peptides identified in CPG-8 samples, some contained a single modified site at either residue 61 or 63 (the tandem mass spectrometry data did not discriminate which site actually carried the chain). Interestingly, other peptides were modified at both serine residues. A similar phenomenon was observed at residues 84 and 88. CPG-9 (Y67D8C.8) has a predicted mass of only 7 kD, and mass spectrometry confirmed two of four attachment sites. Interestingly, none of the CPGs identified using the scheme in showed homology to vertebrate CSPGs. Furthermore, no homologues were found in searches of the genome or EST and genomic databases, suggesting that the core proteins have evolved independently in different organisms. In contrast, homologues for each CPG were present in , another free-living nematode. The glycosylation sites of many CPGs appeared to follow an “SGEX” motif, where X was usually G, A, T, or S (). BLAST searches using this sequence motif identified CPG-2, -3, -4, and -7 but no other proteins in or vertebrates. contains one protein containing this motif (CG6048-PA), but it has not yet been shown to contain a chondroitin chain. The aggrecan family of CSPGs in vertebrates also contains repetitive SG sequences within domains densely substituted with chondroitin chains, but the intervening sequences differ from those found in CPGs. To confirm that the CPGs identified by mass spectrometry could serve as a scaffold for chondroitin synthesis, the cDNA sequences were subcloned into a mammalian expression vector with a COOH-terminal Myc tag and expressed in COS-7 cells. Conditioned media was collected 48 h after transfection, purified by anion-exchange chromatography, treated with chondroitinase ABC, and Western blotted with an antibody to Myc (). CPG1-6 yielded reactive protein bands that did not appear in the absence of chondroitinase ABC digestion. Attempts to express CPG7-9 in COS cells have thus far been unsuccessful, possibly because of the small size of the protein core. None of the proteoglycans migrated at the molecular mass predicted by their primary sequence. Furthermore, a comparison of the bands in to the calculated masses in did not demonstrate obvious correspondence. However, the migration pattern of CPG-2 expressed in COS cells appears similar to the 120-kD core in the crude proteoglycan preparation (). Likewise, CPG-3 migrates at 55–60 kD, the same position as a prevalent band in the worm extract. These observations suggest the presence of higher order structures in the core proteins or that the chondroitin stub oligosaccharides remaining after chondroitinase digestion altered their migration. Mammalian CSPGs also tend to migrate aberrantly, often at twice their predicted size (). Expression of , , , , and is enriched in the germline, suggesting possible activity in germline development and progression, spermatogenesis, oogenesis, or embryogenesis (; ). However, genome-wide RNAi screens have not reported phenotypes for any of the genes, suggesting either functional redundancy or subtle phenotypes (). reported that GLD-1 acts as a translational repressor of several mRNA targets in the germline, including and B0280.5, which were identified in our proteomic screen as and , respectively. They also reported that RNAi of each gene alone had no phenotype, whereas simultaneous depletion of both genes resulted in embryonic lethality. Because this phenotype is reminiscent of the embryonic phenotype resulting from inhibiting the SQV-5 chondroitin synthase, we decided to analyze CPG-1 and -2 in greater detail. To confirm that CPG-1 and -2 contain chondroitin chains in vivo, worms were fed bacteria expressing double-stranded RNA (dsRNA) directed against or , and samples were analyzed by Western blotting. No differences were initially observed in CPG-1–depleted extracts (), but CPG-2–depleted extracts showed a dramatic reduction of the major band at ∼120 kD, as well as the bands at ∼80 and ∼45 kD (, arrowheads), suggesting that all three were related to each other (the predicted mass of the protein was 54 kD; ). The simultaneous reduction of these three bands was confirmed by using a second construct targeting a different region of (unpublished data). The RNAi effect was specific, as the other major core protein band at ∼60 kD remained unaltered. CPG-1 and -2 have in common multiple peritrophin-A chitin binding domains (). To test the functionality of these domains, crude worm extract was incubated with chitin beads. Four CPG bands were able to bind chitin, including those at 120, ∼80, and 45 kD (, filled arrowheads), as well as a new band at ∼150 kD (, open arrowhead). Chitin binding appeared to be selective, as the major band at 60 kD did not interact with the resin. RNAi treatment diminished the three bands at 120, ∼80, and 45 kD, confirming the presence and functionality of one or more peritrophin-A chitin binding domains in these proteoglycans (, filled arrowheads). The band migrating at ∼150 kD was present in vector and CPG-2–depleted extracts but absent in extracts depleted of CPG-1 and CPG-1/CPG-2 (, open arrowhead). Recombinant CPG-1 expressed in COS-7 cells had a similar mass (), suggesting the 150-kD band represents CPG-1. Thus, both CPG-1 and -2 behave as CPGs in vivo and have functional chitin binding domains. To study the function of CPG-1 and -2, we compared the effect of silencing their expression by RNAi to depletion of SQV-5, which encodes the chondroitin synthase. Brood sizes of , , and c animals were comparable to uninjected worms or worms injected with buffer alone (; P > 0.05 by one-way analysis of variance). Depletion of CPG-1 or -2 alone also had no effect on viability, as 96–99% of the embryos hatched into healthy larvae that grew into normal fertile adults (). However, simultaneous depletion of CPG-1 and -2 had a synergistic effect, resulting in penetrant embryonic lethality. Whether this synergy is due to the functional redundancy of CPG-1 and -2, the partially penetrant nature of RNAi, other functions of the proteoglycans besides carrying chondroitin chain, or the cross-reactivity of dsRNA with other targets remains to be determined. treatment also had no effect on egg production or laying but had a hatching phenotype similar to c (95% of progeny failed to hatch). To look for genetic interactions between and or a heterozygous mutation was compounded with single c or . However, no combination affected brood sizes or embryonic viability (unpublished data), possibly because the heterozygous mutation did not reduce chondroitin levels sufficiently to see the effect of depleting a single CPG. To determine the underlying cause for embryonic lethality, we looked more closely at the first cell division, where defects have been reported in chondroitin synthase mutants (). To better visualize the plasma membrane, a strain expressing GFP fused to a pleckstrin homology (PH) domain derived from mammalian phospholipase-C was injected with dsRNA against , , , or . The PLC1δ1 PH domain binds to phosphatidylinositol 4,5 bisphosphate located specifically on the plasma membrane and is an excellent probe for monitoring the changes in cell shape that accompany cytokinesis (). The depleted embryos were simultaneously imaged by differential interference contrast (DIC; , top) and spinning-disc confocal microscopy (, bottom). Embryos were imaged in utero because depletion of either SQV-5 or both CPG-1 and -2 resulted in embryos that were fragile and osmotically sensitive. Depletion of CPG-1 or -2 alone had no effect on fertilization, membrane ruffling, pseudocleavage, pronuclear meeting and rotation, karyokinesis, and cytokinesis (compare Video 2 to buffer-injected worms in Video 1, available at ). Initiation of the cleavage furrow occurred normally (, bottom), and two separate daughter cells routinely appeared (). A four-celled embryo was distinctly seen at the end of filming, demonstrating the fidelity of the second round of cell division (). In contrast, embryonic development was severely perturbed in embryos simultaneously depleted of CPG-1 and -2 (Video 3). In wild-type embryos, fertilization triggers the oocyte pronucleus to undergo two rounds of meiotic segregation, which produces two polar bodies that are extruded from the embryo by small cytokinesis-like events. In CPG-1/CPG-2–depleted embryos, fertilization occurred normally but polar body extrusion failed, resulting in extra nuclear material that remained in the embryo (). The pronuclei migrated and the complex of the pronuclei and centrosomes rotated onto the long axis of the cell as in wild type, but the membrane ruffling that normally precedes and accompanies these events was absent, as was the space between the embryonic plasma membrane and the eggshell that is normally present during cell division (compare , C [bracket] and J [solid white lines]). Spindle formation appeared normal and the chromosomes appeared to segregate, but the cleavage furrow failed to form and ingress to introduce a new cell surface between the segregated chromosomes (). Without the barrier of a new cell membrane, the daughter nuclei fused () and repeatedly attempted cell division without cytokinesis (). An essentially identical phenotype was observed after depletion of SQV-5 (; and Video 4). Since depleting CPG-1 and -2 mimics the defect in embryonic cytokinesis by depletion of the SQV-5 chondroitin synthase, we conclude that CPG-1 and ∼2 are two functionally important CPGs required at this early developmental stage. We report here the identification of a group of novel CPGs in . The identification scheme used a combination of conventional methods to purify proteoglycans and glycopeptides based on the polyanionic character and size of the chondroitin chains coupled with mass spectrometry. Additional selection criteria ensured that the majority of identified proteins had the properties of a proteoglycan. These included chondroitin attachment sites modified with DTT, the presence of a serine residue followed by glycine and one or more nearby acidic residues, and a signal peptide. Although DTT addition was not stoichiometric, the β-elimination step introduced an alternate method to identify the protein via the characteristic mass difference between dehydroalanine and serine (). Attempts to optimize the DTT-addition reaction have not yet yielded higher efficiency, and other methods are needed to improve this step to better assess the stoichiometry of chondroitin attachment. Similarly, adjustment of the proteolysis conditions might improve peptide coverage, as the absence of trypsin cleavage sites and the potential inhibitory effects of closely spaced chondroitin chains could interfere with the analysis. Regardless of its limitations, the current protocol led to the identification of nine proteins that had the hallmarks of CPGs, and six served as substrates for chondroitin sulfate assembly when expressed in mammalian cells. The simplicity of the identification strategy suggests that it should be generally applicable to other organisms, as well as tissues, cells, and secretions from vertebrates for which little information about the proteoglycan composition exists. Interestingly, none of the CPGs identified in this study had predicted membrane spanning domains or consensus sites for glycosphosphatidylinositol-anchor attachment. expresses homologues of the membrane heparan sulfate proteoglycan syndecan, which has a single membrane spanning segment, and the glycosphosphatidylinositol-anchored heparan sulfate proteoglycan, glypican. Because vertebrates express multiple membrane CSPGs (e.g., NG2, CD44, and phosphacan; ), our failure to identify membrane CPGs in could indicate either their absence or that they represent only minor components of the proteoglycan population. Heparan sulfate proteoglycans also were not detected, most likely because expresses ∼250 times less heparan sulfate than chondroitin (; ). Thus, it may be necessary to modify the procedure to selectively enrich membrane proteins and increase sample size to detect rare proteoglycans. The lack of homology between CPGs and vertebrate CSPGs and the observation that none of the CPGs have homologues in two other invertebrates ( and ) raises interesting questions about the evolution of these molecules. All of the core proteins identified in this study have sequence motifs required for initiation of glycosaminoglycan biosynthesis similar to those found in vertebrates (). Furthermore, expresses orthologues of all of the vertebrate enzymes required for assembly of the linkage region tetrasaccharide (xylosyltransferase, galactosyltransferases I and II, and glucuronyltransferase I; ; ), as well as the chondroitin polymerizing system (; ), including the newly discovered chondroitin polymerizing factor (). These findings indicate that the mechanism of chain initiation and polymerization evolved early in metazoans and has been maintained by strong selection. In contrast, the core proteins on which the chains assemble have continued to evolve, presumably to serve specialized functions. For example, CPG-1 and -2 have functional chitin binding domains that could interact with the nematode eggshell, whereas no vertebrate proteoglycans contain this motif and vertebrates lack chitin as a structural component. The presence of chitin in insect exoskeletons and peritrophic matrices that line the gut suggests that orthologues of these CPGs might exist in (), but BLAST searches have not yet uncovered them. Additional studies using techniques like the one reported here are needed to identify the proteoglycans present in and other organisms, which in turn might help define the evolution of proteoglycans during the metazoan expansion. Our preliminary analysis of RNAi-depletion experiments indicated that silencing or had no effect on embryogenesis or morphogenesis. However, depletion of both genes resulted in a strong embryonic phenotype characterized by failure of polar body extrusion after fertilization, loss of membrane ruffling preceding pronuclear fusion, and failure to initiate the cleavage furrow before cytokinesis. The necessity of depleting both CPG-1 and -2 to uncover a phenotype suggests functional redundancy of the proteins, though the only commonality is the presence of chitin binding domains and at least one chondroitin chain (). An alternative explanation is that depletion of each protein by RNAi may not have been fully penetrant, and the phenotype after double depletion could be the result of more extensive depletion of chondroitin. However, this explanation seems unlikely because CPG-1 is a minor component and had to be enriched by chitin affinity chromatography to detect its presence. Instead, the requirement for simultaneous depletion of both proteoglycans could be due to other functional aspects of the CPGs (e.g., their ability to bind chitin or the presence of other glycans). Generation of deletion mutants will help resolve this issue. The strong embryonic lethal phenotype seen with resembles that of the chondroitin synthase mutant described by . How the CPGs mediate these effects is unknown. One of the roles of CPGs may be to fill the space between the eggshell and the embryo, and the presence of a high concentration of polyanions and their counterions might cause sufficient hydrostatic pressure to aid in formation of the extraembryonic space and subsequent ingress of the cleavage furrow. Thus, the loss of CPGs in the perivitelline space surrounding the embryo could result in collapse of the extracellular space and apposition of the plasma membrane to the eggshell (). Interestingly, chondroitin has been shown to be present on both the embryonic cell surface and the eggshell, suggesting that it may function at both locations (). This leads to an alternative explanation: CPG-1 and -2 could act as structural elements of the eggshell or might bridge chitin polymers in the eggshell with other components of the embryonic plasma membrane that result in transmembrane signaling to cytoskeletal components involved in cytokinesis. Regardless of which hypothesis is correct, we have shown that CPG-1 and -2 play a crucial role in cytokinesis, as their simultaneous depletion results in multinucleated single cell embryos. The observation that the regulatory protein GLD-1 may translationally regulate and transcripts indicates a sophisticated level of control over proteoglycan expression during early embryogenesis. GLD-1 is localized to the germline cytoplasm () and translationally regulates the expression of a large number of genes by interacting with mRNA 3′UTR sequences (). It also protects some transcripts from nonsense-mediated mRNA decay by binding to 5′UTR sequences (). One example is , which encodes the enzyme glucosamine 6-phosphate -acetyltransferase. This enzyme plays an essential role in the biosynthesis of the nucleotide sugar UDP--acetylglucosamine (UDP-GlcNAc). The biosynthesis of a variety of glycans depends on UDP-GlcNAc, including chitin (a polymer of GlcNAcβ1,4) and chondroitin by way of UDP-GalNAc. Interestingly, a null allele of exhibits maternal effect lethality and multinucleated embryos (Johnston, W., and Dennis, J., personal communication). The similar phenotype of , , and mutants may indicate a common mechanism of action through formation of CPGs. Posttranscriptional regulation of CPG assembly in via GLD-1 is reminiscent of the regulation of enzymes involved in heparan sulfate formation in vertebrates () and may indicate the general importance of translational control over proteoglycan assembly during development. The OD58 strain (provided by A. Audhya, University of California, San Diego, La Jolla, CA) carries a PH membrane domain of phospholipase-C fused in-frame to a GFP reporter (δ) under the control of the promoter (). All other strains were obtained from the Caenorhabditis Genetics Center and cultivated as described previously (). Worm vectors were provided by A. Fire (Stanford University School of Medicine, Stanford, CA). BLAST searches were performed in the National Center for Biotechnology Information database () or WormBase () using annotated human or mouse CSPG core protein sequences. The presence of signal peptides was confirmed with PSORT II () and SignalP 3.0 Server (). Worm extracts were prepared from 20-g batches of a mixed-stage N2 (Bristol) population by sonication in 50 mM sodium acetate buffer, pH 6. Material was extracted for 48 h at 4°C in three volumes of solution containing 4 M guanidine-HCl, 0.1 M NaCl, 0.3% CHAPS, 50 mM sodium acetate, pH 6, and protease inhibitors (1 μg/ml leupeptin, 1 mM PMSF, and 1 μg/ml pepstatin A; ). The extract was dialyzed three times against 40 volumes of 6 M urea, 0.1 M NaCl, and 50 mM sodium acetate, pH 6. Insoluble material was removed by low-speed centrifugation and filtration through filter paper (No. 1; Whatman). The concentration of protein was assayed by the Bradford method (Bio-Rad Laboratories). Proteoglycans were purified from worm extract (250 mg protein) by anion-exchange chromatography (0.2 M – 1 M NaCl, DEAE-Sephacel) and desalted by gel filtration (). Partially purified material was treated with trypsin (), and glycopeptides were further purified by an additional pass over DEAE-Sephacel and in some experiments by gel filtration HPLC (TSK-2000 column [Tosoh Bioscience]; 1 M NaCl in 10 mM KHPO, pH 6). Fractions containing uronic acid () were pooled and desalted by gel filtration. Glycosaminoglycan chains were removed by β-elimination followed by Michael addition with DTT (BEMAD; ). In brief, glycopeptides were incubated for 3 h at 50°C in 20% ethanol, 1% triethylamine, 10 mM DTT, and 0.1% NaOH. The reaction was quenched by adjusting the sample to 1% trifluoroacetic acid, and the peptides were purified by reverse-phase chromatography on a C18 Sep-Pak cartridge (Waters Corporation) by elution with 70% acetonitrile in 0.1% trifluoroacetic acid. Samples were dried before further analysis. In some experiments, the samples were passed over a thiol-Sepharose column to enrich for DTT-modified peptides (). Tagged peptides were identified by MUDPIT (; ; ; ). Peptides were eluted stepwise from a biphasic capillary column made of strong cation-exchange resin coupled to a reverse-phase resin directly into a tandem mass spectrometer. The identities of the DTT-tagged peptides were determined by searching the tandem mass spectra against a proteome database using SEQUEST software and a computer array. The unique mass signature imparted by DTT (+167 D) and unmodified dehydroalanine residues (−18 D) were used to determine the sites of chondroitin addition. cDNA was prepared from total RNA with the SuperScript III First-Strand kit (Invitrogen). through were amplified from cDNA with primers that included the start codon and the penultimate codon and restriction sites to subclone the products in-frame into the pcDNA3.1(-)MycHis B vector (Invitrogen). Expression constructs were transfected into COS-7 cells with Lipofectamine (Invitrogen) following the manufacturer's instructions. Media was harvested 48 h later and purified over DEAE-Sephacel as described previously (). COS-7 cells were cultured in Dulbecco's modified Eagle's medium (CellGro) supplemented with 10% fetal bovine serum, 100 U/ml penicillin G, and 100 μg/ml streptomycin sulfate. cDNA was amplified by PCR with primers engineered to contain T7 (forward primer) or T3 (reverse primer) bacterial promoter sequence and base pairs 1–649 for and 1–775, 754–1572, and 1190–1572 for . dsRNA was generated with the Megascript T7 and T3 transcription kits (Ambion) according to the manufacturer's instructions. OD58 L4 or young adult worms were injected with dsRNA and allowed to recover at 16°C or 20°C. To count brood sizes, worms were transferred 24 h after injection to individual plates, and 1 d later the number of eggs and hatched L1 larvae were counted (). Viability was measured 24–36 h later by counting the number of hatched larvae and unhatched embryos. Percentage of viability was calculated as number of hatched progeny divided by the total number of eggs laid. Early embryonic cell division was assessed 24 h after injection. Embryos were filmed in utero because and embryos were fragile and osmotically sensitive. Injected animals were anesthetized with 1 mM levamisole in M9 buffer, mounted on an agarose pad, and filmed as described previously (). Images were acquired on a DeltaVision deconvolution microscope (Applied Precision) equipped with a charge-coupled device camera (CoolSnap; Roper Scientific) at 20°C. Images were acquired using a 100×, 1.3 NA U-Planapo objective (Olympus) with a 2 × 2 binning and a 480- × 480-pixel area. DIC and GFP images were acquired at 10-s intervals by sequentially rotating the analyzer and GFP filter set into the light path. Illumination was attenuated with a 10% neutral density filter. Images were analyzed with MetaMorph software (Universal Imaging Corp.). To determine the effect of RNAi on proteoglycan expression, cDNA sequences for and described earlier in this section were subcloned into vector pL4440, which carries dual T7 promoter sites (Fire Lab Vector kit) and drives the formation of dsRNA. The expression vector was transformed into the HT115 bacterial strain, which was then fed to L1 larval worms as described previously (). After 48–60 h, protein extracts were prepared by sonication of either whole worms or embryos and subjected to SDS-PAGE and Western blotting as described in the next section. Proteoglycan samples (10 μL) were digested with 10 mU chondroitinase ABC (Seikagaku) and/or 2.5 mU heparin lyase II (Sigma-Aldrich) for 3–5 h at 37°C. After reduction with β-mercaptoethanol and alkylation with iodoacetamide, samples were analyzed by SDS-PAGE. CPG core proteins were detected with 1B5 chondroitin stub mAb (1:1,000; Seikagaku) followed by goat anti–mouse secondary antibody (1:2,000; Bio-Rad Laboratories). Recombinant proteins expressed in animal cells were Western blotted with a murine anti-Myc mAb (1:5,000; Invitrogen). Blots were visualized with the WestPico Chemiluminescent kit (Pierce Chemical Co.). The videos show the first two rounds of cell division in embryos exposed to the RNAi treatments described in . Video 1 shows normal cell division in buffer-injected embryos. Video 2 shows normal cell division in embryos. Video 3 shows that embryos fail to complete the first cell division. Video 4 shows that produces the same phenotypic defects as . Online supplemental material is available at .

is an extracellular protozoan parasite responsible for a reemerging tropical disease known as sleeping sickness in humans. There are two main proliferative forms of the parasite: the bloodstream form in the mammalian host and the midgut insect stage or procyclic form in the tsetse vector. Changes in the variant surface glycoprotein (VSG) type on the surface allow the bloodstream form of the parasite to elude the host immune antibody response, ensuring a persistent infection (; ; ). The monoallelically expressed gene is always located at the end of a telomeric expression site (ES). Previous estimations suggest the presence of 20 different telomeric ESs that share highly homologous promoter sequences. The ES promoter, which is located 40–60 kb upstream of the telomere, drives the polycistronic transcription of developmentally regulated genes named ES-associated genes (for review see ). In the bloodstream form, only one ES is fully transcribed at a given time so that each cell displays a single VSG type on the surface. Transcriptional switching among ESs results in antigenic variation. In the procyclic form, VSG is not expressed, but an invariant family of glycoproteins called procyclins are constitutively expressed and replace VSG on the parasite surface (). Previous data suggest two distinct mechanisms for ES regulation: a developmental silencing of the ES in the procyclic form and a coupled mechanism for ES activation/inactivation in the bloodstream form (). In eukaryotic cells, RNA polymerase I (pol I) transcribes ribosomal loci (ribosomal DNA [rDNA]) and is highly compartmentalized in the nucleolus (for review see ). Interestingly, in , pol I also transcribes and . Previously, we have proposed a model whereby the recruitment of a single ES to a discrete pol I–containing extranucleolar body (ES body [ESB]) defines the mechanism responsible for monoallelic expression (; for review see ). In this study, we investigate the nuclear localization of pol I–transcribed chromosomal sites in the context of pol I machinery and transcription activity. Our results show that the nonmutually exclusive gene family is transcribed at the nucleolus periphery in contrast to the monoallelically expressed ES, which is associated with the extranucleolar ESB. Furthermore, we address the possible repositioning of bloodstream pol I–transcribed loci during differentiation to the insect procyclic form. We found that upon developmental silencing, the active ES promoter is subjected to nuclear envelope repositioning concomitant with ESB disassembling and is followed by chromatin condensation. Nuclear positioning dynamics of developmentally regulated chromatin domains is involved in coordinating transcriptional activation and repression. For a precise positional analysis of a particular sequence in nuclei, we have adapted the in vivo GFP tagging of chromosomes () to bloodstream and procyclic trypanosomes. By expressing GFP-I in a tetracycline-inducible system (), we are able to localize a particular DNA sequence in the nucleus, as visualized by GFP-I binding to operator sequences inserted in a chromosome site in vivo and in fixed cells, thereby exploiting the advantages of this tool (for review see ). immunofluorescence (IF) analysis has been considerably improved by adapting 3D deconvolution wide-field fluorescence microscopy () to the study of nuclear architecture in this paper. Researchers have reported that heterologous genes transcribed from the locus generate mRNAs that are localized either to the nucleolus (; ) or to the nucleoplasm (), as assessed by RNA-FISH. However, the nuclear position of the chromosomal loci has not been investigated. In this study, we address the nuclear position of the chromosomal locus, which is transcribed by pol I () and is developmentally regulated (). For this purpose, the operator repeats were chromosomally inserted upstream of a promoter within the (GPEET-PAG3) locus. First, to avoid possible fixing artifacts, the position of the procyclin locus was determined in vivo. After DAPI staining of DNA in the nucleus of live cells, the position of the nucleolus was indirectly determined by the absence of DAPI staining, and localization of the GFP-I bound to the locus was visualized upon GFP-I induction. A fluorescent GFP dot was clearly visible upon induction, and its localization was determined to be at the periphery of the nucleolus (). Live cell 3D microscopy confirmed that the locus was confined to the border of the nucleolus (Video 1, available at ). Second, to more precisely determine the position of this sequence in the nucleus with respect to the nucleolus, we performed IF analysis in PFA-fixed cells. The localization of GFP-I was detected using an anti-GFP monoclonal antibody, and pol I was stained using affinity-purified anti–pol I large subunit (anti-TbRPA1) antiserum (). Analysis of deconvolved 3D datasets indicated that the GFP-I–tagged locus is associated with the nucleolus (). Interestingly, this position was observed in 97.8% of GFP dot–positive cells (), suggesting a highly constrained chromosomal position, which was confirmed by time-lapse fluorescence imaging in living cells (Video 2, available at ). Because the gene family is transcribed at a similar level for all allelic variants (), there seems to be no need to associate them with a single extranucleolar body, which is in contrast to the model of the monoallelic expression of ES in the bloodstream form (). Although in vitro differentiation from the bloodstream to the procyclic form is efficient in , the converse is not feasible. Thus, we are unable to address possible nuclear localization changes for the locus in the procyclic form () upon differentiation to the bloodstream stage. The same lac operator construct that was used to tag the locus in procyclics was repeatedly used in bloodstream parasites with no success, suggesting that the procyclin promoter is developmentally down-regulated in the bloodstream form by a chromatin-mediated mechanism, as suggested previously (). IF analysis showed that pol I was exclusively localized to the nucleolus, and no substantial extranucleolar signal was detected (), ruling out the possibility of a specific pol I–containing body responsible for expression (). Interestingly, pol I was found to be subcompartmentalized in the nucleolus, with distinct foci peripherally distributed in a U-shaped pattern that was easily detectable by 3D microscopy (). To investigate this unexpected pol I distribution, we performed BrUTP labeling of nascent RNA in situ () in PFA-fixed procyclic cells to determine the sites of pol I transcription. To exclusively detect pol I transcriptional activity in the nucleus of permeabilized cells, experiments were performed in the presence of high concentrations of α-amanitin (100 μg/ml), which is known to inhibit pol II and III transcription. Indeed, although many transcriptional foci were distributed along the nucleus in the absence of the drug (), in the presence of α-amanitin, nascent RNA was solely detected in the nucleolus (). Furthermore, within the nucleolus, BrUTP-labeled RNA was confined to distinct foci located predominantly in a peripheral position similar to that of the GFP-I–tagged locus (). To further investigate pol I–dependent transcriptional activity, we determined the position of the rDNA in the procyclic form. Several independent clones were analyzed, and all revealed a perinucleolar position for the GFP-I–tagged rDNA chromosomal site ( and Video 3, available at ). Again, the position of the GFP-I bound to the rDNA locus associated with the position of pol I and showed a stable perinucleolar position (98.9% of GFP dot–positive cells) when examined by 3D microscopy (). The peripheral nucleolus location of and rDNA loci, together with pol I transcription foci along the nucleolus periphery () instead of an inner central position, may explain the lack of colocalization of these two loci that were described previously using RNA-FISH (). To determine whether the peripheral distribution of pol I–transcribed loci in the nucleolus is a unique feature of the insect form of the parasite or is also present in the bloodstream form, we addressed the position of the rDNA locus. We performed 3D IF of bloodstream-form cells upon PFA fixation in suspension (), which preserves nuclear structure better than the previously used fixation conditions (). The position of rDNA locus in the bloodstream form localized to the nucleolus and, similar to the procyclic form, was peripheral with an equivalent constrained position (98% of GFP dot–positive cells; ). To investigate a possible nuclear position–dependent regulation of pol I–transcribed chromosomal sites in the bloodstream developmental form, we first analyzed the position of the active ES promoter. Double IF using an anti–pol I antibody and an anti-GFP antibody showed that the active ES tagged with GFP-I localizes to the ESB as previously described (), whereas pol I was present in the ESB as well as in the nucleolus (). We also addressed the nuclear position of the internal chromosomal basic copy (BC) tandem genes. These copies of different genes serve as substrates for recombination events into the active ES telomere, resulting in an antigenic switch. The GFP-I–tagged 121 BC locus showed no association with the ESB, which is similar to an inactive 121 ES promoter region (). Importantly, statistical IF position analysis of both the BC and inactive ES promoter sequences revealed no considerable association to the nuclear envelope (2% of GFP dot–positive cells). We show that in bloodstream form, the telomeric silencing of ES proposed previously () is not associated to either nuclear periphery repositioning or chromatin condensation. We next determined whether the active ES undergoes nuclear repositioning upon developmental differentiation from the bloodstream to the procyclic form, where no VSGs are expressed. For this purpose, the differentiation of bloodstream- to procyclic-form parasites was induced in vitro, and nuclear localization changes were analyzed early (5 h) or late (24 h) during differentiation. To assess the differentiation process, we monitored the developmental expression of the surface glycoprotein procyclin by double IF using anti–EP procyclin and anti-VSG221 antibodies. 22% of the cells displayed procyclin on the surface 5 h upon in vitro differentiation. This value increased 24 h upon differentiation, with 83% of cells exclusively displaying procyclin on the cell surface and 5% displaying a mixed coat of procyclin and VSG. The remaining 10% of cells that solely displayed VSG on the surface can be interpreted as differentiation retarded or defective in the asynchronous differentiation process that occurs in this monomorphic cell line. 3D IF analysis showed that the active ES promoter relocated to the nuclear envelope early during differentiation (5 h; ). Importantly, at the same time, extranucleolar pol I (ESB) was no longer detected, which is consistent with our observation that pol I exclusively localizes to the nucleolus in the established procyclic form (). Statistical analysis of the position indicated that 70% of the nuclei display the GFP-I dot at the nuclear periphery 5 h upon differentiation (). The relocation of the active ES promoter to the nuclear periphery in 70% of the cells was higher than the number of procyclin-positive cells (22%), suggesting that ES nuclear reposition silencing is preceding the full surface expression of Finally, 24 h upon in vitro differentiation, the GFP-tagged active ES promoter was located to the nuclear periphery in 88% of the cells () displaying on their surface. To determine whether such rapid developmental repositioning was a unique feature of the active ES promoter, we determined the localization of various other chromosomal sequences. For example, the rDNA locus showed no change in nuclear localization either 5 () or 24 h upon differentiation and was always detected in a perinucleolar location (100% of GFP dot–positive cells; ). Similarly, statistical analysis on the location of the GFP-I–tagged 121 BC and inactive ES promoter chromosome sites showed no significant nuclear envelope repositioning upon early differentiation (). Together, our data indicate that the active ES promoter sequences reposition to the nuclear periphery concomitantly with the ES transcription silencing during differentiation to the insect form (). Importantly, rapid nuclear repositioning of the promoter detected at 5 h after differentiation induction precedes the full down-regulation of transcription given that mRNA is still clearly detectable at 12 h after differentiation (). This is the case despite that mRNAs are down-regulated by the 3′-untranslated region in the procyclic form (). This mechanism seems to be specific for the active ES promoter, as such rapid repositioning was not observed for the inactive 121 ES promoter or 121 BC loci at early differentiation stages (). Interestingly, although 83% of GFP-positive nuclei tagged at the active ES promoter showed a clear GFP-I dot in an exponentially growing bloodstream culture, upon 24 h of differentiation and nuclear repositioning, only 8% of the GFP-positive nuclei showed a detectable GFP-I dot (). In contrast, detection of the GFP-I bound to rDNA was evident in 98% of the GFP-positive nuclei even 24 h upon differentiation. Cell lines tagged either at the inactive 121 ES promoter region or in the 121 BC region showed an intermediate situation, with 53–76% of the cells displaying a visible GFP dot 24 h upon differentiation (). Similar data were also obtained by in vivo GFP fluorescence direct visualization. In late differentiation (24 h), cells showed a GFP-I dot for the rDNA locus that was easily detectable. In contrast, 24 h upon differentiation, when the active ES was tagged, the GFP-I dot was almost undetectable even though the cells displayed diffuse GFP expression in their nuclei (). These differential results suggest that GFP-I binding to the lac operator sequences inserted into distinct chromosomal positions reflect differences in chromatin accessibility and, thus, allow us to detect changes in chromatin condensation. These data are supported by the previously described ES chromatin remodeling of the bloodstream ES after differentiation to the procyclic form to yield a structure that is no longer permissive for T7RNAP transcription in vivo (; ). Recently, detected an opposing chromatin decondensation event upon gene activation utilizing the accessibility of GFP-I. In this context, changes in chromatin seem to dramatically affect the accessibility of GFP-I to the lac operators inserted in the active ES promoter region, as indicated by the drastic decrease in the number of nuclei with a detectable GFP dot (). Although chromatin in the rDNA locus is not affected at all upon differentiation, a moderate degree of chromatin condensation was also found for the 121 BC and inactive 121 ES promoter regions even though these loci are not transcribed in the bloodstream form. Moreover, an eventual repositioning of inactive ES promoter to the nuclear envelope does occur, as tagging the inactive ES promoter regions in established procyclic form revealed that these chromosomal loci localized to the nuclear envelope in 41.5% of nuclei (). The active ES promoter repositioning in 88% of cells at early stages of the differentiation process is in contrast with the 41.5% of nuclei detected for the promoter locus in established procyclics ( and ). Thus, our results show that nuclear repositioning targets more efficiently at early stages during the differentiation process and suggest that the establishment of silencing requires a transient perinuclear localization. Despite many correlations between nuclear localization and gene activity, it remains unclear whether nuclear repositioning is the cause or the result of such activity. Like yeast (), TbKU80-deficient trypanosomes are unable to halt ES developmental silencing () or the silencing of all ESs but one in the bloodstream form (), but no information on possible nuclear repositioning is available for this mutant. Although we cannot conclude that nuclear repositioning causes silencing, importantly, our data provide new insights into this problem. First, the active ES promoter, located 60 kb upstream of the telomere, is the sole target for nuclear envelope relocation during differentiation, which is in contrast to inactive ES promoters. Second, this rapid repositioning precedes chromatin condensation during differentiation (). Nuclear envelope repositioning and chromatin condensation events have been suggested to affect pol II promoter activities in yeast and mammalian cells (). Our data represent the first example of a pol I–transcribed chromatin domain targeted by a nuclear position–dependent silencing mechanism, indicating that such regulation is not restricted to pol II and that nuclear architecture plays a universal role in the epigenetic regulation of transcription. bloodstream-form (Molteno Institute Trypanozoon antigenic type 1.2 [MITat 1.2]; clone 221a) and 427 procyclic-form DNA transfections and selection procedures were described previously (). For these studies, the bloodstream cell lines were differentiated in vitro to procyclics using standard conditions but with SDM-79 medium (). IF was performed on cells in suspension () except that fixation was performed for 2.5 h on ice with 4% PFA and permeabilized with 1% NP-40 for 1 h at room temperature. IF was performed in 1% blocking reagent (Roche) in PBS (Sigma-Aldrich) using the monoclonal anti-GFP (Invitrogen) and affinity-purified anti–pol I (TbRPA1) rabbit antiserum (1:600; ). AlexaFluor488- or -594–conjugated goat species–specific antibodies (Invitrogen) were used as secondary antibodies, and cells were DAPI stained and mounted as described previously (). Stacks (0.1-μm z step) acquisition was performed with a microscope system (Cell R IX81; Olympus), 63×/100× objectives, illumination system (MT20; Olympus), and camera (Orca CCD; Hamamatsu). Deconvolution of 3D images was performed using Huygens Essential software (version 2.9; Scientific Volume Imaging) using an experimentally calculated point-spread function with 0.2-μm TetraSpeck microspheres (Invitrogen). All images displayed in the figures are maximum intensity projections from digitally deconvolved multichannel 3D image datasets. Pseudocoloring and maximum intensity projections were performed using ImageJ software (version 1.37; National Institutes of Health). Nascent RNA labeling in permeabilized procyclic form was essentially performed as described previously for the bloodstream form () except that cells were fixed with 2% PFA for 20 min. The single-slice deconvolution shown in was performed using Huygens software with 0.3 μm as a z sample size. We have adapted the in vivo GFP tagging of chromosomes (; ) to bloodstream and procyclic trypanosomes. GFP-I was expressed in a tetracycline-inducible manner (). We localized a particular DNA sequence in the nucleus by detection of the GFP-I bound to the operator sequences inserted in a chromosome. Stable transformants in occur by homologous recombination, allowing us to insert a operator tagging cassette by a single crossover. The bloodstream single marker (SM) cell line () and the procyclic cell line 1313-1333 () were used for tetracycline-inducible expression. To express the GFP-I fusion in a tetracycline-dependent manner, we used pMig75, which was described previously (), in procyclic and bloodstream forms (SM-75). These two cell lines were used to obtain all transformants with the operator tagging constructs (described in the next paragraph) in the absence of tetracycline induction. To GFP-I tag any locus of interest, we developed a series of constructs containing variable target DNA upstream of a 256– operator–containing fragment () and downstream of the promoter of the locus under study, which will drive expression of the selectable marker. To GFP tag the () locus, the targeting sequence located 60 bp upstream of the endogenous promoter was a PCR fragment generated using oligonucleotides (5′-CGAGCTCATACCGCTGCCGGCCTAAATGC-3′ and 5′-CAAGCTTCATTTTGCACAAAATGCACTATTG-3′). To drive the expression of the hygromycin selectable marker, we used a promoter obtained by PCR using oligonucleotides (5′-GTGGATCCTCCATTTTGTGGCAGTGATGG-3′ and 5′-CGCCATGGAAAGGGAACGAGGTGCCATTG-3′). To tag the rDNA spacer located between two rDNA repeats, the targeting sequence was a PCR fragment (5′-AATTCGAGCTCATATAGTTGG-3′ and 5′-CGCGAAGCTTCGGTGTGTTGCCAAAGACATTC-3′) using pLew82 as a template (). To drive a bleomycin selectable marker, we used a ribosomal promoter obtained by PCR using oligonucleotides (5′-CGAGGGATCCACCCAGCGCGGGTGCATTC-3′ and 5′-GGCATATGCAGTCCTGCTCCTCGGCC-3′). The 121VSG BC target sequences that we used were the full 121VSG cDNA and the ES promoter described previously (). The constructs to tag the active 221 ES and the inactive 121 ES were previously described (). All constructs were inserted upstream of the promoter of the locus under study, resulting in tandem repeats of similar promoters. However, in the case of the 121 BC targeting construct, we included an ES promoter to drive the selectable marker that is not present in the endogenous locus. GFP-I expression was induced in early exponential cultures with 1–0.1 μg/ml doxycycline (Sigma-Aldrich) for bloodstream and procyclics, for 16 h. Simultaneously with differentiation induction, the expression of GFP-I was induced with 0.1 μg/ml doxycycline. GFP-I expression in SM-75 and 1313-75 cell lines displayed a proportion of nuclei that did not express the GFP-I fusion after induction even without the operator repeats. Thus, this variable expression was not caused by a toxic effect but rather by variegated activity of the promoter driving the expression of GFP-I. Thus, all statistical analyses in both developmental stages described in this paper are based on GFP-I–positive nuclei cells recognized by the unbound GFP- that was detected in a dispersed manner in the nucleoplasm. Statistical analysis of GFP-I–expressing nuclei that were positive or negative for the GFP dot was performed in 100–120 interphase nuclei. The positive ones were grouped in different categories based on the GFP dot nuclear position within the DAPI staining (nuclear periphery, nucleolus, and nucleoplasma) and the relative position between the GFP dot and pol I signals. The scoring was performed by direct optical observation. Questionable cells were analyzed by 2D or 3D digital imaging. At least 20 representative cells were analyzed by 3D deconvolution microscopy. Tagged chromosome position and GFP dot detection probability distributions were compared between categories indicated with an asterisk in using chi-square analysis. Statistical significance was determined by using a 95% confidence interval. Video 1 shows a live cell in which the procyclin chromosomal site is tagged with GFP-I. Video 2 shows in vivo visualization of the highly transcribed procyclin chromosomal site tagged with GFP in a procyclic-form trypanosome. Video 3 shows 3D deconvolved slice animation through the whole fixed nucleus. Online supplemental material is available at .

Mitochondrial DNA (mtDNA) forms nucleoprotein complexes (). In yeast, several candidate mitochondrial nucleoid proteins have been identified by in organello formaldehyde cross-linking experiments (). Several of them associate closely with mtDNA and contribute to its stability (; ; ). Less is known about mammalian mitochondrial nucleoids; they contain Tfam, which is believed to be the major mtDNA packaging protein (), and Twinkle, an mtDNA helicase (), mitochondrial single-strand binding protein, and DNA polymerase γ (). Additional proteins copurify with frog oocyte mtDNA (), although their roles in mtDNA maintenance are uncertain. In mammals, many molecules of mtDNA contain a short triple-stranded region, or displacement loop (D-loop; ; ), located in the major noncoding region (NCR). The third strand of the D-loop, 7S DNA, is ∼0.65 kb long in humans, spanning from approximately nt 16,111 to nt 191 (). D-loops are synthesized via transcription initiating at the light strand promoter and transition to DNA synthesis at the origin of heavy strand replication (). They have been proposed to represent stalled or aborted replication intermediates (). Hitherto, there has been no evidence that mitochondrial D-loops are functional entities. Tfam/Abf2 are members of the HMG family of DNA binding proteins, which bend DNA. Subunit α of bacterial HU is a histone-like protein, which is capable of binding to a variety of nucleic acid substrates () and of complementing Abf2-deficient yeast (). Because HU is simpler than its eukaryotic counterparts and more readily expressed in , we used it as a bait to affinity purify mammalian mtDNA with its associated proteins; this strategy led to the isolation of the protein TOB3 (ATAD3p), found but not characterized previously in a proteomic screen of rat liver mitochondria () and, more recently, in enriched mitochondrial nucleoprotein preparations (). Here, we show that ATAD3p is a DNA binding protein, which is present in mitochondrial nucleoids; siRNA of ATAD3 decreased the number of mtDNA multimers, and in vitro part of the protein binds preferentially to D-loop–containing molecules, suggesting that mitochondrial D-loops play a role in mtDNA organization. Preparations of HU affinity-purified rat liver mtDNA contained six identifiable proteins (Table S1, available at ): only Tfam was a known mitochondrial nucleoid protein. One of the six, ATAD3p, has an AAA domain (), located toward its C terminus, and a conserved hydrophobic region of 20 amino acids (residues 246–265 of human ATAD3B); however, it lacks a canonical N-terminal mitochondrial targeting signal. The sequence of ATAD3 is 85% identical between rat and humans, and homologues are spread throughout animals, plants, and protists (Fig. S1). The hydrophobic region is the appropriate length to span a lipid bilayer, but DAS-TMFilter, Phobius, and Minnou programs do not predict it will form a transmembrane helix; thus, in silico analysis suggested that ATAD3p was not an integral membrane protein. Nevertheless, the protein is tightly associated with mitochondrial membranes based on alkaline carbonate treatment of mitochondria (Fig. S2 A). To determine the cellular location and properties of ATAD3, recombinant human protein was produced. Human ATAD3B consists of 648 amino acids with a calculated molecular mass of 72 kD. The protein expressed poorly in (unpublished data), but two fragments, ATAD3-f1 (residues 44–247) and ATAD3-f2 (residues 264–617), were expressed readily as soluble GST fusion proteins (unpublished data). Antibodies raised against ATAD3-f1 recognized two proteins in human 143B osteosarcoma cells but only one protein (ATAD3A) in A549 adenocarcinoma cells (Fig. S2 B). Immunocytochemistry with antibody to ATAD3-f1 revealed a punctate staining pattern within mitochondria (Fig. S2 C), which frequently coincided with mtDNA (). Notwithstanding, many mitochondrial nucleoids appeared to lack ATAD3p (). Thus, the amount of ATAD3p associating with mtDNA appears to vary from nucleoid to nucleoid. Two rounds of transfection of 143B osteosarcoma cells with double-stranded RNA (dsRNA)–452 targeted to ATAD3 decreased PicoGreen staining of mitochondrial nucleoids markedly (). However, the copy number of mtDNA after ATAD3 siRNA was ∼88% of control values (), suggesting that PicoGreen staining does not provide a direct measure of mtDNA mass, an inference confirmed by immunofluorescent detection of DNA (). Therefore, PicoGreen staining of DNA must depend on the DNA's topological state. Relaxation of supercoiled plasmid DNA in vitro was accompanied by a substantial increase in PicoGreen signal (unpublished data), substantiating this view. Therefore, we conclude that ATAD3p depletion leads to an increase in negative supercoiling and that the more condensed form of mtDNA largely excludes PicoGreen. Mitochondria had an essentially normal morphology in ATAD3 siRNA–treated 143B cells ( and not depicted), implying that ATAD3p has no role in mitochondrial fission or fusion; hence, the observed change in mtDNA associated with gene silencing is not an indirect consequence of mitochondrial disorganization. 2D agarose gel electrophoresis (AGE; ) has been used to characterize mitochondrial replication intermediates (; ). Here, the bulk of the protein was removed by treatment with detergent and phenol without the usual proteinase K digestion. Under these conditions, an AccI restriction fragment (nt 15,255–1,504) of human mtDNA did not enter the gel readily compared with protease-treated samples, and there were numerous prominent spots on the linear duplex arc, indicating that protein remained associated with mtDNA (). In addition, residual protein gave rise to a series of spots resolving well above the linear duplex arc. The position of the first of these (, species 4) was coincident with the apex of an X arc (; ); that is, its mobility implied it comprised two fragments of DNA joined near the center ( [interpreted in ]). The mass of the other spots (species 6, 7, and 8) is consistent with higher order multimers, and the increasing distance from the linear duplex arc implies that these are also joined near their center (as interpreted in ). If ATAD3p is a bona fide component of mitochondrial nucleoids, its depletion via RNAi might alter the mobility of protein-bound mtDNA fragments. Therefore, total cellular DNA was extracted, without a protease step, from control cells and ATAD3 siRNA–transfected cells, and restriction digested, and the fragments were separated by 2D-AGE. After Southern blotting, nylon membranes were hybridized sequentially to probes for three regions of mtDNA covering most of the human mitochondrial genome: nt 15,255–1,504 (a); nt 1,505–6,286 (b); and nt 8,157–15,254 (c). In ATAD3 siRNA samples, there was a significant decrease (P = 0.00005) in signal from mtDNA trapped near the well, and at the boundary of the 1D and 2D gels, specifically for the NCR-containing fragment (a; versus and Fig. S2 D). The iterated spots corresponding to the apex of a simple X arc, and higher multimers thereof, were also significantly decreased (P < 0.0065), whereas the protein-bound fragments of mtDNA resolving on or close to the linear duplex arc were substantially the same as controls (), and no significant alteration was observed in other regions of mtDNA in controls and ATAD3 siRNA–treated samples ( and not depicted). Therefore, ATAD3p binds preferentially within the region defined by the AccI fragment spanning nt 15,255–1,504 of human mtDNA, and ATAD3 is implicated in the maintenance or formation of mtDNA multimers. Although the 2D-AGE data () suggested that ATAD3p binds to multiple molecules of mtDNA, other factors presumably contribute to mitochondrial nucleoid stability in vivo, as there was no apparent nucleoid fragmentation in ATAD3 siRNA–treated cells stained with anti-DNA antibody (). Based on the analysis of AccI fragments of human mtDNA described above, ATAD3p might bind at any number of sites from nt 15,255 to nt 1,504. However, the fact that the depleted species were X-like structures led us to focus on the central portion of the fragment, the NCR. The NCR encompasses the mitochondrial D-loop, a triple-stranded region of ∼600 nucleotides. Therefore, the human NCR was cloned in Bluescript and synthetic D-loops were produced to investigate whether the two recombinant fragments of ATAD3 had specific DNA binding properties. First, however, the f1 and f2 fragments of human ATAD3 were incubated with random pieces of single-stranded and duplex DNA and analyzed by electrophoretic mobility shift assay (EMSA) to determine their nonspecific DNA binding properties. Only ATAD3-f1 was capable of binding to duplex DNA, whereas both fragments bound to single-stranded oligodeoxynucleotides ( and Fig. S2 E). In all cases, the vast majority of linear DNA remained unbound; therefore, neither duplex nor single-stranded DNA is a good substrate for ATAD3. The first synthetic D-loop (C) tested used an oligonucleotide (oligo C) spanning 120 nucleotides, from nt 16,081 to nt 16,200, near the 3′ end of the native D-loop. D-loop C was incubated with RecA protein, or ATAD3-f1 and -f2, and subjected to 1D- AGE. ATAD3-f1 and RecA bound to D-loop C to a much greater extent than ATAD3-f2 (); the pronounced mobility shift effected by hATAD3-f1 compared with RecA suggests either that the former accrues multiple DNA molecules or that each D-loop is bound by a large number of ATAD3p molecules. In competition experiments, all molecules of D-loop C associated with ATAD3-f1 when mixed with a >1,000-fold excess of pUC19, plasmid DNA (). Plasmid including the NCR sequence (pNCR) was a more effective competitor, yet there remained a preference for D-loop structures over supercoiled plasmid without a D-loop, as ∼80% of D-loops bound to protein in the presence of a 100-fold excess of pNCR (). The increased competition from pNCR, relative to pUC19, suggested that there might be an element of sequence- specific binding to the action of ATAD3-f1. However, a second synthetic D-loop (A) corresponding to the other end of the D-loop (nt 72–191) displayed the same competitive advantage in binding ATAD3-f1 as D-loop C (). The sequences of oligonucleotides A and C used to generate the two D-loops are not alike (); therefore, ATAD3-f1 does not display sequence-specific binding. D-loop C occasionally formed two bands in the absence of protein, particularly when lower amounts of RecA were used to generate the synthetic D-loop (, lane 6); both species had greater mobility than open circular DNA (). Differences in supercoiling are more readily apparent when DNA molecules are separated on chloroquine gels (); therefore, the same products were separated on a chloroquine gel. As expected, this amplified the difference in mobility between forms a and b, consistent with band a being a less tightly supercoiled version of synthetic D-loop C than band b (). ATAD3p showed a marked preference for the most highly supercoiled D-loop–containing molecules (, lanes 7 and 8); this feature indicates that D-loop context is crucial to ATAD3p recruitment or retention, which may be necessary to avoid the protein interfering with the processes of replication and transcription. The highest concentrations of protein produced only a modest shift in some of the more relaxed molecules (, lanes 9 and 10). Therefore, tightly supercoiled DNA with a D-loop is the strongly preferred substrate for ATAD3p, and such DNA molecules permit oligomerization of either the protein or DNA, or both. In summary, ATAD3p frequently colocalizes with mtDNA, and supercoiled DNA with a D-loop is the preferred substrate of a recombinant fragment of ATAD3p. The pronounced preference of ATAD3p for triple-stranded DNA predicts it will associate with a specific class of mtDNA molecules, those containing a D-loop. Hence, the key prediction of this report is that D-loop (7S DNA) synthesis occurs to recruit ATAD3p to mtDNA via the f1 portion of the protein. Much remains to be elucidated about the function of ATAD3. Several AAA family members are involved in DNA transactions, including the bacterial nucleoid protein FtsK, clamp loaders, Cdc6, and components of the origin recognition complex. The AAA domain of ATAD3 might well confer on the protein the ability to translocate DNA as elsewhere (). The ATPase of ATAD3 is functional (Fig. S2 F), and ATP but not ADP enables ATAD3-f2 to bind single-stranded DNA, although ATP hydrolysis is not required for binding, as the protein also binds to single-stranded DNA in the presence of PCP (Fig. S2 E). Presumably, the binding of ATP to ATAD3-f2 induces a conformational change that enables the protein to bind to single-stranded DNA. ATAD3p is tightly associated with mitochondrial membranes (Fig. S2 A) and, therefore, is likely to contribute to the association of mtDNA with membranes; yet, the in silico prediction is that ATAD3p lacks transmembrane helices, so other proteins are probably required to tether mtDNA to the inner membrane. The amount of ATAD3p associated with mitochondrial nucleoids is highly variable, and some nucleoids appear to lack the protein entirely (), which suggests that ATAD3p associates only transiently with mtDNA or with a distinct subpopulation of mtDNAs. The limited effect of ATAD3 gene silencing on mtDNA copy number implies it is not required for replication. Therefore, we favor a role for ATAD3p in nucleoid formation or segregation. Rat liver mitochondria prepared as described previously () were treated with 0.2 mg/ml RNase A and DNase I for 2 h at 4°C, washed, and sedimented by centrifugation at 8,000 for 10 min. 40 mg of mitochondria were disrupted by suspension in 4 ml lysis buffer (10 mM Hepes-NaOH, pH 7.6, 0.2 mM PMSF, 1 mM EDTA, 1 mM DTT, and 0.8% n-Dodecyl-β--maltopyranoside) and centrifuged at 1,000 for 10 min, and the supernatant was incubated with 4 ml recombinant-HU–coated beads in lysis buffer supplemented with 200 mM NaCl. After washing the beads with lysis buffer containing 100 mM NaCl and elution with 80 mM glutathione, the eluate was separated on a 20–45% iodixanol gradient, the fractions enriched in mtDNA were pooled, and the affiliated proteins were analyzed by mass spectrometry. Fragments of ATAD3(B) corresponding to amino acids 44–247 (ATAD3-f1) and 264–617 (ATAD3-f2) were amplified from a full-length cDNA (IMAGE Clone ID 3138578; Mammalian Genome Collection) and fused with a GST gene. Expressed protein was purified by sequential chromotography (glutathione Sepharose, HiTrap SP FF/HiTrap Q FF, and Superose 12 gel-filtration [GE Healthcare]). GST-HU protein was expressed similarly, except that induction was at 37°C, and purification was via Ni-Sepharose HP followed by SP Sepharose FF (GE Healthcare). Human cells were grown in DME with 10% fetal bovine serum. For RNAi, 143B human osteosarcoma cells growing on 6-well plates at 25–30% confluency were transfected with 10 nM dsRNA and 3 μl of Lipofectamine 2000 (Invitrogen). Cells were transfected a second time, at 72 h, and examined by confocal microscopy at 144 h or lysed for total RNA or DNA extraction. In preliminary tests, dsRNA-452 (5′-UCAAUGAGGAGAAUUUACGGAAGCAAG-3′; 5′-UAAGUUACUCCUCUUAAAUGCCUUCGU-3′) reduced mRNA levels by 65% after 72 h (based on qPCR analysis). Twinkle and Tim17A siRNA and qPCR were performed as described previously (). 143B cells were washed and live stained with 3 μl PicoGreen reagent (Invitrogen) and 100 nM of MitoTracker orange (Invitrogen) as described previously (; ) or fixed and stained with anti-DNA antibody (PROGEN Biotechnik) and 1:5,000 anti–ATAD3-f1. A confocal microscopy system (Radiance 2000; Bio-Rad Laboratories) was used for cell imaging, and images were edited using Photoshop Element (Adobe). 2D-AGE of AccI-digested human DNA was done as described previously (). The mtDNA copy number was estimated by qPCR, as described previously (). Synthetic D-loops were generated by incubating end-labeled oligo C nt 191–72 or A nt 16,081–16,200 with pNCR as described previously (). pNCR comprised Bluescript plasmid (Stratagene) and a 1.2-kb fragment of human mtDNA encompassing the NCR (nt 16,024–576) of human mtDNA. Purified proteins were incubated with labeled DNA substrates, the products were separated on 1% AGE or 3.5% native PAGE, and phosphorimages were produced using a Typhoon detector (GE Healthcare). The sequences of the oligonucleotides used in the EMSAs are listed below, except for oligos A and C, which were based on the revised human mtDNA sequence (). Oligo24 top, 5′-GATCTTGTACACGGCCGACTAGTG-3′; Oligo24 bottom, 5′-CATGTGCCGGCTGATCACCTAG-3′; Oligo50 top, 5′-ATCCGGAATCTCCACGCAAACGGCGCCTCATTCTTCTTCATCTGTATCTTC-3′; and Oligo50 bottom, 5′-GAAGATACAGATGAAGAAGAATGAGGCGCCGTTTGCGTGGAGATTCCGGAT-3′. Table S1 lists the six proteins identified by mass spectrometry analysis that consistently copurified with mtDNA. Fig. S1 shows that ATAD3 is an evolutionary conserved member of the AAA family. Fig. S2 shows that ATAD3p is a mitochondrial membrane–bound ATPase with two single-stranded DNA binding domains, which stabilizes extracted mtDNA multimers. Online supplemental material is available at .

The microtubule cytoskeleton plays roles in determining cell shape, cell polarity, vesicle trafficking, and cell division. Consequently, microtubule reorganization during differentiation is believed to be essential for morphogenesis. However, in most tissues and differentiated cells, little is known about the organization of microtubules or the mechanisms involved in microtubule reorganization. The epidermis is a stratified squamous epithelium that acts as a barrier between the internal and external environments. The innermost (basal) layer maintains contact with its underlying basement membrane, separating epidermis from dermis. As basal cells move outward, they enter a program of terminal differentiation that involves the assembly of robust intercellular desmosomal junctions and culminates in the formation of flattened dead cells that are sloughed from the skin surface. During terminal differentiation, both the intermediate filament (IF) and the actin cytoskeleton undergo dynamic reorganization in their associations with desmosomes and adherens junctions, respectively (). What happens to the microtubule cytoskeleton in the course of terminal differentiation remains unexplored. We report the first example of a transgenic mouse with fluorescently labeled microtubules. Using these mice, we have uncovered a differentiation-specific reorganization of the microtubule cytoskeleton that, surprisingly, depends on the desmosomal linker protein desmoplakin (DP). Further implicating DP in microtubule organization is the observation that DP is required for the relocalization of the microtubule-anchoring protein ninein from the centrosome to cell junctions. To visualize microtubules in living epidermis and cultured keratinocytes, we engineered transgenic mice expressing the microtubule-binding domain of ensconsin fused to three copies of GFP (EMTB-3GFP). This fusion protein has been previously well characterized and has a fast on-rate for microtubules, but does not significantly alter microtubule dynamics (; ). To drive its expression, we used the epidermal-specific keratin 14 (K14) promoter (; ). Mice expressing were viable and fertile, with no observable detrimental phenotype. Whole-mount epifluorescence microscopy revealed transgene expression throughout the epidermis and in the dots corresponding to developing hair follicles (). At embryonic day (e) 14.5, the microtubule network of basal cells was significantly concentrated in the apical domain, and it extended downward toward the base of these cells (sagittal views in , and Video 1, available at ). Confocal imaging from the surface of living embryos revealed that the microtubule-organizing center (MTOC) of each basal cell localized to its apical domain (). Microtubules were apically organized within the basal epidermal layer whether the epidermis was well stratified (e18.5) or not (e14.5; [inset]). This organization contrasted markedly from that of stratified cells, where microtubule networks were concentrated at cell–cell borders (). Because the K14 promoter is most active in basal cells, these distinctions were best visualized at different exposures (, inset). Another striking difference between basal and suprabasal microtubule networks was the lack of an obvious MTOC in stratified cells. This is most apparent by confocal imaging through the suprabasal cells (Video 2, available at ). Collectively, these data suggested that as epidermal cells differentiate, they undergo a dynamic rearrangement of microtubules, which includes disappearance of the centrosomal array (). The organization of microtubules in the epidermis differed significantly from that reported for simple epithelia, where microtubules exhibit a perpendicular (apical–basal) orientation with no obvious MTOC (). To determine whether the differentiation-associated changes in microtubule organization can be recapitulated in vitro, primary mouse epidermal keratinocytes (1° mouse keratinocyte [mk]) were cultured from both wild-type (WT) and EMTB-3GFP–expressing mice. In low-calcium media, cells remain proliferative and do not form cell–cell junctions. Under these conditions, microtubules extended radially from the nucleus, which appeared to serve as an MTOC for these cells (). This organization was also visualized by immunostaining for α- or β-tubulin in wild-type keratinocytes. When calcium levels were raised to induce cell–cell adhesion and differentiation, a reorganization of the microtubule cytoskeleton was observed. At early times after induction, microtubules concentrated at areas of developing cell–cell contacts (), which were particularly prominent in the apical domains of the cells. By 72 h after induction, cells were stratified, and microtubules were most concentrated at the cell junctions of the suprabasal (apical) cells (). Notably, microtubules no longer displayed the radial association with the nucleus that had been prominent in low-calcium conditions (compare ). As judged by confocal microscopy and z-stack imaging, MTOC-like structures were absent in these differentiating cultures (). Overall, the microtubule rearrangements seen in cultured keratinocytes bore a strong resemblance to those in intact epidermis, and further substantiated that upon differentiation, microtubules undergo a stereotypical reorganization consisting of loss of centrosomal/nuclear MTOC activity and accumulation of microtubules at intercellular junctions. The accumulation of microtubules at calcium-stimulated cell junctions suggested a role for cadherin-mediated junctions in this process. To test this hypothesis, we mated mice to obtain embryos that expressed epidermal EMTB-3GFP and had epidermal-specific loss-of-function mutations in or genes. DP is a linker protein known to connect desmosomal cadherin complexes to the IF cytoskeleton (; ), and it is essential for the formation of desmosomes (; ). In contrast, α-catenin integrates E-cadherin–β-catenin complexes with actin dynamics, and it is essential for the formation of adherens junctions (). In both wild-type and α-catenin cKO epidermis, the microtubule cytoskeleton was concentrated at the junctions of suprabasal cells (). In striking contrast, despite the relatively normal organization of microtubules in DP cKO basal cells, the suprabasal cells displayed a paucity of cortical microtubules ( and Video 3, available at ). Instead, cytoplasmic aggregates of microtubules were observed (, inset). This observation was surprising because desmosomes and DP have not been implicated in microtubule organization. There were small regions of the α-catenin cKO epidermis that displayed reduced cortical staining for microtubules. However, closer inspection revealed that they exhibited a corresponding lack of cortical DP (Fig. S1). Collectively, the cortical microtubule organization observed in suprabasal epidermal layers appeared to be directly attributable to DP. The dependency of cortical microtubules on DP could also be recapitulated in vitro. Thus, in WT and –null epidermal cultures grown in a high-calcium medium, microtubules concentrated at sites of cell–cell contact (). In contrast, DP-deficient keratinocytes in high-calcium media displayed an evenly distributed array of cytoplasmic microtubules with no appreciable border accumulation, even in closely juxtaposed cells, and with no obvious perinuclear organization (). In low-calcium media, where cell junctions do not form, a radial microtubule organization extending from the perinuclear area was observed in WT, α-catenin–null, and DP-null keratinocytes (). These findings revealed that in high calcium, the loss of a perinuclear MTOC and gain of a cortical microtubule network can be uncoupled when DP is absent. Moreover, the results demonstrate that the microtubule rearrangements that occur in differentiating WT epidermal cells can be dissected into DP-dependent and -independent events. In many cells, protein complexes containing γ-tubulin induce local microtubule nucleation at the centrosome/MTOC. In cross sections of WT epidermis, basal cells displayed a concentrated focus of anti–γ-tubulin staining at the apical edge of the nucleus (). These data provided strong support for our prior localization of the putative MTOC in basal cells (). Although no longer apical, γ-tubulin retained its focal perinuclear localization in the differentiating suprabasal cells of the e18.5 epidermis (). This was also the case for a centriolar marker, centrin-GFP (; ). For both γ-tubulin and centrin-GFP, the intensity of labeling was reduced as cells moved to the suprabasal layers. However, the persistence of the centrosomes was surprising, given the loss of a discernable MTOC. In vitro, γ-tubulin and the centriolar proteins centrin and centriolin () also remained associated with centrosomes under both low- and high-calcium conditions (). This was also the case in DP-null keratinocytes (). Thus, both in vivo and in vitro, the differentiation-induced rearrangements in microtubule organization were not attributable to a loss of the centrosome and/or relocalization of γ-tubulin. Centrosomes remained largely intact, but they did not function as MTOCs when epidermal cells enter their program of terminal differentiation. Time-lapse imaging of EMTB-3GFP revealed that centrosomes in differentiating epidermal cells are still able to nucleate microtubules (Video 4, available at ). Additionally, when keratinocytes were released from nocodazole-induced microtubule depolymerization, a burst of microtubule polymerization was detected at centrosomes (, and Videos 5 and 6). The ability of microtubules to be nucleated from centrosomes was observed in both low- and high-calcium conditions. However, although the centrosome remained associated with some of the microtubules in low calcium, this was not the case in high calcium. These data provide further evidence suggesting that when epidermal cells differentiate, centrosome activities become separated; microtubule nucleation continues to occur, whereas microtubule-anchoring activity is lost. Several proteins have been implicated in anchoring microtubules at the centrosome (; ; ; ). One of these is the coiled-coil protein ninein, which localizes to the subdistal appendage of the mother centriole (), a preferred site for microtubule attachment. Ninein is a marker for microtubule anchoring not only in proliferative cells but also in differentiated inner ear hair cells, where it colocalizes with microtubule minus ends (). Functionally, it has been shown that the levels of ninein at the centrosome determine its microtubule-anchoring activity (; ). Interestingly, the only reported experimentally induced separation of centrosomal nucleation and anchoring functions (which resembles what happens physiologically in differentiating epidermis) is by expression of mutants of ninein that prevent its targeting to the centrosome (). To investigate whether ninein might play a role in these microtubule rearrangements, we first examined its localization. In cultured keratinocytes, ninein localized to one of the two centrioles marked by antibodies against γ-tubulin (, inset). Such labeling has been observed before in fibroblasts (). However, keratinocytes differed from fibroblasts in their display of anti-ninein labeling around the nuclear envelope, from which the radial array of microtubules emanated in these cells (, arrowheads). Another distinguishing feature was illuminated when epidermal cultures were induced to terminally differentiate. Associated with this process was a depletion of ninein at the centrioles and nuclear envelope, and concomitant ninein relocalization to cell–cell junctions (). Quantitation of ninein-containing centrioles revealed that both WT and DP-null cells show a characteristic loss of ninein from centrioles upon differentiation (). However, at 48 h after calcium addition, DP-null cells have more ninein-positive centrioles than do WT cells. This might be expected given that DP-null cultures did not stratify efficiently and remained more proliferative than their WT counterparts (mitotic index 0.40 ± 0.06 for DP-null vs 0.06 ± 0.03 for WT). In vivo, for both WT and DP cKO epidermis, ∼90% of basal cells had detectable ninein at the centrosome, compared with <10% of suprabasal cells ( > 200). Localization of ninein to cell junctions was not an antibody artifact, as an exogenous GFP-ninein fusion protein also localized to cell junctions (). This localization did not require either intact microtubules or F-actin (Fig. S2, available at ). To further characterize the localization of ninein to cell–cell borders, we costained high-calcium keratinocyte cultures with antibodies against ninein, E-cadherin, and DP. Although ninein displayed a largely nonoverlapping pattern with E-cadherin (), it showed extensive colocalization with DP (), and with the desmosomal cadherins (not depicted). Based upon these data, ninein's border localization appeared to coincide with desmosomal, rather than adherens junction, markers. To determine whether DP might act to recruit ninein to desmosomes, we repeated our calcium-induced differentiation experiments in -null epidermal cultures. As shown in , ninein remained largely cytoplasmic and was not recruited to cell–cell borders in the absence of DP. Moreover, this was not a general effect of disrupting intercellular adhesion because calcium-treated keratinocyte cultures lacking α-catenin still localized ninein at sites of cell–cell contact (). Using deletion mutagenesis and transient transfections, we narrowed down the desmosomal-colocalization domain of ninein to a 368-aa stretch in the central region of ninein (Fig. S2). The insolubility of both desmosomal proteins and ninein, coupled with the ability of these proteins to aggregate, precluded our ability to conduct further biochemical studies on the nature of the interaction with desmosomal components. In WT skin, ninein localized to the centrosomes in dermal fibroblasts and basal epidermal cells, but to intercellular borders in suprabasal cells (). The loss of ninein from the centrosomes of suprabasal cells was selective, as γ-tubulin and centrin both remained associated with centrosomes in suprabasal cells and were not detected at cell junctions (). These data unveiled a striking correlation between the rearrangement of microtubules in differentiating epidermal cells and the selective redistribution of ninein from the centrosomes to cell–cell borders. Moreover, this redistribution was also seen in esophagus (Fig. S3, available at ), suggesting that it could be a general phenomenon of epithelial stratification/differentiation. The distinctive localization of ninein to cell borders was lost in DP-null epidermis, although E-cadherin and cell–cell junctions were still observed (). In contrast, ninein's colocalization with desmosomal components remained intact in α-catenin–null epidermis (). Given that the number of desmosomes increases dramatically as epidermal cells differentiate, one might expect that ninein's localization is determined by a competition between desmosomes and centrosomes. This did not appear to be the case, however, because in DP-null suprabasal cells, ninein did not remain focused at centrosomes (). These results suggest that distinct signals govern ninein's centrosomal localization. Based upon these data and previous experiments (), depletion of ninein from the centrosome appears to be physiologically relevant for the subsequent loss of microtubule anchoring at the MTOC. Furthermore, the DP-dependent relocalization of ninein to desmosomal cell–cell contacts may be involved in the reorganization of microtubules to the cell cortex. DP has been traditionally viewed as an adaptor protein that links to desmosomes through its head domain and links to the IF cytoskeleton through its tail domain (). To ascertain which of these domains are required for the reorganization of ninein, we transiently transfected DP-null keratinocytes with expression vectors driving either full-length DP or a stable truncated form of DP lacking the C-terminal domain that binds to IFs (; ; ). Both the full-length and tailless forms of DP restored ninein localization to cell junctions, suggesting that tethering of IFs to desmosomes is not required for cortical ninein localization (). A construct containing only the DP head domain failed to recruit ninein, suggesting that the coiled-coil rod domain is required for this function (). These findings suggest a novel function for DP that goes beyond its ability to connect IFs to desmosomes. In this report, we have described a dramatic rearrangement of microtubules that occurs as epidermal cells differentiate. In suprabasal cells, the microtubule-anchoring protein ninein is released from the centrosome and relocalizes in a DP-dependent manner to developing desmosomes at cell–cell junctions. This leads to a robust cortical network of microtubules in the differentiating layers. That said, ninein need not act alone in eliciting its association with desmosomes and triggering the dramatic reorganization of the microtubule network from the centrosome to the cortex. Several other centrosomal proteins have now been shown to be involved in microtubule-anchoring at centrosomes (; ; ), and, intriguingly, they include not only minus-end microtubule proteins like ninein but also plus-end microtubule-binding proteins. In this regard, it is noteworthy that an isolated report over a decade ago showed that the plus-end microtubule-binding protein CLIP170 colocalizes with desmosomes (). Our work also raises questions about the roles of microtubules in terminally differentiating epidermal cells. Our studies documenting the functional importance of DP in governing microtubule reorganization in differentiating epithelia now pave the way for future studies in dissecting the underlying mechanisms involved, and in evaluating the specific contributions that microtubule remodeling makes in dictating desmosomal functions in the epidermis. A cDNA encoding the ensconsin microtubule-binding domain fused to three copies of GFP was generated previously (). This was subcloned into a vector under the control of an epidermal keratin promoter (). This gene was removed from the vector backbone by digestion with EcoRI and SphI and used for injection via established procedures into fertilized mouse embryos. For visualization of microtubules, tissues were fixed in a solution containing 80 mM PIPES, pH 6.9, 50 mM NaCl, 2 mM MgCl, 0.4 mM CaCl, 1% glutaraldehyde, 3% paraformaldehyde, and 0.2% Triton X-100 at 37°C. After fixation, tissues were extensively washed, and then treated with sodium borohydride (0.1%) in PBS for 30 min. Tissues were then washed extensively in PBS before embedding in OCT. For other stainings, tissues were embedded directly in OCT, frozen, sectioned, and fixed in 4% PFA in PBS. Fixation of cultured cells for preservation of microtubules was carried out with the same buffer, except with 0.5% glutaraldehyde and 0.1% Triton X-100. For any stainings involving ninein or γ-tubulin, cells or tissue were fixed in −20°C methanol for 3 min. Antibodies used were rabbit anti-ninein (Rattner Laboratory, University of Calgary, Calgary, Alberta, Canada), mouse anti–γ-tubulin (Sigma-Aldrich), rat anti–α-tubulin (Serotec), mouse anti-DP (ICN Biomedicals), rat anti–β4 integrin (BD Biosciences), rat anti– E-cadherin (Fuchs Laboratory, The Rockefeller University, New York, NY), and rabbit anti-centriolin (Doxsey Laboratory, University of Massachusetts, Worcester, MA). The DP deletion constructs used in this study were previously generated and characterized (; ), with the DP-Head including aa 1–1,020 and the head + rod including aa 1–2,000. For the microtubule-recovery video, primary cells were cultured in low- or high-calcium media for 72 h, and then treated with 10 μM nocodazole for 1 h at 4°C. Cells were thoroughly washed with four changes of media to remove nocodazole before imaging. For analysis of ninein localization, keratinocyte cultures were treated for 1 h with 2 μg/ml latrunculin-B or 2 μM nocodazole before fixation. Images of fixed cells were taken with an Axioplan with Apotome attachment (both Carl Zeiss MicroImaging, Inc.). A 63×/1.4 NA objective was used, and images were collected with Axiovision software (Carl Zeiss MicroImaging, Inc.). Living cells/tissues were imaged with a spinning disk (Perkin-Elmer), using a 63×/1.4 NA objective at 37°C, and acquired with Perkin-Elmer software. Images were collected on cameras (Orca-ER; Hamamatsu). Videos were rendered using AfterFX software. Fig. S1 shows loss of DP and ninein in small regions of the α-catenin cKO epidermis. Fig. S2 shows that neither F-actin, microtubules, nor the IF-binding domain of DP are required for ninein localization. In Fig. S3, ninein localization to cell junctions in suprabasal esophagus is demonstrated. Video 1 is a z-stack imaging microtubules in e14.5 epidermis. Videos 2 and 3 are z-stacks imaging microtubules in e18.5 epidermis in WT (Video 2) and DP cKO (Video 3) epidermis. Microtubule nucleation at the centrosome is shown in unperturbed differentiated cells in Video 4, and after nocodazole washout in proliferating cells (Video 5) and differentiated cells (Video 6). Online supplemental material is available at .

Dynamic changes in chromatin structure are directly influenced by the posttranslational modifications of the N-terminal histone tails (). Specific amino acids within the tails are modified by phosphorylation, ubiquitination, ADP ribosylation, acetylation, and methylation (; ). Methylation of different lysine and arginine residues in histone H3 and H4 tails is associated with actively transcribed or repressed chromatin (). Histone H4 lysine 20 (K20) can be mono-, di-, or trimethylated. PR-Set7 is a histone methyltransferase that specifically monomethylates histone H4K20 (; ; ; ). Trimethylation of the same lysine is controlled by other histone methyltransferases, Suv4-20h1, and Suv4-20h2 (). Coincident with the conservation of the H4K20 methyl modifications in higher eukaryotes, both enzymes show substantial homology in species ranging from flies to humans. A null mutation in suppresses position effect variegation, indicating that H4K20 methylation plays a role in silencing of gene expression (). Several observations suggest that PR-Set7–dependent methylation of H4K20 also plays an important role in cell proliferation. In HeLa cells, expression of PR-Set7 increases during S phase and peaks at mitosis (). In larvae, tissues with higher rates of cell divisions, such as imaginal discs, are severely affected by the depletion of PR-Set7. Homozygous mutant discs are smaller than wild type because they contain only ∼25% as many cells as wild type (). Here, we investigated the function of PR-Set7–dependent methylation in more detail by studying the cell cycle in mutant neuroblasts. Neuroblasts are diploid, and their cell cycle progression has been well documented (). In the –null allele, used in the experiments described here, the PR-Set7 protein is missing from the first-instar larval stage onward, homozygous or hemizygous () animals survive until the larval/pupal transition, and the reduction of methylated H4K20 is only observed in late-stage larvae (). In mutant third-instar larval brains, monomethylated H4K20 was strongly reduced. We found that in the mutant brains, the mitotic index was reduced, progression through early mitosis was delayed, and cyclin B was reduced. The abnormalities in mitotic progression and in the level of cyclin B were rescued when the DNA damage checkpoint was abolished, indicating that the DNA damage checkpoint is activated in . The organization of the third-instar larval brains used in all the experiments here is affected in the mutant (). The wild-type brain hemispheres contain two rings with high rates of cell divisions, called the optic lobes. These regions are clearly disorganized in both homozygous and hemizygous larval brains ( and not depicted). We initiated our studies by determining whether histone H4K20 methylation is reduced in homozygous brains. Western blots of mutant third-instar larval brain lysates showed that mono-, di-, and trimethylated H4K20 and total histone H4 are reduced compared with wild type, when each value is normalized to the value of lamin ( and Table S1, available at ). However, when each value is normalized to the values of both histone H4 and lamin, monomethylated H4K20 is reduced to ∼26% of wild-type levels, whereas di- and trimethylation is only down <15% (, bottom; and Table S1). This result indicates that monomethylated H4K20 is strongly reduced in the mutant brains. The reason the histone H4 level was lower in the mutant extracts is not clear. Staining of neuroblasts with anti–monomethylated H4K20 antibody (anti-mono) gave similar results (). We next observed the distribution of monomethylated H4K20 during the cell cycle by staining wild-type neuroblasts. Anti-mono staining was detected only on condensed DNA or chromosomes (Fig. S1, available at ). Monomethylated H4K20 is first detected in cells in late G2-prophase at the very onset of DNA condensation and peaks at metaphase, similar to the detection of phosphorylated histone H3 Ser 10 (PH3), known to be associated with condensed DNA (; ). Interestingly, the monomethylation mark thought to be associated with repressed chromatin is not increased at the centromere region but distributed evenly along the chromosomes (). Monomethylated H4K20 is present throughout the cell cycle in HeLa cells () and is detected on polytene salivary gland chromosomes (). Salivary gland cells undergo a modified cell cycle, lacking G2 and M (). This staining therefore suggests that monomethylated H4K20 is also present in nonmitotic cells. It is likely that monomethylated H4K20 is present throughout the cell cycle and that it is not immunocytologically detectable until chromosome condensation concentrates the signal. Monomethyl mark may be uniformly distributed throughout the cell cycle and throughout the genome as detected on mitotic chromosomes (). We studied the progression through mitosis of neuroblasts to determine when the mutant cells have a defect. In wild type, the mitotic index (percentage of cells in mitosis) was 2.16%, and it was significantly reduced in to 1.30% (P = 2.75 × 10; and Table S2, available at ). To identify cells in the different mitotic stages, neuroblasts were stained with both anti-PH3 and anti–α-tubulin antibodies. We considered cells in prophase if they were positive for PH3 staining and showed interphase-like organization of microtubules without visible asters. As shown in (Table S3), both and had a fourfold higher frequency of cells in prophase (, 42.5%; , 41.0%) than wild type (10.3%) and a correspondingly lower frequency of cells in the other mitotic phases. After a 1-h treatment with colchicine the mitotic index in the mutant increased 2.7-fold ( and Table S2), indicating that the spindle assembly checkpoint is not disrupted in . However, with cholchicine, the ratio of prophase and prometaphase cells in the mutant was still high (wild type, 2.6%; PR-Set7, 43.6%; Fig. S2 a). These results show that the mutant cells are delayed in early mitotic stages. In wild type, the PH3 signal appears at prophase (), increases with chromosome condensation (), and covers the entire chromosomes at late prometaphase (). The formation of mitotic spindles always correlates with both PH3 staining and chromosome condensation, and two asters are observed before the PH3 signal covers the entire chromosome (). However, in , the PH3 signal covered most of the chromatin before the asters appeared (). In wild type, 79% of prophase cells showed no chromosome condensation, as shown in (see also Fig. S2 b), whereas in , most of the prophase cells (65%) had condensed DNA (or chromosomes), as shown in (h–k). These results indicate that nuclear and cytoplasmic events, namely, chromosome condensation and spindle formation, are uncoupled in . These events may keep mutant cells from entering metaphase or from chromosome separation. The phenotypes described so far could be caused by the failure to accumulate mitotic cyclins. It has been well established that accumulation of mitotic cyclins and the activation of the Cdk–cyclin complexes are essential for entry into mitosis and formation of mitotic spindles (; ). We therefore examined cyclin A and B protein levels by Western blotting and found that cyclin B was reduced in , whereas cyclin A was present at normal levels (). In , the expression pattern of cyclin A is similar to cyclin B, and both accumulate during G2 (). Because the cyclin A protein level was not reduced in the mutant, the lower mitotic index observed in the mutant does not explain the reduction of cyclin B. We also found that the cyclin B protein level was still down-regulated in the mutant when the mitotic index was increased after a 1-h colchicine treatment ( and not depicted). To investigate whether the low level of cyclin B was controlled at the transcriptional level, we measured mRNA levels of cyclin A and B by quantitative real-time PCR. Both mRNAs showed similar reductions in the mutant compared with wild type (), probably because of the lower mitotic index, indicating that the reduction of the mRNA level does not fully explain the reduction of cyclin B protein. The APC/C subunit cdc27 is required for the degradation of cyclin B in (). Hence, we made double-homozygous and mutants and found that the cyclin B level recovered in the double mutant (). These results indicate that the reduction of cyclin B in is mediated by the APC/C proteolysis and is not regulated at the transcriptional level. Mitotic stages other than prophase also showed abnormalities in neuroblasts. Many prometaphase figures had irregular chromosomes (Fig. S2, c and g). Although most metaphase figures looked surprisingly normal in the mutant (Fig. S2, d and h), a large proportion of anaphase and telophase figures contain lagging chromatids (40.9%; and Fig. S2, e, f, i, and j), raising the possibility that has a defect in chromosome condensation. We next looked at the morphology of metaphase chromosomes in wild-type and brains (). We focused on the length and width of chromosomes and boundaries between sister chromatids to assess whether chromosome condensation is affected in the mutant. Chromosome figures were subdivided into categories I–IV, depending on phenotypes. Most of the wild-type chromosomes (85.7%) had clearly defined sister chromatids (normal chromosomes, category I). 97.3% of mutant chromosomes displayed aberrant morphology. 17.9% of chromosomes were thinner and longer than category I chromosomes and showed vaguely defined sister chromatids (category II). In 53.8%, the sister chromatids are not well defined enough to be apparent (category III). Other chromosomes (25.6%) were entangled (category IV). All these observations point to a defect in chromosome condensation in the mutant. We also prepared chromosome spreads after a 1-h colchicine treatment and hypotonic shock to better define the degree of chromosome condensation (). Most of the wild-type chromosomes (92.7%) showed clearly defined sister chromatids as expected (category I). 60.9% of the mutant chromosomes were strikingly abnormal, considerably longer and thinner than wild type, and showed no defined sister chromatids (category II). 10.9% of mutant chromosomes were longer and thicker and lost the sister chromatid borders, similar to what is observed in category III in . Because the length of mutant chromosomes in category IV (17.2%) is similar to normal chromosomes (category I), some mutant chromosomes seem to complete chromosome axis shortening but still have an obvious defect in defining sister chromatids. The simplest explanation of these results is that the PR-Set7–dependent monomethylation of H4K20 is required for proper chromosome condensation. We found that despite the abnormal chromosome organization, the mutant chromosomes contain the condensin component Barren (the fly orthologue of XCAP-H) and DNA topoisomerase II, both important for chromosome architecture (; ; Fig. S3, available at ). To determine whether the abnormal chromosome condensation and separation in the mutant results in polyploidy, we measured the DNA content of wild-type and mutant brain cells using a laser-scanning cytometer (), and found that the number of polyploid cells was not increased in the mutant (). In the cytometer analysis, we noticed some cells with less DNA than 2n in (, arrow). We investigated whether apoptotic cells are increased in the mutant by TUNEL and found that positive cells were not significantly increased in the mutant (wild type, 40/5,012, and , 56/4,056; P = 0.108). The observed abnormalities in cell cycle progression and the reduction of cyclin B protein levels mediated by APC/C-dependent proteolysis in led us to hypothesize that the DNA damage checkpoint is activated. We also checked the S phase index (percentage of cells in S phase) in and found that it was significantly reduced (P = 2.82 × 10; and Table S4, available at ). To investigate whether these phenotypes are caused by the activation of the DNA damage checkpoint, we made a homozygous double mutant of and the orthologue (), essential for the DNA damage checkpoint (; ; ). The allele used here has a defect in the checkpoint, allowing cells with damaged DNA to enter mitosis (). In the double mutant, the mitotic index was rescued, similar to that observed in wild type and the homozygous mutant ( and Table S2). The number of prophase cells was reduced compared with the number observed in homozygotes and became similar to wild type and ( and Table S3), and the “uncoupled cells” observed in () disappeared (not depicted). These results indicate that the abnormalities of mitotic progression in are rescued in the double mutant. The S phase index in the is increased 2.4-fold compared with wild type, suggesting that the mutant has a defect in DNA replication ( and Table S4). Therefore, we cannot determine whether the decrease of the S phase index in is rescued in the double-homozygous mutant. The protein levels of cyclin B also recovered in the double-homozygous mutant (). To confirm this result, we also made a homozygous double mutant of and another allele of , also affecting the G2/M checkpoint, (; ). Further, we constructed a double mutant with and the orthologue (). The gene functions downstream of , and mutants in both genes have similar phenotypes (; ). On Western blots of brain extracts of both double mutants, monomethylated H4K20 is not detected, similar to the extract, but the cyclin B protein levels are rescued (). These results show that in , the DNA damage checkpoint, especially the ATR pathway, is activated and that this activation is responsible for the down-regulation of cyclin B and the abnormal mitotic progression. These results also suggest that a specific mechanism exists in that down-regulates cyclin B through APC/C proteolysis in response to DNA damage. The ratio of anaphase/telophase cells with lagging chromatids was increased when the checkpoint was abolished in the double-homozygous mutant (), indicating that the defect in chromosome condensation is independent of checkpoint activation and that, in the absence of the checkpoint, the severity of the chromosome condensation defect is enhanced. Because we found that the DNA damage checkpoint is activated in , we investigated whether DNA DSBs are increased in the mutant. We stained cells with anti–phosphorylated histone H2Av (PH2Av) antibody. Like H2A.X, H2Av, an essential fly histone variant, becomes phosphorylated at sites of DSBs by DNA damage recognizing factors (). Thus, PH2Av is well established as a marker for DSBs. We found that the PH2Av-positive cells were not significantly increased in the mutant (P = 0.145), indicating that endogenous DSBs do not occur in (). To confirm that the cells react normally to DNA damage, we irradiated both wild-type and mutant brains with x-ray and found that the PH2AV staining increased in both (). Further, after γ-irradiation PH3-positive mitotic cells were drastically decreased in the mutant (), similar to what was shown in wild type (), indicating that the cells react in similar ways as wild-type cells to γ-irradiation. We conclude that DSBs are not increased in . What activates the DNA damage checkpoint in ? Histone methylation has been thought to control gene expression by packaging DNA into open and closed chromatin. indeed suppresses position effect variegation, indicating that it functions as a transcriptional suppressor (). However, abnormal regulation of gene expression in does not explain the activation of the checkpoint. Because the mutation in () abolished the phenotypes caused by the activated checkpoint in , the expression of genes downstream of is apparently unchanged in . Furthermore, ATR is one of the proteins that initiate checkpoint signaling and localize to sites of DNA damage, suggesting that the ATR protein is directly activated by DNA damage (). Also, we showed that the DNA repair pathway after γ-irradiation was activated in ; PH2Av staining is increased and the number of mitotic cells is decreased, similar to what is observed in wild type (; ). Collectively, our results indicate that the expression of genes involved in the DNA damage checkpoint is normal in and that control of gene expression is not involved in activation of the checkpoint. We observed reduction of both S phase and mitotic indexes in the mutant (), suggesting that both G1 and G2 checkpoints are activated. Because the mutant shows many defects in chromosome condensation and separation (), these defects could be one of the reasons the G1 checkpoint is activated after mitosis. What activates the G2 checkpoint? Because DSBs are not increased in the mutant (), they cannot be cause for G2 checkpoint activation. Several checkpoints appear to exist in mammalian cells that monitor chromatin structure as well as DSBs. A decatenation checkpoint that monitors chromatid decatenation has been demonstrated in human cells, with progression from G2 to mitosis being inhibited when chromatids are insufficiently decatenated (). Furthermore, in mammalian cells, ATM (DNA damage checkpoint kinase) activation is not necessarily dependent on direct binding to DSBs but may result from changes in the structure of chromatin (). As in mammalian cells, abnormal higher order structure of DNA or chromatin may activate the DNA damage checkpoint, as observed in . We therefore hypothesize that monomethylated H4K20 is involved in the maintenance of proper higher order structure of DNA or proper chromatin structure. Crystal structural analysis showed that K20 is not part of the N-terminal tail of histone H4, like other sites of methylation (e.g., histone H3 lysine 9; ). K20 lies close to the histone-fold domain and is covered by DNA. The results in agree with the crystal structure. It has been proposed that methylated H4K20 lies normally inside the DNA and that it is exposed only when DSBs occur, creating a binding site for Crb2 (). The structural analysis further showed that H4K20 makes an interparticle contact with an H2A-H2B dimer (), suggesting that H4K20 could be involved in the maintenance of proper histone structure. In , the loss of acetylation of H3 lysine 56 affects nucleosome structure (). H3 lysine 56 is also close to the histone-fold domain and weakens histone–DNA interactions (). Like acetylation of H3 lysine 56, monomethylation of H4K20 may affect nucleosome structure. We showed that the loss of monomethylated H4K20 activates the DNA damage checkpoint, which may be induced by abnormal higher order structure of DNA or abnormal chromatin structure in the absence of DSBs. The abnormal DNA or chromatin structure probably causes the abnormal mitotic chromosomes observed in . Our results suggest that monomethylation of H4K20 has a more global effect on chromatin structure than described so far. The stock was used as the wild-type control. Homozygous larvae were recognized by the absence of the balancer (). , , and were obtained from the Bloomington Stock Center. () and () were obtained from K. McKim (Rutgers University, Piscataway, NJ) and W.E. Theurkauf (University of Massachusetts Medical School, Worcester, MA), respectively. Brains were dissected, and immunoblotting was performed as previously described (). Rabbit polyclonal anti-monomethylated, anti-dimethylated, anti-trimethylated, and histone H4 antibodies (Upstate Biotechnology) were used at 1:1,000 dilution (). Rabbit polyclonal anti-lamin obtained from P.A. Fisher (State University of New York, Stony Brook, Stony Brook, NY) was used at 1:1,000. Mouse monoclonal anti–cyclin B antibody, F2F4 (Developmental Studies Hybridoma Bank), and anti–cyclin A antibody, A12 (Developmental Studies Hybridoma Bank), were used at 1:200. Third-instar larval brains were dissected, fixed, and stained as previously described (). Whenever required, brains were dissected in PBS and incubated in 10 μM colchicine (Sigma-Aldrich) for 1 h before fixation. Rabbit polyclonal anti-monomethylated (Upstate Biotechnology) and mouse monoclonal anti-PH3 antibody (Upstate Biotechnology) were used at 1:200 dilution. Rabbit polyclonal anti-PH3 antibody (Upstate Biotechnology) was used at 1:1,000. Anti–α-tubulin conjugated with FITC (Sigma-Aldrich) was used at 1:50. Rabbit polyclonal anti–PH2Av (Ser139) antibody (Upstate Biotechnology) was used at 1:250. Rabbit polyclonal anti-Barren antibody was obtained from H. Bellen (Baylor College of Medicine, Houston, TX) and used at 1:50, and mouse monoclonal anti-Top2 antibody was obtained from A. Kikuchi (Nagoya University, Nagoya, Japan) and used at 1:5. Secondary antibodies were Cy3-conjugated goat anti–rabbit or anti–mouse (Jackson ImmunoResearch Laboratories) used at 1:500 and Alexa Fluor 489 goat anti–mouse or Alexa Fluor 488 donkey anti–rabbit (Invitrogen) used at 1:400. For TUNEL assay, fixed cells were incubated with TUNEL enzyme (Roche) for 1 h, and the Alexa Fluor 488 Signal-Amplification kit (Invitrogen) was used for detection of the signals. The samples were mounted in Vectashield (Vector Laboratories), and Immersol 518F (Carl Zeiss MicroImaging, Inc.) was used as the imaging medium. Immunostained preparations were studied using a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) with a 63× Plan-Apochromat NA 1.4 and 100× and a 40× Plan-Neofluar NA 1.3 objective lenses (room temperature) equipped with a digital camera (SensiCam; Cooke Corp.). Digital images were collected using the Image Pro Plus imaging software (MediaCybernetics). To measure the DNA content, we stained brain cells with DAPI and used a laser-scanning cytometer, iCys (CompuCyte Corp.; ). Brains were dissected and incubated for 45 min in serum-free Schneider's insect medium containing 100 μg/ml BrdU. They were rinsed three times in PBS, fixed, and stained with the mouse monoclonal anti-BrdU antibody (1:100 dilution; Becton Dickinson) as previously described (). After a 1-h treatment with colchicine, brains were incubated for 3 min in 0.5% trisodium citrate for hypotonic shock. They were then fixed and stained with rabbit polyclonal anti-PH3 antibody to identify mitotic chromosomes. Third-instar larvae were transferred to yeast paste-supplemented agar plates and irradiated with 2000 rad of x-rays. After 15 min, the brains were dissected, squashed, and stained. Fig. S1 shows that monomethylated H4K20 is detected only on condensed DNA or chromosomes. Fig. S2 shows typical mitotic figures in . Fig. S3 shows that chromosomes contain the condensin component Barren and Top2. Table S1 shows the numbers behind the graph shown in b. Table S1 shows the ratio of intensity of the bands from relative to wild type in b. Tables S2 and S3 show mitotic index and parameters in squashed brains of wild type, , , and and , respectively. Table S4 shows S phase index in squashed brains of wild type, , , and . Online supplemental material is available at .

The nuclear envelope (NE) is composed of inner and outer nuclear membranes that are perforated by nuclear pore complexes and supported by the nuclear lamina (). Proteins of the inner nuclear membrane have been implicated in organization of nuclear architecture and cell cycle control. One such family of inner nuclear membrane proteins is lamina-associated polypeptide 2 (LAP2), generated by alternative splicing from a single gene (; ; ). LAPs have only been found in vertebrates; up to six isoforms exist in humans and mice (α, β, γ, δ, , and ζ). Most LAP2 isoforms have a closely related N-terminal nucleoplasmic domain of variable length with a nuclear localization signal sequence, a single-membrane spanning region, and a short luminal domain at their C terminus. LAP2α is structurally and functionally unique. It shares the first N-terminal 187 residues with all other LAPs but contains a unique C-terminal domain of 506 residues, which lacks a membrane spanning domain. As a result, LAP2α is distributed diffusely throughout the interphase nucleus except for nucleoli (). At their N terminus, LAP2 proteins share a LEM (LAP2, emerin, MAN1) domain (aa 111–152; ), which binds to the DNA bridging protein barrier-to-auto-integration factor (; ), and a LEM-like domain (aa 1–85), which binds to DNA and chromosomes (). In addition, LAP2α interacts with chromosomes via its α-specific C-terminal domain in a phosphorylation-dependent manner (; ). Overexpression of C-terminal fragments of LAP2α dominantly inhibits assembly of endogenous LAP2α, nuclear membranes, and A-type lamins in in vitro nuclear assembly assays and causes a cell cycle arrest in interphase, indicating a role for LAP2α in cell cycle progression (). LAP2α is found in stable complexes with the type V intermediate filament proteins lamins A/C in the nucleoplasm (). The LAP2α-specific C terminus associates with the C terminus of lamins A/C (aa 319–566) in vivo and in vitro. Dominant-negative lamin mutants, which cause aggregation of lamins A/C, also cause LAP2α to redistribute to the same aggregates, indicating a functional association between LAP2α and lamin A/C (). Mutations in the gene encoding lamins A/C () cause a spectra of human diseases (termed laminopathies), including muscular dystrophies, lipodystrophies, cardiomyopathies, neuropathies, dermopathies, and premature aging syndromes (). It has been suggested that altered lamin A/C–LAP2α associations might occur in laminopathies () because LAP2α binds to a region of lamin A/C where many different mutations have been found. In support of this hypothesis, a mutation in the LAP2 gene causing cardiomyopathy and affecting the C-terminal domain of LAP2α alters LAP2α interaction with lamins A/C (). The retinoblastoma protein (Rb) has functions in muscle and fat cell differentiation and in coordinating proliferation and differentiation during muscle regeneration (). The C-terminal domain of Rb is involved in its nuclear tethering, which is essential for its growth-suppressing activity (). In early G, hypophosphorylated Rb is tethered in the nucleus and is capable of binding transcription factor E2F-1, which prevents the transcriptional activation of S phase–specific genes and traverse to S phase. Hyperphosphorylation of Rb during late G phase releases it from nuclear tethers, which in turn releases and derepresses E2F and leads to passage through S phase. A-type lamins (; ) and LAP2α (; ) both interact with the C-terminal nuclear anchorage domain of Rb. In line with an Rb tethering function of LAP2α–lamins A/C complexes, LAP2α has recently been shown to be involved in an Rb-mediated negative cell cycle control (). One model for laminopathies proposes that mutations in LMNA impair the control of cell proliferation, particularly, Rb-mediated regulatory mechanisms controlling the exit and reentry into the cell cycle (). mouse show cell cycle defects because of reduced levels of Rb (). Interestingly, despite the proposed growth-suppressive function of LAP2α (), its expression is dramatically down-regulated upon cell cycle exit in human diploid fibroblasts (HDFs) and conversely up-regulated during reentry into the cell cycle (). Therefore, we tested the hypothesis that in HDFs, LAP2α expression is required for cell cycle progression via direct regulation of Rb. Here, we report that expression and lamin A–dependent organization of LAP2α is indeed required for maintaining HDFs in a proliferative state by promoting uniform nucleoplasmic localization of Rb. Our findings support the “cell cycle–proliferating–aging model” for laminopathies and provide a novel biological function for LAP2α in cell cycle regulation of human nontransformed adult cells. We previously showed that the expression of LAP2α is down-regulated as HDFs progress from a proliferating to a G state after serum starvation (). To further investigate LAP2α responsiveness to factors that induce quiescence in HDFs, we grew cultures to confluence and used immunoblotting to investigate the expression of proteins of interest. The level of expression of LAP2α declined dramatically as cultures reached confluence and the protein was undetectable in postconfluent cultures. In contrast, no changes in the level of expression of LAP2β, lamin A/C, or lamin B2 were detected in confluent or postconfluent cultures (Fig. S1 a, available at ). The level of expression of hyperphosphorylated Rb and, in particular, the RbS780 isoform also declined progressively in confluent and postconfluent cultures and correlated precisely with changes in the expression of LAP2α (Fig. S1, a and b). These data confirm that loss of expression of LAP2α occurs as a direct consequence of entry into G and is closely correlated with dephosphorylation of pRb at serine 780. To further investigate the correlation between changes in expression of LAP2α and phosphorylation of pRb, we performed double immunofluorescence microscopy to investigate both the expression and distribution of total Rb and forms phosphorylated at serines 780 and 795. In cultures induced to enter G through confluence, LAP2α as well as total Rb (Ab2) and RbS780 were barely detected in a majority of cells (). Cells still expressing higher levels of LAP2α also expressed relatively high levels of RbS780 and RbS795 (). In contrast to Rb780, RbS795 was still detectable in cells with low levels of LAP2α and distributed in a small number of nucleoplasmic foci (, arrows). To see how the distribution and phosphorylation of Rb is correlated with expression of LAP2α, G cultures were restimulated to reenter the cell cycle. 12 h after serum restimulation, LAP2α was detected in the nuclei of all cells and was distributed uniformly within the nucleoplasm, excluding nucleoli. Total and RbS780 were also expressed in nearly all cells and had a distribution that was very similar to that of LAP2α (). In contrast, RbS795 was located in a restricted number of nucleoplasmic foci (). 18 h after serum restimulation, the intensity of LAP2α, Rb, and RbS780 had increased but remained uniformly distributed throughout the nucleoplasm, whereas RbS795 remained within nucleoplasmic foci. Only at 24 h after serum restimulation, when the staining intensity of LAP2α, Rb, RbS780, and RbS795 was maximal, were all four proteins distributed uniformly throughout the nucleoplasm. To investigate the solubility properties of the total Rb and the two Rb phosphoisoforms more closely, G or serum-restimulated HDFs were extracted in situ. In most HDFs induced to enter G, total Rb, RbS780, and RbS795 were barely detectable (). Between 12 and 18 h after serum restimulation, all three forms of Rb were detectable and resistant to extraction. However, although Rb and RbS780 showed a more uniform staining throughout the nucleoplasm (), RbS795 was located in brightly stained nucleoplasmic foci (). 24 h after serum restimulation, when cells enter into S phase, all isoforms were relatively more soluble after extraction with Triton X-100 (). Collectively, our data suggest that progression of HDFs from G to a proliferating state (and vice versa) leads to a substantial remodeling of lamina proteins correlated with a more complex pattern of Rb nuclear anchorage than had previously been thought. In HDFs entering or exiting a quiescent state, Rb appears to be present in at least two different compartments depending on its phosphorylation status. Only when cells enter S phase (24 h after serum restimulation) are all forms of Rb detected in an apparently uniform distribution and largely soluble. The nature of the foci detected by antibodies against RbS795 was examined by colocalization with a range of antibodies previously reported to detect nuclear foci. There was no substantial colocalization between RbS795 and dense chromatin compartments (detected by DAPI; Fig. S2 a, available at ), phosphohistone H2A.X, or PML bodies (Fig. S2, c and d). In contrast, there was almost complete colocalization of RbS795 foci with splicing factor SC-35, which resides within splicing speckle compartments (Fig. S2 b), indicating that in cells entering G, this form of Rb is localized in splicing speckle compartments. To further investigate the distribution and nuclear anchorage of RbS795, we performed double immunofluorescence and confocal microscopy on HDFs that had been restimulated from G, before and after nuclear matrix extraction. Between 6 and 12 h after serum restimulation, Rb795 colocalized with splicing speckle compartments and remained insoluble within this compartment. However, after 18 h, colocalization between RbS795 and SC-35 was lost, but RbS795 was still resistant to extraction. Only upon entry into S phase (24 h) did RbS795 become completely soluble (). Thus, associations of Rb with the splicing speckles are both growth dependent and related to the expression of LAP2α. HDFs just entering the cell cycle and expressing low levels of LAP2α have RbS795 associated with splicing speckles. As LAP2α levels increase, RbS795 is no longer associated with speckle compartments but codistributes with LAP2α and other Rb forms within the nucleoplasm. To investigate whether lamins A/C directly influence the distribution of Rb within speckles, we performed double immunofluorescence on HDFs from a patient with lethal fetal akinesis, which harbors a homozygous mutation Y259X in the LMNA gene and is null for lamins A/C (). When grown for 2–4 d, an increasing number of Y259X HDFs (20–40%) displayed an abnormal accumulation of LAP2α into aggregates typically at one pole of the nucleus (). These cells stained negatively with antibodies against the proliferation marker Ki67 (). In addition, cells that were either negative for LAP2α or in which LAP2α was entirely located within aggregates had greatly reduced levels of total Rb and RbS780 (). In contrast, nearly all cells containing LAP2α aggregates expressed RbS795, which was mainly located in nuclear speckles () colocalizing with anti-SC35–positive splicing speckles (Fig. S3, available at ). Therefore, the absence of lamins A/C is correlated with aggregation of LAP2α and accumulation of RbS795 in speckle compartments. To investigate the influence of lamins A/C on nuclear anchorage of the different Rb isoforms, control or Y259X HDFs were subjected to extraction with detergents before staining with different Rb antibodies. In exponentially dividing control HDFs, the nuclei of most cells stained positively for total Rb and RbS780 before extraction. After extraction, ∼40% cells had greatly reduced or absent staining for total Rb and RbS780, whereas ∼60% remained strongly positive (), reflecting the different cell cycle stages. RbS795 was uniformly distributed in the nucleus of most cells before extraction, whereas it was found in speckles in ∼60% of cells and greatly reduced or absent in ∼40% of cells after extraction (). In contrast to control cells, the majority of Y259X HDFs were negative or only weakly positive for total Rb and RbS780 before extraction, and nearly 100% of cells were negative for Rb and RbS780 after extraction (), whereas ∼90% of Y259X HDFs still contained RbS795 restricted entirely to speckles after extraction (). These data suggest that nuclear anchorage of RbS780 was lamin A/C dependent. To test this hypothesis, Y259X HDFs were transfected with either GFP–lamin A or GFP–lamin C and prepared for immunofluorescence before or after detergent extraction. RbS780 was strongly retained in most cells that expressed GFP–lamin A after extraction (). Surprisingly, RbS780 was not retained in cells expressing GFP–lamin C (). Our data suggest that anchorage of nucleoplasmic forms of Rb is dependent on expression of lamin A, whereas anchorage of forms of Rb located in speckles is independent of lamins A and C. In Y259X HDFs, aggregation of LAP2α and cell cycle arrest occurs in only 40% of cells, suggesting that a compensatory mechanism might be abrogating the effects of loss of lamin A/C in culture. To investigate the consequences of loss of lamin A/C or LAP2α more directly, we used siRNA to knock down each protein independently. HDFs were transfected with siRNA targeted to LAP2α or lamins A/C or as a control, scrambled siRNA. We observed a ∼70% reduction in the level of LAP2α or lamins A/C expression 48 h after transfection with specific siRNA as compared with control siRNA, whereas expression of LAP2β was unaffected. Surprisingly, lamin A expression decreased by ∼30% in cells transfected with LAP2α RNAi, and LAP2α decreased similarly in lamin A/C RNAi–treated cells. The total amount of Rb and, more dramatically (by ∼70%), the levels of RbS780 were reduced in both LAP2α and lamins A/C RNAi cultures (). To investigate the effects of siRNA knock down of LAP2α or lamins A/C on cell proliferation, we tested the expression of Ki67. After transfection with scrambled siRNA, only 10% of cells were negative for Ki67. In contrast, ∼70 and ∼60% of cells transfected with LAP2α-specific siRNA or lamin A/C–specific siRNA, respectively, were negative for Ki67 (). We concluded that knock down of LAP2α or lamins A/C in HDFs leads to cell cycle arrest. To confirm these findings, we performed DNA flow cytometry on HDFs. Cells were arrested in G by serum starvation, transfected with siRNA, restimulated by addition of complete serum medium, and harvested 0, 24, 48, or 72 h after transfection (). After 24 h, a large proportion of cells had entered S phase, and there was no difference in the cell cycle profiles between any of the transfected cultures. 48 h after transfection (the time at which knock down of LAP2α and lamins A/C were detected), control RNAi cultures proceeded normally through G/M and displayed a cell cycle profile similar to asynchronously dividing fibroblasts. In contrast, cultures in which LAP2α or lamin A/C had been knocked down had a greater proportion of cells in S phase and an abnormally high proportion of cells in G/M phase. In view of previous data, showing a negative effect of LAP2α on the G–S phase transition (), our result may indicate that loss of LAP2α or lamins A/C promotes more rapid progression through G and premature entry into S phase, leading to a checkpoint arrest in G. However, 72 h after transfection, the LAP2α RNAi–treated cultures appeared to accumulate in G phase of the cell cycle. At the same time point, many cells in lamin A/C knockdown cultures had accumulated in G, but some remained arrested in G. Control cultures were dividing asynchronously at this time point (). These data are consistent with a cell cycle arrest of LAP2α- and lamin A/C–deficient cells in the G phase, presumably by activating a checkpoint to overcome defects due to a premature S phase entry in the previous cell cycle round. We previously reported that down-regulation of LAP2α after cell cycle arrest during myoblast differentiation is correlated with relocation of lamins A/C from the nucleoplasm to the NE (). To investigate the effects of siRNA knock down of LAP2α on lamins A/C distribution, fibroblasts were costained with antibodies against LAP2α and either lamin A or lamin C. In cells in which LAP2α was reduced or absent, both lamin A () and lamin C (not depicted) concentrated at the NE. In contrast, in control fibroblasts or in those fibroblasts in knockdown cultures that still expressed LAP2α, both lamins were found in the nucleoplasm and at the NE. To investigate the influence of LAP2α on the nuclear distribution of Rb, control and LAP2α siRNA–transfected cells were costained with antibodies against LAP2α and different forms of Rb. Although most LAP2α-positive control cells had Rb, RbS780, and RbS795 distributed uniformly throughout the nucleoplasm (), in LAP2α knockdown cultures, Rb was greatly reduced and had a granular appearance () and RbS780 was largely absent (). In contrast, RbS795 was present in all LAP2α-deficient cells but was restricted to nuclear speckles (), whereas it had more uniform distribution in LAP2α-expressing cells. Our data suggest that knock down of LAP2α or lamin A/C expression in HDFs leads to cell cycle arrest, dephosphorylation, and migration of Rb into nuclear speckles. Here, we show that entry of HDFs into G is correlated with loss of expression of LAP2α. Furthermore, in HDFs that are null for lamins A/C, LAP2α accumulates in aggregates, which is correlated with cell cycle arrest. Finally, siRNA knock down of LAP2α or lamins A/C in HDFs results in cell cycle arrest, and this is correlated with accumulation of Rb into speckles. Our data suggests that the expression and normal distribution of LAP2α allows proper regulation of Rb and thus maintains a proliferative state in HDFs. Our findings are consistent with other reports that show that HDFs lacking lamins A/C grow very slowly in culture () and that HDFs harboring mutant lamins enter a senescent state prematurely (). Other papers, however, appear to partly contradict these findings. mouse (−/− mouse embryonic fibroblasts [MEFs]) display a rapid growth phenotype with characteristics of Rb-null fibroblasts resulting from proteosomal degradation of Rb (). MEFs displayed a rapid growth phenotype but suggested that this resulted from inhibition of TGFβ signaling, leading to increased phosphorylation of Rb (). We have shown that in mouse fibroblasts and adipocytes, LAP2α represses E2F activity via an Rb-dependent pathway and that knock down of LAP2α in HeLa cells leads to rapid growth (). Although there are discrepancies between each paper as to the precise mechanism involved, all three agree that a loss of lamin A/C or LAP2α function leads to rapid proliferation through pathways involving Rb. The apparent discrepancies between the findings reported here and those reported previously (; ; ) probably reflect fundamental differences between the model systems used. Our study exclusively made use of HDFs, whereas the previous papers used either MEFs or transformed human cell lines. HDFs are able to respond to stimuli, such as genotoxic stresses, by inducing a checkpoint arrest transiently in G and eventual G arrest. This checkpoint is nonfunctional in MEFs as well as in transformed human cell lines with inactivated pocket proteins, which continue to divide or enter apoptosis after exposure to genotoxic agents (; ). Interestingly, FACS analysis of LAP2α and lamin A/C knockdown HDFs reveals that, initially, cells accumulate in G, before entering a G arrest. Therefore, it appears that HDFs respond to a loss of LAP2α or lamin A/C function as if they were treated with a genotoxic stress. Alternatively, the loss of Rb repressor activity upon LAP2α or lamin A/C down-regulation may prematurely drive the cells into S phase, causing activation of an incomplete S phase cell cycle arrest initially in G and eventually in G. Therefore, the reason that previous studies have not found that a loss of either lamins A/C or LAP2α leads to cell cycle arrest is most likely because these studies used cell lines in which this checkpoint pathway was abrogated. It has been shown that Rb binds to lamin A in vitro and associates with intermediate filament-like structures in the nucleus (; ). MEFs (), these findings are widely interpreted as evidence that anchorage of Rb within the nucleoplasm is dependent on lamins A/C. Our current data suggest that anchorage of Rb within the nucleoplasm depends on both lamin A and LAP2α. We show that nucleoplasmic forms of Rb are entirely absent from cells that contain either no LAP2α or in which LAP2α is entirely restricted to aggregates. These findings are consistent with our previous observations that Rb is anchored within the nucleoplasm via its C-terminal pocket C domain, which also binds to LAP2α in vitro (). Therefore, our results suggest that functional complexes of lamin A and LAP2α are required for anchorage of Rb within the nucleoplasm of HDFs. We also show that absence of LAP2α or its accumulation in aggregates is correlated with the preferential association of Rb with speckle compartments. Previous studies have suggested that Rb can bind speckle-associated protein p84 via sequences within its N-terminal domain (). These data are entirely consistent with our current findings because the N-terminal domain of Rb does not bind to lamins A/C or LAP2α (), and we show that Rb anchorage in speckles is independent of lamins A/C or LAP2α. The cell cycle arrest caused by down-regulation of LAP2α and lamin A/C in HDFs is correlated with rapid dephosphorylation of pRb. In a related study, we have shown that the introduction of dominant-negative lamin A mutants into C2C12 myoblasts causes loss of expression of LAP2α, which is also correlated with the absence of RbS780 (). Dephosphorylated forms of Rb bind to and inhibit the transcription factor E2F, thereby suppressing growth in both normally dividing cells and during differentiation of several mesenchymal cell types (for review see ). Therefore, it seems likely that the growth arrest caused by knock down of LAP2α results from dephosphorylation of Rb. Whether association of Rb into speckle domains is also a prerequisite for growth arrest is unclear. Certainly, the N-terminal region of Rb, through which Rb can associate with speckle-associated protein p84, is crucial for both terminal differentiation and growth suppression (). We propose that functional LAP2α–lamin A nucleoplasmic complexes might be required to anchor Rb in a nucleoplasmic compartment. This may allow correct regulation of Rb by cyclin-dependent kinases and protein phosphatases, which in turn makes cells responsive to environmental stresses, such as genotoxic agents. In conclusion, the corollary of this hypothesis is that a loss of function of lamin A or LAP2α, in cells with functional checkpoint pathways, might lead to irreversible cell cycle arrest and possibly cellular senescence because Rb can no longer be maintained in a phosphorylated state within the nucleoplasm and instead enters speckle compartments by default. This could in turn explain why mutations in lamins A/C and LAP2α both cause diseases associated with premature aging (). HDFs from a needle biopsy of the forearm were cultured in DME (Invitrogen) supplemented with 10% newborn calf serum (NCS) and 10 U/ml penicillin plus 50 μg/ml streptomycin, at 37°C in humidified incubators containing 5% CO. HDFs from a patient with a homozygous Y259X LMNA mutation were obtained as an autopsy sample after an informed consent. Cultures were grown to 70–80% confluence and subcultured thereafter at a seeding density of 3 × 10 cells per 75 cm flasks. In this study, control and Y259X HDFs were used between passages p8 and 12. To induce quiescence by serum starvation, control HDFs were grown for 3 d in complete medium and maintained in starvation medium (0.5% NCS) for 5 d. To induce cell cycle reentry, quiescent HDFs were serum restimulated (10% NCS) for 6, 12, 18, or 24 h and prepared for immunofluorescence microscopy. To induce quiescence by contact inhibition, HDFs were grown in complete medium (10% serum) for 7 d (confluent stage) or for 10 d (postconfluent stage). Immunofluorescence was performed according to established laboratory procedures (). The primary antibodies used and their dilutions are described in . Secondary antibodies were donkey anti–mouse and anti–rabbit IgG conjugated to rhodamine (TRITC) or fluorescein (FITC; Strata-Tech) and, for viewing, DNA in cell coverslips were mounted in DAPI. For imaging cells, a confocal microscope imaging system (Radiance 2000; Bio-Rad Laboratories) with LaserSharp software (Bio-Rad Laboratories) or confocal microscope imaging system (LSM 510 META; Carl Zeiss MicroImaging, Inc.) with LSM510 image browser software (Carl Zeiss MicroImaging, Inc.) were used at ambient temperature, equipped with 40×/1.3 and 63×/1.4 oil-immersion lens and nonimaging photodetection device (photomultiplier tube; Carl Zeiss MicroImaging, Inc.). The imaging medium used was immersion oil (Immersol 518; Carl Zeiss MicroImaging, Inc.). A dynamic range adjustment was used to optimize the signal for the fluorophores, and images were collected in sequential mode (Bio-Rad Laboratories) or multitrack mode (Carl Zeiss MicroImaging, Inc.). Any brightness and contrast adjustments were performed in Photoshop (Adobe). In situ nuclear matrix extractions using sequential treatment with detergents, nucleases, and salt were performed as described by . After extraction, cells were prepared for immunofluorescence microscopy. HDF cell pellets were washed with ice-cold PBS and lysed in 0.1 ml of ice-cold hypotonic buffer per 10 cells (10 mM Tris, pH 7.4, 10 mM KCl, 3 mM MgCl, and 0.1% Triton X-100), containing protease inhibitor cocktail and 100 U/ml of RNase-free DNase I (Sigma-Aldrich) for 10 min on ice. Cell lysates were analyzed by SDS-PAGE. Alternatively, before the above, cell pellets were subjected to sequential extraction according to the protocol of using ice-cold buffers and freshly added protease inhibitor cocktail. 1D SDS-PAGE was performed according to . For immunoblotting, proteins separated on gels were electrophoretically transferred onto nitrocellulose membranes (Schleicher & Schuell) using the Mini Trans-Blot system (Bio-Rad Laboratories) and processed according to standard protocols (). Secondary antibodies were donkey anti–mouse or donkey anti–rabbit IgG conjugated to HRP (Jackson ImmunoResearch Laboratories). For the immunological detection of proteins, membranes were incubated in ECL reagents (GE Healthcare) and visualized using either LAS-1000 intelligent dark box (FujiFilm) or autoradiography. Densitometry of signals obtained for the protein bands was performed using Image Gauge analysis software (FujiFilm). LAP2α- and lamin A/C–specific siRNA duplexes were obtained from Ambion. The sequences were selected from the open reading frames to obtain 21-nt sense and 21-nt antisense strand with symmetric 2-nt 3′overhangs of identical sequence. The sequences of each strand of siRNA oligos were as follows: LAP2 sense, 5′-GCUAAGAAAGUACAUACUUtt-3′; LAP2 anti-sense, 5′-AAGUAUGUACUUUCUUAGCtg-3′; lamin A/C sense, 5′-CUGGACUUCCAGAAGAACAtt-3′; and lamin A/C anti-sense, 5′-UGUUCUUCUGGAAGUCCAGtt-3′. RNAi transfection procedure was modified from . On the day of transfection, cells were seeded at 5 × 10 cells/well in 6-well plates in the presence of 10% NCS and no antibiotics and transfected in tandem with specific or control (scrambled) siRNAs using Oligofectamine reagent (Life Technologies). 24 h after transfection, medium was replaced by fresh medium (10% NCS) without antibiotics. Cells were assayed 48–72 h after transfection. Transfection efficiency was determined by immunofluorescence microscopy and immunoblotting. Specific silencing of LAP2α or lamin A/C was confirmed by four independent experiments. HDFs were synchronized by serum starvation and restimulation. Cultures were trypsinized, counted, and transfected with LAP2α, lamin A/C, or control siRNA. Transfected cells were harvested by trypsinization after 0, 24, 48, or 72 h and resuspended in PBS and methanol prechilled at −20°C (1:9 ratio). Subsequently, washed cell pellets were incubated in PBS containing 100 μg/ml RNase and 25 μg/ml propidium iodide, washed in PBS, centrifuged, and diluted in PBS for cell cycle analysis on a FACSCaliber flow cytometer (Becton Dickinson). Data were collected as DNA histograms from 5,000 single-cell events, and cell cycle phase distribution (percentage of G, S, and G/M cells) was determined by the Dean/Jett/Fox model using FlowJo software. Fig. S1 shows expression and solubility properties of confluent HDFs Fig. S2 shows colocalization of RbS795 foci with splicing speckles in confluent HDFs. Fig. S3 shows colocalization of RbS795 foci with splicing speckles in Y259X HDFs. Online supplemental material is available at .

In mammalian cells, the interrelationship between centrosomes and cell cycle is multifaceted. Because the activities of the centrosome are temporally linked to, and dependent on, cell-cycle progression, the centrosome was traditionally thought to be controlled by the cytoplasmic changes accompanying progress through the cell cycle. For example, the number of microtubules that are nucleated by centrosomes start out low in early interphase and increase markedly as the cell approaches mitosis. The amount of γ-tubulin at the centrosome, and the number of microtubules that grow from it in vitro in lysed cell models, increases as the cells approach mitosis (; ; ; ). Also, the precise duplication of the centrosome is initiated at the onset of S phase by the rise in the activities of cyclin-dependent kinase 2 coupled to cyclin E and/or A, the kinase complexes that drive the cell into S phase (). However, the centrosome is more than just a follower of the cell cycle. Evidence has been accumulating that the centrosome has an activity that is essential for the cell to progress through G1 and enter S phase. The first indication came from the finding that microsurgical removal of the interphase centrosome from BSC-1 cells did not prevent the acentrosomal cells from entering mitosis, but almost all of them arrested in G1 thereafter (; ). Similarly, after laser ablation of one centrosome at metaphase, CV-1 cells divided but the daughter cells that inherited no centrosome arrested in G1 (). Subsequent work indicated that acentrosomal BSC-1 cells after mitosis arrest with elevated levels of p21, an absence of the Ki-67 proliferation antigen, and hypophosphorylated retinoblas toma protein, which imply an early G1 arrest involving p53 (; unpublished data). In contrast, removal of centrosomes from HeLa cells does not block G1 progression (). However, these are transformed cells with dysfunctional G1 controls caused by the expression of human papillomavirus proteins E6 and E7 (for review see ). Importantly, several recent studies report that the knockdown or displacement from the centrosome of a variety of proteins associated with the centrosome leads to a p53-dependent G1 arrest of a large proportion of the cell population (for reviews see ; ,; ). Together, these studies point to a role for the centrosome in the mechanisms that control the untransformed cell's progress through G1 into S phase. The way in which the centrosome influences G1 progression in untransformed cells is a mystery because such a wide variety of seemingly disparate experimental perturbations all lead to a G1 arrest. Possibilities include a novel checkpoint that monitors centrosome absence or damage, disorganization/dysfunction of the interphase cytoskeleton, and disabling of the centrosome's possible role in promoting the efficiency of signaling reactions that may be necessary for G1 progression (; ; ,; ). These possibilities, however, are presently ideas awaiting experimental investigation. Given our previous observation that HeLa cells can progress through G1 without a centrosome (), we investigate the consequences of centrosome removal in normal human cells. We were particularly interested in determining if untransformed human cells without a centrosome can progress through G1. We identified RPE1 cells that were in G1 by the presence of two bright focal centrin-1/GFP spots (centrioles); in phase contrast, we cut between the nucleus and the centrioles with a glass needle, as previously described (; ), to form an acentrosomal cell and a centriole-containing cytoplasmic fragment called a cytoplast (). 15–30 min after the operation, cytoplasts were examined for ∼1 s in fluorescence to confirm the presence of the centrioles (). Individual acentrosomal cells were followed with phase-contrast time-lapse video recordings after the coverslips were transferred from micromanipulation preparations to closed chambers, which allow cells to proliferate normally for at least 100 h, or until confluency is reached (). Before we fully describe the cell cycle progression of acentrosomal RPE1 cells (see next section), we first describe our observation that untransformed cells assemble centrioles de novo. Surprisingly, we found that acentrosomal RPE1 cells progressed through interphase and divided one or more times. During the first interphase and after the first or second mitosis, the cells contained 0–6 puncta of centrin-1 (, and ). The bright focal appearance and variable number of centrin foci is characteristic of de novo centriole assembly (; ). Serial section electron microscopy of three acentrosomal cells previously followed in vivo revealed that the bright centrin foci assembled de novo corresponded to morphologically normal centrioles (). HMEC cells also formed bright centrin foci in the first interphase after G1 ablation of both centrioles. Serial section electron microscopy of two of these cells confirmed the de novo formation of centrioles (unpublished data). To determine when acentrosomal cells start to assemble centrioles de novo, we cut RPE1 cells in G1, added BrdU to the medium, and examined them at various times thereafter. We observed the formation of 2–7 centrin foci (precentrioles) starting ∼9 h after the microsurgery (), which was temporally coincident with S phase as determined by BrdU incorporation. To test if the early, and perhaps invisible, formation of precentrioles occurs during G1 in RPE1 cells, we microsurgically removed the centrosome from 15 cells that were pretreated with 1 mM mimosine to arrest them in G1 (; ). Because precentriole maturation into morphological centrioles is a time-dependent process in HeLa cells (), there would be sufficient time for nascent precentrioles to mature and become readily visible in acentrosomal cells arrested in G1. 11 acentrosomal cells remained arrested in G1 for at least 24 h and, with one exception, none contained any visible centrin foci. The other four progressed into S phase, and all contained two or four centrin foci; these serve as internal controls, demonstrating that mimosine does not have an activity that shuts down the de novo centriole assembly pathway. Separately, we laser ablated the centrosome in five G1 RPE1 cells, and then hit the nucleus with the laser to induce DNA damage to hold the cells in G1. All arrested in interphase for at least 72 h, and none formed centrin foci. Together, these observations indicate that precentriole formation occurs in S phase, and thus, the G1 progression of acentrosomal cells reported in the next section is not supported by the presence of precentrioles. In 38 trials, all acentrosomal cells, even those produced 1 h after the completion of mitosis, progressed through interphase, entered mitosis (mean 15 h after cut), and completed cleavage into two daughters in a normal fashion (). However, instead of arresting in G1 after mitosis, as we would have expected from previous studies (), out of 72 daughters that stayed in view, 2 arrested in G1, 5 progressed into S phase, and 65 divided at least one more time within 48 h (). To determine if G1 progression after centrosome removal is peculiar to microsurgery, we used laser ablation of the centrosome, which destroys only a small volume of the cell. Using a spinning disk confocal at low laser power setting (107 μW output at the objective lens) to visualize the GFP-tagged centrioles, we found that 7/8 RPE1 cells progressed through interphase to mitosis after G1 ablation of both centrioles. We also conducted G1 centrosome ablations on a p53-positive clone of HMECs expressing centrin-1/GFP to tag the centrioles. Using the same confocal power setting to visualize the GFP-tagged centrioles, we found that after ablation of both centrioles during G1, all 16 cells progressed through interphase into mitosis (, top, line A). In these experiments, we used pairs of sister G1 cells in which one received a cytoplasmic control ablation and the other a directed centrosome ablation; we found that the time from ablation to mitosis was the same (control irradiated cell mean = 26.9 h, = 15; experimental cell mean = 26.8 h, = 16). Together, these results demonstrate that G1 progression without a centrosome is not specific to the type of untransformed cell or the means used to remove the centrosome. Furthermore, when we induced physical damage to the centrosome by ablating one G1 centriole, three fourths of the cells progressed through interphase to mitosis (, top, line A). It is formally possible that we removed the centrosome from G1 cells after the point at which the centrosome becomes dispensable for cell cycle progression. To directly test whether or not RPE1 acentrosomal cells can progress through G1 in its entirety without a centrosome, we removed one of the two centrosomes from cells during late S–G2 (after centriole replication). The de novo pathway is inhibited as long as cells contain even a single centriole (), and cells that enter mitosis with a single centrosome will divide into two daughters, one inheriting a centrosome and the other entering G1 without a centrosome (; ). 4 out of 20 acentrosomal daughter cells arrested in G1 after the first mitosis, as determined by lack of BrdU incorporation. Six acentrosomal daughters progressed into S phase, but did not enter mitosis within 36 h. The remaining 10 progressed through interphase and through the next mitosis (). An example of such an acentrosomal cell progressing from one mitosis to the next is shown in (an additional example is shown in Fig. S1, available at ). These observations reveal that 80% of the acentrosomal daughter cells progress through the entirety of G1 without a centrosome. Because precentrioles do not form during G1 (; this study), the G1 progression we observed in this study is not supported by the assembly of precentrioles. This notion is further supported by one case of a G2 microsurgery in which we observed that the daughter cell “born” without centrioles progressed through the next mitosis, and no centrin foci were observed in the granddaughter cells. These observations are in clear contrast to previous reports that centrosome removal or the knockdown/displacement of a wide variety of centrosomal proteins lead to a G1 arrest in untransformed cells (for reviews see ; ,). Insight into a possible reason for these fundamentally different observations was first suggested by our anecdotal observations that prolonged exposure of acentrosomal cells to 488-nm blue light, which is used to excite GFP, correlated with a G1 arrest, whereas control cells in the same microscope field continued through multiple cell cycles. This observation led us to ask if loss of the centrosome, by itself, is a stress for the cell, and if any additional stress (in this study, blue light) causes it to arrest in G1. To test this notion, we used microsurgery of G1 cells to produce acentrosomal cells, and, in the same microscope field, performed control amputations of equivalent cell areas on other cells. The untouched cells in the same fields served as controls. 30 min after the cutting operations, we exposed the field of cells to various durations of 488-nm blue light (18 nW/μm at the field plane; 580 μW output at the objective lens) to controllably stress all the cells. The field was then followed by time-lapse microscopy to determine which cells progressed through interphase into mitosis and which did not. We used blue light as an exogenous stress, because it is deleterious to cells and dosages can be precisely controlled by computer control of the shutter on the epifluorescence pathway. Our results do not depend on knowing the details of how blue light stresses a cell; we use blue light only as an experimental tool. In this regard, other stressors, such as pH and composition of the media, could, in principle, be used in our application. Our results, which are summarized in , reveal that the acentrosomal cells are most sensitive to blue light–induced stress, the control-amputated cells are sensitive, but less so, and the untouched control cells are not affected by blue light exposures within the range we used. The brevity of the blue light exposures that lead to a G1 arrest of acentrosomal cells and control-amputated cells reveals how sensitive they are to blue light, relative to the untouched controls. All untouched control cells exposed in G1 to 20–40 s exposures of blue light progress through interphase to mitosis ( = 34). Stresses such as UV light, heat, and osmotic shock result in activation of the MAP kinase p38, which in turn leads to G1 arrest by influencing cyclin D1 stability, as well as the phosphorylation of p53 and pRb (for reviews see ; ; ). To assess the involvement of the p38 MAPK pathway in a G1 arrest of acentrosomal cells, we added SB 203580, which is an inhibitor of the p38 stress kinase (), to the medium 30 min before the G1 cuts were made. 30 min after the microsurgery, the fields were exposed to blue light for either 2 or 4–5 s. We found that none of the acentrosomal cells or control amputees arrested in G1, even after 4–5 s exposures (, bottom). We also investigated how blue light exposure influences the G1 cell cycle progression of acentrosomal HMEC cells when two intensities of blue light were used to image centrosomes and monitor their ablation (the percentage of cells arresting in interphase under various conditions is summarized in ). At the lower blue light level (107 μW output at the objective lens), G1 laser irradiation of the cytoplasm adjacent to the centrosome (no damage to centrosome), damage to the centrosome in the form of ablation of one centriole, or the ablation of both centrioles did not give a substantial incidence of cell cycle arrest. Only one cell arrested (, top, line A). In contrast, at an approximately fourfold higher blue light level used to observe the cells (450 μW output at the objective lens for the same total amount of time), laser irradiation during G1 of the cytoplasm adjacent to the centrosome (control irradiation) led to an interphase arrest in approximately one third of the cells (, top, line B). Ablation of one or both centrioles during G1 arrested two thirds of the cells in interphase. To determine if activation of the p38 stress-activated kinase plays a role in these observed interphase arrests at the higher blue light level, we repeated these ablations with cells continuously exposed to the p38 inhibitor SB 203580. Cytoplasmic laser irradiations, ablation of one centriole, or ablation of both centrioles did not lead to a G1 arrest in any of the cells (, top, line C). Together, these results indicate that physical damage to the centrosome, or its complete ablation, promotes a p38 stress-activated kinase–mediated interphase arrest when the HMEC cells are additionally stressed by blue light. We previously reported that after microsurgical removal of the centrosome during interphase, BSC-1 cells progressed through mitosis and 88% arrested in G1 after that mitosis (). To test how our previous results fit with our current findings, we reinvestigated the consequences of microsurgical removal of the interphase centrosome from BSC-1 cells using our current methodology. Our current methods involve several system upgrades, such as the use of a mechanically more stable micromanipulator and more sensitive video cameras that allow ∼64-fold lower green light (546 nm) intensities for time-lapse imaging (4.7 nW output from the condenser vs. 302 nW condenser output previously used). Also, after microsurgery, we now remount the cell bearing coverslips into sealed observation chambers () for time-lapse observations, rather than leaving them in oil-capped micromanipulation preparations. The sealed chambers contain an approximately threefold higher volume of medium (600 μl). We cut a BSC-1 cell to remove the centrosome, and we performed a control amputation of cytoplasm from another in the same field of view. The untouched cells served as controls. For some experiments, the coverslips were transferred after the microsurgery to sealed observation chambers, as we have done after the microsurgery of RPE1 cells. For other experiments, we left the cells in the oil-capped micromanipulation preparations for time-lapse observations. Using our current observation conditions, we found that 14% of the acentrosomal cells arrested in interphase after mitosis, whereas none of the control amputation or untouched controls arrested in interphase (). When the cells were left in the micromanipulation chambers for time-lapse filming, 33% of the acentrosomal cells and 13% of the control-amputated cells arrested in interphase after mitosis; none of the untouched controls arrested. To test if acentrosomal BSC-1 cells are sensitive to the level of continuous green light used for time-lapse observations, we performed the same experiments, but raised the illumination intensity to 1,170-nW condenser output (3.8-fold higher than the study). Our results () show that for both the sealed and oil-capped micromanipulation chambers used for filming, a higher percentage of the acentrosomal cells and control cut cells arrest in interphase after mitosis under these higher green light conditions. Notably, none of the untouched control cells arrested under any of these conditions. De novo centriole assembly after centrosome removal has been, as of yet, observed only in transformed cells (; ). The fact that de novo centriole assembly was not found in untransformed cells suggested that transformation abrogates the normal limits on spontaneous centrosome assembly (; ; ). We demonstrate that de novo centriole assembly is a general phenomenon for mammalian somatic cells, not a peculiarity of cell transformation. In RPE1 and HMEC cells, this process appears to have the same characteristics as in HeLa cells. For centrosome removal during G1, a variable number of GFP fluorescent centrin foci (called precentrioles) appear at the time of first S phase, become brighter with time, and eventually develop into morphologically recognizable centrioles. Although the number of centrioles assembled de novo in RPE1 cells is variable, we note that fewer (up to 6) form in these cells than in HeLa cells (up to 14). For RPE1 cells, we observed 40 cases in which the daughters of acentrosomal cells contained two centrioles after the second mitosis. This could point to the existence of a mechanism that controls centriole copy number or reflects the de novo assembly of a single centriole that later duplicates. Lastly, we observed four acentrosomal RPE1 cells and seven acentrosomal HMEC cells that did not form bright centrin foci in the first cell cycle, something not seen in the HeLa experimental system (). A surprising aspect of our study was the finding that normal human cells progress through G1 in its entirety without a centrosome, as long as they are not subjected to exogenous stress, such as 488-nm light. Cells progress through G1 to mitosis whether the centrosome is removed early in G1 or the cells are “born” at the end of mitosis without a centrosome. The identical behaviors of RPE1 and HMEC cells after centrosome removal indicate that our results are not specific to cell type or the means used to remove the centrosome. Although these cells assemble centrioles de novo during S phase after centriole removal, we tested for the early stages of centriole formation during G1 by grossly prolonging this cell cycle phase, and we could not find the formation of any centrin foci indicative of precentrioles. Thus, it does not appear that progression through G1 in acentrosomal cells was supported by the formation of precentrioles soon after mitosis. Our finding that 1 RPE1 and 5 HMEC acentrosomal cells went through more than a complete cell cycle without any centrin foci and no centriole structures were found by serial section electron microscopy (HMEC; = 3; unpublished data) provides additional evidence that G1 progression was not supported by precentriole assembly. To gain insight into the apparent difference between our present results and those of , we characterized the behavior of microsurgically produced BSC-1 acentrosomal cells and control-amputated cells using our current experimental conditions. Our observations reveal that acentrosomal BSC-1 cells behave in a qualitatively similar fashion to acentrosomal RPE1 and HMEC cells. BSC-1 cells can progress through G1 without a centrosome under our current conditions, but not at as high a frequency as RPE1 cells. Together, our observations indicate that G1 progression in BSC-1 cells may be more sensitive to the loss of the centrosome than in RPE1 cells, and micromanipulation chambers provide a less favorable environment than sealed chambers for the G1 progression of BSC-1 cells that have been stressed by microsurgery and loss of the centrosome. Also, the previous use of higher green light intensities for time-lapse observations than we currently use may have contributed to the previously observed G1 arrest of acentrosomal cells. Our finding that the centrosome is not needed for G1 cell cycle progression raises the question of why knockdown or displacement of a variety of centrosomal proteins in untransformed cells leads to a G1 arrest in a substantial proportion of the population of cells (; ,; ). We envision two possibilities to initially consider separately. The first is that the cell has a mechanism that can detect damage to the centrosome and/or sense-compromised centrosome function (; ,). If some portion of the centrosome must remain present to act as a signaling platform to trigger the p53 pathway, complete centrosome removal would not be sensed by the cell as centrosome damage or dysfunction. Although this reasoning can explain why experimental centrosomal protein knockdowns could arrest cells in G1, one wonders if cells in an organism would face the loss or disabling mutation of centrosomal proteins often enough to drive the evolution of a distinct checkpoint. Thus, we raise the speculative possibility that a centrosome-damage–sensing mechanism could have evolved to deal with other cellular defects, and that it uses the centrosome as a device to stop the cell cycle until the problem is resolved. For example, when cells are exposed to heat, which is a common environmental hazard, the proteotoxic stress causes centrosomes to lose some of their proteins, which can lead to abnormal mitosis (). Also, irreversibly damaged proteins can accumulate at the centrosome as an “aggresome” to be degraded by proteosomes concentrated there (). There is evidence that excess protein accumulation at the centrosomes leads to the fragmentation of the centrosomal microtubule-organizing center and the consequent generation of multipolar spindles if the cell enters mitosis (). In this scenario, possible disruption of the centrosome by denatured protein accumulation could lead to a G1 arrest until the damaged proteins are degraded and the centrosome is restored to its intact state. The second possibility is that damage to the centrosome or its removal leads to centrosome dysfunction, which is a stress for the cell that can act additively with other stresses to trigger a p38-p53 response. Normal mammalian cells in G1 monitor a variety of intracellular and extracellular conditions that, together, lead to a commitment, or not, to enter the cell cycle (; ; ; ; ). For example, the cell is sensitive to serum growth factors, cell–cell contacts, and substrate adhesion. In addition, even slight perturbations of the actin cytoskeleton can cause a durable G1 arrest (; ; ; ). Restrictions on cell spreading and disassembly of the interphase microtubule array have also been reported to arrest untransformed cells in G1 in a p53-p21–dependent fashion (; Sablina et al., 2001; ). We tested this stress hypothesis and obtained functional evidence that centrosome removal significantly sensitizes both RPE and HMEC cells to exogenous stress, leading to a p38-dependent arrest. Acentrosomal RPE1 cells are more sensitive to blue light–induced stress than control-amputated cells, and both are substantially more sensitive than the untouched controls in the same fields. Neither the loss of the centrosome nor short exposures to blue light acting singly is sufficient to cause a G1 arrest, but they can work additively to tip the balance toward such an arrest. Furthermore, the use of two intensities of blue light to observe HMEC cells after laser ablation of the centrosome provides additional evidence that partial or complete centrosome removal makes G1 progression more sensitive to exogenous stress, in this case blue light. Together, our results provide assurance that the phenomena we observe are not dependent on the means used to remove the centrosome or the specific cell type. We used blue light as the exogenous stressor, but recognize that a wide variety of other suboptimal conditions could also act in concert with centrosome loss or damage to cause a G1 arrest. Indeed, the recent finding that siRNA depletions of the centrosome-associated proteins PCM1 or pericentrin lead to a p38-dependent G1 arrest of only ∼50% of the cells () led these authors to propose that the arrest is not a specific centrosome-dependent cell cycle control, but rather, a stress-driven response to centrosome defects. Centrin-1/GFP in pEGFP-N1 vector () was transfected into telomerase-immortalized normal human cells (hTERT-RPE1; CLONTECH Laboratories, Inc.), and stable clones were isolated via G418 selection and limited-dilution cloning. Cells were cultured as described in . HMEC clone 184A1 (wild-type for p53 and pRb) were obtained from M. Stampfer (Lawrence Berkeley National Laboratory, Berkeley, CA). These cells were transfected with centrin-1/GFP in LentiLox3.7 vector as directed (). HMEC cells were grown according to protocols available at . BSC-1 cells were obtained from the American Type Culture Collection. The cells were cultured in MEM containing 12.5 μM Hepes, 10% fetal calf serum (Invitrogen), 100 U penicillin (Invitrogen), and 100 μg streptomycin (Invitrogen). BrdU (Sigma-Aldrich) was added to a final concentration of 5 μg/ml immediately after centrosome removal from G1 cells or just after first mitosis after centrosome removal from G2 cells. The incorporation of BrdU was determined as previously described (). 1 mM mimosine (Calbiochem) was prepared from a 10-mM stock solution in culture media (; ). SB 203580 (10 μM [Sigma-Aldrich] for RPE1 and 20 μM [Calbiochem] for HMEC) was prepared by dilution of a 10-mM DMSO stock into culture media. Glass-needle microsurgery and phase-contrast time-lapse recording were conducted as described in and . In brief, coverslips bearing cells were assembled into open-faced micromanipulation chambers filled with culturing media kept in the CO incubator for several hours to equilibrate the pH, and capped with mineral oil (Sigma-Aldrich). Microsurgery was performed at 37°C with a custom-built piezoelectric micromanipulator on an ACM microscope (Carl Zeiss MicroImaging, Inc.) equipped with phase-contrast optics and epifluorescence. First, the position of centrin-1/GFP dots was confirmed by a few exposures to 488-nm blue light. The cell was cut between the nucleus and the centrosome with a glass microneedle under phase-contrast optics. 15–30 min after the operation, the removal of centrosome was confirmed by, at most, a 1-s exposure to 488-nm blue light. After the microsurgery, the acentrosomal cell was circled with a diamond scribe, and the coverslip was removed from the manipulation chamber and assembled into a closed chamber () with buffered culture media containing BrdU, and then followed at 37°C with Universal (16×, NA 0.32 objective; Carl Zeiss MicroImaging, Inc.) or BH-2 (10×, NA 0.3 objective; Olympus) microscopes equipped with phase-contrast optics. Images were recorded with Orca ER (Hamamatsu), Orca 100 (Hamamatsu), 1300 (Retiga), 2000R (Retiga), or EXi (QImaging Corp.) cameras. Image sequences were written to the hard drives of PC computers using C-imaging software (Compix, Inc.) and were exported as .avi movies. Centrin-1/GFP foci in the acentrosomal cells were characterized by z series (20 optical planes separated by 0.2 μm or 8 optical planes separated by 0.5 μm) fluorescent imaging with a microscope (DMR; Leica; 63×, NA 1.32 objective or 100×, NA 1.30). Maximal intensity projections were compiled with SlideBook software (Intelligent Imaging Innovations, Inc.). The light intensities at each wavelength were measured as total output at the stage level with a LaserMate-Q laser power meter (Coherent, Inc.). Laser microsurgery was conducted on a custom-assembled microscopy workstation centered on a microscope (TE2000-E2; Nikon). 532-nm, 8-ns laser pulses were generated by a Q-switched Nd:YAG laser (Diva II; Thales Lasers, Paris, France) run at 20-Hz repetition rate. Collimated laser beam was expanded to ∼8 mm to fill the aperture of a 100× 1.4 NA PlanApo lens and delivered through a dedicated epi-port. It takes ∼10 laser pulses (1 s) to destroy the centrosome. Fluorescence images were recorded with a Cascade512B back-illuminated EM-CCD camera (Photometrics) attached to the left microscope port (100% transmission) in confocal mode (spinning disk confocal; Perkin-Elmer). 3D datasets were taken at 0.25-μm z steps. All light sources were shuttered by either fast mechanical shutters (Vincent Associates) or AOTF (Solamere Technology Group) so that cells were exposed to blue light only during laser operations and/or image acquisition. The system was driven by IP Lab software (BD Biosciences). After laser ablation of the centrosome, the position of the experimental cell was marked with a diamond scribe and filmed as previously described (). Fixation, embedding, and serial sectioning were performed according to established procedures (). 100-nm sections were examined in a microscope (910;Carl Zeiss MicroImaging, Inc.) at 100-KV and photographed on film. Film negatives were subsequently scanned and contrast-adjusted in PhotoShop CS (Adobe). Fig. S1 shows the progression of a RPE1 cell born without a centrosome through G1. Online supplemental material is available at .

The loss of functional adenomatous polyposis coli (APC) is an early event in the development of most polyploid colorectal cancers (; ). Mutated APC has been linked to genetic instability, which is an early hallmark of colorectal tumors and a factor that greatly promotes tumor development (; ; ). Exactly how APC is involved in maintaining chromosomal stability remains unknown. APC performs multiple roles, including negative regulation of the Wnt signaling pathway (; ), organization of the cytoskeleton, and regulation of cell migration (; Zumbrunn et al., 2001; ; Nathke, 2005). This functional diversity places APC in an important position in maintaining gut epithelia, but it is not clear how it relates to the function of APC in safeguarding a normal karyotype. The formation of a protein complex between APC and the spindle assembly checkpoint proteins Bub1 and BubR1 and the ability of these kinases to phosphorylate APC in vitro () raise the intriguing possibility that APC “talks” directly to the mitotic checkpoint machinery. Insufficient microtubule plus end attachment resulting from the expression of dominant, truncated fragments of APC inhibits chromosome congression at metaphase and results in abnormal chromosome segregation (; ). Consequently, the overexpression of N-terminal APC fragments (like those commonly found in tumors) in cells with wild-type APC can lead to premature exit from mitosis and aneuploidy (). The loss of heterozygosity in the APC locus, which initiates most colorectal tumors, has two direct consequences: the loss of normal functional APC and the expression of a truncated N-terminal APC fragment. It is likely that both the absence of functional APC and the presence of N-terminal APC fragments contribute in separate but interactive ways to the phenotype of APC-deficient cancers. The majority of previous work examined the dominant effects of N-terminal APC fragments commonly found in tumors (; ; ). In most cases, the effects of N-terminal fragments were assessed in cells that, unlike tumors, also express wild-type, full-length APC. The interpretation of these data requires an understanding of the effects produced by the absence of APC in order to distinguish the effects of N-terminal APC fragments from the effects that result from loss of APC itself. Therefore, we specifically addressed the role of the full-length APC molecule by depleting full-length APC in several different systems. Using egg extracts, we previously showed that lack of APC causes mitotic spindles to have several defects, including a disorganized microtubule network with reduced total microtubule mass, particularly in the midspindle area (). In the present study, we show that in addition to spindle defects, the mitotic spindle assembly checkpoint is not functioning properly in cells lacking APC. Consequently, we find that the loss of APC leads to the accumulation of tetraploid cells. Additionally, we observed decreased apoptosis in APC-deficient cells. We propose that this combination of defects can aid in the longevity of cells with abnormal DNA content to promote polyploidy. Importantly, we provide direct evidence for the appearance of tetra- and polyploid cells extremely early after APC is inactivated not only in cultured cells but also in gut tissue. Additionally, we eliminate a major role for β-catenin signaling in the observed effects by showing identical defects in HCT116 cells, which carry an activating mutation in β-catenin. To address how the lack of APC affects mitotic spindle function in cells, we removed APC from human osteosarcoma U2OS cells using RNAi. Transfection of 5 nM siRNA targeting APC reduced the level of APC protein to 14.7 ± 8.6% of that in cells transfected with nontargeting control siRNA (). The increase in β-catenin in cells treated with anti-APC siRNA further confirmed the reduction of functional APC in such cells (). Mitotic spindles in cells lacking APC were not grossly perturbed. However, the inability of such spindles to create full tension at kinetochores showed compromised spindle function (). The distance between sister kinetochores in prophase, when there is no bipolar attachment and therefore no tension, was almost identical in APC-inhibited and control cells (0.74 ± 0.03 μm and 0.70 ± 0.03 μm [mean ± SEM], respectively). In control cells, the distance between sister kinetochores increased to 1.38 ± 0.02 μm in metaphase, indicating the development of tension. Notably, interkinetochore distance in metaphase chromosomes of APC-deficient cells only reached 1.15 ± 0.02 μm, which is significantly different (P < 0.001; Mann-Whitney rank sum test and test) from the control metaphase dataset. This demonstrates that metaphase kinetochores in APC-deficient spindles were under reduced tension. In mouse fibroblasts in which APC expression can be conditionally inactivated (), APC removal produced the same effect (Fig. S1, available at ). Thus, the loss of APC leads to defects in the mitotic spindle that are translated into reduced tension on metaphase kinetochores. Insufficient interkinetochore tension induced by APC loss should lead to the accumulation of mitotic checkpoint proteins like Bub1 and BubR1 on kinetochores. Therefore, we quantitated the amount of kinetochore-associated Bub1 and BubR1 in APC-deficient and control mitotic cells by measuring the intensities of Bub1 and BubR1 immunofluorescence at kinetochores, which were defined by costaining with CREST (; also see Materials and methods; ). Bub1 and BubR1 accumulate at kinetochores at prophase and gradually disappear throughout mitosis, with only minimal kinetochore localization at metaphase and anaphase (; ). Our measurements confirmed these dynamics of Bub1 and BubR1 at kinetochores (). However, despite insufficient tension at kinetochores, unsynchronized U2OS cells transfected with APC-targeting siRNA accumulated 1.84-fold less Bub1 per cell (mean reduced by 45.6%, which is significant; P < 0.001 in a two-tailed test) and 1.66-fold less BubR1 (mean reduced by 39.7%, which is significant; P < 0.01 in a two-tailed test) at kinetochores during prometaphase than cells transfected with control siRNA (). The difference was still detectable at the prometaphase→metaphase transition and in metaphase cells; however, it was not statistically significant at these stages. We also used fibroblasts from mice with constitutively reduced APC protein () and compared them with fibroblasts isolated from wild-type littermates (). Again, we found that APC-deficient cells had less Bub1 and BubR1 at kinetochores than control cells (). In prometaphase, the mean kinetochore-associated Bub1 intensity per cell in APC-deficient fibroblasts was reduced 1.65-fold (39.3% of that in wild-type cells, which is significant; P < 0.05 in a two-tailed test; mean kinetochore-associated BubR1 per cell was reduced 2.37-fold [58.3% of wild-type counterpart], which is significant; P < 0.001 in a two-tailed test). Thus, two independent APC-deficient cell systems showed a decrease in the amount of both Bub1 and BubR1 at kinetochores. Monitoring the mitotic progression of logarithmically growing APC-inhibited and control U2OS cells revealed that inhibiting APC resulted in a decrease in the time from entry into mitosis to the onset of anaphase (27.9 ± 1.1 min in control and 23.3 ± 0.6 min in APC-inhibited cells), which was statistically significant (P < 0.005 by a Mann-Whitney rank sum test; ). To assess mitotic checkpoint function, we treated cells with microtubule poisons that induce mitotic arrest if the mitotic spindle checkpoint is intact and counted phosphohistone H3–positive cells. Histone H3 is specifically phosphorylated in mitosis (), and this event is commonly used as a marker for mitotic cells. We observed that in U2OS cells, both nocodazole and taxol treatment led to a dose-dependent accumulation of cells in mitosis (; iContr). Inhibition of APC reproducibly resulted in a reduction in the number of mitotically arrested cells after 20 h of treatment with a broad range of concentrations of both taxol and nocodazole compared with cells transfected with control nontargeting siRNA (). This was not caused by a G2 block or delay, which would prevent cells from entering mitosis, as G2 progression in presynchronized cells was not altered by APC removal (Fig. S2, available at ). Thus, APC inhibition does not completely obliterate but substantially compromises the mitotic checkpoint in U2OS cells. One of the consequences of a damaged mitotic checkpoint is the inappropriate exit of cells from mitosis into G1 to produce tetraploid cells. To distinguish cells that had exited mitosis with a 4n DNA content from the rest of the 4n DNA population, representing a mixture of cells in G2 and M, we costained cells with 7AAD (for DNA profile) and FITC-labeled anti– cyclin B1 antibody (). The cyclin B1 level is lowest in G1, gradually accumulates throughout S and G2, and peaks in mitosis (). Thus, the 4n DNA/cyclin B1–negative population (, red) represents cells that inappropriately exited mitosis with double chromosome content. This population was clearly increased in APC-deficient cultures even in the absence of mitotic poisons (). Mitotic arrest induced by taxol or nocodazole increased the number of 4n DNA/cyclin B1–negative cells. Importantly, such tetraplolid G1 cells were 1.5–2 times more abundant after APC depletion (). Comparing cyclin B1 profiles for control and APC-negative cells arrested by taxol after synchronization (see Fig. S2 A for protocol) revealed a massive degradation of cyclin B1 in APC-negative cells when the level of cyclin B1 in control cells was only slightly decreased (). Thus, premature cyclin B1 degradation is likely to be involved in the increased mitotic slippage of APC-deficient cells. To determine the fate of APC-deficient cells that exit mitosis prematurely, we imaged U2OS cells expressing H2B–red fluorescent protein (RFP). H2B-RFP localizes to chromatin at all times, making it possible to observe chromatin structures in vivo. Observation of control ( = 175) and APC-inhibited ( = 172) cells arrested in taxol for 20 h revealed several differences between these cells. First, we found that as expected, APC inhibition shortened the time that cells remained arrested in mitosis (). The number of cells arrested for >10 h was 25% lower when APC was inhibited (88% of all mitoses in control and 66% in APC-deficient cells), and the number of cells arrested in mitosis for >15 h was halved by APC inhibition (52% in control and 27% in APC-deficient cells). This confirmed that the loss of APC induces spindle checkpoint defects. Second, careful monitoring of cells revealed two types of mitotic exit. One type () was accompanied by prolonged membrane blebbing that coincided with mild chromosomal decondensation and was followed by the tightening and fragmentation of chromatin, which are typical signs of apoptosis. Eventually, such a cell acquired an irregular shape, lost all movements, and floated (, bottom). The second typeof mitotic exit () displayed mild, if any, membrane blebbing but slight and short waves of cell reshaping. This coincided with chromatin decondensation and was followed by cell spreading. After cell spreading, decondensed chromatin remained partially fragmented for some time and contained holes that were gradually sealed (, bottom). This phenotype suggested that cells exited mitosis with a 4n DNA content and remained alive. Note that even in the absence of mitotic poison, we frequently observed misshapen interphase chromatin in APC-deficient cultures that could be the result of such alternative mitotic exit (). Quantitating these types of mitotic exit in control and APC-deficient cells revealed striking differences (). APC-deficient cells exited mitosis by the second route (spreading) twice as often as control cells, with 85% of APC-negative cells exiting in this manner, whereas only 41% of control cells spread after mitotic exit. Additionally, in a few cases, APC-deficient cells that spread after exiting mitosis initiated furrow formation and chromosome stretching () despite the presence of taxol, which prevented spindle function. The prevalence of cell spreading as a route of mitotic exit in APC-negative cells suggested that these cells survived taxol treatment better than control cells. Therefore, we asked whether APC inhibition altered apoptotic response to microtubule poisons. Consistent with this idea, we found that inhibiting APC invariably decreased the relative size of the sub-G1 DNA fraction, which is an indicator of apoptosis (). This was true for both untreated cells and cells treated with taxol at various concentrations and for different amounts of time (10 h in and 20 h in ). To measure apoptosis directly in APC-negative cells, we stained cells with fluorescently labeled antibody directed against the cleaved version of caspase 3 that is specific for cells in apoptosis and counted the number of cells positive for active caspase 3 using flow cytometry (, R2 area). All APC-negative cells described here, constitutively APC-deficient mouse fibroblasts (, top), anti-APC siRNA-treated U2OS cells (, bottom), and mouse fibroblasts with conditionally inactivated APC (Fig. S1) responded to staurosporine with a dramatic increase in the number of apoptotic cells. These data validated the antiactive caspase 3 antibody as a useful tool to detect apoptosis in these cells and indicated that the apoptotic machinery is generally functional in the absence of APC. Using this method, we found that inhibiting APC reduced apoptosis in untreated and nocodazole- or taxol-treated U2OS cells (). In addition, the baseline level of apoptosis in mouse fibroblasts that are constitutively deficient for APC (ΔAPC) was reduced in comparison with that in the wild-type control cells (). Decreased apoptotic response to virus treatment was also detected in fibroblasts in which APC expression was conditionally inactivated (Fig. S1). Together, these results show that inhibiting APC results in a decrease in apoptosis. Tetraploid cells are often eliminated by apoptosis (; ), and failure to do so would allow such cells to progress to polyploidy. The relative number of polyploid cells (>4n DNA; , blue shaded area) was up to three times higher in APC-deficient than in control cultures, which is consistent with the predicted consequence of an increased tetraploid G1 pool and decreased apoptosis. This was true for untreated or drug-treated cells (). Notably, deleting APC caused an overproportional increase in the number of polyploid cells relative to the increase in the tetraploid G1 cells (; compare with ), suggesting that relatively more tetraploid cells proceeded to cycle. This is consistent with an apoptotic deficiency in these cells and confirms that lack of APC promotes polyploidy. Using a β-catenin/T cell factor (TCF)–responsive promoter element fused to luciferase () to measure the effect of APC inhibition on the activation of β-catenin signaling, we confirmed that in HCT116 cells, APC inhibition produced a minute effect (3.8× activation) on β-catenin–activated transcription compared with the massive (1,903×) activation in U2OS cells (Fig. S3, available at ). If β-catenin/TCF target genes make a major contribution to controlling mitotic progression, the effect of APC inhibition in HCT116 cells should be minimal. However, APC inhibition in HCT116 cells resulted in quantitatively nearly identical phenotypical changes to those observed in U2OS cells. Mitotic index in both taxol- and nocodazole-arrested HCT116 cells was decreased by APC inhibition (), indicating a mitotic spindle checkpoint defect. The relative size of both the tetraploid G1 and polyploid populations were increased in mitotically arrested and unsynchronized HCT116 cells upon APC removal (). Additionally, similar to results in U2OS cells, the sub-G1 fraction was decreased by APC depletion in unsynchronized and taxol/nocodazole-treated HCT116 (), indicating apoptotic deficiency. Thus, the effects of APC on the mitotic checkpoint, apoptosis, and, consequently, on the loss of euploidy can be exerted independently of downstream targets of β-catenin. To test whether the loss of euploidy demonstrated by increased tetra- and polyploidy in APC-deficient cells in vitro was also detectable in vivo, we determined whether a similar phenomena occurred when APC was deleted from intestinal cells using Cre-lox technology. Mice bearing a lox/P-flanked APC and an inducible form of Cre-recombinase (AhCreApc) lose APC expression specifically in the gut () upon the induction of Cre-recombinase expression. At day 8 after Cre induction, we observed a subset of cells within the AhCreApc intestine with up-regulated p21, a marker for the loss of euploidy (). This was apparent as early as 3 d after induction (). The average nuclear area of these cells at day 3 () or 8 () showed an approximately twofold increase (P = 0.01; Kolmorov-Smirnov test), implying polyploidy. To confirm ploidy changes in these cells, the DNA content of intestinal epithelium from induced CreApc and CreApc mice was examined at day 6 by flow cytometry (). In CreApc epithelium, only 0.63% of cells had a DNA content >4n, whereas 9.1% of cells extracted from CreApc mice had a >4n DNA content. Thus, the loss of euploidy is also an early consequence of APC depletion in the intestinal epithelium. Together, our data provide evidence that the loss of APC in several systems rapidly and directly leads to tetra- and polyploidy both in vitro and in vivo. A combination of defects is likely to be responsible for this: first, abnormal spindles; second, a compromised spindle assembly checkpoint; and, third, a decrease in apoptosis that allows inappropriate survival of defective cells. We propose that incapacitating one single molecule, APC, creates this unique combination of defects that leads to the simultaneous disruption of several layers of cellular control () and, thus, provides a selective advantage for APC-negative tumor cells. Tetraploidy has been correlated with poor cancer prognosis (; ; ). Recent evidence suggests that it is not just a harmless side effect of cancer-promoting chromosomal instability but could be tumorigenic in its own right (). Thus, it is important to understand the mechanisms involved in producing and maintaining tetraploidy in cancer and the mechanisms that normally prevent cells from becoming tetraploid. In this study, we show that depletion of the APC tumor suppressor is sufficient to produce tetraploidy. Our data are consistent with the idea that this occurs, at least in part, via mitotic slippage with nonseparated chromosomes. When cells divide, the spindle checkpoint machinery ensures that all kinetochores are attached to spindle microtubules and are under tension. When this process is completed, the anaphase-promoting complex is activated to initiate a proteolytic cascade that, on one hand, results in chromosome separation because of the release of separase and, on the other hand, initiates mitotic exit that begins with inactivation of the CDK1–cyclin B complex and leads to cytokinesis (; ). If the mitotic checkpoint cannot be satisfied because of spindle damage, cells arrest at this stage for a limited length of time that depends on many factors and varies between cell types (). At this point, cells are either cleared by apoptosis or escape mitotic arrest to reenter G1 with a 4n DNA content. Our live imaging data () suggest that in the absence of APC, the second route for mitotic exit dominates. In U2OS cells depleted of APC by RNAi, 85% of cells that exited a taxol-induced arrest spread again and remained alive. Although U2OS cells have a relatively high spontaneous slippage rate, this number was more than doubled when APC was removed. At this time, we do not understand the mechanism of mitotic slippage well enough to clearly dissect the role of APC in this process. However, there are several clues. First, we show that the mitotic spindle checkpoint is compromised after APC removal. Mitotic progression of logarithmically grown cells was faster despite spindle damage induced by APC loss (), and mitotic checkpoint proteins Bub1 and BubR1 were not localized efficiently to the kinetochores. Importantly, the accelerated exit from drug-induced mitotic arrest in APC-deficient cells reflects a defect in this checkpoint ( and ). The reduced accumulation of Bub1 and BubR1 at kinetochores precedes and, thus, might be the cause of premature mitotic exit. Indeed, we found that the depletion of Bub1 by RNAi from U2OS cells caused a similar decrease in mitotic arrest as the depletion of APC (unpublished data). However, it is also formally possible that these events occur independently of each other. Experiments from another laboratory did not reveal a mitotic checkpoint defect in APC-depleted HeLa cells (). This discrepancy is likely caused by differences in the experimental approach used to measure such defects. Our experimental setup was able to detect a compromised mitotic checkpoint, whereas that used by was designed to only detect the complete absence of a mitotic checkpoint. Second, we observed extensive degradation of cyclin B1 in these cells after prolonged mitotic arrest. Of course, this could be a consequence of the mitotic checkpoint defect, as its abrogation would induce the cyclosome-driven degradation of cyclin B as cells exit mitosis. On the other hand, the destruction of cyclin B was previously reported to accompany mitotic slippage in mammalian cells with an active mitotic checkpoint (). At this point, we cannot distinguish whether increased cyclin B degradation in APC-deficient cells is a cause or a consequence of premature mitotic exit. Third, the failure of APC-deficient cells to die after they had escaped mitotic arrest could be related to the decreased apoptosis we detected in all APC-negative cells (). Bax-dependent apoptosis can be induced by mitotic slippage after prolonged mitotic arrest; however, this requires an intact mitotic checkpoint (). Thus, apoptosis deficiency of APC-negative cells could be partially related to the mitotic checkpoint defect in these cells. However, this can only be part of the story because we detect a decreased rate of apoptosis in the absence of APC even in cells that were not exposed to mitotic poisons. An alternative explanation places both apoptotic and mitotic checkpoint defects downstream of one event, namely Bub1/BubR1 insufficiency. The proapoptotic function of the mitotic checkpoint proteins BubR1 and Bub1 (kinetochore-associated pools of both of them are decreased in APC-deficient cells) can trigger apoptosis specifically in polyploid cells (). It is possible that a defect in this function of Bub1 and BubR1 in APC-depleted cells is responsible for their inability to undergo apoptosis efficiently. Consistent with this idea, the relative number of polyploid cells is overproportionally increased in APC-negative cultures (). Importantly, overexpressing recombinant Bub1 reduced polyploidy that was induced by APC loss and rendered the mitotic spindle checkpoint less sensitive to APC inhibition (Fig. S4, available at ). In this context, it is important to note that we found decreased apoptosis in all of the APC-deficient systems we tested, including fibroblasts with a floxed APC allele that allows the conditional inactivation of APC (Fig. S1). These particular cells lack a functional mitotic checkpoint; both floxed fibroblasts and their wild-type counterparts did not maintain a mitotic arrest when treated with mitotic poisons (unpublished data). Interestingly, in these cells, only Bub1 but not BubR1 was reduced at the kinetochores when the APC level was reduced. Thus, the apoptotic defect in these cells could not be mediated by changes in BubR1. Importantly, data from these cells suggest that the downregulation of Bub1 at kinetochores is a primary consequence of APC loss and that the lack of BubR1 in the two other cell types may be secondary to the Bub1 deficiency. Indeed, BubR1 was shown to require Bub1 to be recruited to kinetochores (). APC is a known negative regulator of the Wnt pathway. To determine whether this function contributes to the effect of APC on ploidy, we used HCT116 cells as a model. These cells have constitutively active β-catenin that is insensitive to APC depletion. Using these cells, we showed that APC is likely to affect the mitotic checkpoint and apoptosis independently of β-catenin targets (). This is in contrast to recent data that implicate the upregulation of conductin downstream of activated β-catenin in establishing genomic instability in the absence of APC (). However, activation of Wnt signaling by APC inhibition in HCT116 cells used in this study was not well described, making a direct comparison difficult. Our current data are consistent with the idea that the effect of APC on the spindle checkpoint and possibly on apoptosis involves Bub1 and, in some cases, BubR1. Recently, caspases were implicated in the removal of Bub1 and BubR1 from kinetochores during prolonged mitotic arrest and the resulting mitotic slippage (). Because APC is found at kinetochores (; ), where many spindle checkpoint components accumulate, we considered the possibility that APC could protect checkpoint proteins from degradation by caspases at these sites. However, the defect in mitotic arrest induced by APC depletion was not altered by caspase inhibitors (unpublished data), suggesting that this is not the case. Furthermore, the removal of APC did not alter the phosphorylation of Bub1 and BubR1 in mitotically arrested cells (unpublished data). Therefore, the molecular mechanism of Bub1 and/or BubR1 regulation by APC remains unresolved. The combination of spindle, spindle checkpoint, and apoptosis defects allows APC-negative cells in culture to become polyploid. Importantly, our data are supported by in vivo evidence showing that the inhibition of APC in intestinal mouse epithelia also leads to the loss of euploidy 3 d after APC loss is induced. It is intriguing that in this case, both prominent tetraploidy and a decrease in the number of apoptotic bodies in intestinal crypts (relative to increased apoptosis in the rest of the tissue; unpublished data) were restricted to the same area of the epithelium. Furthermore, a section of intestine from an APC mouse that produces β-catenin–positive adenomas () after the loss of wild-type APC also revealed a striking difference in the nuclear size between cells in small β-catenin–positive lesions and the surrounding APC heterozygous (APC) normal tissue. Consistent with the idea that the mitotic checkpoint is compromised in the absence of APC, APC- mouse adenomas were arrested in mitosis by vinorelbine treatment less efficiently than normal tissues in these mice (our preliminary data). In summary, we show that the loss of APC has immediate consequences: spindle defects together with a compromised mitotic checkpoint can produce tetraploid cells; combined with decreased apoptosis, this promotes the expansion of the polyploid population (). Furthermore, we demonstrate the loss of euploidy in the mouse gut after the inactivation of APC and in adenomas from mice, suggesting that tetraploidy is indeed a common feature of APC-deficient cancers. Lack of APC induces a spectrum of cell cycle defects that can amplify each other to promote genetic instability. Furthermore, in the context of deregulated β-catenin, which maintains cells in a proliferative state inappropriately, the resulting defects in genetic stability could be particularly effective in producing tumors. Human colon carcinoma HCT116 cells and mouse embryonic fibroblasts constitutively deficient or wild type for APC (a gift from R.A. Weinberg, Whitehead Institute, Cambridge, MA) were cultured in DME supplemented with 10% FCS and 1% penicillin-streptavidin stock solution (MP Biomedicals). For human osteosarcoma U2OS cells, nonessential amino acids (1:100; Sigma-Aldrich) and 1% -glutamine were added. For mitotic arrest, U2OS cells were treated with the indicated amount of nocodazole or taxol for 20 h (unless stated otherwise), and HCT116 cells were treated with 10 ng/ml taxol or 100 ng/ml nocodazole for 12 h. For synchronization, U2OS cells received two rounds of 2 mM thymidine for 22 h with a 10-h interval. When needed, 1.25 μg/ml taxol or nocodazole was added at the time of release from the second thymidine block. For time-lapse imaging of mitotic progression (), U2OS cells were presynchronized with a single 18-h thymidine treatment and were released 6 h before taking images to enrich for cells dividing at the time of filming. For apoptosis induction, cells were treated with 0.1 μM staurosporine (Calbiochem) for 12 h before analysis. Cells were fixed in warm 1.85% PFA in PHEM buffer (60 mM Pipes, 4 mM MgSO, 25 mM Hepes, and 10 mM K-EGTA, pH 6.9) for 15 min, washed in PBS, and blocked in blocking buffer containing 0.1% Triton X-100, 2% BSA, 5% donkey serum, and 0.02% NaN in PBS supplemented with 50 mM NHCl for at least 15 min. Alternatively, cells were fixed in ice-cold methanol for 5 min, rehydrated in PBS, and blocked in the aforementioned blocking buffer for at least 30 min. Primary antibodies were diluted in blocking buffer as follows: anti-Bub1 and -BubR1 (gift from S. Taylor, University of Manchester, Manchester, UK) were used at 1:130 for mouse cell lines and at 1:1,000 for human cell lines, anti-Bub1 mAb (Chemicon) was used at 1:500, and CREST was used at 1:100 for mouse cell lines and at 1:300 for human cell lines (gift from B. McStay [Biomedical Research Centre, Ninewells Hospital, Dundee, Dundee, UK] and W. Earnshaw [University of Edinburgh, Edinburgh, UK]). Secondary antibodies raised in donkey (Jackson ImmunoResearch Laboratories) conjugated with either FITC, Texas red, or Cy5 were used at a 1:250 dilution. Cells were counterstained with DAPI at 1 μg/ml for 2 min. High resolution images were collected with an imaging system (DeltaVision Restoration; Applied Precision) built on an inverted microscope or stand (Eclipse TE200; Nikon; or 1X70; Olympus) using a 100× NA 1.4 objective lens. Images were acquired at 0.2-μm intervals in the z dimension and were deconvolved, and, where required, projections of multiple sections were built using SoftWoRx software (Applied Precision). Interkinetochore distances were measured in 3D images using SoftWoRx. H2B fluorescence/brightfield time lapses were acquired on a DeltaVision Restoration imaging system that was built on a stand (1X70; Olympus) equipped with a 37°C chamber using a 40× NA 1.4 dry objective lens. We collected five fluorescent sections at 2-μm intervals in the z dimension and one brightfield reference image in the middle of the stack every 10 min for 19 h for mitotically arrested U2OS cells or every 3 min for 6 h for the mitotic progression of logarithmically grown U2OS cells. Fluorescent H2B projections were built from the sections containing in-focus images using SoftWoRx software. Deconvolved 3D images of cells stained for CREST (kinetochore marker) and either Bub1, BubR1, or both were cropped around the nuclear area using SoftWoRx Explorer (Applied Precision) and analyzed in an open source microscopy image management system (Open Microscopy Environment [OME]; ; ; ). The FindSpots algorithm available within OME was used to automatically analyze entire datasets of images. This algorithm identified as kinetochores any 3D objects with a volume of >30 voxels and intensities of CREST staining above the threshold, which was set up as μ + σ, where μ is the mean, σ is the standard deviation of all voxels in the image stack, and is either three or four but is constant throughout each experiment. The total Bub1 or BubR1 kinetochore intensity in each cell was determined by adding the integrated intensity of Bub1 or BubR1 signal for all 3D objects defined by the thresholded CREST signal in the image. After subtracting the background, which was calculated as × μ, where μ is the mean voxel intensity in the Bub1 or BubR1 channel and is the total CREST-positive (kinetochore) volume of that image, this figure was standardized using the total kinetochore volume to yield a background-corrected kinetochore-specific intensity: K* = ( − × μ)/. This analysis was achieved through mostly automated data processing, which facilitated the analysis of a large number of images (). The data was summarized visually using box and whisker plots. Each shows the normalized per cell kinetochore levels of Bub1 or BubR1 through a five-point summary: the median (thick middle line), lower quartile (bottom boundary of box), upper quartile (top boundary of box), and the lower and upper extents of the data (bottom and top whiskers drawn from the box, respectively). This occurred after excluding outliers as defined by the standard 1.5 interquartile range (IQR) rule: mild outliers were defined as those points lying >1.5 times the IQR away from the lower or upper quartiles and are indicated by small circles; extreme outliers (lying more than three times the IQR) are shown by asterisks (). Box plots were generated using SPSS software version 11 (SPSS, Inc.). Two-tailed tests were performed using the Analysis ToolPak in Excel 2004 (Microsoft) to determine whether there was a statistically significant difference in the mean of the kinetochore intensities after removing the outliers identified by SPSS. Alternatively, tests and Mann-Whitney rank sum tests were performed using the SigmaSTAT program (Systat Software, Inc.). Cells were lysed in MEBC buffer (50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 5 mM EGTA, 5 mM EDTA, 0.5% NP-40, and 40 mM β-glycerol phosphate) supplemented with 10 μg/ml each of leuptin, pepstatin A, and chymotrypsin. Soluble fractions were run on 4–12% gradient SDS gels (Invitrogen) using MOPS running buffer and transferred to nitrocellulose membrane (Protran) with a 0.1-μm pore size (Schleicher & Schuell). For APC detection, Ali mouse monoclonal antibody (against N terminus; Cancer Research UK; ) or crude serum of rabbit polyclonal antibody raised against the middle portion of APC (anti-APCII; Nathke et al., 1996) were used at 1:1,000 in blocking solution (TBS containing 5% nonfat milk, 1% donkey serum, and 0.02% Triton X-100). Anti–α-tubulin mouse monoclonal antibody DM1a (Sigma-Aldrich) was used at a dilution of 1:2,000, anti-actin (C4 clone; MP Biomedicals) was used at 1:2,000, anti–glyceraldehyde-3-phosphate dehydrogenase (GAPDH; Abcam) was used at 1:5,000, anti-Bub1 and -BubR1 sheep polyclonal antibodies (gifts from S. Taylor) were diluted 1:500, and anti–β-catenin polyclonal antibody against C terminus () was diluted 1:1,000. Secondary anti–rabbit, –mouse or –sheep HRP-labeled antibodies (Scottish Antibody Production Unit) or IRDye 800/700–conjugated anti–sheep and –mouse secondary antibodies (either Invitrogen or Rockland) were used at 1:5,000 dilutions. To induce recombination, AchCreAPC and AchCreAPC mice () were given either single or multiple intraperitoneal injections of 80 mg/kg β-napthoflavone. At each time point, mice were killed, and the small intestine was removed and flushed with water. The proximal 10 cm of intestines was divided into 1-cm lengths, bundled using surgical tape, and fixed in 4% formaldehyde at 4°C for no more than 24 h before processing. Immunohistochemistry of mouse intestines was performed using two different antibodies for p21, which gave the same pattern of expression (1:5, Neomarkers; and 1:25, Santa Cruz Biotechnology, Inc.). Nuclear area was assessed using standard protocols () after image capture with analySIS software (Soft Imaging Systems). Immunohistochemistry with β-catenin antibody (dilution of 1:50; Becton Dickinson) on a small intestine of an APC mouse () was performed as described previously (). Fig. S1 shows analysis of mitotic and apoptotic defects caused by APC reduction in mouse fibroblasts with conditional APC expression. Fig. S2 presents a comparison of cell cycle and G2 progression of presynchronized APC-negative and control U2OS. Fig. S3 shows comparative activation of the β-catenin/TCF-responsive promoter by APC depletion in U2OS and HCT116 cells. Fig. S4 shows the rescue effects of recombinant Bub1 overexpression in U2OS cells on mitotic checkpoint damage and polyploidy induced by APC depletion. Online supplemental material is available at .

The Pumilio-Fem-3 binding factor (PUF) proteins are defined by the presence of a Pumilio (PUM-HD) domain. This domain is crucial for PUF protein function and has the capacity to bind to the 3′ untranslated region (UTR) of mRNAs and to regulate transcript localization, translation, and/or decay (; ). PUF proteins are found in eukaryotic cells, from yeast to humans. In , the PUMILIO protein binds to mRNA, to repress its translation at the posterior pole during early embryogenesis (). In , PUF proteins regulate the switch from spermatogenesis to oogenesis through effects on translation (), and germ line stem cell propagation through effects on expression (). Recent studies indicate that the PUF proteins of are components of “posttranscriptional operons,” punctate, cytoplasmic structures that interact with RNAs encoding proteins that localize to the same subcellular location, are part of the same protein complex, or act in the same cellular pathway (). Our recent studies support an unexpected role for Jsn1p/Puf1p in mitochondrial motility and inheritance in budding yeast. During cell division, equal segregation of mitochondria between the mother cell and bud occurs by regulated, region-specific mobilization and immobilization of the organelle. That is, mitochondria are actively transported to the opposite poles of the yeast cell, i.e., the bud tip and mother cell tip. During poleward movement, mitochondria exhibit linear movement either in the anterograde direction, toward the bud tip, or in the retrograde direction, toward the mother cell tip. Thereafter, they are retained at the poles until the end of the cell division cycle (for review see ). The poleward movement of mitochondria during inheritance occurs using actin cables, bundles of F-actin that align along the mother-bud axis and serve as tracks for movement (; ; ). Binding of mitochondria to F-actin in vitro and association of mitochondria with actin cables during poleward movement and inheritance in vivo require the mitochore, an integral mitochondrial membrane protein complex consisting of the proteins Mmm1p, Mdm10p, and Mdm12p (, ; ). Our studies support a role for Jsn1p/Puf1p in recruiting the Arp2/3 complex to mitochondria, for force generation during anterograde movement along actin cables. In budding yeast, the Arp2/3 complex localizes to endosomes (; ) and mitochondria (, ), where it stimulates actin nucleation and generates force for intracellular movement. Mitochondria-associated Arp2/3 complex is required for normal mitochondrial morphology and for anterograde, but not retrograde, movement of the organelle during inheritance (; ). Jsn1p/Puf1p localizes to the cytoplasmic face of the mitochondrial outer membrane and interacts with mitochondria-associated Arp2/3 complex. Moreover, deletion of results in a decreased association of Arp2/3 complex with mitochondria, defects in mitochondrial morphology, and reduced levels of anterograde, but not retrograde, mitochondrial movement (). Together, these studies support the model that Jsn1p contributes to recruiting the Arp2/3 complex to mitochondria in budding yeast. This, in turn, allows for Arp2/3 complex–mediated actin polymerization and force production for anterograde movement of mitochondria, using actin cables as tracks, from mother cells to buds during yeast cell division. In the current work, we found that another PUF protein, Puf3p, interacts with the machinery for mitochondrial motility and inheritance. Previous studies indicate that Puf3p binds preferentially to mRNAs for nuclear-encoded mitochondrial proteins. Specifically, 87% of the 154 transcripts that bind to Puf3p encode proteins that localize to mitochondria and contribute to mitochondrial protein synthesis, respiration, organization, and/or biogenesis. Moreover, Pum-HD of Puf3p binds to a consensus motif in the 3′ UTR of mRNAs that is found in the 3′ UTR of many nuclear-encoded mitochondrial proteins (). Other lines of evidence indicate that Puf3p can promote the deadenylation (polyA-tail shortening) and decay of mRNA in vitro and that deletion of results in a decrease in mRNA deadenylation and a twofold increase in the half-life of mRNA in vivo (; ). These findings support the model that Puf3p affects mitochondrial biogenesis through effects on the stability and/or targeting of mRNAs for nuclear-encoded mitochondrial proteins. Here, we report that Puf3p localizes to mitochondria, where it regulates not only mitochondrial biogenesis but also mitochondrial motility during inheritance. Conventional two-hybrid screens have had limited success identifying proteins that interact with integral membrane proteins. Because mitochore subunits are integral membrane proteins, we used an unconventional two-hybrid screen that tests for protein–protein interactions at the plasma membrane (; ). The system takes advantage of the fact that Ras must be activated at the plasma membrane to stimulate cell proliferation by using a strain, , that carries a temperature-sensitive mutation in , the guanyl nucleotide exchange factor for Ras. Incubation of at 37°C results in a loss of Cdc25p function and traps Ras in its inactive GDP-bound form, producing growth arrest. Expression and targeting of hSos, the human guanyl nucleotide exchange factor for Ras, to the plasma membrane rescues the growth defect of the strain at 37°C. We fused the bait—full-length Mmm1p, Mdm10p, or Mdm12p—to hSos, and the target, a yeast cDNA library, to the plasma membrane–targeting myristylation signal of Src (see Materials and methods). If fusion proteins containing the bait and target interact when coexpressed in the strain, hSos is recruited to the plasma membrane, where it activates the Ras pathway and supports growth at 37°C. We did not detect any reproducible two-hybrid interactions with Mmm1p or Mdm10p as bait. However, expression of -hSos resulted in growth at 37°C in nine clones from a pool of 42,000 transformants. One of them, pMyr-, contained amino acids 635–879 of Puf3p, which encodes the C terminus and part of the PUM-HD RNA binding domain of the protein (). pMyr- (635–879) rescued the temperature-sensitive growth defect of cells when coexpressed with pSos- but not with the pSos alone (). Thus, amino acids 635–879 of Puf3p interact with Mdm12p in the two-hybrid assay. Finally, we found that Mdm12p can bind to the full-length PUF repeat region on Puf3p (aa 549–828), and to aa 635–879, a region encompassing the C-terminal 2/3s of the PUF region and the C terminus of Puf3p (). Previous studies indicate that PUF proteins localize predominantly to the cytoplasm in higher eukaryotes (; ). In budding yeast, Puf3p localizes to punctate structures in yeast (; ). However, it was not clear whether Puf3p-containing structures colocalized with mitochondria. With the improved spatial resolution of deconvolution microscopy combined with visualization of Puf3p and mitochondria in the same cells, we found that Puf3p-GFP localizes to punctate and tubular structures that colocalized with mitochondria (). Although single projections of the 3D volumes are shown, colocalization of Puf3p with mitochondria was confirmed by examining the projections at multiple angles. Given the localization of Puf3p to punctate structures in these and previous studies, it is likely that the tubular Puf3p-containing structures consist of multiple Puf3p puncta that are closely spaced on mitochondria. For subcellular fractionation studies, whole cell extracts from cells expressing Puf3p-GFP were fractionated into mitochondria, microsomes, and cytosol by differential and Nycodenz gradient centrifugation (). (Cyb2p); that is, Puf3p was enriched in the mitochondrial fraction and depleted in the fractions containing microsomes and cytosol to the same extent as Cyb2p upon subcellular fractionation. Carbonate extraction and protease-sensitivity studies were performed to determine the disposition of Puf3p on mitochondrial membranes. Puf3p could be extracted from mitochondrial membranes with NaCO but remained associated with mitochondria after washes with KCl (). Moreover, Puf3p was degraded upon treatment of Nycodenz-purified mitochondria with trypsin and chymotrypsin (). This protease treatment degraded a mitochondrial surface protein (Tom70p) without affecting the integrity of the organelle, as assessed by the stability of Cyb2p, an intermembrane space protein. Together, these results indicate that Puf3p is a peripheral mitochondrial membrane protein that is associated with the cytosolic face of the mitochondrial outer membrane. Previous findings () indicated that deletion of had no effect on growth rates on fermentable carbon sources and produces a subtle decrease in growth rates on nonfermentable carbon sources. We find that overexpression of resulted in cell growth defects on glycerol at 37°C (). Using the DNA binding dye DAPI, we detected mitochondrial DNA (mtDNA) in 90 and 85% of wild-type and overexpressing cells examined, respectively (). Thus, the observed defect in respiration-driven growth at elevated temperatures in yeast overexpressing is not due to a loss of mtDNA. Previous studies revealed that Puf3p binds to and affects the stability and/or targeting of mRNAs for nuclear-encoded mitochondrial proteins. Here, we studied the effects of deletion or overexpression of on the steady-state protein levels of Pet123p, porin (Por1p), and Cyb2p. Pet123p is a subunit of the mitochondrial ribosome that is encoded by a transcript that binds to Puf3p (). Por1p and Cyb2p localize to mitochondria but are encoded by mRNAs that do not exhibit high-affinity binding to Puf3p (; ). The steady-state level of mitochondria-associated Pet123p was elevated in cells and reduced in cells overexpressing . Overexpression of a Puf3p mutant that bears a deletion in the PUF repeat region has no effect on Pet123p levels (unpublished data). In contrast, Por1p levels were not affected by deletion or overexpression of . These findings are consistent with the proposed function of Puf3p in binding to and stimulating the decay of specific mRNAs for nuclear-encoded mitochondrial proteins. Although deletion of also had no effect on the steady-state levels of Cyb2p, overexpression resulted in a reduction in the amount of Cyb2p. Overexpression of a Puf3p mutant that bears a deletion in the PUF repeat region has no effect on Cyb2p levels (unpublished data). Because Puf3p does not exhibit high-affinity binding to mRNA, it is possible that Puf3p binds to nonrelevant targets when overexpressed. Alternatively, because Puf3p binds to transcripts that encode mitochondrial protein import proteins (e.g., Tim9p, Tim44p, Tim17p, Tom22p, Tom6p, Ssc1p, and Hsp60p), overexpression may affect Cyb2p levels through indirect effects on transcripts for proteins that affect Cyb2p import and/or stability. Here, we tested whether Puf3p protein levels are altered in two situations in which mitochondrial biogenesis is up-regulated: the diauxic shift and the shift in growth from a fermentable to a nonfermentable carbon source. Upon growth of budding yeast on glucose as the sole carbon source, budding yeast undergo two sequential exponential growth phases (). The first growth phase is largely a fermentative phase. The second phase is mostly driven by the aerobic metabolism of ethanol produced during fermentative growth. The transition between fermentation- and respiration-driven growth phases is the diauxic shift. During this transition, there is an up-regulation in mRNAs for enzymes involved in gluconeogenesis, the glyoxylate cycle, the tricarboxylate cycle, and respiration (), and a large (up to 10-fold) increase in mitochondrial abundance (). A similar induction is observed when yeast cells are shifted from growth on a fermentable to a nonfermentable carbon source. Because respiration-driven growth is slower than fermentation-driven growth, the diauxic shift is detectable by monitoring growth rates as a function of time in glucose-based liquid medium (YPD). Under our growth conditions, this occurs after 10 h of growth (). We found that the steady-state level of Puf3p-GFP declines during the diauxic shift () and that there is little detectable Puf3p-GFP in cells that are adapted to growth on lactate, a nonfermentable carbon (). Together, these findings provide additional support for a role of Puf3p in the down-regulation of mitochondrial biogenesis. Given the localization of Puf3p described here, it appears that this regulation occurs on the surface of mitochondria in budding yeast. Finally, we studied the level of Cyb2p and Pet123p in cells grown in different carbon sources. First, we confirmed a previous study that Cyb2p levels are increased by 15-fold in cells grown on nonfermentable carbon source compared with that observed in glucose-repressed cells (). In addition, we found that Pet123p levels are elevated by 30% in wild-type and cells grown on a nonfermentable carbon source compared with wild-type, glucose-grown cells (unpublished data). These results are consistent with the established up-regulation of mitochondrial biogenesis upon release from glucose repression and provide correlative evidence for a role of Puf3p in this process. We identified two links between PUF proteins and the mitochondrial motility machinery. First, Jsn1p/Puf1p interacts with mitochondria-associated Arp2/3 complex and contributes to anterograde mitochondrial movement through effects on recruiting the Arp2/3 complex to the organelle (). Second, Puf3p interacts with the mitochore, the protein complex that is required to link mitochondria to actin cables for anterograde and retrograde movement during cell division. Here, we studied these protein–protein interactions and their role in mitochondrial motility and morphology. To determine whether Puf3p is an Mdm12p binding partner, we tested whether Mdm12p coimmunoprecipitates and colocalizes with Puf3p. Both studies were performed using a yeast strain in which the chromosomal genes and were tagged with GFP and multiple copies of the Myc epitope, respectively. Because cells expressing Puf3p-GFP and Mdm12p-Myc exhibited normal growth rates on fermentable and nonfermentable carbon sources and normal mitochondrial morphology (unpublished data), both fusion proteins appear to be fully functional. Puf3p-GFP was recovered in the pellet obtained upon immunoprecipitation of Mdm12p-Myc using a monoclonal anti-Myc antibody (). Puf3p-GFP was not recovered in the immunoprecipitated pellet from yeast expressing untagged Mdm12p or in antibody-free immunoprecipitation controls (). Moreover, OM45p, an abundant, unrelated integral mitochondrial outer membrane protein, did not coimmunoprecipitate with Mdm12p-Myc or Puf3p-GFP (unpublished data). Thus, Puf3p-GFP coimmunoprecipitates with Mdm12p-Myc. Consistent with this, punctate structures stained with Puf3p-GFP frequently colocalized with Mdm12p-Myc. Colocalization was greatest in punctate structures that exhibited high levels of Mdm12p-Myc and Puf3p-GFP () and was detected when the imaging threshold was adjusted to eliminate structures with low GFP or Myc signal. These results indicate that the association between Puf3p and Mdm12p detected in our unconventional two-hybrid screen is both specific and physiologically relevant. Genome-wide affinity purification screens revealed that Puf3p can bind to several subunits of the Arp2/3 complex (). We found that Arp2p coimmunoprecipitated with Puf3p-GFP upon immunoprecipitation of mitochondrial extracts from Puf3p-GFP–expressing cells but not from cells expressing untagged Puf3p or from antibody-free control samples (). Under these conditions, OM45p, an abundant mitochondrial outer membrane protein, was not immunoprecipitated. Thus, Puf3p interacts with mitochondria-associated Arp2p. Finally, we tested whether the mitochore coimmunoprecipitates with the Arp2/3 complex in a wild-type cell and a mutant. Mdm12p-Myc coimmunoprecipitated with Arp2p from mitochondrial extracts in wild-type cells (). Deletion of results in a 60% reduction in the amount of Arp2p that is recovered with Mdm12p-Myc after coimmunoprecipitation. Thus, we obtained evidence that mitochondria-associated Arp2/3 complex interacts with the mitochore and that Puf3p contributes to this interaction. Because Jsn1p/Puf1p contributes to recruiting the Arp2/3 complex to mitochondria in budding yeast, it is possible that the reduced interactions between the mitochore and the Arp2/3 complex detected in cells are due to a decreased association between the Arp2/3 complex and mitochondria. To address this issue, we compared the amount of Arp2p that is recovered with mitochondria during subcellular fractionation in wild-type cells, deletion mutants, and cells that overexpress . Deletion or overexpression of had no effect on the recovery of Arp2p with mitochondria (), indicating that Puf3p and Jsn1p are not functionally redundant. In contrast, our results indicate that Puf3p contributes to recruiting the Arp2/3 complex to the mitochore. Previous studies indicate that mutations in the mitochore and Arp2/3 complex/Jsn1p/Puf1p have different effects on mitochondrial morphology and motility. Mutations of mitochore subunits result in an accumulation of large, spherical mitochondria that fail to move in either the anterograde or retrograde directions (; ). In contrast, mitochondria from yeast carrying mutations in or Arp2/3 complex subunits are largely tubular, not spherical. However, they are fragmented and aggregated and exhibit defects in anterograde, but not retrograde, mitochondrial movement. If Puf3p contributes to recruiting the mitochore to the Arp2/3 complex, then mitochondrial motility and morphology in deletion mutants should resemble those observed in the Arp2/3 complex or Jsn1p mutants. Analysis of mitochondrial morphology in deconvolved 3D projections revealed that 14% of wild-type cells exhibited aggregation or fragmentation of mitochondria ( = 55). In contrast, 44% of cells in a -null population showed abnormal mitochondrial phenotypes, with 32% of cells containing aggregated mitochondria and 32% fragmented mitochondria ( = 63; ). The observed defects in mitochondrial morphology were not due to defects in actin organization or in the interaction between mitochondria and actin cables (unpublished data). Thus, mitochondrial morphology in a mutant is similar to that observed in and Arp2/3 complex mutants. Consistent with this, cells show defects in mitochondrial motility that are similar to those observed in yeast bearing a deletion in or mutation of Arp2/3 complex subunits (). In wild-type yeast, 40% of mitochondria moved in the anterograde direction, and 24% move in the retrograde direction during 1 min of analysis ( = 123). Deletion of had no obvious effect on retrograde mitochondrial movement. In contrast, we observed a 35% decrease in the amount of anterograde mitochondrial movement in a population as compared with wild type ( = 109; ). These findings support a role for Puf3p in recruiting the mitochore to the Arp2/3 complex to promote anterograde mitochondrial motility and normal mitochondrial morphology in budding yeast. Because Puf3p has been implicated in mitochondrial biogenesis and mRNA stability, it is possible that the motility defects observed in cells are a consequence of Puf3p effects on mRNA turnover that are critical for mitochondrial morphology or biogenesis. To address this issue, we measured mitochondrial motility in yeast bearing mutations in or (). Mas37p is a subunit of the SAM/TOB complex, which mediates assembly of newly imported β-barrel proteins into the mitochondrial outer membrane and promotes the segregation of Mdm10p from the SAM/TOB complex (; ). Tom7p is a subunit of the protein-translocating mitochondrial outer membrane pore (). Deletion of results in defects in mitochondrial protein import and produces defects in mitochondrial morphology that are similar to those observed in cells (, ). Deletion of or had no major effect on the level of anterograde and retrograde mitochondrial movement. Therefore, mutations that affect mitochondrial biogenesis or morphogenesis do not produce motility defects that resemble those observed in mitochore or mitochondria-associated PUF family proteins. Finally, we studied poleward mitochondrial movement in wild-type and cells upon growth on media containing fermentable or nonfermentable carbon sources (). Because Puf3p is present in cells grown in glucose-based media and severely down-regulated in cells grown on glycerol-based media, mitochondria motility in yeast grown on nonfermentable carbon sources should be similar to that observed in cells. We did not detect any carbon source effects on the extent of retrograde mitochondrial movement. In contrast, we found that growth on glycerol results in a 50% decrease in anterograde mitochondrial motility compared with that observed in yeast grown on a fermentable carbon source. Thus, two conditions that result in severe down-regulation of Puf3p levels, i.e., deletion of the gene or growth on nonfermentable carbon source, produce a decrease in anterograde but not retrograde mitochondrial motility. Consistent with this, mitochondrial motility in wild-type cells that were grown in a glycerol-based medium was not appreciably different from that observed in deletion mutations that were grown in media containing either glucose or glycerol. Puf3p, one of the six PUF proteins found in , binds preferentially to cytoplasmic mRNAs for nuclear-encoded mitochondrial proteins and promotes the decay of bound mRNA by enhancing mRNA deadenylation (; ). Another PUF protein of budding yeast, Jsn1p/Puf1p, localizes to mitochondria, where it contributes to the recruitment of the Arp2/3 complex to the organelle for anterograde movements that lead to inheritance (). Here, we report that Puf3p also localizes to mitochondria in budding yeast, where it contributes to both mitochondrial motility during inheritance and mitochondrial biogenesis. Puf3p cofractionates with a mitochondrial marker protein during subcellular fractionation and localizes to punctate structures that colocalize with mitochondria in living yeast cells. Carbonate extraction, salt washes, and protease-sensitivity studies indicate that Puf3p is a peripheral membrane protein that is associated with the cytoplasmic face of the mitochondrial outer membrane. Thus, two of the six PUF proteins of budding yeast, Jsn1p/Puf1p and Puf3p, localize to mitochondria and to the same submitochondrial compartment. Yeast strains used in this study are listed in . Strains were derivatives of BY4741 or BY4743 (Open Biosystems). Yeast cells were cultivated and manipulated according to . Standard molecular techniques for cloning procedures were used (). In all cases, PCR was performed using Ultra HF DNA polymerase (Stratagene) according to the manufacturer's instructions. Baits used for the two-hybrid screen were encoded on plasmids (pSos-, -, and -) that express fusion proteins of Mmm1p, Mdm10p, or Mdm12p with the hSos (; ). Primers used for construction of these plasmids are listed in . In all cases, the entire open reading frame of each gene of interest was amplified by PCR from genomic DNA. The primers used for these amplifications contained restriction sites (BamHI or MluI). To subclone amplified genes into the pSos plasmid, PCR products were cut with BamHI and MluI and inserted into the polylinker region of the pSos plasmid after linearization with BamHI and MluI. The nucleotide sequences of the cloning junctions were sequenced to verify that the bait proteins were expressed in frame with the hSos protein. DNA encoding the bait proteins was sequenced to verify the absence of mutations. pRS423- is a multicopy plasmid containing the complete coding sequence flanked by 300 kb upstream of the ATG start codon and 300 kb downstream of the TGA stop codon. The open reading frame was amplified by PCR from genomic DNA. Primers used for this cloning were as follows (underlined sequences correspond to XhoI and BamHI restriction sites sequences): forward, 5′-TGCAACCATTAGCACACTTGAGAATGTATATTGG-3′, and reverse, 5′-TGCATTCTCTATCTGTTGCAGAAATAAGAAGAGCG-3′. The PCR product was cut with XhoI and BamHI and integrated into the polylinker of the pRS423 plasmid after linearization with XhoI and BamHI. LGY015 (pRS423-[1–548]-GFP) transformed cells overexpress a chimeric protein in which the PUM-HD domain of Puf3p has been replaced with GFP. For this construction, part of the open reading frame of (aa 1–548) was amplified by PCR from genomic DNA with the following primers: forward, 5′-TGCAACCATTAGCACACTTGAGAATGTATATTGG-3′, and reverse, ATGCGATCTTTGCAAAACTCTAAGG-3′. The underlined sequences in the forward and reverse primers correspond to XhoI and BamHI restriction sites, respectively. The residue shown in bold font is a G that was inserted into the reverse primer and allows GFP to be expressed in frame. GFP was amplified from the plasmid pFA6a-GFP (S65T)- () with the following primers: forward, CAACATGGTCCCGGGTTAATTAA-3′, and reverse, 5′-ATGCGAGCTCGTTTAAAC-3′. The underlined sequences in the forward and reverse primers correspond to BamHI and EcoRI restriction sites, respectively. PCR products were cut with XhoI and BamHI (for truncated ) and BamHI and EcoRI (for GFP). Both fragments were ligated, and the resulting product was inserted into the polylinker of the pRS423 vector as an XhoI–EcoRI fragment. The nucleotide sequence of the cloning junction between and the was sequenced to verify that the GFP was expressed in frame with the Puf3p. The carboxy terminus of Puf3p was tagged GFP using PCR-based insertion into the chromosomal copy of the locus (; ). The carboxy terminus of Mdm12p was tagged with 13 tandem copies of the Myc epitope (13Myc) using the same technique. lists primers used to tag these genes. Yeast cells were first transformed with the PCR products using the lithium acetate method (). PCR was used to confirm the proper integration of tags into the target locus. Expression and localization of GFP- and 13Myc-tagged proteins were analyzed via Western blot. GFP and 13Myc localization were visualized in cells directly (GFP) or by immunofluorescence staining using a monoclonal anti-Myc antibody (13Myc). The two-hybrid screen was performed essentially as described previously (; ). After cotransformation of a strain with the bait (pSos-, -, or -) and the Cytotrap XR yeast cDNA library (Stratagene), single colonies were allowed to grow on solid media consisting of synthetic complete, glucose-based medium -Leu and -Ura (SC glucose–Leu–Ura) for 3–4 d at 23°C. Colonies that grew at 23°C were replica-plated onto synthetic complete, galactose-based medium -Leu and -Ura (SC galactose–Leu–Ura). Cotransformants that grew on SC galactose–Leu–Ura plates at 37°C were retained for further characterization. Library plasmids were isolated from clones that showed consistent galactose-dependent growth at 37°C. Isolated plasmids were transformed into cells in combination with the bait (pSos-, -, or -) or the pSos vector. Plasmids that supported growth of the strain only in the presence of bait containing , , or were sequenced. Mitochondria were visualized using two different fusion proteins. For analysis of mitochondrial morphology and motility, a centromeric plasmid containing the open reading frame of the citrate synthase 1 fused to GFP (pCS1-GFP; ) was used. For Puf3p-GFP and mitochondria colocalization studies, mitochondria were visualized using a fusion protein expressed from the plasmid pTDT104GAL1+PreFATPase-(subunit9)-DsRed (). Both hybrid proteins were expressed from plasmids using their own promoters. Cells were grown to midlog phase (OD 0.3–0.9) in SC-Ura media and placed on an SC-Ura–containing agarose pad on a microscope slide. A coverslip was applied to the slide, and the slide was sealed with Valap (1:1:1 Vaseline/lanolin/paraffin). Images were collected with a microscope (E600; Nikon) using a Plan-Apochromat 100×, 1.4 NA objective lens and a cooled charge-coupled device camera (Orca-ER; Hamamatsu). Illumination with a 100-W mercury arc lamp was controlled with a MAC5000 shutter controller and Ludl filter wheel (Ludl Electronic Products Ltd.). Ludl filter wheels or a Dual View image splitter (Optical Insights) were used for two-color imaging. Hardware control and image enhancement were performed using Openlab software (Improvision, Inc.). The bicinchoninic acid assay (Pierce Chemical Co.) was used for protein concentration determinations. Gel electrophoresis and Western blot analysis were performed as described previously (). Yeast mitochondria were isolated as described previously (). Immunofluorescence and visualization of DNA using DAPI was performed as described previously (). Immunoprecipitation, carbonate extraction, and protease-sensitivity studies were performed as described previously ().

Cell fusion is an important developmental event, from sperm–egg fusion during fertilization to syncytium formation in the development of placenta, muscle, and certain hematopoietic cell types. Although detailed mechanistic characterizations have been performed for virus–cell fusion and vesicle–organelle fusion, the molecular events mediating cell–cell fusion are poorly understood. In virus and vesicle fusion, a protein machine— a fusase—assembles between the fusing bilayers such that it spans both membranes (). For influenza virus, the fusase is the hemagglutinin protein, which is anchored in the viral membrane and inserts itself into the target membrane (; ), whereas for vesicle–organelle fusion, the interaction of cognate SNARE family transmembrane proteins results in the assembly of a multiprotein complex anchored in both vesicle and target membranes (). Hemagglutinin and the SNARE complex each adopt a coiled-coil structure that undergoes a series of conformational changes to winch the two bilayers into close proximity (; ). During this process, the bilayer structure becomes distorted, and water separating the apposing membranes is squeezed out, initiating fusion (; ). A similar fusase mediates cell fusion during placental development: syncytin, a protein encoded by a retrovirus-derived gene, is necessary and sufficient for placental cell fusion (). However, no analogous fusases have been identified in muscle precursors, sperm or egg, or other cells that fuse. Over a dozen proteins required for myoblast or osteoclast/macrophage fusion have been identified, but many of these proteins promote early steps, including cell migration and adhesion, rather than the later step of cell fusion (; ). Likewise, in sperm–egg fusion, the fertilin complex was initially recognized as bearing hallmarks of a fusase (it contains a hydrophobic peptide capable of inserting into a membrane, and experimentally blocking fertilin function prevents fusion), yet fertilin knockout mice are primarily defective in sperm migration into the oviduct and binding to the zona pellucida that surrounds the egg, with a much weaker defect at the final step of cell fusion (; , ). A few proteins likely to act late in cell fusion, possibly at the ultimate step of membrane fusion, have been identified. EFF-1, a single-pass transmembrane protein, is required for syncytia formation in the hypodermal cells of () and, when ectopically expressed, is sufficient to fuse cells that do not normally fuse (Shemer et al., 2004; ; ), thus making it an excellent candidate fusase. Two proteins, CD9 and CRISP-1, are important for sperm–egg fusion and seem to act after the initial steps of cell adhesion. CD9 is a multispanning membrane protein in the oocyte plasma membrane, and oocytes from mice lacking CD9 adhere normally to sperm but do not fuse with them (; ; ). CRISP-1 is a peripherally associated membrane protein on the surface of sperm that, when blocked, prevents sperm–egg fusion but not adhesion (). Yeast mating offers a genetically powerful system in which to identify factors controlling the late steps of cell fusion. During yeast mating, haploid cells of mating types MATa and MATα secrete pheromone (MATa cells make a-factor, and MATα cells make α-factor), which is detected by a G-protein–coupled receptor on the complementary cell type, initiating a MAPK signaling cascade that results in G1 cell cycle arrest, polarized growth in the direction of highest pheromone concentration, and transcriptional up-regulation of ∼100 genes (). The mating partners adhere to one another through interactions in the cell wall to produce a mating pair. Finally, in a process whose molecular details have only begun to come to light, a small region of the cell wall at the interface between the mating partners is degraded, the mating partners' plasma membranes become apposed, and, finally, cell fusion occurs. Numerous attempts to identify the cell fusion machinery have identified factors that are required for cell wall degradation at multiple steps, from regulating cell wall remodeling and secretory vesicle trafficking to the maintenance of osmotic integrity (; ; , ; ). However, none of these genetic screens have identified genes that seem to act at the final step in cell fusion: the merging of plasma membrane bilayers. Previously, we designed a reverse genetic approach aimed at uncovering the fusion machinery (). We reasoned that the cell fusion machinery that acts during mating probably includes a transmembrane protein expressed specifically in response to mating pheromone. We began studying pheromone-regulated membrane proteins (PRMs), and, using the data-mining program Webminer (), we identified the membrane protein most induced by pheromone, Prm1, and characterized its role in membrane fusion (). Prm1 is a multispanning membrane protein that is not expressed under standard growth conditions but is induced in both mating types in response to pheromone (). It localizes to the site of cell fusion. If either mating partner lacks Prm1, ∼10% of mating pairs fail to fuse, but if both mating partners lack Prm1, ∼50% of mating pairs fail to fuse (). When we examined × mating pairs by electron microscopy, we observed a morphology never before seen. In some mating pairs, the cell wall had been degraded, and the plasma membranes had become apposed yet failed to fuse (). This result indicates that Prm1 facilitates the final step in cell fusion, that of plasma membrane fusion (). However, Prm1 cannot constitute the complete machinery. Even in its absence, about half of all mating pairs still fuse, indicating that Prm1 either facilitates the action of a yet unidentified fusase or that Prm1 is itself a fusase and one or more alternative fusases exist. Intriguingly, mating pairs frequently lyse when attempting to fuse, suggesting the remaining presence of an active but dysregulated fusase (). Among mating pairs that are capable of fusion, the initial permeance and expansion rate of the fusion pore are slightly decreased, indicating a role for Prm1 in fusion pore opening; however, the subtlety of this defect again points to the presence of a redundant fusion activity (). The notion of an additional fusion machinery that is regulated by or acts in parallel to Prm1 implies that the disruption of additional components should create more severe blocks to membrane fusion than can be achieved by disrupting alone. In this study, we have exploited this prediction to design a genetic screen that led to the identification of a gene acting in conjunction with to promote cell fusion. To identify factors required for Prm1-independent cell fusion, we screened for mutants that enhance the × mating defect. We performed random mutagenesis of a MATa strain bearing a selectable marker. We then plated the mutants, allowed them to form small colonies, and replica plated them to a lawn of MATα cells bearing a different selectable marker. We allowed mating to occur and replica plated to a medium selective for auxotrophic markers of both parent strains, thus allowing the growth only of diploids that arose during mating. Each mutant colony from the original plate resulted on the final selective plate in a small patch with many diploid microcolonies emerging from it as papillae (). The density of diploid papillae within each patch reflected the mating efficiency of the mutant that gave rise to it. Using this replica mating assay, we screened for mutants in the background that mated poorly to a partner. In addition to mutants in the -independent fusion pathway, we expected to find sterile mutants not relevant to this study. To distinguish these classes, we tested the ability of each mutant to mate to a wild-type (WT) partner. Mutants that mated poorly to a WT partner were considered sterile and were discarded. To further characterize the remaining mutants, we performed a backcross to ensure that the observed phenotypes segregated as single mutations. To our surprise, 4/10 mutants revealed a new phenotype after backcrossing. MATα progeny bearing these mutations, but not MATa progeny, displayed complete sterility whether mated to a WT or partner. Therefore, we assumed that a set of mutations enhancing the phenotype in MATa cells causes sterility in MATα cells. Because sterility was easier to score, we used complementation cloning to isolate the gene responsible for the MATα-specific sterility in one of the mutants. The remaining mutants were not characterized further. We recovered four genomic fragments that restored mating to this mutant. These fragments overlapped in a region containing the coding sequence of . Kex2 functions as a protease in the Golgi apparatus that processes several proteins traversing the secretory pathway, including the α-factor mating pheromone (; ). This essential role of Kex2 in the processing of prepro–α-factor readily explains why MATα mutants are sterile. In contrast, Kex2 does not process the a-factor mating pheromone, and MATa mutants do not display pheromone response defects, making it unlikely that the observed mating defect results from impairment of the pheromone signaling pathway (). As expected, a MATα mutant was sterile in our assay (unpublished data). mutant mated efficiently to a WT partner but poorly to a partner. × mating was readily apparent (). To learn whether Kex2 acts in cell fusion, we used a quantitative cell fusion assay as previously described (). Mating partners carrying deletions in , , both, or neither were mixed and allowed to mate. One partner expressed soluble cytoplasmic GFP to serve as a marker for cytoplasmic mixing. Mating pairs were examined by fluorescence microscopy. Mating pairs with GFP throughout their volume were scored as fused, whereas mating pairs in which GFP remained restricted to one partner were scored as unfused (). By counting the ratio of fused to total mating pairs, we quantitated the efficiency of cell fusion. This assay differs from replica mating in that it scores only the cell fusion step of mating rather than the entire mating process. In agreement with our previous results (), we observed in control mating reactions that the deletion of from both mating partners creates a substantial block to cell fusion compared with WT (, compare bar 1 with 7), whereas the deletion of from either mating partner alone produces a barely perceptible decrease in fusion efficiency (, compare bars 1, 3, and 5; ). Interestingly, the loss of in the MATa partner alone decreases fusion by 15% compared with WT (, bars 1 and 2), thereby demonstrating a role for Kex2 in MATa cells in cell fusion. Although the defect was small, it was highly reproducible. Because of the role of Kex2 in α-factor processing, we could not reciprocally assay MATα mutants. We observed a markedly greater Kex2 dependency of cell fusion in mating reactions in which both partners lacked Prm1. × mating pairs is 70% lower than that in × mating pairs (, bars 7 and 8). Thus, the mutation unilaterally and potently enhances the otherwise weakly penetrant fusion phenotype. The Kex2 dependency of mating reactions in which only one partner expresses Prm1 proved more complicated. Mating pairs in × mating reactions fuse with a much reduced efficiency compared with WT × mating reactions (, bars 3 and 4). × WT mating reactions do not differ greatly from × WT matings (, bars 5 and 6). In other words, the mutation produces a much stronger effect when placed in trans rather than in cis to the mutation. The Kex2 protease has been extensively characterized (). In brief, Kex2 acts as a furin-type endopeptidase that cleaves substrate proteins at dibasic sequence LysArg sites as the proteins traverse the Golgi apparatus. For many substrates such as α-factor, the initial Kex2 cleavage is followed by the action of two exopeptidases, which trim the newly exposed ends: Kex1, a carboxypeptidase, removes the LysArg sequence from the C terminus of the N-terminal fragments, whereas Ste13, an aminopeptidase, removes pairs of residues (preferring X-Ala sequences) from the N terminus of the C-terminal fragments. To test whether Kex1 or Ste13 also affects cell fusion, we subjected and mutants to the same genetic analysis we used with mutants. We conducted mating reactions in which the partners lacked either Prm1 or Kex1 in all combinations or Prm1 or Ste13 in all combinations and assayed the resulting mating pairs for fusion using the GFP mixing assay. As shown in , a mutant displays a slight but reproducible fusion defect when crossed to a WT partner (, bars 1 and 2). This defect was enhanced when we introduced a mutation in trans but not in cis (, bars 3 and 4 and bars 5 and 6, respectively). Finally, the most severe defect occurred when we introduced a mutation into a × cross, which reduced the number of successful fusions by more than half (, bars 7 and 8). Thus, the effects of the mutation qualitatively phenocopy those of the mutation, although the mutation produces slightly milder fusion defects. In contrast, the deletion of from a WT × WT mating reaction produced no substantial difference in cell fusion (, bars 1 and 2). Furthermore, did not enhance the fusion phenotype when placed in trans or in cis (, bars 3 and 4 and bars 5 and 6, respectively). Finally, when introduced into a × mating, the mutation did not appreciably reduce mating (, bars 7 and 8). These results demonstrate that the complement of proteases required to promote cell fusion in MATa cells is distinguishable from that required for α-factor processing. Among known Kex2 substrates are cell wall glucosidases such as Scw4 and Scw10 (; ; ) and cell wall structural components such as Hsp150 (). We systematically generated deletions in eight known Kex2 substrates and mated each mutant to a WT or mating partner (). If proteolytic activation of a given substrate is required for fusion, we expected the loss of that substrate to phenocopy the loss of Kex2; it should display a mild decrease in fusion when crossed to a WT partner and a more severe decrease when crossed to a partner. As shown in , none of the mutants displayed such a fusion defect. Thus, the fusion defect does not result from the inactivation of any one of these substrates singly. Some of these Kex2 substrates may act redundantly and only show a phenotype when removed in combination. For example, it has been shown that the lack of Scw4 or Scw10 alone causes a very mild cell wall defect, but the loss of both results in extreme weakening of the cell wall and a mating defect (). double mutant in our fusion assays and saw no effect with a WT or mating partner () or with a mating partner (unpublished data). It remains possible that inactivation of some other combination of known Kex2 substrates would recapitulate the fusion defect. We hypothesized that there might be an additional, unidentified Kex2 substrate that mediates Kex2-dependent fusion. We designed a bioinformatics screen to attempt to identify such a substrate. In brief, we developed a scoring matrix based on the cleavage site sequences of known substrates and used it to rank potential cleavage sites in all other proteins, discarding high-scoring candidate sites that are not conserved among closely related yeasts or that are predicted to be cytoplasmic (Table S1, available at ). We tested 11 proteins with the highest ranked candidate sites by generating epitope-tagged alleles of each in WT and backgrounds and performing SDS-PAGE and Western blotting of cell lysates. With this approach, we identified two new Kex2 substrates, Prm2 and Ykl077w (). Prm2 is predicted to be a pheromone-regulated multispanning membrane protein with a topology similar to Prm1 and was identified in the bioinformatics screen that led to the characterization of Prm1 (). Ykl077w is an uncharacterized protein predicted to have a large (∼300 amino acid) extracellular/lumenal domain and a single transmembrane segment. Both proteins showed a shift in apparent molecular weight in a mutant background that is consistent with Kex2-dependent proteolysis (). We generated and mutants and tested them in our fusion assays. Neither mutant showed a defect with WT or mating partners (). Thus, although we were able to identify two novel Kex2 substrates, neither appears to be the hypothetical substrate relevant to fusion. It may be that another, currently unidentified substrate or a combination of redundant Kex2 substrates acts during cell fusion. We asked whether the sum of physiological effects resulting from a lack of processing of Kex2 substrates might explain the fusion defect. For example, cells lacking Kex2 display a weakened cell wall phenotype, as assayed by the up-regulation of cell integrity pathway target genes and hypersensitivity to the cell wall–binding dye Congo red (). Cell wall stress is known to induce the PKC signaling pathway, which can inhibit cell fusion (). Thus we next asked whether cell wall stress could explain the cell fusion defect caused by the loss of Kex2. and mutant cells. Cell wall stress and low osmolarity signals activate Pkc1 through the Bck1 MAPK module, eventually leading to activation of the transcription factor Rlm1, which activates the transcription of many genes, including (). Thus, an - reporter gene has been used to detect activation of the PKC cell integrity pathway (; ). We grew strains bearing the - reporter overnight with or without the addition of 1 M sorbitol as osmotic support, harvested cultures in exponential phase, and assayed for reporter activity during vegetative growth or after exposure to α-factor pheromone (). WT cells show enhanced activation upon α-factor treatment, as expected from the cell wall remodeling that accompanies the pheromone response. The presence of osmotic support slightly decreased activation in WT cells (, black bars). Note that mutant cells were indistinguishable from WT in the absence and presence of α-factor, strongly suggesting that the deletion of does not affect cell wall structure (). In contrast, and mutants showed enhanced baseline activation (2.5- and 6-fold greater than WT levels, respectively), which is consistent with a cell wall structural defect. The enhanced activation was exacerbated by pheromone treatment and weakly mitigated by the presence of osmotic support (). mutants did not show these effects (unpublished data). As a further measure of cell wall integrity, we assayed each mutant for Congo red sensitivity. Mutants with compromised cell walls generally do not grow on media containing Congo red. (; ). In contrast, the viability of neither nor was affected by Congo red. Thus, as assayed by activation and Congo red sensitivity, cell wall defects are severe in mutant cells, mild in mutant cells, and undetectable in mutant cells. Most cell wall defects manifest themselves as a result of a failure of the cell wall to provide a rigid support counteracting the outward force on the cell membrane caused by the osmotic imbalance between cytoplasm and the growth medium. Thus, we next asked whether osmotically stabilized medium (i.e., growth medium formulated at an osmolarity closer to that of cytoplasm), which relieves many phenotypes resulting from cell wall defects, could suppress the cell fusion defect. Mating reactions were performed under standard conditions with or without 1 M sorbitol, and fused mating pairs were counted in the quantitative cell fusion assay. As controls, we showed that bilateral crosses of the classical cell wall remodeling mutants and were partially suppressed by mating on osmotic support (). Surprisingly, we found that cells displayed a decreased fusion efficiency in the presence of osmotic support (; a similar observation was reported by ). The explanation for this decrease is not clear, but it suggests that the defect is distinct from cell wall stress. In contrast, the mild × WT defect was suppressed by osmotic support (), as was the × WT defect (not depicted). The strongly deficient × mating was partially suppressed, but, importantly, fusion was not restored to WT levels. Thus, and behave similarly to other fusion mutants known to affect cell wall degradation, whereas does not. If cell wall defects caused by the loss of Kex2 are responsible for strongly enhancing the fusion defect, we expect other mutants with similar cell wall defects also to synergize with in a fusion assay; alternatively, if the failure to process Kex2 substrates that act specifically with Prm1 causes the enhanced fusion defect, other cell wall mutants will not synergize with . To distinguish these possibilities, we mated strains bearing a , , or (a gain of function allele that mimics constitutive cell wall stress; ) mutation to a partner and scored mating pairs with the cell fusion assay. Unlike × , no combination of mutants in these mating reactions showed a synergistic defect (). × () similarly did not produce a synergistic defect. Thus, collectively, mutant cells experience cell wall stress and concomitantly increase activation. However, those defects are not sufficient to explain the unique synergy we observe between and mating partners. Therefore, these results strongly suggest that the synergistic effect of the and mutations on cell fusion results from combined defects in the cell fusion machinery. To characterize ultrastructurally the cell fusion intermediate at which × WT mating reactions arrest, we examined fusion-arrested mating pairs using electron microscopy. In the majority (80%) of unfused × WT mating pairs, we observed novel bleblike structures in the cell wall separating the two mating partners. Such cell wall–embedded blebs appear disconnected from both cells ( and ). The blebs are bounded by a visible lipid bilayer ( and , F and M; in other views, the bilayer is harder to discern because of the angle of the section relative to the plane of the bilayer). A gap of a relatively consistent width of ∼8 nm separates the blebs from the plasma membrane that they appear adhered to (, and E; and 7, F, J, and M). About 90% of the blebs appear preferentially linked to one mating partner, but ∼10% of the blebs closely approach the plasma membrane of the other mating partner as well (, B and C; and 7, F and M). In any given section, we observed numbers ranging from one bleb () to one primary bleb with others clearly above or below it (), to two blebs side by side with their surfaces tightly apposed (), to a cascade of blebs spread out across the diameter of the cell–cell interface (). About 75% of unfused mating pairs have one to five blebs, with 5% having more and 20% having none. In serial sections, we never detected a clear cytoplasmic continuity between a bleb and either mating partner. The texture of the staining inside the blebs often appears fibrous, unlike the regular punctate staining of ribosomes observed in normal cytoplasm (). We examined the 3D structure and arrangement of blebs in more detail by serial section analysis. A representative set of serial sections is shown in . At one end of the series, the cell–cell interface appears restricted, and secretory vesicles are sparse, indicating the sections come from a region where the cells are just beginning to make contact off center of the long axis of the mating pair (). As the sections approach the center of the mating pair, the contact zone widens, the number of secretory vesicles in the cytoplasm increases, and a cell wall–embedded bleb appears (). Moving more to the center of the cell–cell interface, the bleb broadens and appears to push slightly into the mating partner on the left () before disappearing from view (). A second bleb appears in a lower section and widens (); a third and possibly a fourth bleb appear still farther along the stack of sections (). The bleb in (C–F) almost contacts both plasma membranes; in (magnified in ), it appears only ∼10 nm from the partner on the right. Other structures of unknown function are also frequently observed in these images. A dark, unclosed circle reminiscent of the formation of autophagic structures by the fusion of small vesicles () appears to begin enclosing a region of cytoplasm (; magnified in N). Similarly, a spherical lipid bilayer enclosed in a second, equidistant bilayer contains dark-staining cytoplasm (; magnified in O) and suggests a mature form of the first structure. Both structures are surrounded by a zone of ribosome exclusion. Similar structures appear often in sections of mutant–derived mating pairs (). We have previously described the formation of bubbles as characteristic features of fusion-arrested × mating pairs (). Bubble formation appeared to result from a block in fusion of the mating partners after the intervening cell wall had been removed and plasma membranes had tightly adhered to each other, often buckling as a double membrane into either cell. × mating pairs would reflect the order of and function in the fusion pathway. Rather than observing a single epistatic phenotype, however, we saw a more complex heterogeneous mixture of three classes of structures. × mating pairs similar to those seen in × mating reactions. A characteristic bubble in such mating pairs is shown in (A and B). In this example, the mating partner on the bottom forms an extension past the midline of the mating pair and well into the space previously occupied by the mating partner on the top. The plasma membranes appear tightly apposed but unfused. The cytoplasmic continuity between the bubble and the bottom cell is obvious, and the texture of the staining within the bubble matches that of normal cytoplasm. Second, we observed cell wall–embedded blebs similar to × WT mating reactions. × bleb are shown in . Several blebs extend over the full width of the cell–cell interface. No cytoplasmic continuity between the blebs and either mating partner can be found. Additionally, a double bilayer-bound structure appears in the upper mating partner of this pair. In some mating pairs, blebs of an enormous size accumulated (both mating pairs in ; magnified in D and E). × mating pairs display a unique morphology that was not previously observed, which is referred to here as enormous barren bubbles. Enormous barren bubbles appear similar to × bubbles but are essentially devoid of the staining of ribosomes and vesicles that populate normal cytoplasm (serial sections; ). These structures also lack the fibrous pattern typical of blebs. Instead, they present the appearance of empty, organelle-free cytoplasm despite the presence of clear continuities to one mating partner (). One section shows an enormous barren bubble that may be folded back onto itself, thus giving the appearance of two separate structures (). The lack of cytosol might reflect a lysis event, which a fraction of mating pairs undergo (). Similarly, barren areas have been observed in ultrastructural studies of myoblast fusion pores and within membrane sacs subsequent to cell–cell fusion (see in ). It remains an intriguing mystery how a portion of the cytoplasm could become so distinctly different without a visibly delimiting barrier. The molecular machine that fuses cells during yeast mating has remained elusive. In this study, we describe the discovery of a role of Kex2 during cell fusion. To date, Kex2's only known function in mating involved an earlier step, namely the proteolytic processing of the pheromone α-factor in MATα cells (). In contrast, Kex2 is not required in MATa cells for pheromone processing, which allowed the discovery of its role in cell fusion. This duality mirrors the Axl1 protease, which processes a-factor pheromone in MATa cells and is required in MATα cells for efficient fusion (; ). However, as Axl1 activity is cytoplasmic and Kex2 activity is lumenal/extracellular, it is unlikely that this curious parallel reflects a shared mechanism. MATa cells lacking the exopeptidase Kex1 display a cell fusion defect similar to that of cells lacking Kex2, strongly suggesting that it is the lumenal/extracellular proteolytic activities of Kex2 and Kex1 that are required in this process. Therefore, we propose that Kex2 and Kex1 proteolytically activate at least one (yet to be identified) substrate protein that comprises part of the fusion machinery. By analogy, furin, a Kex2 family protease, proteolytically activates the fusases of several viruses. Mechanistically, the postulated Kex2 substrate could form a complex with Prm1 and possibly other components to constitute the membrane fusion machine. Alternatively, the Kex2 substrate and Prm1 could act at distinct yet mechanistically coupled sites in the membrane to promote fusion. Genetic analysis of and shows a synergistic interaction between these genes. To achieve efficient cell fusion, at least one mating partner must carry active and . × ), whereas only mild defects are observed whenever one mating partner is WT (WT × WT, × WT, WT × , × WT, and × WT). Thus, a simple model emerges from the genetic data: Prm1 and Kex2 (the latter likely acting by proxy through a substrate) are both important for the same step in cell fusion, and this step can be performed by either mating partner. This model also accounts for the finding that and mutations synergize in trans but not in cis. Therefore, this definition of 's role in cell fusion illuminates another layer of genetic redundancy in the process. Originally, eluded detection in traditional screens because a mutant only displays a cell fusion phenotype when mated to a partner also lacking . A mutant likewise displays a strong cell fusion phenotype only when mated to a partner. Consequently, mutants would be isolated for their cell fusion phenotype only in a sensitized screen, such as the one described here. This strategy can now be extended to identify other genes in the pathway, including, but by no means limited to, the postulated and eagerly sought-after Kex2 substrate. Although the Kex2 substrate relevant to cell fusion remains unknown, one especially interesting candidate is Prm2. Prm2, a protein of unknown function, is topologically similar to Prm1, is expressed only during mating, and is a Kex2 substrate. However, the deletion of Prm2 causes no fusion defect. It is possible that Prm2 acts redundantly with another Kex2 substrate or that the deletion of Prm2 fails to mimic the presence of unprocessed Prm2. On the other hand, it also remains possible that Kex2 acts indirectly during fusion (for example, through general effects on the stability of the cell wall). Consistent with this hypothesis, the magnitude of the fusion defect is reduced by osmotic support. However, other mutants that affect cell wall integrity ( ), cell wall remodeling ( and ), or hyperactivate the cell wall stress pathway () do not synergize with , arguing that the defect is uniquely linked to the Prm1-dependent step of membrane fusion. Likewise, the electron microscopy phenotype of unfused zygotes resulting from matings of MATa cells suggests a specific and novel defect resulting from attempted fusion. Although it is unlikely that the morphological features observed in arrested mating pairs reflect bona fide intermediates in the fusion pathway, the morphological consequences of blocking the fusion reaction are nevertheless intriguing. Rather than arresting at the same end point as one might naively expect, and mating pairs show unique morphologies at the electron microscope level. However, in many respects, the blebs observed here resemble bubbles seen previously in × matings, which is consistent with the notion that and act at similar steps. Like bubbles, blebs are membrane-bounded structures that are often found apposed to a nearby plasma membrane separated by a regular gap of ∼8 nm, and both bubbles and blebs appear to push into the space occupied by one mating partner. In contrast to bubbles, however, blebs are extracellular entities that show no continuity to either parent cell. This difference shows that the loss of Prm1 and the loss of Kex2 are not equivalent. If both Prm1 and the postulated Kex2 substrate are components of a single fusion machine, which becomes partially inactivated when either component is compromised, the residual machines in the respective mutant cells preferentially stall in the pathway at different points, thus leading to the characteristic and distinct morphological phenotypes. Consistent with this notion, stalling can occur at either end point when both Prm1 and Kex2 are missing in mating cells. Unfortunately, the effects of disruption can currently only be observed in MATa cells because of the requirement for the Kex2 processing of α-factor. One possible mechanism for the formation of blebs is that a -like bubble forms first but then becomes severed from the partner that forms it (); an intermediate suggesting this state is seen in mutants deficient in the a-factor transporter Ste6 (). Alternatively, the delivery of exocytic vesicles may be misregulated, thus producing the blebs (), or blebs could derive from the unusual circular structures observed (). According to both of these latter possibilities, blebs would not be derived from bubbles because in either case, the membrane surrounding the bleb would come from the same cell that provides the apposing plasma membrane. Precedence for the mechanism shown in comes from our knowledge of the sperm acrosome reaction. In this system, a repository of fusogenic material is delivered to the sperm surface in a burst of exocytosis. As part of this process, membrane-bounded cytoplasmic fragments are excised from the sperm as a result of rapid exocytosis at many points along the plasma membrane (). To distinguish between these models, it will be helpful to determine in future studies from which of the two parental cells the blebs originate. Strains used in this study are listed in . Gene replacements were generated with the PCR transformation technique (Longtine et al., 1998). Strains MHY398 and MHY427 were derived from KRY18 (a gift from R. Fuller, University of Michigan Medical School, Ann Arbor, MI; ). The plasmid pDN291 was used to express soluble cytosolic GFP and contains the gene as previously described (). The plasmid pRS314 is a standard vector containing the gene and was used in conjunction with pDN291 to create a set of mating type–specific selectable markers (). The plasmid pJP67 is used to express the hyperactive allele (; ). Congo red plates were prepared as previously described () by adding a 20-mg/ml stock solution of Congo red to <70°C autoclaved YPD (yeast extract/peptone/glucose) agar to a final concentration of 100 μg/ml. The plasmid was a gift from K. Cunningham (Johns Hopkins University, Baltimore, MD). MATa cells were grown to log phase, and 4 AU were washed once in a buffer of 10 mM potassium phosphate, pH 7.4 (10 ml; Sigma-Aldrich), and resuspended in the same solution. 300 μl of the mutagen ethyl methane sulfonate (Sigma-Aldrich) was added. Cells were vortexed and incubated for 30 min at 30°C. At that point, a 15-ml solution of 10% sodium thiosulfate (Sigma-Aldrich) was added to quench the reaction. Cells were washed twice in YPD medium and allowed to recover in YPD for 90 min at 30°C to fix any mutations that were induced. Serial dilutions of this stock were plated to medium lacking tryptophan, and the titer of colony-forming units was calculated; meanwhile, the stock was kept at 4°C. For screening, the stock was plated to 100 plates lacking tryptophan at a density of ∼120 colonies/plate. Colonies were allowed to grow for 40 h at 30°C. After ∼25 h, a stationary overnight culture of α was plated to 100 plates of YPD at 100 μl/plate and was incubated at room temperature for the remaining 15 h to form lawns. These lawns were respread with 100 μl/plate of water to a dull matte appearance indicative of homogeneity. Colonies of the mutagenized MATa cells were replica plated to mating lawns and incubated for 8 h at 30°C. The plates were then replica plated to medium lacking tryptophan and uracil to select for diploids. Phenotypes were scored on plates incubated for 2 d at 30°C. The clarity of the phenotypes critically depended on having homogeneous lawns of the proper density. MATα-specific sterility appeared in several of the enhancer mutants. We aimed for complementation of this phenotype because it was easier to score. After backcross to a strain, the sterile α was transformed with a pRS316-based library (a gift from S. O'Rourke, University of Oregon Institute of Molecular Biology, Eugene, OR; ). 15,000 transformants were subjected to a replica mating assay as described in with a tester strain as partner. The cell fusion assay was performed as described previously (). Cells of opposite mating types with the α strain expressing soluble cytosolic GFP were grown overnight to log phase, and 1 AU of each were mixed and vacuumed to a nitrocellulose filter. The filter was placed cell-side up on a YPD plate, and the plate was incubated for 3 h at 30°C. Cells were then scraped off the filter, fixed in 4% PFA, and incubated at 4°C overnight. This mixture was then spotted on a slide and observed with a fluorescent microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) using a 63× plan-Apochromat oil-immersion objective (Carl Zeiss MicroImaging, Inc.). First, a field was selected randomly using transmission optics. Then, groups of zygotes and mating pairs within that field were identified by brightfield microscopy and were subsequently scored as fused zygotes or unfused mating pairs by switching between brightfield and fluorescence. This procedure was continued until all the zygotes and mating pairs in the field were scored, at which point a new field was chosen and the procedure was repeated. To capture images, a single optical section was taken by both brightfield and fluorescence microscopy using a confocal microscope (TCS NT; Leica) with a 100× oil-immersion objective (Leica), a 150-mW, 488-nm argon excitation laser (Uniphase), and a 510–550-nm band-pass emission filter to visualize GFP. These images were then superimposed and contrast enhanced. Yeast strains containing the reporter were grown to log phase in SC-URA with or without 1 M sorbitol. For pheromone induction, log-phase cultures were incubated with 10 μg/ml α-factor for 2 h. Reporter activity was quantified as previously described () using 0.8 mg/ml -nitrophenol β--galactoside (Sigma-Aldrich). Reactions were incubated at 32°C for 10 min and stopped by adding an equal volume of 1 M NaCO. A scoring matrix to predict Kex2 cleavage sites was generated based on previously reported Kex2 substrates (Kex2 []; Mfα1 and Mfα2 [; ]; Ccw6/Pir1 and Ccw7/Hsp150 []; Ccw8/Pir2, Ccw11, Scw3/Sun4, and Scw4 []; Scw6/Exg1 []; and Scw10, Scw11, and killer toxin [Bostian et al., 1984; ]). The scoring matrix consisted of 10 protein sequence positions centered on the cleavage site, with the score for each residue at each position reflecting its prevalence among the known substrates at that position (Table S1). To obtain an overall score for a candidate sequence, the scores at each position were multiplied. For comparison purposes, we took the negative natural log of this value. Using a perl script, we searched the yeast genome for high-scoring potential cleavage sites. From the list of proteins that contained high-scoring sites, we discarded those that did not have a predicted transmembrane domain or signal sequence. Finally, candidates were selected that had high-scoring sites and in which the P2 and P1 positions were conserved among fungal homologues. Mating reactions were performed identically to the method described for quantitative fusion assays but at room temperature. During the mating, plates were taken to the University of California, Berkeley, electron microscopy labaratory and subjected to high-pressure freezing after ∼3 h of total incubation (; ). Samples were fixed, stained, and embedded (). High-pressure freezing was found to yield superior contrast between membranes and surrounding areas and a smoother curvature to membranes than we had observed by conventional fixation (; ). Sections of ∼60-nm thickness were cut, poststained with uranyl acetate and lead citrate (Ted Pella Inc.), and imaged with an electron microscope (Tecnai-F20; Philips) equipped with a 200-kV LaB6 cathode and a bottom-mounted four-quadrant 16 million–pixel CCD camera (UltraScan 4000; Gatan). The scoring matrix used in the bioinformatic screen for potential Kex2 substrates is available as Table S1. Online supplemental material is available at .

The Na,K ATPase, or Na pump, is primarily involved in the generation of Na and K gradients across plasma membranes and in the determination of cytoplasmic Na levels. Its function is tightly coupled to the regulation of cell volume and intracellular pH and Ca levels through the activities of the Na/H and Na/Ca exchangers (for reviews see ; ). In addition to maintaining the osmotic balance of the cell, Na,K ATPase functions as a scaffold for proteins involved in different functions, ranging from signal transduction to the cytoskeleton (). Studies in invertebrate models and tissue culture systems have suggested an involvement of Na,K ATPase in junctional complex formation. In , the Na,K ATPase α and β subunits are essential for barrier function of the septate junction, which is a structure that functions as a paracellular diffusion barrier similar to the vertebrate tight junction (; ). Colocalization and functional studies demonstrated that Na,K ATPase is essential for correct localization of several other septate junction proteins without affecting epithelial cell polarity. In vertebrates, a Ca-switch assay in MDCK cells suggested that Na,K ATPase activity may be essential for tight junction formation and development of polarity (). Blockage of the pump function using ouabain and by K depletion (which abolishes the steep K gradient required for pump function) caused the specific and reversible inhibition of tight junction formation, whereas adherens junctions were not affected. Similarly, the inhibition of Na,K ATPase affected the tight junction structure and transmembrane permeability of human retinal pigment epithelial cells (). The zebrafish α1B1 subunit of Na,K ATPase is encoded by the () locus and is required for heart morphogenesis and brain ventricle formation during embryonic development of the zebrafish (; ; ). The mutant heart phenotype is characterized by a delay in heart cone formation and transformation of this cone into the elongated heart tube, a phenotype reminiscent of the ()/ι (ι) and ()/ () mutant and morphant phenotypes (; ; ; ). Has/aPKCι is required for the establishment of apical–basal polarity of epithelial cells and is an essential component of a tight junction–associated protein complex that includes the PDZ domain, containing scaffolding proteins Par3 and Par6. Nok/Mpp5 is a tight junction–associated scaffolding protein of the apical Crumbs polarity complex (). Physical interactions between the Crumbs–Nok/Mpp5 and the Par3–aPKC–Par6 protein complexes have been previously described (; ; ). Consistent with a function in apical–basal cell polarity, loss of /ι and / using morpholino antisense oligonucleotides (MO) causes a disruption of Zonula occludens-1 (ZO-1)–positive apical junctions within the embryonic myocardium (). The role of Na,K ATPase during the development of vertebrate epithelia is poorly understood. We investigate function and regulatory mechanisms of this important ion pump during zebrafish heart morphogenesis. Our analysis demonstrates genetic interactions between / and in the maintenance of ZO-1–positive junction belts within myocardial cells. Functional analysis of regulatory and catalytic ATPase mutants, as well as pharmacological inhibition experiments, demonstrate the importance of correct ionic balance controlled by the Na pump for the maintenance of myocardial integrity. Zygotic and mutants display similar heart tube elongation defects, which led us to evaluate the possibility that both genes interact in the maintenance of apical junctions within myocardial cells. To visualize the effects of / and on myocardial development, we introduced a transgene that expresses GFP under the control of the () promoter region ([]; ) into the mutant background and injected clutches of these fish with MO ( mutants were recognized by their prominent retinal pigment epithelial phenotype). An antibody against the junctional protein ZO-1 was used to assess the integrity of apical myocardial junctions. In comparison to wild type (wt), both mutants and morphants displayed strongly shortened heart tubes by 36 h postfertilization (hpf; ), but displayed intact apical ZO-1 junction belts (; , = 12/12 hearts with intact ZO-1 junction belts; morphants, = 10/10 hearts with intact ZO-1 junction belts). Loss of both genes (; double mutant/morphants) resulted in a severe cardiac elongation defect that was stronger than the individual loss of function phenotypes (). In some cases, the heart was small and positioned at the midline, suggesting that morphogenesis was arrested at the heart cone stage, a phenotype reminiscent of /ι mutants. This phenotype correlated with severely disrupted apical ZO-1 junction belts (; = 0/8 hearts with intact ZO-1 junction belts). In comparison, 16-somite stage embryos of different genetic backgrounds (including ; double mutant/morphants) exhibited intact apical ZO-1–positive junction belts (Fig. S1, available at ). We next investigated whether interaction between Had/Na,K ATPase and Nok/Mpp5 is via regulation of each other's subcellular localization. To test this possibility, we analyzed morphants using an antibody against aPKCι and ζ as a marker for the apical Nok/Mpp5–Par6–aPKC protein complex (; ) and detected normal localization at the membrane at 34–36 hpf (). For the converse analysis, we characterized the subcellular localization of Myc-tagged Had/Na,K ATPase in wt (), mutant (), and mutant backgrounds (). In both mutants, the fusion protein was correctly localized to the membrane. Therefore, Had/Na,K ATPase and Nok/Mpp5 do not affect each other's membrane association. However, the squamous morphology of cardiomyocytes prevented an unambiguous characterization of protein distribution along the apical–basal axis. At the 20-somite stage, myocardial cells exhibit cuboidal shapes and are highly polarized. As shown in , morphants displayed correctly localized aPKC and ZO-1 junction belts (), suggesting that apical–basal polarity was not impaired. However, whereas aPKC was strongly localized to apical junction belts in wt cardiomyocytes (), it was clearly displaced in morphant cardiomyocytes, indicating a loss of apical–basal polarity (). In addition, we visualized the subcellular localization of Had/Na,K ATPase by analyzing the distribution of the exogenous Myc-tagged Had/Na,K ATPase. Although we consistently detected low levels of Myc-tagged Had/Na,K ATPase localized to the membrane of wt cardiomyocytes (), high levels of Myc-tagged Had/Na,K ATPase were detected around the circumference of myocardial cells in both and morphants (; five embryos analyzed for each genotype). These findings suggest that one way by which Nok/Mpp5 and Has/aPKCι affect Had/Na,K ATPase could be by directing its subcellular localization. The finding that Had/Na,K ATPase and Nok/Mpp5 interact in the maintenance of apical myocardial junctions raised the intriguing possibility that the ionic balance produced by the Na pump is critical in this process. To functionally characterize the role of the ion pump function, we first investigated the mechanisms by which Had/Na,K ATPase activity is regulated during heart morphogenesis. We initiated this characterization by a sequence comparison of the two functionally divergent α1B1 (Had) and α2 subunits of Na,K ATPase, which share a high degree of similarity throughout the entire peptide (84% identity), except for the first cytoplasmic domain (67% identity among the first 98 residues). In contrast to mutants, embryos deficient for the α2 subunit display cardiac laterality rather than primitive heart tube formation defects (). To test the relevance of N-terminal regulatory elements for α1B1 subunit function, we generated two expression constructs that encode chimeric proteins between the α1B1 and α2 subunits. The first construct encodes a chimeric protein containing the 98 N-terminal residues of the α1B1 subunit fused to the α2 C-terminal rest (N1C2), whereas the second construct encodes the reciprocal chimeric protein containing the N-terminal domain of the α2 subunit fused with the α1B1 C-terminal rest (N2C1). Although 77% of mutants injected with the N1C2 mRNA developed a normal heart tube ( = 35/45 embryos; rescue efficiency is similar to wt mRNA [81%, = 34/42 embryos]; P > 0.5), mutants could not be rescued by injection of N2C1 mRNA ( = 0/41 embryos; 0% rescue efficiency). This structure function analysis points at regulatory elements important for α1B1 subunit function within the N-terminal end of the protein. Intriguingly, there are several conserved aPKC consensus phosphorylation sites within the 25 most N-terminal residues of the α1B1 subunit that are missing within the α2 subunit (). To assess whether heart tube elongation depends on the regulation of Had/Na,K ATPase via one or more of these three N-terminal aPKC consensus phosphorylation sites, we generated expression constructs with point mutations that encode for phosphorylation-deficient forms of the protein (exchanges of Ser residues for Ala: , , and ), as well as a triple-mutant nonphosphorylatable form (; ). We assayed the effects of these mutations on heart tube elongation by their ability to rescue the mutant phenotype and found that injection of , , and a phosphomimetic mutant mRNA with an exchange of Ser25 to Glu () yielded a robust rescue of the heart tube elongation defects in most embryos that was comparable to mRNA (), whereas and mRNAs displayed significantly reduced rescue efficiencies (). Thus, some of the biological activity of Had/Na,K ATPase during heart tube elongation critically depends on Ser25. The zebrafish mutation /ι prevents normal heart tube elongation (), and our data showed that Had/Na,K ATPase is mislocalized in morphants, raising the possibility that Had/Na,K ATPase is a direct target of Has/aPKCι. Therefore, we conducted a direct phosphorylation assay, but did not detect any phosphorylation among the first 98 amino acids of Had/Na,K ATPase by either human recombinant aPKCζ or zebrafish Has/aPKCι (Fig. S2, available at ). This finding suggests that the N-terminal regulatory domain Had/Na,K ATPase is not a direct target of Has/aPKCι. Direct phosphorylation of rat Na,K ATPase by aPKC at Ser23 of the α subunit is an important mechanism by which the pump activity is regulated (, , 1999; ; , ). Phosphorylation at this residue causes the increased internalization of the Na pump into endocytic vesicles, effectively resulting in a reduced ion pump function at the plasma membrane. To assess whether reduced activity of Had could be the consequence of a changed subcellular protein distribution compared with the wt form, we analyzed the membrane association of the mutant and wt Myc-tagged fusion proteins. Western blot analysis demonstrated that the nonphosphorylatable form of the protein was enriched within the membrane and cytoskeletal fraction of embryonic extracts similar to the wt form (). To further quantify and compare the relative membrane distribution of wt and nonphosphorylatable forms of Had/Na,K ATPase, we measured the relative amounts of protein from three independently prepared protein fractionations (one of which is shown in ). Indeed, the relative membrane fractions accounted for 65.5 ± 5.3% SD for wt and 64.5 ± 6.7% SD for the nonphosphorylatable mutant forms of the protein, and demonstrated that the relative membrane association of Had/Na,K ATPase is not affected by Ser25 phosphorylation. In immunohistochemical stainings, both wt and nonphosphorylatable forms of the protein were largely associated with the outer cell membrane of myocardial cells (). Quantification of the relative membrane distributions of protein based on these stainings accounted for 60.8 ± 4.4% SD for wt ( = 45 myocardial cells) and 61.9 ± 4.7% SD for the nonphosphorylatable mutant form ( = 50 myocardial cells). These observations suggest that, during zebrafish embryogenesis, regulation via Ser25 does not cause a substantial removal of the Na pump from the outer cell membrane; rather, they suggest that phosphorylation of Ser25 positively controls ion pump activity at the outer cell membrane. To further explore the possibility that the ionic balance produced by the Na pump is critical in the maintenance of myocardial apical junction belts, we first produced a mutant form of Had/Na,K ATPase that specifically affects the ATPase catalytic activity, which is required to pump Na across the plasma membrane and tested whether this activity is essential for regulating primitive heart tube formation. We replaced the aspartic acid at position 379 by an asparagine (Had) to produce a mutant protein that has previously been shown to abolish the binding of ATP to Na,K ATPase in cultured cells (). Injection of mRNA into mutant embryos could not rescue the primitive heart tube phenotype (only 4% of injected mutants develop a heart tube, = 40). Nevertheless, the protein was stable and could be detected on Western blots (Fig. S3, available at ). Therefore, ATPase catalytic activity is required for heart tube elongation. Next, we analyzed ATPase catalytic and regulatory mutants in the genetic interaction with . Low levels of MO (2.5 ng) were coinjected together with ∷, ∷, or ∷ mRNA into mutants, and antibodies against the junctional protein ZO-1 were used to assess the integrity of apical myocardial junctions at 32 hpf. Indeed, unlike the control ; single-mutant embryos, ; and ; double mutant/morphants displayed disrupted ZO-1 junctional belts (). Similarly, embryos treated with ouabain, which is a potent inhibitor of Na,K ATPase activity, resulted in a loss of ZO-1 junction belts (). Together, these findings suggest that the interaction between Nok/Mpp5 and Had/Na,K ATPase in the maintenance of myocardial ZO-1 junction belts requires the Na pump function, and that correct ionic balance contributes to the maintenance of myocardial integrity. Our study has demonstrated an essential role of the ion pump function of Had/Na,K ATPase for the maintenance of apical ZO-1 junction belts. Independent mutations that affect Na,K ATPase regulation and catalytic activity, which is required to pump Na across the plasma membrane, provide strong evidence for the importance of correct ionic balance of myocardial cells for cell polarity and heart morphogenesis. We have shown that impaired ion pump function enhances the loss of apical ZO-1–positive junction belts in a mutant background (). These findings are corroborated by previous studies in MDCK cells that suggested a critical requirement of the Na pump in the establishment of epithelial cell polarity and in the formation of tight junctions in a calcium switch assay (,). These studies also provided evidence for synergistic activities between the adherens junction protein E-cadherin and Na,K ATPase in the formation of continuous ZO-1–positive tight junction belts that are similar to our observation of genetic interactions between and during zebrafish heart morphogenesis (, ). There has been considerable interest in Na,K ATPase from a physiological standpoint. However, little is known about the regulation or cell biological functions of the Na pump in the context of early vertebrate development. Our study reveals that Ser25 within the α1B1 subunit positively affects its function. Loss of rescue activity associated with a Ser25Ala mutation and normal rescue activity associated with a Ser25Glu phosphomimetic mutation suggest a positive regulatory mechanism for Na,K ATPase function during zebrafish heart development. Studies using D1-transfected OK cells (a cell line derived from Opossum kidney) and oocytes suggest that an important mechanism of regulation of Na pump function is via aPKC-mediated inhibition involving phosphorylation of Ser23 of rat. This phosphorylation event results in the internalization of the Na pump via increased endocytosis and, therefore, effective inactivation (, ; , , ). According to , there are multiple potential phosphorylation motifs present within the zebrafish Na,K ATPase α1B1 subunit that may be phosphorylated by various isoforms of aPKC. Alternative potential phosphorylation events could be the basis for variable effects of aPKC on ion pump activity reported in the literature, and also for the seemingly different results obtained in our study (). We find that during early zebrafish development, the phosphorylation state of the N-terminal regulatory residues of Had/Na,K ATPase does not affect the relative membrane association. The experimental model system established in this study provides a framework for detailed approaches to dissect the molecular components involved in the regulation of the Na pump during heart morphogenesis and development in general. In this study, we found that the apical–basal polarity is lost in myocardial cells of morphants at the 20-somite stage, and that Had/Na,K ATPase is mislocalized along the entire outer cell membrane. These findings are consistent with a previous study showing that the polarized distribution of Na, K ATPase in MDCK cells is disrupted upon treatment with a dominant–negative form of aPKC () and support the notion that proper subcellular distribution of Had/Na,K ATPase is dependent on the activity of Has/aPKCι. Because our data demonstrate that the phosphorylation of Ser25 is critical for heart tube elongation, it was reasonable to hypothesize that Has/aPKCι regulates the subcellular distribution of Had/Na,K ATPase by directly phosphorylating this residue. In an in vitro assay, we could not detect direct phosphorylation of Had/Na,K ATPase, which does not rule out the possibility that Has/aPKCι needs cofactors that were missing in our assay system. Furthermore, studies on whether or not Has/aPKCι phosphorylates the C-terminal domains of Had/Na,K ATPase and whether or not there is an indirect effect of Has/aPKCι on the phosphorylation state of Had/Na,K ATPase will provide additional insights into the mechanistic relationship of Has/aPKCι and Had/Na,K ATPase during early heart morphogenesis. The phenotypic similarities of and mutants in heart morphogenesis and the more severe defects observed in ; double-deficient embryos suggest a common defect underlying the loss of myocardial morphogenetic potential. Our data indicate that weakening of tight junctions and ionic imbalances in myocardial cells contribute to the severe heart elongation phenotype observed in embryos deficient in , , and both. It is possible that correct ionic gradients, modulated by Had/Na,K ATPase, stabilize the integrity of the tight junction, and that weakening of the tight junction, the paracellular diffusion barrier, may then enhance ionic gradient imbalances. We have previously shown that the heart beat rate is reduced by 20% in mutants (). A similarly reduced heart beat rate is noted in mutants (118.3 ± 16.9 beats per minute [ = 46] vs. 142.8 ± 11.5 beats per minute in wt siblings [ = 32]), indicating a defective ionic balance or reduced Na pump activity. Interestingly, a junctional defect was also noted in Na,K ATPase mutants, suggesting that the role of Na,K ATPase in maintaining the integrity of cellular junction barriers is evolutionarily conserved from to fish (; ). Further studies are required to address the possible link between tight junctions and ionic gradients in the control of epithelial morphogenesis. Zebrafish were maintained at standard conditions (). Embryos were staged at 28.5°C () and according to somite number. The following fish strains were used: wt AB, () (), , , and Embryos were treated with ouabain as previously described (). Both wt and mutant forms of / were produced by PCR amplification from a full-length cDNA template, introducing the XhoI and XbaI restriction sites 5′ and 3′, respectively. Site directed mutagenesis was performed using the Quick Change kit (Stratagene) and constructs were subsequently subcloned into the pCS2 + HisMyc expression vector (). Primer sequences are available upon request. The shorter wt and mutant forms ( and ) were produced by PCR amplification of only the first 294 bp of the respective template clones, introducing a 3′ stop codon and XhoI and XbaI restriction sites 5′ and 3′, and were subsequently subcloned into pCS2 + HisMyc. For the generation of chimeric constructs between and , the first 294 bp of both genes were PCR amplified and blunt-end ligated into the reciprocal PCR fragments lacking the 5′-end sequences. Constructs were transcribed using the SP6 mMessage mMachine kit (Ambion). () and () embryos were injected with 3.8–5 ng and 2.5 ng of MO (). For rescue experiments, 100 pg of mRNAs were injected. For overexpression, 150 pg of or mRNA and 150 pg of or mRNA were used. The heart tube phenotype was evaluated at 24 hpf. Data presented are the means of at least three independent experiments. Statistical comparisons of rescue efficiencies of wt against different mutant forms of protein were made by tests, and P < 0.05 was considered statistically significant. Whole-mount antibody stainings were performed as previously described (). Samples were embedded in SlowFade Gold antifade reagent (Invitrogen). Transverse sectioning was performed according to . Sections were embedded in 1.5% low melting Agarose. All images were obtained at RT. Confocal images were obtained with a confocal microscope (TCS SP2; Leica) using 40× objective and 4× zoom, or with a confocal microscope (LSM 510 Meta; Carl Zeiss MicroImaging, Inc.; and ) using 63× objective and 2× zoom. TCS SP2 and LSM 510 software were used to capture the images. Images were processed using Photoshop (Adobe). The following antibodies were used: rabbit anti-aPKCζ (1:100; Santa Cruz Biotechnology, Inc.), mouse anti-ZO-1 (1:200; Zymed Laboratories), mouse anti-Myc (1:200; Invitrogen), goat anti–rabbit RRX (1:200), and anti–mouse Cy5 (1:200; Jackson ImmunoResearch Laboratories). Immunohistochemical quantification was performed by measuring membrane and cytoplasmic intensity with ImageJ freeware (W.S. Rasband, National Institutes of Health, Bethesda, MD; ). Membrane area was defined using aPKC membrane staining as a marker. Relative membrane versus cytoplasmic distribution of wt or 3A mutant Had protein was calculated by comparing the intensity of membrane area with the corresponding sum of total cytoplasmic plus membrane areas for each individual myocardial cell. Whole-mount in situ hybridization was performed as previously described (). The antisense RNA probe used in this study was (a gift from D.Y.S. Stainier, University of California, San Francisco, San Francisco, CA). Embryos were embedded in glycerol. Images were obtained at RT with a SV11 stereomicroscope (Carl Zeiss MicroImaging, Inc.) using the 1.6× objective and 6.6× zoom with the Axiocam camera and Axiovision software (Carl Zeiss MicroImaging, Inc.). Photos were processed using Photoshop. Fractionation of 24 hpf zebrafish embryos was done essentially as previously described (). Membranes were probed with mouse anti-Myc antibody (1:1,000; Invitrogen). For loading and fractionation control, membranes were stripped and tested for acetylated tubulin (mouse antiacetylated tubulin; 1:1,000; Sigma-Aldrich). For genotyping of 16-somite stage embryos, we made use of a SalI restriction site that is deleted by the mutation and performed PCR on tail tissue, followed by the SalI digest. DNA primers used for genotyping are available upon request. We also used RFLP to genotype the rescued mutant embryos as previously described (). or mRNA at the one-cell stage. Embryo (24 hpf) extracts were prepared in lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 5 mM β-mercaptoethanol, 20 mM imidazole, 1% C12E10, 0.2 mM PMSF, protease inhibitor cocktail [Roche]). HisMyc-tagged fragments were purified using Ni-NTA columns (QIAGEN). In vitro kinase assays were performed on the eluted HisMyc-tagged protein. Samples were incubated at room temperature for 30 min with 50 ng human recombinant aPKCζ (Calbiochem) or Has/aPKCι in kinase reaction buffer (20 mM Tris-HCl, pH 8.0, 5 mM MgCl, 100 mM imidazole, 15% glycerol, 30 mM NaHPO, pH 8.0, 0.05% Tween 20, 50 μM ATP, 5 μCi γ[P]-labeled ATP). Kinase reaction was stopped by adding 4× SDS loading buffer and boiling at 95°C for 5 min. We used the commercially available peptide epsilon (25 ng loaded per lane) as positive control for the kinase assay (Calbiochem). Samples were split in two, for autoradiography and for Western-blot analysis, and resolved in 18% acrylamide/bisacrylamide (29:1) gels. For detection of P incorporation, the gel was dried and visualized by autoradiography. For Western blot analysis, the following antibodies were used: mouse anti-Myc (1:1,000; Invitrogen), rabbit anti-aPKCζ (1:1,000; Santa Cruz Biotechnology, Inc.), goat anti–rabbit HRP (1:5,000; Pierce Chemical Co.), and goat anti–mouse HRP (1:10,000; Jackson ImmunoResearch Laboratories). Recombinant Has/aPKCι was generated by injecting embryos with 300 pg of ι mRNA. Embryo (24 hpf) extracts were prepared in lysis buffer. HisMyc-tagged Has/aPKCι protein was purified using Ni-NTA columns, and eluted in kinase elution buffer (20 mM Tris-HCl, pH 8.0, 300 mM NaCl, 5 mM MgCl, 100 mM imidazole, 30 mM NaHPO, pH 8.0, 0.05% Tween 20, and 30% glycerol). Fig. S1 shows that the establishment of ZO-1–positive tight junction belts is not affected in mutants, morphants, and ; double mutant/morphants at the 16-somite stage. Fig. S2 shows that the N-terminal tail of Had/Na,K ATPase is not directly phosphorylated in vitro by Has/aPKCι. Fig. S3 shows that the ATPase catalysis mutant Had is stable and correctly associates with membranes.

Neutrophils are one of the first lines of defense against invading microbes (; ). These cells are terminally differentiated, and they have a short life span and low levels of gene expression. When they reach the circulation, they are already equipped with the proteins required to kill microorganisms (). Neutrophils in circulation are directed by cytokines into infected tissues, where they encounter invading microbes. This encounter leads to the activation of neutrophils and the engulfment of the pathogen into a phagosome. In the phagosome, two events are required for antimicrobial activity. First, the presynthesized subunits of the NADPH oxidase assemble at the phagosomal membrane and transfer electrons to oxygen to form superoxide anions. These dismutate spontaneously or catalytically to dioxygen and hydrogen peroxide. Collectively, superoxide anions, dioxygen, and hydrogen peroxide are called reactive oxygen species (ROS; ). Second, the granules fuse with the phagosome, discharging antimicrobial peptides and enzymes. In the phagosome, microorganisms are exposed to high concentrations of ROS and antimicrobial peptides. Together, they are responsible for microbial killing (). Patients with mutations in the NADPH oxidase suffer from chronic granulomatous disease (CGD; ). CGD patients are severely immunodeficient, have recurrent infections, often with opportunistic pathogens, and have poor prognosis. Recently, we described a novel antimicrobial mechanism of neutrophils. Upon activation, neutrophils release extracellular traps (neutrophil extracellular traps [NETs]; ). NETs are composed of chromatin decorated with granular proteins. These structures bind Gram-positive and -negative bacteria, as well as fungi (). NETs provide a high local concentration of antimicrobial molecules that kill microbes effectively. NETs are abundant at inflammatory sites, as shown for human appendicitis and an experimental model of shigellosis. Recently, NETs were shown to be relevant in vivo in human preeclampsia () and streptococcal infections (), causing necrotizing fasciitis () and pneumococcal pneumonia (). The release of intact chromatin decorated with cytoplasmic proteins into the extracellular space is unprecedented. We describe that activated neutrophils initiate a process where first the classical lobulated nuclear morphology and the distinction between eu- and heterochromatin are lost. Later, all the internal membranes disappear, allowing NET components to mix. Finally, NETs emerge from the cell as the cytoplasmic membrane is ruptured by a process that is distinct from necrosis or apoptosis. This active process is dependent on the generation of ROS by NADPH oxidase. In an infection, ROS formation may contribute to the following two antimicrobial pathways: intraphagosomal killing in live neutrophils and NET-mediated killing post mortem. To analyze NET formation, we monitored individual neutrophils with live-cell imaging through four different channels. First, we recorded the phase-contrast image to determine the morphology. Second, to assess cell viability, neutrophils were loaded with calcein blue, a dye that is retained in the cytoplasm of living cells and rapidly lost upon cell death. Third, the neutrophils were incubated in the presence of Annexin V, which binds to phosphatidylserine (PS). PS is localized to the inner leaflet of the cell membrane. Annexin V can only bind to PS of cells undergoing apoptosis, when PS is transferred to the outer leaflet, or after membrane rupture, when Annexin V can enter into the cell. Thus, if the plasma membrane breaks, the cells lose the vital dye and are stained with Annexin V simultaneously. If a cell undergoes apoptosis, it will first become Annexin V–positive and later lose the vital dye. Fourth, to detect the appearance of NETs, we used fluorescently labeled Fab fragments of monoclonal antibodies against the complex composed of histone 2A, histone 2B, and DNA ( and Video 1, available at ; ) or neutrophil elastase (Fig. S3 and Video 2). In viable neutrophils, neither Fabs nor Annexin V have access to their targets. When NETs emerge or cells die, Fabs and Annexin V can bind; because of the increase in the local concentration, they become detectable. Purified peripheral blood neutrophils were activated with PMA and monitored by live-cell imaging for 240 min. Initially, neutrophils flattened and formed numerous intracellular vacuoles ( and Video 1). After 80 min, nuclei lost their lobular shape, and they expanded and filled most of the intracellular space (). At this time, the cells were viable because they retained calcein blue (, blue) and excluded Annexin V (green). After 220 min of activation, progressively more cells lost the calcein blue marker and concurrently stained with Annexin V (), indicating rupture of the plasma membrane. NETs only became detectable when the cell membrane was ruptured. As shown in , NETs (, red) appeared exactly when cells lost the vital dye and simultaneously became positive for Annexin V (). This clearly demonstrates that NETs emerge from dying neutrophils. This also shows that NETs are not released by apoptotic neutrophils because PS was not exposed before the rupture of the plasma membrane, which is indicated by the loss of calcein blue. In contrast, apoptotic neutrophils became Annexin V positive before the integrity of the plasma membrane was compromised (unpublished data). Controls with Fabs against irrelevant antigens did not stain the NETs (unpublished data). The process of NET formation was identical when neutrophils were seeded on plastic, glass, or collagen-coated glass. To investigate the morphological changes leading to the cell death and NET formation, we fixed neutrophils at different time points after activation and analyzed their morphology by transmission electron and fluorescence microscopy. 60 min after stimulation, the nuclei started to lose their lobules, and the chromatin began to decondense while the nuclear membrane remained intact (). At this time point, the space between the inner and outer nuclear membrane dilated (0 min, 18.8 ± 3.9 nm; 60 min, 27.9 ± 5.8 nm). 120 min after stimulation, the nuclear membranes formed distinct vesicles (), and by 180 min the nuclear envelope disintegrated into numerous small vesicles and the chromatin decondensed (). These vesicles originated from the nuclear envelope, as demonstrated by immune detection of the lamin B receptor, which is a component of the inner nuclear membrane (; ), and by immunofluorescence staining of nuclear membrane proteins (). By 180 min, most granules disappeared (). With the loss of nuclear and granular membranes, the decondensed chromatin came into direct contact with cytoplasmic and granular components. We showed by immunofluorescence that 120 min after activation the granular marker neutrophil elastase colocalized with chromatin, although some granular punctuate staining remained (). Later in the activation, the granular staining was lost and all the neutrophil elastase was associated with chromatin (). The intracellular mixing of granular components and chromatin was supported by the observation that most of the neutrophil elastase released by the cell remained bound to NETs (). Interestingly, the plasma membrane integrity of activated neutrophils was not affected by disruption of internal membranes, allowing the intracellular mixing of NET components (). Similar morphological changes were observed when neutrophils were infected with bacteria (). We compared the cell death that leads to NET formation to apoptosis and necrosis. We induced apoptosis () by incubating neutrophils with anti-Fas antibodies for 18 h (). In these conditions, ∼70% of the cells were dead. Apoptotic neutrophils showed the classical morphology, including condensation of chromatin and fragmentation of nuclei without rupture of the nuclear envelope, as well as cytoplasmic vacuolization. Organelles in the cytoplasm remained intact. Neutrophils incubated with high concentrations of secreted pore-forming toxins from for 15 min when >70% of the cells were dead showed typical necrotic morphology (; ). The nuclei lost their structure, and lobules fused into a homogenous mass without segregation into eu- and heterochromatin and did not make NETs, even in long incubation periods. In contrast to cells forming NETs (), the nuclear membranes remained intact and clearly separated nucleoplasm from cytoplasm. The structure of organelles was not affected. shows a neutrophil activated with PMA for 4 h to promote NET formation. The most salient morphological differences in comparison to the apoptotic and necrotic cells are disintegration of the nuclear envelope and mixing of nuclear and cytoplasmic material, loss of internal membranes, and disappearance of cytoplasmic organelles. Neutrophils activated with PMA did not proceed to have a necrotic morphology, even after longer incubation. DNA fragmentation, as detected by TUNEL, is a hallmark of apoptosis and was only apparent after stimulation with anti-Fas antibodies (), but not when the cells died by necrosis after incubation with toxins. Neutrophils activated to make NETs were also TUNEL negative. We quantified NET formation under these different conditions by digesting the DNA scaffold of NETs with an endonuclease and measuring the content of released DNA in the supernatant (Fig. S1, available at ). NETs were detected after PMA activation (), but not after incubation with Fas antibody or treatment with toxins. Together, these data indicate that neither apoptosis nor necrosis lead to NET formation, and that NET-inducing cell death is different from both apoptosis and necrosis by morphological and molecular criteria. Stimulation of neutrophils with interleukin (IL)-8, lipopolysaccharide (LPS), or PMA (; ) induced NETs as early as 45 min after activation. IL-8– or LPS-activated neutrophils formed NETs above background (Fig. S1 a). The number of NETs were detected microscopically, but were under the detection level of the nuclease method described in Fig. S1 (b–f). We also observed that infection of neutrophils with bacteria produced a robust amount of NETs ( and S1 a) that increased with time. To investigate the mechanism of NET formation, we stimulated neutrophils with the potent activator PMA, as well as with viable bacteria, as a biologically relevant stimulus. As expected, we observed differences in NET formation between different donors (Fig. S5), and experiments are representative and done in triplicate or quadruplicate with neutrophils from a single donor. Neutrophils infected with () or PMA () formed NETs. Typical morphological features, like decondensed nuclei merging with granular components in the cytoplasm, as well as extracellular fibers consisting of DNA, histones, and elastase, were detected after stimulation with both stimuli (), but not in unstimulated cells (). Quantitation of NETs showed that ( and ) induced more NETs, and after 45 min the kinetics are even faster than after stimulation with PMA (). It is important to note that in these infections was grown in conditions where the toxins are not expressed (see Materials and methods). In these experiments, all cells were incubated for the same amount of time, but stimulated for the indicated periods. That is, cells that were not stimulated (time 0 min) were in fact incubated for the duration of the experiment. Accordingly, cells stimulated for the entire duration of the experiment are shown as 180 () or 240 min (). Stimulation with live bacteria or PMA triggers the assembly and activation of NADPH oxidase and the production of ROS (). Therefore, we tested if this enzyme was required to make NETs. The NADPH oxidase inhibitor diphenylene iodonium (DPI) prevented NET formation (), ROS production (), and cell death () upon activation with or PMA. To test whether stimulation of neutrophils downstream of NADPH oxidase produces NETs, we generated hydrogen peroxide exogenously using glucose oxidase (GO) and showed that stimulation with hydrogen peroxide, which is membrane permeable, induces NETs (). At the concentrations used, GO produced ROS comparable to – or PMA-stimulated neutrophils (). As expected, because hydrogen peroxide stimulates downstream of NADPH oxidase, neutrophils incubated with GO died () and made NETs () even when NADPH oxidase was inhibited. Control experiments demonstrate that NET formation was specific to neutrophils, as stimulation of peripheral blood mononuclear cells (PBMCs) with PMA (unpublished data) or GO did not release any DNA (). To determine whether ROS were regulating the process of NET formation we tested the role of catalase in this process. Catalase converts hydrogen peroxide into water and dioxygen. Accordingly, the presence of exogenous catalases reduced NET formation in response to PMA activation (). Catalases completely inhibited NET formation after GO activation, confirming that stimulation by GO is exclusively caused by the production of hydrogen peroxide (). Conversely, the inhibition of endogenous catalases with 3-amino-1,2,4-triazole (AT), dramatically increased NETs in response to different stimuli (). Interestingly, serum has antioxidant activity (), and, consistently, NET formation is inhibited in a concentration-dependent fashion by serum (Fig. S2, available at ). Therefore NET induction is optimal at low serum concentrations (≤2%). Mutations in the phagocyte NADPH oxidase cause CGD. We tested neutrophils isolated from five patients that have mutations in the NADPH oxidase. We confirmed that neutrophils from each of these patients are unable to generate ROS upon PMA activation (). The morphology of naive neutrophils isolated from CGD patients is indistinguishable from that of neutrophils isolated from healthy donors (). Interestingly, neutrophils isolated from CGD patients activated with () or PMA () did not show NETs and lacked the morphological changes characteristic of NET formation, such as breakdown of the nuclear envelope and the mixing of NET components within the cytoplasm. However, neutrophils from these patients, when stimulated with GO, formed NETs that were similar to those made by neutrophils from healthy donors (; and Fig. S4 and Video 3, available at ). Quantification of DNA revealed that CGD neutrophils do not make NETs upon and PMA activation, but release NETs at normal levels after GO stimulation (). Together, the data with ROS inhibitors and from neutrophils isolated from CGD patients show that NADPH oxidase is required to trigger a signal that culminates in the formation of NETs. We compared the antimicrobial activity of viable, naive neutrophils and neutrophils that formed NETs. Although microbes induce NETs by themselves, in this experiment we synchronized NET formation by preactivating the cells with either PMA () or GO (). After isolation, the stimulus was added to the neutrophils so that by 240 min after the initiation of the experiment, the cells would be stimulated for 0, 60, 120, 180, and 240 min. Next, the stimulus was removed and the cells were incubated in the presence or absence of DNase before infection with . In the absence of DNase (, untreated), neutrophils can kill the bacteria by both phagocytosis and through NETs. In contrast, in the presence of DNase the NETs are dismantled (Fig. S1 c) and bacterial killing is only mediated by phagocytosis. The differences in bacterial killing in cultures with and without DNase reflect the antimicrobial activity of NETs. Efficient bacterial killing was observed with naive neutrophils (0 min), and the presence of DNase had no effect, indicating that the antimicrobial activity was exclusively through phagocytosis. 60 min after stimulation, when the neutrophils start to undergo morphological changes but the cells have not made NETs yet (), phagocytic killing is decreased and little, if any, NET killing is observed because there is no difference between untreated cultures and cultures incubated with DNase. By 120 min, neutrophils start dying and forming NETs, and from this time point on most of the antimicrobial activity in the culture is NET mediated. Infection of neutrophils that were activated for 240 min, where the cells were dead and had formed the maximal amount of NETs, showed that NET killing can be as effective as phagocytosis (compare to time 0). Similar results were obtained if either PMA or GO were used for prestimulating the cells. As expected, neutrophils from CGD patients, whether PMA stimulated or not, showed impaired killing of because ROS are required for phagosomal killing and to make NETs (). Interestingly, when neutrophils from CGD patients were stimulated with GO for 240 min, bypassing the requirement of NADPH oxidase for NET formation, was killed very efficiently. This killing activity was mediated by NETs because it was completely abrogated by DNase treatment. These data further support our finding that NADPH oxidase is essential not only for phagosomal killing (; ) but also for NET formation. The additive roles of this enzyme likely explain the severe clinical infections of patients with CGD. NET formation is a potent antimicrobial mechanism of neutrophils. NETs have been documented in vivo in several pathological conditions, including appendicitis, experimental models of shigellosis (), and preeclampsia (). Furthermore, the importance of NETs in host defense was recently demonstrated in models for necrotizing fasciitis () and pneumonia (). The release of intact chromatin into the extracellular space is unprecedented. We describe that after activation, the neutrophils become highly phagocytic and eventually undergo morphological changes that lead to NET formation ( and and Videos 1 and 2, available at ). The changes follow a particular pattern that is initiated by the loss of nuclear segregation into eu- and heterochromatin. Simultaneously, the characteristic lobular form of the nucleus is also lost. At this point the nuclear membranes () start to separate from each other, but the morphology of cytoplasm and organelles seem intact. At later time points, the nuclear envelope disintegrates into vesicles () and the granular membranes disappear, allowing the mixing of nuclear, cytoplasm, and granular components. Throughout this process the cell membrane is intact, and only after the chromatin and granular components are mixed does it break, allowing the extrusion of the NETs. The mechanism of NET formation is clearly distinct from apoptosis () because there is no DNA fragmentation, PS is not exposed before cell death, and the morphological characteristics of these two forms of active cell death are very different (). Furthermore, caspases, which are proteases that are the executioners of apoptosis, do not seem to be involved in the process of NET formation (unpublished data). There are several reports about apoptosis in activated neutrophils (; ; ). Apoptosis was shown to depend on the production of ROS, independent from caspases, and does not feature DNA fragmentation. These observations seem inconsistent with the data presented here. The discrepancy is likely caused by several differences in the experimental setup. We performed all our experiments with adherent neutrophils and in the absence or low concentrations of serum, which mimics conditions at inflammatory sites. Many of the published results on induction of ROS-dependent apoptosis were performed with neutrophils in suspension and the presence of high concentrations of serum, which inhibits NET formation (Fig. S2). Furthermore, cells that have lost their membrane integrity were not considered (; ). This might have excluded neutrophils that had potentially made NETs. It would be interesting to reanalyze NET formation under the conditions used by other investigators. In addition, our data are consistent with previous studies showing that the cytotoxicity induced by PMA and hydrogen peroxide (; ; ) in neutrophils is ROS dependent and different from apoptosis because it is caspase independent, does not induce DNA fragmentation, and has unique morphological features, such as the loss of nuclear membrane. Necrosis is also distinct from the process of NET formation. The most apparent difference is the morphological change of the nucleus preceding the formation of NETs. While in necrosis, the nuclear envelope remains intact (), whereas early in the process that leads to NETs, the nuclear membranes disintegrate into a multitude of vesicles, allowing the mixing of nuclear components and granular material. Another clear difference is the requirement for specific cellular activation and the assembly and activity of the NADPH oxidase in the case of NET formation. The conclusion of our results is that NET formation is the consequence of a novel form of active cell death. We consistently observe that, depending on the stimuli and donor (Fig. S5, available at ), only a certain percentage of the neutrophils in a culture make NETs. The significance of this observation is not clear. Experiments with neutrophils seeded at very low numbers show that the majority of neutrophils in circulation are capable of making NETs (unpublished data). This suggests that in neutrophil cultures, and maybe also in vivo, there are mechanisms to regulate which neutrophils start a program that culminates in NET formation. To investigate the signaling pathway contributing to NET formation, it was important to exclude preactivation of cells during the isolation procedure. Initial observations of NETs were from neutrophils isolated by a dextran/ficoll method (). As was also observed by others (), some neutrophil dextran/ficoll preparations were preactivated, which lead to high background of NETs in unstimulated controls. Interestingly, this isolation method may cause ROS production in neutrophils (). The data presented here was generated with neutrophils isolated by percoll density-gradient centrifugation. This method avoids steps like dextran erythrocyte sedimentation and lysis (), which activate cells. Using this method, we routinely isolated >98% pure and unstimulated neutrophils from human blood. We also observed that IL-8 and LPS induce NETs less efficiently than bacteria. We propose that optimal NET formation requires activation through multiple receptors. Further experiments should address this point. In our initial description of NETs (), we suggested, based on indirect evidence, that NETs are actively released from living neutrophils. We show, using live-cell imaging (, S3, and S4), the status of individual neutrophils from activation to NET formation, and demonstrate that NET formation is the final step in a program of active cell death and that NETs are released in the moment an activated neutrophil dies. We show, pharmacologically and genetically, that ROS produced by the NADPH oxidase are required, and that hydrogen peroxide is sufficient for the formation of NETs ( and ). Indeed, it is interesting that, although PMA has many effects in the cells besides the activation of NADPH oxidase through PKC activation, the morphology of cells treated with PMA and GO share morphological changes like nuclear membrane disintegration, chromatin relaxation, mixing of cytoplasmic and nuclear components, and other features of cell death that result in NETs. Because catalases decompose hydrogen peroxide to water and oxygen, they regulate NET formation. The inhibition of neutrophil catalases resulted in a dramatic increase of NET formation, whereas exogenous catalases had an inhibitory effect. Because catalases are commonly found in bacteria, it is conceivable that microbes use catalases to regulate NET formation (). Our data show that ROS function as signaling molecules (; ; ), but the nature of the downstream signaling is still unclear. It has been shown that the oxidation of the active cysteine in phosphatases () to sulfenic acid transiently inactivates these enzymes. This ROS-dependent modification happens at ROS concentrations that are present in activated neutrophils. The oxidized form of proteins is extremely short lived; therefore, it is difficult to determine the specific targets of ROS in neutrophils. Also, suggested that ROS production in neutrophils might function as second messengers that alter the granular physiology. Indeed, they show that ROS induce modifications in the phagocytic vacuole that are reminiscent of the initial changes in the neutrophils before formation of NETs. The implications of our results are that the severe immunodeficient phenotype of CGD patients might be linked to NET formation (). The lack of NADPH oxidase activity might have different effects, including the reduction in the direct antimicrobial activity of ROS (), the lack of granule content release (), and the inability to make NETs. The catastrophic consequences of mutations in NADPH oxidase might be a combination of these three effects. At this point, it is unclear why only granular, and not cytoplasmic, proteins were observed decorating the NETs (). Our results suggest that the chromatin is also in contact with the cytoplasm (), but it is probable that the interaction between granular proteins and chromatin are based on charge because most of the granular proteins observed on NETs are highly cationic and might avidly bind to DNA. Further investigations will have to comprehensively determine the molecular composition of NETs and clarify whether, indeed, there are cytoplasmic proteins present in them. Neutrophils are terminally differentiated when they leave the bone marrow, and they have a short life span in circulation (). Their fate is either to be cleared from circulation or recruited to an inflammatory site. Upon microbial challenge, neutrophils are activated, and, initially, they become phagocytic. In our in vitro models, during the first 60 min, phagocytosis is the main mechanism to clear microbes (). 120 min after stimulation, their phagocytic capacity is restricted. At this point, the second antimicrobial mechanism takes over. Nuclear and granular components mingle in the cytoplasm and are rapidly released when the neutrophils finally die, filling up the extracellular space with antimicrobial NETs. A hitherto unknown form of active cell death apparently evolved to allow neutrophils to kill microbes post mortem. In this form of cell death, the potent cationic antimicrobial peptides and proteins of neutrophils are mixed with chromatin and released to form NETs. Interestingly, the generation of ROS by NADPH oxidase is required for efficient phagocytic killing, and ROS act as a second messenger to trigger NET formation. Importantly, in this form of cell death DNA fragmentation is not activated, allowing the chromatin to unfold in the extracellular space. NETs can bind and kill microbes by providing a high local concentration of antimicrobial peptides and, at the same time, minimize tissue damage by sequestering the noxious granule enzymes. The novel form of active cell death allows neutrophils to pursue their antimicrobial battle even beyond their life span. Human neutrophils and PBMCs were isolated from healthy donors or patients with CGD using density gradient separation (). For control experiments, we isolated neutrophils by a dextran/ficoll method (). Patients with CGD were recruited in the Centre for Patients with Primary Immunodeficiency (Charité Medical University, Berlin). Diagnosis of CGD was determined by DHR test and genetic analysis of NADPH components, and was confirmed by chemiluminescence. All patients and controls gave written consent and the study was approved by the ethics committee of the Charité. If not stated otherwise, cells were resuspended in RPMI medium (phenol red–free) supplemented with 10 mM Hepes and 2% human serum albumin. 5 × 10–10 neutrophils/ml were seeded into tissue culture plates, glass, collagen-coated glass, collagen-coated glass for live cell imaging, or on glass coverslips for immunofluorescence pretreated with 0.001% polylysine. Incubations were performed at 37°C in the presence of 5% CO. 8324–5 (provided by R. Novick, New York University, New York, NY) was grown to exponential phase in brain heart infusion medium with aeration. Bacteria were washed twice with chilled PBS. 2 × 10 bacteria/ml were added to 10 neutrophils/ml in the presence of 0.5% heat-inactivated serum. Bacteria were centrifuged onto the neutrophils at 200 g and 37°C. The following stimuli were used at indicated concentrations: 1 μg/ml LPS in combination with 500 ng/ml of LPS-binding protein (Biometec), IL-8 at 100 ng/ml, PMA at 10–20 nM, and 100 mU/ml of the HO-producing enzyme GO (Worthington Biochemical Corp.). Neutrophils were incubated with inhibitors 30 min before stimulation. The NADPH oxidase inhibitor DPI (Calbiochem) was used at 5–10 μM and the catalase inhibitor AT (Calbiochem) at 1 mM. Catalase (Worthington Biochemical Corp.) was used at 100 U/ml. Heat-inactivated fetal calf serum was used at concentrations ranging from 5–20% to inhibit NET formation. For fine structural analysis, cells were fixed with 2.5% glutaraldehyde, postfixed with 1% osmiumtetroxide, contrasted with uranylacetate and tannic acid, dehydrated, and embedded in Polybed (Polysciences). After polymerization, specimens were cut at 60 nm and contrasted with lead citrate. For immunodetection, cells were fixed with 4% PFA and embedded in a mixture of 25% sucrose/10% PVA. Ultrathin sections were cut at −105°C, blocked, and reacted with primary antibodies (anti-LBR; Epitomics), followed by secondary antibodies coupled to 6- or 12-nm gold particles. Specimens were analyzed in a Leo 906E transmission electron microscope (Oberkochen) using digital cameras (Ultraview and Morada; SIS). Neutrophils were seeded on glass coverslips treated with 0.001% polylysine, allowed to settle, and treated with PMA (25 nM) or left unstimulated. Cells were fixed with 4% PFA, blocked (3% normal donkey serum, 3% cold water fish gelatin, 1% bovine serum albumin, and 0.05% Tween 20 in PBS) and incubated with primary antibodies antinuclear membrane (ab12365; Abcam); anti–H2A–H2B–DNA complex (); anti-neutrophil elastase (in-house); anti– (Biodesign), which were detected with secondary antibodies coupled to Cy2 or Cy3 (Dianova). Controls were done with isotype-matched controls. For DNA detection, DRAQ5 (shown) Bisbenzimide 33342 (Hoechst 33342), Sytox, and ToPro3 were used. Specimens were mounted in Mowiol and analyzed with a PlanApo 63×/1.32 NA objective on a confocal microscope (TCS-SP; Leica). 2–5 × 10 neutrophils were stained with calcein blue AM (Invitrogen) and seeded into culture dishes equipped with glass bottoms (Mattek). The medium contained Annexin V (1:50; Invitrogen, labeled with Alexa Fluor 488; or Abcam, labeled with Cy3) and Fabs against the H2A–H2B–DNA complex () directly coupled to Atto 555 or against neutrophil elastase (in house) directly coupled to Atto 488 at a concentration of 2 μg/ml. Cells were stimulated with 20 nM PMA and recorded at 37°C on a microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) with a Plan-Neofluar 100×/1.3 NA objective over a period of 4 h. Each minute, a set of four images (phase contrast, blue, green, and red fluorescence) was taken with a camera (Orca ER; Hamamatsu). The system was controlled by the Openlab software (Improvision). Individual frame overlays and videos were prepared using Volocity software (Improvision). NETs generated by activated neutrophils were digested with 500 mU/ml micrococcal nuclease (MNase; Worthington Biochemical Corp.). The nuclease activity was stopped with 5 mM EDTA and the culture supernatants were collected and stored at 4°C until further use. Total DNA was isolated from naive neutrophils with DNazol supplemented with 1% polyacryl carrier (Molecular Research Center) according to the manufacturer's instructions and solubilized in TE buffer. NETs or genomic DNA was quantified using Picogreen dsDNA kit (Invitrogen) according to manufacturer's instructions. Total DNA of neutrophils and PBMCs was isolated from four different donors and quantified. 10 neutrophils had 6.2 μg DNA (±0.27 SD) and 10 PBMCs contained 6.3 μg DNA (±0.81 SD). Percentage of released NET-DNA was calculated by dividing the amount of isolated NET-DNA through the average genomic DNA content. Neutrophils were activated for indicated times with PMA. Neutrophil elastase in suspension was collected in culture supernatants. NET-associated neutrophil elastase and NET-DNA was isolated by incubating the cells in medium containing 500 mU/ml MNase for 10 min. Samples were resuspended in 1 M NaCl 0.02% Triton X-100. Total neutrophil elastase was measured from unstimulated neutrophils lysed with 0.02% Triton X-100 in 1 M NaCl. Neutrophil elastase activity was quantified with 100 μM of the peptide substrate N-(Methoxysuccinyl)-Ala-Ala-Pro-Val 4-nitroanilide for 15 min at room temperature. Optical density was measured at 405 nm (Microplate reader EL800; BIO-TEK Instruments). Neutrophils were incubated with 20 ng/ml anti-Fas (Millipore) antibodies for up to 18 h to induce apoptosis. We induced neutrophil necrosis with secreted toxins from Strain 8325–4 was grown overnight in BHI medium. Bacteria were pelleted, supernatants were collected, and the filter was sterilized. Protein concentration was determined using RC DC Protein Assay (Bio-Rad Laboratories). Necrosis was elicited by treating neutrophils with 1 mg/ml of secreted proteins for 15 min. Cells were stained by the addition of 2 μM Sytox Green (Invitrogen) and were analyzed by fluorescence microscopy. Apoptosis was detected by TUNEL staining (Promega) according to manufacturer's instructions. At least 250 cells per sample were analyzed. ROS production was measured by chemiluminescence (). Neutrophils were activated in the presence of 50 μM luminol and 1.2 U/ml horseradish peroxidase (Calbiochem) and chemiluminescence was detected using a Monolight 3096 (BD Biosciences). was grown to exponential phase in brain heart infusion medium with aeration and washed twice with chilled PBS. 10 human neutrophils/ml were stimulated for the indicated time points up to 240 min with PMA or GO or left unstimulated. The cells were stimulated at different time points, so all wells were incubated for 240 min, but stimulated for the indicated periods. The cells were further incubated in fresh medium without PMA or GO containing 2% heat-inactivated pooled human serum with or without 100 U/ml DNaseI. The culture was infected with 10 and centrifuged for 10 min at 800 and further incubated for 20 min. After adding 5 mM EDTA, the samples were vortexed for 5 min and the cell suspensions were diluted in chilled PBS with 0.1% Triton X-100. The microbicidal effect observed was the same if the cells were lysed with 0.1% Triton X-100, 0.1% Saponin, or distilled water. Aliquots were plated to determine CFUs. Absence of bacterial killing was defined as the number of bacteria recovered after infection of DNaseI-treated neutrophils that were stimulated for 240 min with PMA or GO. At this time point, all the cells were dead and unable to phagocytose (), and degradation of NETs prevented extracellular killing. Fig. S1 shows the method for quantification of NETs. Fig. S2 shows that serum inhibits NET formation. Video 1 (key frames are depicted in ) shows that NETs detected by a chromatin marker are released when the cell membrane ruptures. Fig. S3 and Video 2 show NET formation by using a granular marker. Fig. S4 and Video 3 show that neutrophils isolated from CGD patients make NETs in response to GO. Fig. S5 shows variation in NET formation between neutrophils from different donors. Online supplemental material is available at .

Modification of otherwise soluble proteins with hydrophobic moieties, such as myristoyl or isoprenyl groups, is essential for their targeting to cell membranes. These modifications occur in the cytosol and are irreversible. In contrast, thioester linkage of palmitate, a C16 saturated fatty acid, to cysteine residues (S-palmitoylation) is a reversible modification catalyzed by membrane-bound palmitoyl transferases (PATs). Thus, palmitoylation can be viewed as a secondary signal for membrane association, as other primary signals must bring the protein to the membrane to allow access to PAT enzymes. In some cases, palmitoylation provides a stable membrane anchor to proteins that have arrived at a membrane compartment via weak or transient interactions, including protein–protein interactions or prenylation/myristoylation. However, palmitoylation also occurs on proteins that are already tightly associated with membranes, including transmembrane proteins, indicating that palmitate is more than just a membrane tether. The recent identification of a large family of PATs containing a signature DHHC cysteine-rich domain has brought about a renewed interest in the mechanisms and functions of protein palmitoylation (). There is now accumulating evidence supporting a role for palmitoylation in regulating many aspects of protein trafficking within the cell. In this mini-review, we focus on specific studies that highlight the diversity of palmitoylation as a signal for protein sorting, before ending with a discussion of the possible mechanisms that underlie palmitoylation-dependent sorting. The effects of palmitoylation on protein sorting are not easily predicted, and indeed modification of different cysteines in the same protein can have distinct effects on trafficking. This is the case for the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor, a ligand-gated cation channel that mediates the fast component of glutamate-induced excitatory postsynaptic currents. All AMPA receptor subunits, GluR1-GluR4, are palmitoylated; palmitoylation sites are located at the intracellular face of the second transmembrane domain (site 1) and in the C terminus of the protein just downstream of the fourth transmembrane domain (site 2) (). Palmitoylation of site 1 was enhanced by coexpression of the PAT DHHC-3 (also called GODZ), promoting accumulation of the receptor in the Golgi and decreasing cell surface expression levels. As DHHC3 is localized to the Golgi (), this implies that palmitate addition to this cysteine residue promotes retention of the receptor at this compartment. Interestingly, several other proteins exhibit a similar intracellular retention upon overexpression of specific PAT enzymes (; ); however, the mechanism for this is not clear. In contrast to the effects of palmitoylation of site 1 in GluR subunits, palmitoylation of site 2 did not appear to regulate steady-state cell surface levels of the receptor. However, mutation of this cysteine residue in GluR1/2 inhibited activity- dependent internalization. Thus, regulated palmitoylation/depalmitoylation of both sites in GluR subunits is likely to play a key role in regulating surface expression of the AMPA receptor, albeit by different mechanisms. Many other transmembrane proteins rely on palmitoylation for correct sorting in mammalian cells. Recent examples of this include the human δ opioid receptor, a G protein–coupled receptor, which required palmitoylation for efficient biosynthetic delivery to the plasma membrane (PM; ), and the mucin-like MUC1 protein in which palmitoylation of two cysteine residues, although not required for biosynthetic delivery to the cell surface, was linked to efficient trafficking from recycling endosomes to the PM (). The role of palmitoylation in regulating the sorting of transmembrane proteins is also apparent in lower eukaryotes, including the yeast . Chs3 is a chitin synthase involved in cell wall growth that localizes to the tip and neck of the bud and also to an intracellular compartment. The polytopic Chs3 protein is palmitoylated by the ER-localized DHHC protein Pfa4, and preventing this palmitoylation caused Chs3 to be retained in the ER (). Interestingly, unpalmitoylated Chs3 displayed an increased level of aggregation, consistent with the idea that palmitoylation may stabilize membrane interactions of the transmembrane helices of Chs3. Palmitoylation also plays an important role in the sorting of proteins lacking transmembrane peptide sequences that are tethered to the cytosolic surface of membranes. In many such cases, a key function of palmitate is to serve as a membrane “trap” by increasing relative membrane affinity, and this is an important difference compared with palmitoylation of transmembrane proteins. A well-characterized example of this is palmitoylation-dependent sorting of H- and N-Ras. The primary signal for membrane association of these proteins is C-terminal farnesylation (isoprenylation), which mediates the association of Ras with ER and Golgi membranes (). However, such single lipid modifications provide only a weak membrane affinity (; ), and farnesylated Ras undergoes diffusional exchange between the cytosol and endomembranes (; ). In contrast, two tandem lipid modifications provide a relatively stable membrane anchor (); thus, subsequent palmitoylation of Ras increases the strength of membrane interaction (). This membrane trapping of Ras facilitates targeting to post-ER compartments, particularly the PM and Golgi. In a manner similar to AMPA receptor trafficking, each of the two palmitoylated cysteines in H-Ras (cys-181 and -184) was suggested to differentially affect protein localization (). Monopalmitoylation of cysteine-184 led to accumulation of H-Ras in the Golgi. As Ras PAT (DHHC9/GCP16) is predominantly localized to the Golgi in mammalian cells (), this implies that palmitate addition to cys-184 may act as a membrane trap but without conferring any additional targeting information. Another possibility is that the small amount of DHHC9/GCP16 present on ER membranes mediates palmitoylation of H-Ras at this compartment and, hence, that modification of cys-184 supports transport from the ER to the Golgi. In contrast, monopalmitoylation of cys-181 more effectively directed PM delivery of H-Ras. Clustering of AMPA receptors at postsynaptic sites is regulated by postsynaptic density protein-95 (PSD-95). The exclusive targeting of PSD-95 to postsynaptic clusters in neurons is dependent on dual palmitoylation of a specific amino acid sequence, MDCLCIV (). Replacing this sequence with the palmitoylation code from the axonally localized GAP43 protein (MLCCMRR) disrupted the exclusive postsynaptic targeting of PSD-95 and redistributed a fraction of the protein to axons (). Features of the palmitoylation domains of PSD-95 and GAP-43 that were important for postsynaptic and axonal targeting, respectively, included the spacing of palmitoylated cysteines and the presence of basic amino acids downstream of the cysteines. It is not clear how these specific features contribute to sorting, but they may be important for directing the proteins to distinct transport vesicles (). Alternatively, the different palmitoylation motifs may form recognition sites for distinct PAT enzymes, with the localization of specific PATs dictating the final distribution of GAP-43 and PSD-95 (). For example, the GAP-43 PAT may reside in a compartment (or specific microdomain of a compartment) that links to an axonal trafficking pathway, whereas the PSD-95 PAT may link to a postsynaptic pathway. In addition to regulating polarized trafficking in neurons, there is also evidence implying a role for palmitoylation in sorting to the myelin membrane in oligodendrocytes () and to tight junctions in epithelial cells (). The effects of palmitoylation on sorting to PM domains are not restricted to polarized cells and may direct the nanometer-scale microlocalization of proteins within the same membrane. Although this topic is beyond the scope of this mini-review, it is worth noting that palmitoylation of cys-184 of H-Ras (see previous section), although not required for PM delivery, was essential for regulating the correct distribution of GDP- and GTP-bound forms of the protein between cholesterol-rich microdomains and domains that were insensitive to cholesterol extraction (). The previous sections highlight the role of palmitoylation in regulating membrane retention or directly influencing protein sorting. In contrast, palmitoylation of the yeast SNARE protein Tlg1 appears to play a more indirect role in membrane targeting of this protein. SNAREs are a family of proteins located on various membrane compartments that regulate intracellular membrane fusion events; Tlg1 regulates membrane traffic between the endosomes and Golgi. Palmitoylation by the DHHC protein Swf1 ensured that Tlg1 is retained on TGN/endosome membranes by protecting the protein from ubiquitination (). Indeed, mutation of the palmitoylation sites in Tlg1 or genetic inactivation of Swf1 led to Tlg1 ubiquitination by the ubiquitin ligase Tul1, causing Tlg1 to be routed to the vacuole and degraded. However, it is interesting to note that cysteine mutants of Tlg1 displayed a subcellular distribution similar to wild-type Tlg1 when ubiquitin ligases were inactivated. This implies that palmitoylation is not required for membrane targeting of Tlg1 per se, but simply to prevent ubiquitination. Palmitoylation was suggested to fix the position of the transmembrane domain of Tlg1 relative to the bilayer, perhaps ensuring that membrane-proximal acidic residues are not exposed to membrane lipids, a situation that would be predicted to lead to ubiquitination by Tul1 (). Palmitoylation is a reversible process, and several cellular proteins undergo dynamic palmitoylation (). The reversibility of palmitoylation can enhance the versatility of this lipid modification as a protein sorting signal. AMPA receptors exhibit stimulation-dependent changes in palmitoylation status as well as regulated internalization (). Interestingly, mutation of palmitoylation site 2 (refer to previous sections) in GluR1/2 subunits inhibited activity-dependent internalization. Furthermore, palmitoylation of this site was associated with an inhibition of GluR1/2 binding to the cytoskeleton-associated 4.1N protein. Binding to 4.1N was suggested to trap GluR1/2 at the cell surface, suggesting that palmitoylation of site 2 enhances internalization of GluR1/2 by regulating this interaction ().Interestingly, activity-dependent changes in palmitoylation also regulate surface distribution of PSD-95, a protein that modulates synaptic clustering of AMPA receptors (). Indeed, these palmitoylation changes in PSD-95 were also suggested to be required for glutamate-mediated internalization of AMPA receptors. Thus, dynamic palmitoylation plays a key role both directly and indirectly (via PSD-95) in regulating surface distribution of AMPA receptors and hence synaptic activity. Recent work also uncovered a dynamic palmitoylation pathway that regulates Ras trafficking. This pathway operates constitutively and is essential for maintaining correct Ras localization (; ). The results of these recent studies also implied that the half-life of palmitate on Ras proteins may be far less than the originally reported values of 20 min for N-Ras () and 2 h for H-Ras (; ), suggesting an incredibly fast rate of palmitate turnover. This dynamic palmitoylation of Ras proteins results in a constant flux of the proteins between endomembranes and the PM. In this system, palmitoylation at Golgi membranes directs Ras to the PM, whereas depalmitoylation at the PM releases the protein into the cytosol, allowing it to rebind to Golgi membranes, where it is once again palmitoylated and trafficked to the PM (). This palmitoylation cycle appears to be essential for maintaining the appropriate subcellular distribution of Ras, as the addition of palmitate-like hexadecylated groups with noncleavable thioether bonds to N-Ras caused a marked missorting of the protein (). This observation suggests that although palmitoylation of Ras drives PM delivery, it may not be sufficient to maintain Ras at the PM during ongoing membrane remodeling, for example, exocytosis and endocytosis. However, these experiments may be more difficult to interpret, as much of the microinjected hexadecylated protein will presumably associate with intracellular membranes independently of the secretory pathway and thus not experience the same initial trafficking cues as endogenous Ras. Identification of this rapid Ras cycling pathway also introduces an extra dimension to trafficking analyses of monopalmitoylated H-Ras mutants. Intriguingly, similar palmitoylation cycles were also observed for short N-terminal sequences of other palmitoylated proteins (), suggesting that dynamic palmitoylation may regulate the intracellular distribution of many proteins. However, this study examined relatively short palmitoylated peptides, and the presence of other domains in the full-length proteins is likely to regulate the intrinsic palmitoylation/depalmitoylation cycle (); indeed, many proteins, such as SNAP25, appear to be stably palmitoylated (). Palmitoylation could affect protein trafficking by several distinct mechanisms. It is important to stress that the role palmitoylation serves in the trafficking of integral membrane proteins may be different from that of proteins with a weak membrane affinity, such as H- and N-Ras. In the simplest case, palmitoylation could trap proteins with a weak membrane affinity on an appropriate intracellular membrane by enhancing the strength of membrane interaction (). This membrane trapping is well characterized for palmitoylated proteins like Ras, where the addition of a second lipid modification leads to a large increase in membrane residency time (; ). This enhanced membrane association would allow the protein to associate more efficiently with budding vesicles and ensure that the protein does not dissociate from the membrane during vesicle transport. In this model, sorting is regulated by the strength of membrane interaction and not palmitoylation per se. The specificity of palmitoylation-dependent sorting, in this case, may be dictated by the localization of the appropriate PAT enzyme. For example, although Ras proteins can presumably “sample” a variety of intracellular membranes by virtue of a hydrophobic farnesyl modification, the restricted distribution of Ras PAT to Golgi/ER membranes may ensure the correct sorting of the proteins to the PM and Golgi. In addition, the specific subcellular localization of Ras thioesterases may also contribute to maintaining the appropriate distribution of Ras proteins. By analogy, it may be the intracellular localization of PAT enzymes that modify the specific palmitoylation sequences of PSD-95 and GAP-43 that ensures correct polarized sorting of these proteins in neuronal cells. In addition to this passive role in intracellular trafficking, palmitoylation may also actively drive the association of proteins with budding vesicles or specific microdomains that facilitate sorting. This model applies both to proteins with a weak membrane affinity (such as Ras) and to integral membrane proteins. There is a steep concentration gradient of cholesterol within cells, with the highest levels in endosomes and the PM. In contrast, the membrane of the ER, where cholesterol is synthesized de novo, contains very little cholesterol. The majority of synthesized cholesterol is transported to the PM directly by nonvesicular traffic; however, a significant fraction is also transported via vesicular traffic through the Golgi (). Cholesterol forms tightly packed domains with saturated phospholipids, and palmitoylated proteins are thought to have a high affinity for these ordered domains. Thus, the association of palmitate groups with cholesterol-rich domains may allow protein movement from the early secretory pathway by vesicular transport. An additional intriguing possibility is that palmitate groups interact directly with cholesterol. Although phospholipids with saturated acyl chains (such as palmitate) exhibit a high affinity for cholesterol in model membranes, the phospholipid head groups are thought to play an important role in this association (). Therefore, it is not clear whether palmitoylated proteins could have an affinity similar to saturated phospholipids for cholesterol, although there is some evidence to support this possibility (; ). Another important mechanism underlying palmitoylation-dependent protein sorting involves the regulation of protein–protein interactions. These interactions may be with sorting receptors or cargo or with specific proteins such as 4.1N in the case of the AMPA receptor. Palmitoylation could regulate many of these interactions by controlling the conformation of the modified protein. For example, cysteine palmitoylation and subsequent membrane integration may force flanking residues into closer membrane proximity (); if these residues form part of a protein binding pocket, then this would inhibit binding. Similarly, the cysteine residue itself may mediate protein interactions that would be inhibited by palmitoylation. Palmitoylation may also bring a protein binding domain into closer proximity to a membrane receptor, enhancing the possibility of productive interactions. In addition to these direct effects on protein interactions, palmitoylation may regulate protein interactions by spatially coupling or segregating proteins within lipid microdomains. The above discussion has highlighted how association with specific membrane microdomains and the regulation of protein–protein interactions could mediate palmitoylation-dependent protein sorting. These models apply equally to integral membrane proteins and to proteins with weak membrane affinities that have been trapped on membranes by palmitoylation. In addition, palmitoylation may specifically affect sorting of integral membrane proteins by regulating interactions of transmembrane domains with the lipid bilayer. An interesting possibility in this regard is that palmitoylation regulates trafficking of transmembrane proteins by enhancing hydrophobic matching between transmembrane domains and the lipid bilayer (; ). Hydrophobic mismatch occurs when a discrepancy exists between the thickness of the hydrophobic region of the phospholipid bilayer and the length of the hydrophobic transmembrane helix (), and this mismatching is thought to be energetically unfavorable. In relation to intracellular trafficking, the exposure of hydrophobic domains by mismatching could lead to protein aggregation or ER retention (as observed for Chs3) or association with ubiquitin ligases (as seen for Tlg1). Palmitoylation may affect the extent of hydrophobic mismatching by inducing lateral movement of the protein into distinct membrane microdomains (as discussed previously in this paper). For example, when there is a positive mismatch (hydrophobic domain of protein longer than the thickness of the hydrophobic part of the membrane), palmitoylation may move the protein into cholesterol-rich domains of the membrane (). Cholesterol plays an important role in controlling bilayer thickness, and the addition of 30% mol/mol cholesterol to C16:0/C18:1 phosphatidylcholine bilayers increased the thickness of the hydrophobic core of the bilayer by ∼15% (). Thus, association with cholesterol-rich domains would be predicted to alleviate the positive mismatch. Another idea is that palmitoylation relieves hydrophobic mismatch by altering the tilt of transmembrane helices within the bilayer (; ); indeed, palmitoylation of hydrophobic membrane-spanning peptides was suggested to modify peptide orientation in lipid vesicles (). It is important to note that although palmitate groups may preferentially partition into cholesterol-rich domains, bulky transmembrane helices are generally excluded from these tightly packed domains. An interesting possibility, therefore, is that the affinity of palmitate for ordered lipid domains, coupled with the exclusion of the transmembrane helix from these regions, promotes tilting of the helix as it resists entry into ordered domains occupied by the adjacent palmitate. The effects on transmembrane tilting may be particularly relevant for proteins that have palmitoylation sites in membrane-spanning regions. Finally, palmitoylation may affect the orientation of transmembrane domains by regulating the interfacial localization of aromatic and basic amino acids that typically flank the hydrophobic segments of transmembrane helices. c e n t w o r k h a s h i g h l i g h t e d t h e d i v e r s e n a t u r e o f p a l m i t a t e a s a p r o t e i n s o r t i n g s i g n a l . A n i m p o r t a n t a r e a o f i n v e s t i g a t i o n n o w i s t o i d e n t i f y c o m m o n m e c h a n i s m s t h a t r e g u l a t e p a l m i t o y l a t i o n - i n d u c e d p r o t e i n s o r t i n g a n d t o “ c r a c k ” t h e p a l m i t o y l a t i o n c o d e s w i t h i n p r o t e i n s t o r e v e a l h o w a p a l m i t o y l a t e d p e p t i d e s e q u e n c e r e l a t e s t o t h e f i n a l d e s t i n a t i o n o f a p r o t e i n i n t h e c e l l . L i k e t h e s e a r c h f o r p a l m i t o y l a t i o n c o n s e n s u s s e q u e n c e s , t h e s e q u e s t i o n s a r e l i k e l y t o p r e s e n t a s i g n i f i c a n t c h a l l e n g e .

Formation of the first compartment of the Golgi apparatus involves recognition events between several different membranes. These include tethering of both anterograde COPII vesicles and retrograde COPI vesicles to the cis-Golgi. In addition, intermediate compartment structures that formed in the cell periphery fuse with the cis-Golgi membranes after movement to the cell center, and Golgi stacks undergo homotypic fusion to form the elongated ribbon characteristic of vertebrate cells. Finally, the cis-Golgi is attached to an adjacent medial compartment in the stack. Several proteins have been proposed to guide these membrane recognition events, including the coiled-coil proteins GM130 and p115, the multisubunit complexes transport protein particle (TRAPP) and conserved oligomeric Golgi (COG), and the PDZ-like protein GRASP65 (; ). This latter protein was identified using an in vitro assay for the postmitotic reassembly of mammalian Golgi stacks (). It consists of two PDZ-like domains flanked by a C-terminal Ser/Pro-rich domain and an N-terminal myristoylation site, which is required for its association with the cis-Golgi (; ). GRASP65 recruits the coiled-coil protein GM130 to the cis-Golgi by binding to the latter's C terminus (). Golgi association of the two proteins appears mutually interdependent, as mutation of the interfacial residues in either protein causes a loss of localization (). GRASP55, a GRASP65 paralogue unique to vertebrates, is found on the medial Golgi but may also contribute to GM130 recruitment (; ). GM130 also interacts with the coiled-coil protein p115 and the small GTPase Rab1, and it has been proposed that the GRASP65–GM130 complex acts in a variety of tethering interactions at the cis-Golgi (; ; ; ). Despite these compelling physical interactions, analysis of the importance of GRASP65 in vivo has not yet reached a clear consensus. Removal of the protein by RNAi has been reported to affect the formation of Golgi ribbons, the number of cisternae, the structure of the cisternae themselves, and even the formation of the mitotic spindle (; ; ). However, in all cases, transport through the Golgi appeared relatively normal, and a similar result has been reported for the loss of GM130 (). These rather variable and perhaps surprisingly mild phenotypes may reflect a degree of redundancy in membrane traffic steps at the cis-Golgi. Not only are there other large coiled-coil proteins and tethering factors that could compensate for the removal of GRASP65–GM130, but there may also be redundancy among the multiple membrane fusion events that generate the cis-Golgi (; ; ). The understanding of membrane traffic in mammalian cells has been helped by studies of model organisms and, in particular, of the budding yeast . However, the GRASP65–GM130 complex has not been investigated by this route, in part because an orthologue of GM130 is not detectable by similarity searches. Moreover, the one protein in yeast that is related to GRASP65, Grh1, does not have a myristoylation site at its N terminus and is not essential for growth. Indeed, one study suggested that it could act in a mitotic checkpoint, although this has not been subsequently investigated (). We have investigated Grh1 in more detail, as we noticed that it has an N-terminal amphipathic helix that is a likely target for N-terminal acetylation by the NatC N-terminal acetyltransferase. We had previously found that a similar helix on the Golgi-localized GTPase Arl3 is responsible for targeting it to Golgi membranes (; ). We report here that Grh1 is on the cis-Golgi, and its acetylated N-terminal amphipathic helix appears to replace the membrane-associating role provided by the N-terminal myristoyl group for GRASP65. We find that Grh1 forms a complex with a previously uncharacterized coiled-coil protein, which, although not related in primary sequence, shares several structural features with GM130. Thus, it appears that the role of GRASP65 is better conserved in evolution than previously thought, and our data suggest that this role is in membrane traffic even if it is not essential for secretion in either yeast or mammalian cells. To identify proteins whose membrane targeting might be dependent on N-terminal acetylation by the NatC complex, we searched for yeast proteins that had a NatC consensus (F, Y, I, L, or W at the second position; ) and also an N-terminal amphipathic helix. We then determined whether both the hydrophobic residue at position 2 and the amphipathic helix were conserved in other yeasts and filamentous fungi. This left Grh1, the yeast orthologue of the mammalian protein GRASP65 (). The putative N-terminal amphipathic helix is followed by a region of poorly conserved length and sequence, which could act as a flexible linker. The metazoan orthologues lack this N-terminal extension and instead have a short region with a glycine in position 2, which forms part of the consensus sequence for N-terminal myristoylation. To determine whether NatC action is required for the membrane association of Grh1, we initially examined the subcellular localization of Grh1. When was tagged in the genome with GFP at the C terminus, Grh1-GFP was found to be on punctate structures, which colocalized with the early Golgi marker RFP-Rud3 (). When the gene encoding the Mak3 catalytic subunit of the NatC complex was deleted, Grh1-GFP was much more diffuse, with only a few faint dots still visible (). This altered distribution of Grh1-GFP did not reflect a general perturbation of the early Golgi, as other markers of this compartment, such as Rud3 and Sed5, were apparently unaffected by the deletion of ( and not depicted). The use of fractionation to examine membrane association was precluded by the membrane association of Grh1 being readily reversible (>95% being in the soluble fraction after cell lysis and centrifugation at 10 for 30 min). However, photobleaching of the diffuse pool of Grh1-GFP in the Δ strain showed that it diffuses much more rapidly than a GFP-labeled membrane protein in a strain known to accumulate Golgi-derived vesicles (Fig. S1, available at ). This indicates that the loss of Mak3 results in a reduction in the membrane association of Grh1 rather than a vesiculation of Grh1-positive membrane structures. In addition, when Phe2 of Grh1 was mutated into an alanine, which precludes the generation of an acetylated N terminus by NatC, the mutant protein was substantially mislocalized to the cytosol, indicating that the precise structure of the N terminus is important for Grh1 localization (). To examine the modification state of the N terminus of Grh1, Grh1-Flag3 was purified from both wild-type and Δ strains. Mass spectrometry of proteolytic products from the protein obtained from the wild-type strain revealed a peptide corresponding to an acetylated N terminus, with this being replaced in the Δ-derived protein with a nonacetylated peptide beginning with Met1 (Fig. S2, available at ). A similar NatC-dependent N-terminal modification is found on the Golgi GTPase Arl3, and, in this case, the modification is required for recognition by the Golgi membrane protein Sys1 (; ). However, deletion of the gene encoding Sys1 did not affect the localization of Grh1 (). Thus, the localization of Grh1 to the cytosolic face of the early Golgi membranes is Mak3 dependent but Sys1 independent. Mammalian GRASP65 interacts with the coiled-coil protein GM130, but there is no obvious homologue of GM130 outside of metazoans. Thus, binding partners for Grh1 were sought using the protein as bait in a yeast two-hybrid screen. Of 25 positives, two were Grh1 itself, and the remainder was an uncharacterized open reading frame YDL099w that we named (bound to Grh1). Bug1 is a 342-residue protein with an N-terminal basic region and a potential coiled-coil domain (see ), which is a structure reminiscent of that of GM130. When was tagged in the genome with GFP at either the C or N terminus, the resulting fusion proteins localized to punctate structures ( and not depicted), but the N-terminally tagged version gave less cytosolic background and was used for the rest of the experiments. GFP-Bug1 colocalizes with Grh1-RFP but is completely mislocalized to the cytosol in the absence of Grh1 () and is diffuse, with only a few dots remaining in the absence of Mak3 (not depicted). Interestingly, Grh1-GFP, in turn, is mislocalized in a Δ strain (). Thus, it appears that Grh1 and Bug1 are interdependent for their localization to the Golgi. To confirm the interaction between Grh1 and Bug1, HA2-Bug1 was immunoprecipitated, and endogenous Grh1 was found to be enriched in the precipitate (). Consistent with this result, recent high throughput screens of yeast protein–protein or genetic interactions have indicated that Bug1 coprecipitates with Grh1-TAP (; ; ). We also found that Bug1-HA3 could be immunoprecipitated with Flag3-Bug1 and that untagged Grh1 could be coprecipitated with Grh1-HA3 (). Thus, both of the proteins form at least dimers, and, therefore, Bug1–Grh1 is a heterooligomer composed of at least a dimer of Grh1 and a dimer of Bug1. Indeed, by gel filtration, Grh1 runs at ∼400 kD, which corresponds to the mass of two molecules each of Bug1 and Grh1 (unpublished data). We next generated truncated forms of Bug1 and compared their ability to interact with Grh1 with their effect on the latter's localization. Removal from Bug1 of the N-terminal basic region either alone or with the adjacent poorly conserved region (amino acids 44–340 and 185–340) had no effect on Grh1 binding or localization (). A construct also lacking the coiled-coiled region (i.e., just the last 65 amino acids of Bug1 [275–340]) could only be detected after immunoprecipitation but was clearly able to bind to Grh1. Moreover, a construct lacking the last 30 amino acids of Bug1 (1–310) was no longer able to bind Grh1. Thus, the binding site for Grh1 is located at the well-conserved C terminus of Bug1. However, recruitment of Grh1 to membranes in vivo appears to require not only this region but also the adjacent putative coiled coil (). Therefore, the C terminus of Bug1 interacts with Grh1, but its coiled-coil region also contributes to Golgi membrane association. Together, this results in a robustly localized complex. During the purification of Grh1-Flag3 to examine the N terminus, we noticed several proteins of 80–100 kD in the precipitate from wild type that were absent from that prepared from the Δ strain (). Mass spectrometry of tryptic peptides identified these proteins as the components of the COPII coat, Sec23 and Sec24, along with the two relatives of the latter, Sfb2 and Sfb3. Probing with anti-Sec23 antibodies confirmed this identification (), and this interaction was also detected in recent high throughput screens (; ; ). Sec23 forms stable dimers with Sec24 or its relatives even when the COPII coat is disassembled (). These results suggest that Grh1 can interact with this dimer and that this interaction requires N-terminal acetylation. To investigate the role of Grh1 and Bug1 in ER to Golgi transport, we examined COPII-dependent budding and transport reactions in an in vitro assay based on semi-intact cells (). ER-derived transport vesicles were produced from washed membranes after incubation with purified COPII proteins (Sar1, the Sec23/24 complex, and the Sec13/31 complex). Grh1 was detected on COPII vesicles produced from wild-type membranes (). The level of Grh1 associated with vesicles relative to total membranes was less than Sec22 (a vesicle SNARE protein) but above the level of an ER-resident protein, Sec12. In the Δ strain, Grh1 was absent as expected, and, in the Δ strain, the amount of Grh1 that associated with vesicles was reduced. To quantify COPII-dependent budding and transport to the Golgi complex in the reconstituted assays, we measured the amount of [S]gpαf packaged into vesicles and the amount of Golgi-modified [S]gpαf (). The level of COPII-dependent budding of gpαf was not substantially influenced by the Δ, Δ, or Δ deletions (). In the additional presence of Uso1 (the yeast homologue of the coiled-coil tethering protein p115), Sec18, and LMA1, such vesicles fuse with Golgi membranes, as detected by the acquisition of Golgi-dependent carbohydrate modifications (), and this showed a small but reproducible decrease for all three deletions (). The Δ deletion produced the strongest decrease and reduced overall transport by ∼60%. When just Uso1 is present, the vesicles are known to become tethered without fusing, as assayed by a reduction in the population of freely diffusible vesicles (). Strikingly, this Uso1-dependent tethering was greatly diminished for all three deletions (). These results indicate that semi-intact cell membranes lacking Grh1 or Bug1 produce COPII vesicles normally but are partially compromised in the fusion stage of ER-derived vesicles with Golgi membranes. This appears to reflect, at least in part, Uso1-dependent tethering being reduced to the point where it is, at most, no longer sufficient to maintain a tethered state through the centrifugation step used to separate tethered from diffusible vesicles. The lethality of yeast strains lacking the GTPase Ypt1 or the tether Uso1 can be suppressed by , a dominant mutation in Sly1, which is a member of the Sec1/Munc18 family of SNARE regulators (; ). This suggests that tethering of COPII vesicles before consumption can occur by Uso1-independent processes, and we hypothesized that these processes might also depend on the Grh1–Bug1 complex. shows that yeast cells lacking or cannot be rescued by if either or is absent. The deletion of also affected the ability of to rescue the loss of , although growth was not completely impaired with a few larger colonies appearing after several days, perhaps corresponding to suppressor mutations. This partial effect is consistent with Grh1-GFP not being completely mislocalized in a Δ strain. Thus, the rescue of strains lacking Uso1 or Ypt1 by relies on the Grh1–Bug1 complex. The Grh1–Bug1 complex appears to contribute to membrane traffic at the cis-Golgi, and biochemical and genetic interactions suggest a role in ER to Golgi transport. Recruitment of the complex to membranes is dependent on an N-terminally acetylated amphipathic helix on Grh1, indicating that N-terminal acetylation has a relevance to membrane targeting beyond the two Arf-like GTPases for which it has been previously shown to be important (; ). The role of Grh1 and Bug1 may be to improve the efficiency of both Uso1-dependent and independent tethering events. Formation of the cis-Golgi is likely to involve several heterotypic and homotypic fusion events between nascent cis-cisternae, COPI vesicles, and COPII vesicles, and such multiplicity could allow cell growth even when one or more fusion pathways are compromised. Alternatively, it may be that multiple interactions are used to increase the accuracy of recognition of a specific fusion target such as the cis-Golgi and that a partial reduction in accuracy does not prevent viability. Interestingly, in higher eukaryotes, p115 but not GM130 or GRASP65 are required for secretion in cultured cells even though the latter clearly interact with p115 in assays that reconstitute fusion with cis-Golgi membranes (; ; ; ). This suggests that the GM130–GRASP65 complex serves to optimize rather than allow membrane traffic in the early Golgi and that the constraints of the in vitro systems highlight a role for components whose loss can normally be bypassed in cells growing in optimal conditions. Given that this mammalian machinery now appears to have yeast analogues, it seems likely that further examination of Grh1 and Bug1 and other putative cis-Golgi tethers will shed light on the precise function of the mammalian proteins. Unless otherwise stated, yeast strains were based on BY4741 (MATa ΔΔΔΔ), BY4742 (MATα ), or disruptions in this background (Open Biosystems), with further genes disrupted or epitope tagged by PCR-based homologous recombination (Table S1, available at ). Immunoprecipitations were performed from the protease-deficient strain c13-ABYS-86 (MATα Δ ) unless otherwise stated. The strains for suppression analysis were based on parental strains CBY903 (MATa ΔΔ Δ carrying pSK54) and CBY1381 (MATα Δ Δ Δ Δ Δ Δ carrying and pSLY1-20) and on the plasmids pSK54 (μ), pSLY1-20 (μ), pSK47 (μ), and pRB320 (μ), which were described previously (). An anti-Grh1 antiserum was generated in sheep, and anti-Sec23 antibodies were described previously (). For small-scale immunoprecipitations, 50–100 mg of yeast pellets were lysed by the addition of 200 μl of glass beads (425–600 μm; Sigma-Aldrich) and 200–400 μl of lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM KCl, 5 mM MgCl, and 1% Triton X-100) containing protease inhibitors (Roche), and the tubes were vortexed twice in a RiboLyser (Thermo-Hybaid) at speed setting six at 4°C. After centrifugation for 15 min at 12,000 , the supernatants were incubated with 20 μl anti-HA F-7 agarose beads (Santa Cruz Biotechnology, Inc.) or anti-Flag M2 agarose beads (Sigma-Aldrich) for 2 h at 4°C. The beads were washed in lysis buffer and eluted with SDS sample buffer. Grh1-Flag3 was expressed on a CEN plasmid under the control of a constitutive promoter in the wild-type strain BY4741 or the same lacking , and protein was precipitated from 1 g of cells as described previously (). For reasons that are not clear, coprecipitation of COPII coat components with Grh1-Flag3 was more efficient from 1 liter rather than 100-ml cultures. For mass spectrometry, the gel was stained with Coomassie blue, and bands were excised, digested with trypsin or Lys-S, and peptides were subjected to matrix-assisted laser desorption ionization mass spectrometry. Yeast semi-intact cells were prepared and analyzed in reconstituted cell-free budding and transport assays as previously described (; ). Yeast strains CBY740 (α ), CBY2009 (CBY740 Δ), CBY2028 (CBY740 Δ), and CBY2029 (CBY740 Δ) were purchased from Invitrogen and are isogenic with BY4742. Fig. S1 shows that Grh1-GFP is displaced from membranes by the loss of Mak3. Fig. S2 shows that the N terminus of Grh1 is acetylated in wild-type but not in Δ cells. Table S1 provides information about the yeast strains generated during this study. Online supplemental material is available at .

xref #text We chose the Δ/Inv strain of serovar Typhimurium (. Typhimurium) as our model to generate phagosomes. The InvA protein is a structural component of the pathogenicity island (SPI)–1–encoded type III secretion system (TTSS; ). The TTSS is a needle-like device that . Typhimurium uses to translocate bacterial effector proteins into the host cell to manipulate the actin cytoskeleton and drive invasion through a ruffling mechanism (). The Δ strain is unable to secrete effector proteins into the host cell and is therefore noninvasive. Expression of the () gene of allows the bacteria to enter the host cell via an alternative mechanism. The Inv protein induces uptake of the bacteria through binding to β1-integrin (). Inv-mediated uptake of leads to degradation of these bacteria in lysosomes (). Similarly, the Δ/Inv strain of . Typhimurium is efficiently internalized and traffics to late endocytic compartments where it fails to replicate (). To ensure that the phagosomal maturation of Δ/Inv . Typhimurium is not manipulated by virulence factors apart from the SPI-1 TTSS, such as the PhoP/PhoQ regulon or the SPI-2 encoded TTSS (), we inhibited bacterial protein synthesis through addition of tetracycline 15 min after internalization. This resulted in the trafficking of the bacteria to a lysosome, leading to their degradation (). To follow the trafficking of the model phagosome we separately transfected HeLa cells with 48 distinct GFP- or CFP-tagged Rab GTPases. The Rabs chosen for our study constitute a broad representation of the Rab family. We then introduced Δ/Inv . Typhimurium and, after phagocytosis, tracked Rab association with the model phagosome over 3 h. At each time point investigated, we determined Rab association for at least 100 internalized bacteria, ensuring we only counted cells where expression of the Rab was relatively low. To ensure we counted only internalized bacteria, we immunostained for . Typhimurium before and after permeabilization of the cellular membrane. In this way we were able to determine which bacteria had invaded and which remained extracellular. We counted a positive association of a Rab with the model phagosome as a distinct ring around the bacteria and in the same focal plane as the bacteria (). In parallel, we infected cells with wild-type . Typhimurium and monitored its association with the same 48 Rab GTPases. Wild-type . Typhimurium use the SPI-1–encoded TTSS to invade epithelial cells in a manner distinct from receptor-mediated phagocytosis (). After invasion, the bacteria reside in a -containing vacuole (SCV) that, like a phagosome, interacts with the early endosomal pathway and acquires Rab5. The SCV then undergoes a maturation process that is characterized by the recruitment of lysosomal-associated membrane protein-1 (LAMP-1) and other lysosomal glycoproteins but does not fuse with lysosomes. The manner in which wild-type . Typhimurium is able to manipulate SCV trafficking is unknown; however, there is evidence it can directly manipulate Rab function (). The results of our screen on the model phagosome correlated well with previously characterized Rabs found on phagosomes (). Rab3, 4, 5, 7, 9, 10, 11, and 14 have all been previously identified on purified latex bead phagosomes () and were all present, with the exception of Rab3, at a level of 20% or higher on model phagosomes at some time during the first 3 h post infection (p.i.; and Table S1, available at ). In addition, the association of Rab5A early (15 min p.i.) and Rab7 later (2–3 h p.i.) matched previous results (), signifying that our model system is an effective tool to study phagosome maturation. Importantly, our studies determined several novel Rab associations (>20% association) with our model phagosome (). Some phagosome-associated Rabs interacted only with the model phagosome and not with the SCV of wild-type bacteria ( and ). These included Rab8B, 13, 23, 32, and 35. These Rabs also associated with the model phagosome in a macrophage cell line with similar kinetics and with phagosomes containing IgG-coated Δ mutant bacteria (lacking Inv expression) after Fcγ receptor–mediated phagocytosis (Table S2, A and B, available at ). To test whether these Rabs have a role in phagosome maturation, Texas red–labeled dextran was preloaded into HeLa cells, followed by transfection with CFP-tagged dominant-negative (DN) Rab constructs or a CFP control vector. Before infection, noninternalized dextran was removed, allowing internalized dextran to label lysosomes. At 3 h p.i., cells were fixed and the number of bacteria in lysosomes was enumerated. For wild-type . Typhimurium, ∼10% of bacteria fused with lysosomes compared with ∼40% of Δ/Inv bacteria (). Expression of DN mutants of Rab8B, 13, and 32 had no effect on phagosomal fusion with lysosomes compared with the control. However, DN mutants of Rab23 and 35 reduced fusion by >25% compared with the control, demonstrating these Rabs play a role in phagosome–lysosome fusion (). Both Rabs are expressed in many cell types, including HeLa cells, and are thought to play a role in endosomal recycling (; ). Thus, it is possible that Rab23 and Rab35 promote phagosome maturation by mediating recycling from this compartment. Interestingly, we observed some Rabs that, although associated with the model phagosome, had greater association with the wild-type SCV ( and ). These are Rab5A, 5B, 5C, 7, 11A, and 11B. It is possible the bacteria specifically mediate recruitment of these Rabs to the SCV. Indeed, our recent experiments indicate that a bacterial factor promotes recruitment of Rab5 to early SCVs by promoting their fusion with early endosomes (unpublished data). Conversely, the lack of association of some Rabs with the SCV suggests that . Typhimurium can block their recruitment. This is consistent with the finding that . Typhimurium can modulate the activity of host GTPases, including members of the Rab family (). Additionally, Typhimurium may mimic the function of host GTPases, obviating a need for their recruitment. Such a scenario was recently suggested for TTSS effectors present in other bacterial pathogens (). Approximately 3 h p.i., . Typhimurium activates a second TTSS encoded in SPI-2 and mediates translocation of a distinct set of effector proteins across the SCV and into the host cell. Deletion of Δ, an essential component of the SPI-2 TTSS apparatus, had no effect on the recruitment of Rabs to the SCV during the first 3 h p.i., which is consistent with the kinetics of SPI-2 TTSS expression (Table S1). We also observed that the Δ mutant did not associate with Rab8B, 13, 23, 32, and 35 up to 10 h p.i., indicating this mutant did not traffic to lysosomes (unpublished data). These observations are consistent with the findings of , who demonstrated that the PhoP/Q regulon mediates avoidance of lysosomes and that the SPI-2 TTSS plays no role in this aspect of SCV trafficking. One of the cellular phenotypes attributed to the SPI-2 TTSS is the formation of -induced filaments (Sifs). Sifs are long tubular structures extending from the SCV that are characterized by their enrichment with LAMP-1 (). The purpose of Sif formation is unclear; however, mutants defective for Sif formation are also defective for intracellular growth and for virulence in animal models of infection. The source of membrane from which Sifs are formed is also unclear; hence, we screened the 48 Rabs to determine which are present on Sif membranes. We found both Rab7 () and Rab9 () on the membrane but none of the other Rabs tested (not depicted). Rab7 controls the acquisition of LAMP-1 by the SCV () and has been previously localized to Sifs and shown to be required for Sif formation (). Rab9 regulates late endosome to Golgi traffic (Lombardi et al., 1993) and its localization to Sifs is a novel observation. Therefore, we determined if Rab9 activity is required for Sif formation by transfection of DN Rab9 in wild-type . Typhimurium–infected HeLa cells (). For these experiments, we transfected cells with DN constructs 2 h after infection with wild-type . Typhimurium. In this way, we could selectively modulate Rab activity after bacterial establishment of a favorable intracellular niche and focus our analysis on Sif formation, which occurs between 6–8 h p.i. (Birmingham et al., 2005). In CFP control-transfected cells, ∼30% of infected cells had Sifs after an 8-h infection. When transfected with DN Rab7, 12% of infected cells had Sifs after 8 h, confirming the requirement of this GTPase in Sif formation. Transfection of DN Rab5 had no effect on Sif formation, suggesting this effect was not caused by a general inhibition of the endosomal system (unpublished data). In cells transfected with DN Rab9, 11% of infected cells had Sifs after 8 h (). Under similar infection conditions, expression of DN Rab5, 7, or 9 did not affect intracellular bacterial growth up to 10 h p.i. (unpublished data). Thus, inhibition of Sifs by DN Rab expression was not caused by an effect on bacterial growth. Together, these studies reveal a novel role for Rab9 in Sif formation. The presence of Rab9 on Sifs suggested that fusion of late endosomes with the SCV can occur at late times p.i. and is consistent with the presence of cathepsin D and lysobisphosphatidic acid on Sifs (). To further test this possibility, we examined other late endocytic markers to determine their presence on Sifs. We found that the late endosome-localized syntaxin7 (), as well as the late endosome/lysosome-localized transporter Niemann-Pick C1 (NPC1; ), were both present on Sifs (). However, we did not detect Sif fusion with lysosomes when examined by addition of fluid phase markers (unpublished data), confirming previous results (). Thus, our data suggests that the SCV fuses with late endosomes and that this compartment could provide a source of membrane for Sifs. We determined when fusion of the SCV with late endosomes occurs during infection. HeLa cells were infected with wild-type . Typhimurium and fixed at various times up to 8 h p.i. Bacteria positive for LAMP-1 (a marker for SCVs) were scored for colocalization with NPC1 (). At 1 h p.i., 80% of LAMP-1 SCVs did not associate with NPC1, indicating that fusion with late endosomes was minimal at this time. However, by 4 h p.i. the majority of LAMP-1 SCVs had accumulated NPC1, indicating late endosome fusion had taken place. Our data demonstrates a two-step maturation of the SCV occurs where there is a delay in fusion with late endosomes. This is reminiscent of the “pregnant pause” mechanism used by and other intracellular bacterial pathogens (), which temporarily halt maturation of their vacuole so they can form a replicative niche before the vacuole matures to the next state. Our experimental approach permitted us to study the interactions of a large number of Rabs with both a model phagosome and the SCV of virulent Typhimurium. The results we obtained correlated well with previous data describing the association of Rab GTPases with both compartments. Importantly, we were able to localize several uncharacterized Rabs to each compartment. A network of 18 different Rabs was present on model phagosomes and 16 were found on the SCV. The finding that up to 12 different Rabs were present on each compartment at a given time is remarkable and consistent with the hypothesis that endosomes are composed of membrane domains (). Our data indicate that the process of phagosome maturation is far more complex than a single Rab5 to 7 transition and provide many insights and new avenues of study. Indeed, we found a novel role for both Rab23 and 35 in phagosome maturation. Furthermore, we were able to monitor the trafficking of a pathogenic bacterium and characterize the divergence of its vacuolar compartment from the normal degradative pathway. Our observations reveal the ability of Typhimurium to undergo a unique two-step maturation process that involves a delayed interaction of the SCV with late endosomes, but not lysosomes (). Future studies are required to define the molecular mechanisms by which Rab GTPases control phagosome maturation and how intracellular pathogens like . Typhimurium modulate their function. HeLa and RAW 264.7 cells were obtained from American Type Culture Collection. Cells were maintained in DME (HyClone) supplemented with 10% FBS (Wisent) at 37°C in 5% CO without antibiotics. Cultures were used between passage numbers 3–25. Wild-type SL1344 (), Δ SL1344 (), Δ (), and Δ/pRI203 (Δ/Inv) 14028S () . Typhimurium strains were used in this study. Plasmids used in this study are described in the Online supplemental material. GeneJuice transfection reagent (Oncogene Research Products) was used for transient transfection of cells. Rabbit polyclonal antibodies to . Typhimurium O antiserum group B were obtained from Difco. Murine monoclonal and rabbit polyclonal anti-GFP antibodies were obtained from Invitrogen. Rabbit polyclonal antibodies to c-myc were purchased from Santa Cruz Biotechnology, Inc. Murine monoclonal anti–human LAMP-1 antibodies (clone H4A3) developed by T. August (University of Iowa, Iowa City, IA) were obtained from the Developmental Studies Hybridoma Bank. Rabbit polyclonal antibodies to NPC1 were obtained from D. Manhuran and R. Bagshaw (Hospital for Sick Children, Toronto, Canada). For wild-type and Δ mutant infections, late-log bacterial cultures were used and prepared using a method optimized for bacterial invasion (). After infection, extracellular bacteria were removed by extensive washing with PBS and addition of growth medium containing 100 μg/ml gentamicin. After 90 min of bacterial infection, the gentamicin concentration was decreased to 10 μg/ml. For Sif studies, cells were transfected 2 h p.i. with wild-type . Typhimurium (see Plasmids and transfection). Cells were fixed 8 h after addition of bacteria. Intracellular growth assays were performed as previously described (). For Δ/Inv infections, bacteria were grown for ∼16 h at 37°C with shaking, subcultured (1:33) in Luria-Bertani broth for 3 h, and diluted to an OD of 1.0 in Luria-Bertani. Bacterial inocula were prepared by pelleting at 10,000 for 2 min, directly resuspended and diluted in PBS, pH 7.2, and added to cells at a dilution of 1:20. Cells were spun at 1,000 rpm for 1 min followed by incubation at 37°C for 15 min. After infection, extracellular bacteria were removed by extensive washing with PBS and addition of growth medium containing 200 μg/ml tetracycline. For Δ infections, bacteria were incubated in 2 ml PBS in the presence of 2 mg/ml of human IgG for 1 h at 37°C with shaking before addition to cells, as mentioned for the Δ/Inv strain. For dextran loading, HeLa cells were incubated with 1 mg/ml Texas red–dextran (Invitrogen) for 24 h, and then chased for 1 h before use. Cells were transfected 16–20 h before use and cells were infected with wild-type and Δ/Inv bacteria as above. Fixed cells were immunostained as previously described (). Immunostaining before permeabilization was used when determining the presence of intracellular bacteria. A fluorescence microscope (DMIRE2; Leica) equipped with a 100×/NA 1.4 oil objective (Plan Apochromat; Leica) was used to enumerate association of different proteins with the SCV. Colocalization was determined visually, with distinct signal surrounding the bacteria considered positive, for at least 100 bacteria. The mean ± the range ( = 2) or the SD ( ≥ 3) is presented. Images of fixed cells were obtained using an inverted microscope (Axiovert 200M; Carl Zeiss MicroImaging, Inc.) equipped with a laser scanning confocal imaging system (LSM510; Carl Zeiss MicroImaging, Inc.) and a 63×/NA 1.4 oil immersion objective lens (Plan Apochromat; Carl Zeiss MicroImaging, Inc.). Images were imported into Photoshop (Adobe) and assembled in Illustrator (Adobe). All experiments were repeated two to four times. Statistical analyses were performed using a two-tailed unpaired test. P-values <0.05 were considered statistically significant. Provided as online supplemental material are descriptions of the plasmids used in this study. Table S1 provides raw data for Table S2 provides data for colocalization of Rab8B, 13, 23, 32, and 35 with phagosomes containing Δ/Inv bacteria (A) or IgG-coated Δ bacteria (B) in RAW 264.7 macrophages. Table S2 C provides the primers used to generate plasmids in this study. Online supplemental material is available at .

Neuronal processes can be one million times as long as they are wide. This elongated shape places extraordinary demands on cell integrity when axons or dendrites are placed under strain. In the vertebrate peripheral nervous system, axons are exposed to strains generated by length changes during movement (). Strain has also been proposed to assist in wiring the central nervous system () and to underlie axon extension in response to growth cone migration (). Strains may also result from external forces, such as impact experienced in traumatic head injury. However, the mechanism for the elasticity of axons and dendrites is unknown, as is the response of neurons to breaks caused by the loss of elasticity. How do neurons maintain their structural integrity when challenged by mechanical strain? In this study, we demonstrate that β-spectrin is essential for neuronal strain resistance. The spectrin-based membrane skeleton is a cytoskeletal structure that is found in most cells, including neurons. The membrane skeleton is primarily composed of α/β-spectrin heterodimers, which associate with each other and with short actin filaments to form a 2D mesh. This submembranous structure anchors ion channels and other transmembrane proteins in the plasma membrane (). In neurons, spectrin is found at high levels in growth cones, and the injection of spectrin peptides prevents neurite extension in cultured cells (; ). These data suggest that the spectrin-based membrane skeleton functions in growth cone behavior. However, in erythrocytes, the spectrin-based membrane skeleton is required for membrane integrity (). In theory, the spectrin scaffold could allow stretching of the long and thin extensions of neurons. A role for the membrane cytoskeleton in neuronal structural integrity can be tested by characterizing β-spectrin mutants. In the nematode , β-spectrin is encoded by a single gene, (). mutants exhibit axonal morphology defects that range from truncated processes to elaborate branching (; ). These defects in axonal anatomy are consistent with the proposed role of spectrin in neuronal development. These defects are also consistent with a loss of axonal integrity caused by mechanical stress (; ). In this study, we demonstrate that β-spectrin is required specifically to maintain normal axon morphology. Spectrin function is dispensable for neuronal development and axon migration. However, the loss of β-spectrin results in spontaneous breaks in neuronal processes. These results suggest that β-spectrin is an essential contributor to neuronal strain resistance. To determine whether β-spectrin functions in axon outgrowth, we imaged growth cones of DD motor neurons in wild-type and β-spectrin mutant embryos (). In all experiments, only the progeny of homozygous null mutants were analyzed so that there was no maternal contribution of β-spectrin. The DD motor neurons send commissural axons from the ventral side to the dorsal side of the worm during embryogenesis (; ). To reach its target muscles in the dorsal quadrant, each DD growth cone pioneers a distinct path across the epidermis rather than following a preexisting axon tract. In wild-type animals, migrating DD growth cones display four stereotypical shapes at specific points in their migratory trajectory (; ). A growth cone is round and has radial projections when migrating across open epidermis. When a growth cone encounters the dorsal muscle quadrant, it spreads laterally into an anvil shape and extends multiple fingers between the muscle and epidermis toward the dorsal nerve cord. Finally, when a growth cone reaches the dorsal cord, it retracts all protrusions along the commissure and extends anterior and posterior processes along the cord. We found that growth cones in β-spectrin mutant animals displayed normal morphology and that changes in the shape of the growth cone were observed at the correct point of the migratory trajectory (). Quantitation of these structures in an age-matched cohort of embryos demonstrated that there was a similar distribution of shapes in our wild-type and mutant samples (; see Quantitation of embryonic DD neuronal phenotypes). Finally, we found that a similar fraction of neurons had successfully completed their migration (formed T shapes on the dorsal cord) in wild-type and mutant embryos (fraction of neurons forming T shapes: wild type = 61%; = 61%; P = 0.91 in a two-tailed Fisher's exact test; ). Together, these experiments demonstrate that β-spectrin is not essential for normal axon outgrowth of the DD neurons. To determine whether β-spectrin is dispensable for the extension of other neuronal processes, we examined the morphology of sensory dendrites, interneurons, and acetylcholine motor neurons in newly hatched mutants (). These neurons were essentially normal. Commissures could be distinguished in all animals ( = 20), and the dorsal and ventral nerve cords appeared intact. The dendritic processes of the sensory neurons were also similar to the wild type. Thus, in general, the extension of axons and dendrites does not require β-spectrin. Because axon and dendrite outgrowth does not require β-spectrin, neuronal defects that were previously observed in animals lacking β-spectrin must occur after outgrowth is complete. We confirmed the accumulation of commissural defects in neurons that had completed development by documenting DD axon morphology at three time points: during embryogenesis, just after hatching, and at 24 h after hatching (; see Quantitation of larval DD neuronal phenotypes). We found that neurons with defects—wandering, branched, or broken commissures—were rarely observed in wild-type animals at any time point. Defects were also rare in β-spectrin mutant embryos (fraction of neurons with defects: wild type = 1.3%; = 3.1%; P = 0.31 in a two-tailed Fisher's exact test). However, β-spectrin mutant animals accumulated defects with time. At hatching, the percentage of neurons with defects had increased to 26% (P < 0.0001 compared with embryos in a two-tailed Fisher's exact test). At 24 h after hatching, this percentage had further increased to 60% (P < 0.0001 compared with hatching in a two-tailed Fisher's exact test). The defects that accumulated in β-spectrin mutants were of several types, none of which were ever observed in wild-type animals (). First, we observed broken axon commissures that were detached from their process in the dorsal cord. These were clearly breaks rather than retractions, as both the proximal and the corresponding distal fragments could be identified as belonging to a single neuron. Second, we observed postembryonic growth cones. Finally, we observed aberrant branching. The percentage of all three types of defects increased between hatching and at 24 h after hatching (broken axons, 12 to 21%; growth cones, 2 to 17%; and branching, 13 to 22%). To determine whether the processes of other neurons also accumulated defects, we examined the morphology (in adult mutants; = 20) of acetylcholine phasmid neurons, interneurons, and motor neurons (). In contrast to newly hatched animals of this genotype, commissures were never visible, and all animals displayed defects in the dorsal cord and sensory dendrites. Thus, β-spectrin prevents the postdevelopmental accumulation of defects in neuronal processes, including sensory neurons as well as motor neurons, in processes oriented on the anterior-posterior axis as well as those on the dorsal-ventral axis, and in dendrites as well as axons. The breaks, growth cones, and branches in the neuronal processes of β-spectrin mutants could be the result of at least two different functions of β-spectrin. First, β-spectrin could act to prevent breaks, and growth cones and branches could be a consequence of breaking. Alternatively, β-spectrin could act to inhibit growth in mature neurons, and breaks and branches could be a consequence of initiating a new growth cone on an established process. To distinguish between these possibilities, we performed longitudinal studies in which we observed 12 individual animals at 24-h intervals (). In these studies, 25 individual nerve processes (that were initially normal) were later found to be broken. In some cases, we observed neurons that broke, regrew, and then broke again at a later time point (). Subsequent breaks occur at different locations along the axon. Although breaks are most easily observed in the commissure, breaks can also be observed in the nerve cords. Importantly, new growth cones were never observed on unbroken processes in these experiments (or in hundreds of additional single observations of β-spectrin mutants). Rather, growth cones appeared only on the proximal end of newly broken axons. Furthermore, every broken process in which the proximal end could be identified (21/25 breaks) reinitiated growth. Breakage and reinitiation of growth were observed at all larval stages. Thus, the primary defect in mutant axons is that they break, and the appearance of growth cones and abnormal branches is a secondary effect of breakage. These results demonstrate that β-spectrin is required to prevent spontaneous breaks in neuronal processes. Although neurons in animals that lack β-spectrin are capable of sprouting new growth cones, regrowth is error prone in comparison with initial outgrowth and generates pathfinding errors and aberrant branching (). However, these defects are not likely caused by the lack of β-spectrin. A large fraction of wild-type neurons that have been artificially severed also fail to regrow normally, demonstrating that abnormal regrowth is a property of regenerating neurons rather than of spectrin mutants (). Thus, neurons that lack β-spectrin have a specific defect in strain resistance. Breaks in neuronal processes that lack β-spectrin could be the result of two potential functions of β-spectrin in neurons. First, β-spectrin might be involved in the addition of membrane to axons during growth of the organism (). Because worms increase in length and circumference during development, failure to insert membrane into axons could cause them to break. Alternatively, β-spectrin might protect neurons against the acute strains caused by movement. To distinguish between these possibilities, we assayed axotomy in β-spectrin mutant animals that were paralyzed. encodes a muscle myosin (), and RNAi of this gene results in completely paralyzed animals. Furthermore, RNAi of in β-spectrin mutants results in animals that are longer than controls (). Thus, if movement is causing axotomy, RNAi of this muscle myosin should alleviate the phenotype because the animals move less. Conversely, if growth is causing axotomy, /myosin RNAi should exacerbate the phenotype because the animals are longer. We found that RNAi of /myosin largely rescued the neuronal defects of animals lacking β-spectrin (; see section RNAi in Materials and methods). L4-stage larvae had a mean of 4.8 axonal defects per animal. In contrast, L4-stage larvae animals treated with double-stranded RNA had a mean of 1.3 defects per animal (P < 0.0001 compared with control plasmid in a two-tailed test). RNAi of /myosin not only reduced truncated and broken axons but also reduced the incidence of growth cones and branching. The lack of new growth cones further supports the idea that excess and aberrant neuronal growth in animals that lack β-spectrin is a secondary effect of breakage rather than hypertrophic axon growth. RNAi of /myosin had no effect on the nervous system of wild-type animals. To demonstrate that the suppression of axon breakage was the result of paralysis rather than a specific effect of perturbation or the process of RNAi, we tested the effect of mutations. encodes the muscle protein Twitchin, and mutant animals are unable to initiate coordinated movement or deep body bends (). We found that the genetic loss of Twitchin in the β-spectrin mutant background resulted in paralyzed animals with fewer neuronal defects than β-spectrin mutants alone (number of normal commissures: = 2.7 ± 0.4; = 6.7 ± 0.5; = 10 each; P < 0.0001 in a two-tailed test). Thus, neurons in animals that lack β-spectrin are sensitive to strain caused by movement rather than growth. These data suggest that β-spectrin does not function in the process of neuronal membrane addition during organismal growth. Rather, β-spectrin protects neurons against breakage caused by movement-induced strain. In , β-spectrin is not essential for many neuronal functions, including growth cone migration and axon elongation. Disorganization of axon architecture is caused by breaks in axons followed by error-prone outgrowth. Similar to the worm mutants, β-spectrin mutants in exhibit disorganized axon tracts (). note that pioneer neurons initially grow out normally but that later, axon projections exhibit midline crossing defects. The loss of βII-spectrin in mice also results in nervous system defects, and the homozygous mutants die in utero (). It is interesting to speculate that these defects could be caused by breaks followed by error-prone outgrowth as in the worm. In , breaks occur in processes of β-spectrin mutants because of sensitivity to acute strain generated by muscle contraction. Neuronal processes may be particularly vulnerable to membrane lesions compared with other cells because a membrane rupture of such a slender process is more likely to result in a break that cannot be repaired by conventional membrane repair (). Thus, β-spectrin may be particularly important in neurons for strain resistance. Consistent with this, muscle and epithelial cell membranes do not appear to degenerate in animals that lack β-spectrin, and β-spectrin is expressed more strongly in neurons than other tissues (; ). The elastic properties of cells may arise from the intrinsic ability of spectrin to reversibly unfold in response to mechanical pulling (). Red blood cells must withstand deformation and shear as they pass through arterioles, and β-spectrin is required for this membrane elasticity (; ). Similarly, neurons can exhibit a remarkable amount of elasticity, stretching up to 65% of their length without breaking (). Thus, a role for spectrin in neuronal elasticity is consistent with its known function in membrane elasticity in erythrocytes. β-Spectrin is abundant in the vertebrate brain and is found in both the axonal and dendritic processes of all neurons (; ). Interestingly, intermediate filaments, which were long thought to function in neuronal integrity, do not play a major function in integrity but rather determine neuronal caliber (). Thus, the elasticity of vertebrate neurons may also be determined in large part by β-spectrin and the membrane skeleton. In sick or injured neurons, activated calpain targets the spectrin-based membrane skeleton for proteolysis (; ); thus, it is possible that neuronal degeneration after traumatic injury may be caused secondarily by the loss of the spectrin membrane skeleton (). Finally, some inherited neuronal diseases are the direct result of mutations in β-spectrin (; ). For example, a form of inherited progressive spinocerebellar ataxia (SCA5), which affected the family of former President Abraham Lincoln, has been recently linked to mutations in βIII-spectrin (). Mutations in βIII-spectrin are associated with a mislocalization of the glutamate transporter EAAT4 and the receptor GluRδ2, which suggests that the disease results in a defect in protein trafficking (). On the other hand, it is possible that this neurodegenerative disorder is caused by axonal breakage similar to the progressive disorder observed in mutants. It is particularly interesting that spinocerebellar ataxia is marked by a delayed onset; the slow progression and variable phenotype of the disease could be caused by the slow accumulation of axonal breaks. Although we have emphasized the neuroprotective role of spectrin in the differentiated nervous system, this does not preclude a role for spectrin during development. It is possible that targeted removal of the spectrin cytoskeleton is regulated during normal development to initiate the degenerative mechanisms used to prune unneeded axons (). All animals that were assayed were the progeny of homozygous hermaphrodites of the appropriate genotype. , , and were previously described (; ; ; ; ). expresses 40 amino acids of GAP-43 fused to GFP under the control of the promoter (a gift of J.-L. Bessereau, Institut National de la Santé et de la Recherche Medicale, Paris, France). expresses UNC-70 in the hypoderm under the control of the promoter (). It was integrated by x-ray irradiation and mapped to linkage group IV. Strains were obtained from the Genetics Center and were maintained at 20–23°C on plates seeded with HB101. Embryos between threefold and hatching were mounted on 3% agarose pads under a coverslip and were imaged with a confocal microscope (Microradiance; Bio-Rad Laboratories) using a 60× NA 1.4 planApo lens (Nikon). Each visible neuron was scored for morphology and assigned to one of the following five classes: reached dorsal cord and formed a T; reached dorsal cord and formed an L; anvil-shaped growth cones; round growth cones with radial protrusions; and defects (breaks, aberrant branching, or wandering growth). Scoring was performed blind to genotype (wild-type or ) and was repeated twice with similar results. For wild-type animals, 32 embryos and 155 neurons were scored, yielding the following counts: T = 94, L = 12, anvil = 35, round = 12, and defects = 2. For animals, 55 embryos and 192 neurons were scored, yielding the following counts: T = 118, L = 35, anvil = 20, round = 13, and defects = 6. Worms of the appropriate age and genotype (wild-type and ) were mounted on 3% agarose pads, paralyzed with 5 mM sodium azide in M9 under a coverslip, and imaged by epifluorescence microscopy or confocal microscopy. Each neuron was scored for defects in axonal morphology and assigned to one of the following four classes of increasing severity: intact commissure, broken commissure, ectopic growth cone, and hypertrophic branching. In a few cases, neurons displayed multiple phenotypes. To avoid double counting these neurons, the phenotypes were ordered in presumed developmental time from early to late: first, broken commissure; second, ectopic growth cone; and last, hypertrophic branching. Neurons with multiple phenotypes were then assigned to the latest class only. In addition to collecting data on commissures, we also collected data on migration along the dorsal cord. We observed a substantial increase in persistent L morphologies in larvae. Although these morphologies are common in wild-type embryos, they are rare in newly hatched animals and were never observed in the wild type by 24 h after hatching. However, we could not distinguish whether this phenotype was independent of commissure defects and, therefore, limited our analysis to the commissures. To determine whether degeneration also occurred in acetylcholine neurons and to ensure that degeneration was not caused by the transgene, we also scored commissures in acetylcholine motor neurons. expresses GFP in the DB acetylcholine motor neurons that extend commissures to the dorsal nerve cord. In addition, expresses GFP in the lateral CAN neuron, in the FLP and SAB neurons, and in other neurons in the head and tail. The processes of these neurons are oriented along the anterior-posterior axis unlike the dorsal-ventral commissures. We found that although some normal commissures were visible in newly hatched L1 animals, by the L4 stage, the commissures had degenerated ( > 10 for each larval stage). We observed a similar degree of degeneration in the lateral processes and in the sensory dendrites. Thus, degeneration occurs in acetylcholine as well as in γ-aminobutyric acid (GABA) neurons, in processes oriented on the anterior-posterior axis as well as those on the dorsal-ventral axis, and in sensory dendrites as well as axons. Because animals are very sick, we performed these experiments in an strain that carries the transgene, which expresses β-spectrin under the control of a skin promoter. These animals are healthier than alone but have similar neural defects. Newly hatched L1 worms were mounted individually on 3% agarose pads under a coverslip in 5 μl of 10 mM muscimol in M9; muscimol was used to anesthetize worms because worms recovered from these slides quickly and resumed normal behavior and growth. Images of VD and DD neurons were taken with a confocal microscope. After imaging, worms were recovered from slides and returned to small agar plates seeded with HB101. At 24-h intervals, the process was repeated for a total of 3–5 d. In all, 12 worms were analyzed. Animals of genotypes and were raised at 22°C on the bacterial strain HT115 containing an empty RNAi feeding plasmid (−; L4440) or an RNAi feeding plasmid (+; L4129; 1999 Fire Lab Vector kit; Addgene) as described previously (). L4 F1 hermaphrodites were selected and scored blind to genotype by counting DD and VD neuronal defects. 15 wild-type animals were scored under each condition; 44 animals were scored on RNAi, and 36 were scored on empty vector. This experiment was performed twice with similar results. The effect of RNAi on length was determined by measuring L4 animals at their midline. RNAi significantly increased length (L4 length of = 488.9 μM, = 16; L4 length of RNAi = 555.3 μM, = 18; P = 0.0003 using a two-tailed test). We also paralyzed animals of genotypes and by introducing the allele. Although animals twitch—that is, experience minor, random contractions of individual muscles—they cannot coordinate contraction and, therefore, do not generate deep body bends. We found that although animals are usually observed with deep body bends, animals have an extended posture. To assess the effect of –dependent paralysis of neuronal morphology, L4 animals were scored by counting the number of commissures that extended from the ventral to the dorsal cord without wandering or branching. Counts were performed blind to genotype. We also attempted to paralyze animals with genetic mutations in , , and and by using the drug levamisole (; ). However, these treatments were lethal in combination with , precluding these experiments.

During embryonic development in the mammalian peripheral nervous system (PNS), bundles of growing axons are surrounded by Schwann cell processes (). These processes sort larger caliber axons to the periphery of the bundles, where they adopt a 1:1 relationship with Schwann cells and are myelinated (). The transition to individual axon ensheathment is associated with, and presumed to depend upon, extensive Schwann cell proliferation (; ; ). Schwann cell proliferation during development is stimulated by cell surface axonal molecules, called the neuregulins, acting via ErbB receptors (; ; ; ). Laminins in the basal lamina secreted by Schwann cells have also been implicated in promoting Schwann cell proliferation at the early stage of axon sorting (; ). These stimuli may not be mutually exclusive because crosstalk between their signal transduction pathways occurs in oligodendrocytes in the central nervous system (CNS; ). Murine Schwann cells lacking the laminin γ1 chain lose all their known laminins and display reduced proliferation during sorting, increased postnatal apoptosis, and decreased ErbB2 phosphorylation (). In contrast, the absence of laminin-2 and -8 influences proliferation and radial axonal sorting, but cell death is unaffected, at least up to 2 wk after birth (; ). A major laminin receptor in the Schwann cell plasma membrane is α6β1 integrin (). Myelination in culture is blocked by anti–β1-integrin antibodies, and Schwann cells lacking β1-integrin display impaired radial sorting (; ). Nevertheless, neither the proliferation nor the survival of Schwann cells is affected in these mice, and they can go on to sort and myelinate axons, albeit inefficiently. Hence, the loss of different laminins and their receptors may have distinct effects during axon sorting. We have investigated the signaling pathways that might regulate axon sorting. FAK is a nonreceptor tyrosine kinase that is central to several signal transduction pathways, including those that stimulate proliferation (; ). FAK associates with β1 integrin in Schwann cells, and the presence of basal lamina activates FAK (). Neuregulin also causes FAK to associate with the Schwann cell ErbB2–ErbB3 complex (). Significantly, Schwann cells in mice deficient in neuregulin not only produce thinner myelin sheaths but also display defects in axon defasciculation (; ). Constitutive inactivation of FAK is lethal (); hence, we have investigated the role of Schwann cell FAK by targeted deletion using the Cre-P system. We find that FAK signaling in Schwann cells during axonal sorting is necessary to stimulate proliferation, and that without this late embryonic burst of cell division defasciculation of axons is highly impaired. We generated mice without functional FAK in myelinating Schwann cells, as described in Materials and methods. / mice were born in expected numbers, had no clinical phenotype at birth, and were of normal weight (wild-type, 1.39 ± 0.04 g; mutant, 1.37 ± 0.03 g; mean ± SEM; = 3 each). Cre-mediated recombination in this mouse line has already been shown to prevent expression of either FAK or its truncated forms in the epidermis and in neurons derived from embryonic stem (ES) cells (; ). First, we demonstrated Cre recombinase–mediated inactivation of in mouse sciatic nerve during late embryonic development, when radial sorting of axons occurs. Genomic PCR analysis showed that Cre-mediated recombination at the P sites at embryonic day (E) 18.5 was very efficient (). This is consistent with the fact that the regulatory elements of the CNP gene are robustly active in all perinatal Schwann cells in the sciatic nerve (). The low level of PCR product corresponding to the residual unrecombined floxed gene probably originates from perineurial fibroblasts and some Schwann cells that escape recombination. The absence of FAK from most Schwann cells in mutant nerve was made clear by immunofluorescence (). In the case of those few Schwann cells in the mutant nerve that had ensheathed axons and went on to myelinate, it is possible that only one allele had been inactivated because these cells were always positive for FAK by immunofluorescence (). By 2 wk after birth, / mice were distinctly less active than normal littermates, and they began to display a tremor by 3–4 wk that progressed to hindlimb paralysis after 3 mo in those few animals that were allowed to reach that age (Videos 1 [wild-type] and 2 [mutant], available at ). By E18.5, axons are organized into bundles and enveloped by Schwann cells (; ). There were considerable variations in axon caliber and bundle size at this age, but no discernible differences between control and mutant nerves (). Furthermore, both control and mutant Schwann cells were able to extend processes into the bundles, showing that FAK-null cells could still initiate the first steps in axonal sorting (). To confirm that mutant Schwann cells interdigitated into bundles normally, we counted the number of Schwann cells with visible nuclei that were associated with bundles and determined the percentage that extended processes into bundles, and they were essentially identical in wild-type and mutant cells (wild-type, 97.9 ± 2.1%; mutant, 98.6 ± 0.7%; mean ± SEM; = 3 each). Schwann cells that had adopted a one-to-one relationship with axons were detectable in both control and mutant nerves (, arrowheads, and ). Although loss of FAK has been shown to cause aberrations in basal lamina structure caused by altered laminin organization in the CNS (), basal lamina surrounding axon bundles in the wild-type nerve (, inset) appeared intact and identical to that in the mutant (, inset). To confirm that mutant embryonic Schwann cells had differentiated, we analyzed the localization of Periaxin. Periaxin is first detectable in the nuclei of embryonic Schwann cells in the sciatic nerve, but relocalizes to the cytoplasm of Schwann cells around E17.5–18.5 (). At postnatal day (P) 1, Periaxin was primarily cytoplasmic in both control and mutant nerves, indicating that mutant Schwann cells were not arrested in their embryonic development (). Extensive axonal sorting and the establishment of a 1:1 relationship between Schwann cells and axons had occurred by P3 in control sciatic nerve (). Mutant nerves were considerably smaller in cross-sectional area than control nerves (wild-type, 13,405 ± 640 μm; mutant, 7,548 ± 573 μm; mean ± SEM; = 3 each), which reflected the retardation of Schwann cell ensheathment at axons in the absence of FAK (). Axon bundles in mutant nerves remained larger, with much fewer ensheathed axons compared with control nerves (wild-type, 284 ± 26; mutant, 19 ± 3; mean ± SEM; = 3 each; ). In contrast to control axon bundles, where large-caliber axons were peeled off by Schwann cells, leaving behind axons with a range of diameters (), mutant nerves were characterized by the arrested sorting of large-caliber axons to the edges of the bundles (). This provided further evidence that mutant Schwann cells were capable of inserting processes into bundles and sorting axons radially. Subsequently, smaller bundles of large-caliber axons appeared to be sorted away from mixed bundles (); a similar phenomenon has been observed in β1 integrin–deficient nerves (). We asked if the abnormalities in axon sorting in the mutant might be caused by deficiencies in the number of Schwann cells. After E17.5, the number of cells in mutant nerves was significantly reduced in comparison to control nerves up to P4 (). This could be caused by decreased proliferation, increased cell death, or both. Labeling with BrdU revealed that Schwann cells lacking FAK proliferated much less than control cells from E17.5 to P4 (). FAK signaling can prevent apoptosis in vivo (); nevertheless, there was no demonstrable increase in apoptosis in mutant nerves at either E18.5 or P4, as detected by activated caspase-3 expression, and activated caspase-3–positive cells always comprised <0.1% of all cells in wild-type and mutant nerves (). These results are in marked contrast to FAK deletion in the developing forebrain, where although neuronal apoptosis was unaffected, proliferation was also unaffected; in contrast, the absence of FAK affected cell migration in the CNS, probably because of alterations in basal lamina organization; furthermore, the morphology of neuronal dendrites was abnormal (). It seems that the effect of losing FAK depends very much on cell type. In a keratinocyte-restricted FAK knockout, the absence of FAK resulted in fewer keratinocyte precursors, which may be caused by defects in mitosis, although apoptosis was unaffected, whereas FAK-null keratinocytes proliferated and migrated normally (). Interestingly, these FAK-null keratinocytes undergo massive apoptosis if placed in culture, which underscores the importance of evaluating the consequences of deleting FAK in vivo when studying complex tissues. Because signaling by axonal neuregulin via ErbB2–ErbB3 receptors has been linked to FAK recruitment (), it was important to determine if any effects on radial sorting that we might attribute to loss of FAK were caused by aberrant ErbB2–ErbB3 function. Western blots showed that ErbB2 phosphorylation was not impaired in the mutant at P1, indicating that axonal neuregulin could still activate the receptor (). Furthermore AKT phosphorylation was also normal, indicating no major derangement to the PI3 kinase–AKT pathway downstream of ErbB receptors (). Interestingly, this also provides support for the view that the absence of FAK does not cause gross structural defects in laminin organization, as seen in the CNS (), because the absence of laminin from Schwann cell basal lamina causes major reductions in ErbB2 phosphorylation (). Because the PI3 kinase pathway has been implicated in both cell survival and proliferation in Schwann cells (; ), this demonstrates that the defect in proliferation during perinatal axonal sorting is not a result of deficits in FAK signaling via the PI3 kinase pathway. Furthermore, because AKT phosphorylation is unaffected during the active phase of perinatal Schwann cell proliferation, FAK does not appear to influence the signaling pathway from ErbB receptors via PI3 kinase to AKT, thus supporting the view that FAK acts via an independent pathway, probably originating with laminin. AKT may still be necessary to promote proliferation during radial sorting, but it is clearly not sufficient during radial sorting. Although β1 integrin and FAK are functionally linked, Schwann cells lacking β1 integrin do not display reduced proliferation (; ). Nevertheless, ablation of these proteins can have distinct effects in the same cell type. FAK-null ES cells can differentiate, whereas the differentiation of β1 integrin–null ES cells is severely retarded (; ). Furthermore, the fact that FAK can either suppress or promote growth in the same cell line indicates that diverse signaling pathways may have distinct roles to play in regulating proliferation at different stages of Schwann cell differentiation (). In contrast to control nerves at 6 wk after birth, where axon ensheathment and the formation of a multilamellar compact myelin sheath were very advanced, mutant nerves still retained bundles of unsorted axons (). Typically, some axons that had been sorted to the edge of bundles were myelinated, but they were still attached to a bundle (), and some bundles appeared to be surrounded by fibroblasts (). There was no substantial progress in axon sorting up to 6 mo after birth, and bundles of large-caliber axons persisted with possible further evidence of perineurial fibroblast infiltration (). Hence, mutant nerves were unable to recover from the perinatal deficit in Schwann cells numbers. All animal work conformed to UK legislation (Scientific Procedures) Act 1986 and the Edinburgh University Ethical Review policy. Generation of mice carrying targeted P sites in the gene, and the genotyping of these mice, has been previously described (). Targeted ablation of FAK in myelin-forming glia was achieved by crossing mice with mice heterozygous both for the floxed allele and for inserted into the locus (/; ). Both mice and heterozygous mice were phenotypically indistinguishable from wild-type mice, as previously shown (; ). Cre-mediated recombination at E18.5 in sciatic nerves was assessed by PCR analysis of genomic DNA, followed by HindIII digestion, as previously described (). Mean weights of newborn animals were measured using seven animals per condition. For Western blotting, we used anti-ErbB2 (1:250); anti–phospo-ErbB2 (Tyr877; 1:1,000); anti-AKT (1:1,000); and anti–phospho-AKT (Ser473; 1:1,000). All rabbit antibodies were obtained from Cell Signaling Technology, and a mouse monoclonal anti–β-actin (IgG1 clone AC-15) was purchased from Sigma-Aldrich. For immunofluorescence, we used rabbit anti-FAK obtained from UBI (BC3, 1:100) and Santa Cruz Biotechnology, Inc. (C-20, 1:50); mouse monoclonal anti–neurofilament-H purchased from Sigma-Aldrich (IgG clone N32; 1:5,000); rabbit anti-activated caspase-3 (1:100) obtained from Cell Signaling Technology; and mouse monoclonal anti-BrdU clone B44 purchased from Becton Dickinson (IgG1; 1:6). Rabbit antibodies against Periaxin () and the neurofilament triplet proteins () have been previously described. Nuclei were stained with either 0.1 μM TOTO-3 obtained from Invitrogen or 2 μM DAPI obtained from Sigma-Aldrich. Immunofluorescence labeling of cryostat sections have been previously described (). Unless otherwise specified, all analyses were performed on the tibial branch of the sciatic nerve. We used a confocal microscope (TCL-SL; Leica) and Leica proprietary software. Conventional fluorescence microscopy was carried out using a microscope (BX60; Olympus), images were captured using a camera (Orca-ER; Hamamatsu), and morphometry was done with Openlab software (Improvision). Thin sections of nerves for electron microscopy were prepared as previously described (). Photographic negatives were scanned and digitized. All figures were prepared using Photoshop version 7.0 (Adobe). Timed-pregnant females (for E17.5 and 18.5 embryos) or individual pups (P2 and 4) were injected intraperitonealy or subcutaneously, respectively, with BrdU (100 μg/g body weight), and animals were killed by decapitation 70 min later. Nerves were fixed for 30 min in cold 4% paraformaldehyde and embedded in OCT; then 10-μm transverse sections were cut, and at least five sections per animal were counted. For Schwann cell quantitation, sections were stained with antineurofilament to identify nerves and with DAPI to identify cell nuclei. For BrdU labeling, rehydrated sections were treated with 0.5% Tween-20/PBS for 5 min, followed by a 1:1 mixture of 10 M HCl/0.5% Tween-20/PBS for 45 min. Sections were washed twice with PBS and twice with 0.5% Tween-20/PBS, and then coincubated with monoclonal anti-BrdU and rabbit antineurofilament antibodies for 2 h, followed by appropriate secondary antibodies. DAPI- or BrdU-labeled nuclei were only counted if they lay within the boundary of the neurofilament-positive tissue. The percentage of BrdU-positive nuclei was the mean BrdU count divided by the mean total cell count for that time point. Cross-sectional areas of 1 μm toluidine blue–stained sections of nerve, excluding the perineurium, were measured. The numbers of myelinated axons were quantitated in the same sections. For quantitation of the percentage of interdigitating Schwann cells, images were acquired for every axon bundle that had an associated Schwann cell with a distinct nucleus in a complete nerve section. A minimum of 30 Schwann cells was counted per sample. 10-wk-old wild-type (Video 1) and mutant (Video 2) mice were filmed to show the severe gait problems, especially in the hind limbs, displayed by demyelinated FAK-null mice. Online supplemental material available at .

During the postmeiotic maturation of male haploid germ cells, or spermiogenesis, the DNA is repackaged in a process involving a dramatic chromatin reorganization (). At the onset of spermiogenesis, round spermatids inherit a nucleosome-based chromatin organization, which is progressively restructured while the genome undergoes condensation. A wave of global histone acetylation marks the initial steps of this process in elongating spermatids and precedes their replacement in condensing spermatids, first by transition proteins, TP1 and TP2, and then by protamines. The latter ensure tight DNA packaging by the establishment of multiple intraprotein cross-links (; ). This textbook vision of mammalian spermiogenesis is now challenged by findings suggesting that the DNA is actually not homogeneously packed within the spermatozoa (). Approximately 10–15% of histones are retained in the human sperm nucleus, heterogeneously distributed within the genome, with an enrichment in specific loci such as imprinted genes or genes expressed during early embryogenesis (; ). Moreover, a large diversity of somatic-type or testis-specific histone variants become associated with the DNA in germ cells (; ; ), some of them presenting a heterogeneous distribution within the spermatids or mature sperm nucleus (; ). In addition, heterochromatin regions seem to maintain a distinct organization during spermiogenesis, as telomeres were shown to be enriched in somatic-type core histones and H2B variants (; ; ; ), and centromeres maintain some of their somatic-specific marks, such as an enrichment in the histone H3 variant CENP-A in spermatozoa (). These observations suggest that the genome undergoes a regional differentiation during mammalian spermiogenesis. The exact nature of this differential reorganization of the genome and the molecular mechanisms driving it are unknown. One possibility is that this process could initially be built from differential marks inherited from early germ cells. In fact, pericentric heterochromatin is characterized by a specific histone code including a K9 trimethylation of histone H3 (H3K9me3) and its association with nonhistone proteins including the HP1 family members (). These large regions surrounding the centromeres are mainly composed of satellite repeats, named the major satellites in the mouse, generally assembled in clusters known as “chromocenters” (). After the completion of meiosis, the pericentric heterochromatin still harbors somatic features (; ). However, in mouse round spermatids, it undergoes a very unique reorganization characterized by the assembly of all pericentric regions into a single large chromocenter. A very interesting and unanswered question is whether this distinct feature could be the first step of a specific reprogramming of pericentric heterochromatin in male haploid germ cells. To investigate this issue, we undertook a step-by-step exploration of the chromocenter organization during mouse spermiogenesis. An increase in histone acetylation had previously been observed in early elongating spermatids (). A close inspection of histone modifications during spermatid elongation reveals that pericentric heterochromatin exhibits very unusual characteristics combining active and repressive histone marks. Moreover, at later stages of spermiogenesis, nucleosomal structures containing acetylated histones are retained on the major satellites when most histones have been removed elsewhere. Finally, the investigation of nucleoprotein structures organizing the genome in condensing spermatids has led to the identification of several new H2A and H2B histone variants. Two of them, named H2AL1 and H2AL2, were found in new DNA packaging structures, which specifically reorganize the major satellite DNA in condensed spermatids. Altogether, these data highlight specific processes activated after meiosis and establish a differential organization of pericentric heterochromatin during mouse spermiogenesis. During postmeiotic chromatin reorganization in male germ cells, a global hyperacetylation of histones occurs, which precedes their removal (). To better characterize these events, the kinetics of the core histones' hyperacetylation and disappearance has been followed by immunofluorescence (IF) on microdissected squash preparations of seminiferous tubules, using an antibody recognizing the tetra-acetylated H4 N-terminal tail (H4ac). Round spermatids (steps 2–6) are weakly stained by the antibody (). In early elongating spermatids (step 8), H4 acetylation clearly increases and is homogeneously distributed throughout the whole nucleus. At later stages (steps 9–10), the signal for acetylated H4 globally decreases, except in a central domain, which is also intensely stained with DAPI (, arrows). Finally, in step 11 condensed spermatids, acetylated H4 has completely disappeared. In elongating step 9–10 spermatids, this restricted central area of the nucleus, where acetylated H4 remains, could correspond to a region where the genome is differentially reorganized. Because of its intense DAPI staining, we hypothesized that this domain could correspond to the A/T-rich pericentric constitutive heterochromatin, composed of the mouse major satellites repeats. To investigate this point, immuno-FISH assays were performed, where acetylated H4 was first detected by IF and the major satellites were then localized by FISH on the same germ cells. As expected, the major satellites were detected in the identifiable chromocenter of round spermatids (, ). The DAPI-dense regions of step 9–10 elongating spermatids were also completely stained by the major satellites probes (, 6 and 10), confirming that they indeed correspond to pericentric heterochromatin. A close analysis of pericentric heterochromatin shows that it undergoes important changes during spermatid maturation. In round spermatids, it is underacetylated and colocalizes with the round chromocenter. In elongating spermatids, it becomes enriched in acetylated histone H4 while undergoing decompaction and spreading within the nucleus (, compare 1 to 5 and 9; note that the acquisition of the H4 acetylation signal in round spermatids was enhanced compared with that in to give better evidence of the unacetylated state of pericentric heterochromatin in round spermatids compared with the hypoacetylated state of the rest of the genome). We further aimed to identify the H4 lysines targeted by acetylation within pericentric heterochromatin. We had previously observed that the global acetylation increase in elongating spermatids mainly affects K5, K8, and K12 residues of histone H4, but not K16 (unpublished data). Here, IF with specific antibodies shows that all three acetylated lysines—acK5, acK8, and acK12—are associated with the major satellite region in elongating spermatids (). This unusual accumulation of acetylated histones in spermatids' pericentric heterochromatin led us to investigate the fate of known heterochromatin marks, such as HP1 binding and H3K9 trimethylation in these cells. HP1β was present in the chromocenter of round spermatids but disappeared at later stages, when H4 acetylation accumulated (), showing a tight relationship between the presence of acetylated H4 in pericentric heterochromatin and the removal of HP1β at the beginning of the elongation process. In contrast, trimethylation of H3K9, easily detectable in the round spermatids' chromocenter, does not disappear in elongating spermatids when H4 acetylation takes place (). Interestingly, during a short period of their developmental stage, corresponding to step 9 spermatids, both marks were localized in pericentric heterochromatin of all cells. A detailed analysis of both modifications in these cells was performed by confocal microscopy. The intensity of H4Ac and H3K9me3 signals, revealed by Alexa 546 and Alexa 488 fluorochromes, respectively, were quantified: for each detection, a region containing values >50% of maximal fluorescence was delimited (, red and green borders, panels 4–8), and a quantification of fluorescence was shown along an axis and reported on a diagram (). Similar experiments were performed by reversing the fluorochromes (H4Ac signal in green and H3K9me3 signal in red), so that artifacts due to differential bleaching or sensitivity could be ruled out (Fig. S1, available at ). It shows that the signals for H4 acetylation and H3K9 trimethylation actually do not strictly colocalize ( and Fig. S1 B, diagrams). This observation suggests the existence within the analyzed domain of subregions where nucleosomes harbor either one or the other modification, but probably not both simultaneously. This partial colocalization of the two histone marks might reflect a dynamic process reorganizing pericentric heterochromatin at this specific stage of maturation (step 9 spermatids), with a progressive invasion of the major satellite region by H4 acetylation. We further aimed to determine whether pericentric regions would also harbor specific chromatin features at later stages of spermiogenesis, after disappearance of acetylated H4 and trimethylated H3K9 in condensing spermatids (, 10). Indeed, at this stage, either the modifications could be removed, leaving the histones in place, or the modified histones themselves could be displaced and degraded. To investigate the possibility of a nucleosomal structure remaining in the pericentric heterochromatin of condensing spermatids, we chose a biochemical approach. Step 12–16 condensing spermatids were purified from mouse testes. Their extreme compaction renders their nuclei completely resistant to MNase digestion by the classical assays developed for somatic cells. Therefore, we first had to partially decondense them in a detergent-containing buffer, to obtain a soluble nuclear extract, and a chromatin-containing pellet, which was then submitted to MNase fractionation (Fig. S2 A, protocol 1, available at ). The MNase-solubilized fraction was then separated by centrifugation from an insoluble pellet. The analysis of the DNA present in the MNase-solubilized fraction shows that a prolonged digestion by MNase produces a single MNase-resistant DNA fragment of the expected nucleosomal size, ∼150 bp, indicating that at least part of the nucleosomal chromatin of condensing spermatids was indeed obtained with this procedure (). Accordingly, core histones were abundantly released by MNase digestion (Fig. S2 B, H2A, H2B, and H4). Surprisingly, this procedure also generated an additional DNA band that migrated faster than the nucleosomal DNA fragment and that was more sensitive to MNase digestion (). An electrophoretic analysis on a concentrated agarose gel with small size DNA markers confirmed that this band essentially corresponds to a discrete DNA fragment of ∼60 bp (). It is of note that this smaller DNA fragment was not produced from somatic cells (, bottom) or from germ cells at earlier stages of spermiogenesis (not depicted), after MNase treatment in the same conditions. We concluded, therefore, that this fragment could correspond to a condensing spermatid-specific DNA packaging structure. We then further investigated the nature of the DNA associated with the nucleosomal and the new spermatid-specific DNA packaging structures. For this purpose, the two types of MNase-resistant DNA fragments were purified and used as probes in FISH assays, on mouse metaphase chromosomes, and on mouse spermatogenic cells. Interestingly, both fragments detected the pericentric regions and perfectly colocalized with the major satellite regions both on metaphase chromosomes and in spermatogenic cells (; and Fig. S4 A, available at ). As controls, probes obtained either by sonication of spermatid total DNA or after MNase digestion of chromatin of mouse somatic cells using the stringent conditions of protocol 1 or of a whole testis cell suspension did not label any specific domain (; and Fig. S4 A). The quantification of the fluorescence signals given by each probe show that ∼72% of the nucleosomal fragment and 51% of the small fragment correspond to major satellite sequences (Fig. S4 B). These results indicate that in condensing spermatids, nucleosomes remain in pericentric heterochromatin regions, where they coexist with another MNase-resistant structure of an unknown nature. We therefore hypothesized that these new DNA packaging structures may contain condensing spermatid-specific proteins, which remained to be identified. To find new proteins involved in the organization of the genome in late spermatids, a proteomic approach was undertaken. Basic proteins were extracted with acid from nuclei of purified step 12–16 spermatids and compared with basic proteins of germ cells at earlier maturation stages on SDS-PAGE. Coomassie staining revealed global changes in the basic protein constitution of condensing spermatids compared with that of germ cells at earlier stages, with a decrease in histone content, and the appearance of two bands ∼8 and 18 kD, corresponding to TP1 and TP2, respectively (). The proteins present in this particular cell fraction were identified by mass spectrometry (MS). As expected, transition proteins TP1 and TP2 and four canonical core histones were found, as well as known somatic and testis-specific histone variants. This proteomic analysis also revealed for the first time the presence of the mouse testis histone variant H3t (available from GenBank/EMBL/DDBJ under accession no. ). The mRNA encoding the human orthologue was previously shown to be expressed in spermatocytes and spermatids (). More important, three new proteins containing substantial sequence similarity with the H2A or H2B histone folds were also identified. These new histone variants were named H2Alike1 (H2AL1; available from GenBank/EMBL/DDBJ under accession no. ), H2Alike2 (H2AL2; accession no. ), and H2Blike1 (H2BL1; accession no. ), respectively. H2AL1 and H2AL2 are related proteins and show a better homology to each other than to H2A (). H2BL1 is very similar to the bovine testis–specific H2B variant, known as SubH2Bv (). In addition to the MS analysis, an in silico search in testis ESTs was performed and led to the identification of two supplementary H2A and H2B variants named H2AL3 (accession no. ) and H2BL2 (accession no. ), respectively (). The expression pattern of the genes encoding the five new histone variants was analyzed by RT-PCR on RNAs from various mice tissues. shows that, like the H3t variant, all the new histone variants are mainly expressed in the testis. To test whether all these new variants could be postmeiotically expressed, RT-qPCR (quantitative PCR) was performed on male germ cells fractionated according to their maturation stages on a BSA gradient. shows that H2AL1, H2AL2, H2AL3, and H2BL1 mRNA were strongly enriched in round and elongating spermatids (compare to pachytene spermatocytes, used as a reference), in contrast to TH2B used as a control, when mRNA was most abundant in meiotic cells. H2BL2 mRNA was detected at a very low level in meiotic, as well as postmeiotic, germ cells. H2BL2, although present in whole testis extracts, was not detected in germ cells ( and not depicted), and H2AL3 was not detected either in whole testis extracts or in germ cells (not depicted). These data suggest that these in silico identified histones probably do not play a role in the postmeiotic chromatin reorganization of male germ cells. In contrast, H2AL1/L2 and H2BL1 are strongly enriched in step 12–16 spermatids compared with the whole testis () and accumulate during late spermiogenesis, in condensing spermatids (). Their expression pattern is therefore different from that of TH2B, detected in meiotic cells and afterward, but similar to that of TP1, TP2, Protamine2, and H1t2, suggesting that H2AL1/L2 and H2BL1 could, like these proteins, be involved in chromatin organization during spermatid condensation. Finally, and very interestingly, H2AL1/L2 and H2BL1 remain present in mature spermatozoa isolated from epididymis, whereas all of the other known spermatid proteins, such as TP1, TP2, and TH2B, disappear (). The subnuclear distribution of H2AL1/L2 in germ cells was further analyzed by IF on testis imprints (). These proteins first appear in step 9 elongating spermatids and strongly accumulate in early condensing spermatids (step 11) after the disappearance of the acetylation signal and before and during protamine incorporation (). These observations are in good agreement with the Western blot data () and with immunohistochemistry (IH) data showing H2AL1/L2 accumulation in step 9–11 spermatids (). Interestingly, H2AL1/L2, in contrast to TP1 and TP2, show a heterogeneous distribution in early condensing spermatids, with a preferential localization in the intensely DAPI-stained region, previously identified as pericentric heterochromatin (). These data highly suggest that during late spermiogenesis, H2AL1/L2 could differentially organize pericentric heterochromatin after the disappearance of acetylated nucleosomes. However, although clearly detectable in Western blots, H2AL1/L2 could not be observed by IF or IH in both step 12–16 spermatids and mature spermatozoa, probably because of the high genome compaction of these cells. Moreover, H2BL1 could not be detected in germ cells by these in situ approaches, probably also because of limited antibody accessibility. Because H2AL1/L2 were localized to pericentric heterochromatin, we checked for their presence in the major satellite-organizing nucleoprotein structures previously isolated from step 12–16 spermatids. In preliminary analyses, we observed that H2AL1/L2, as well as somatic-type core histones and TH2B, were released in the MNase-digested fractions of condensing spermatids prepared according to protocol 1 (Fig. S2 B), suggesting an association of these histones with at least one of the two MNase-resistant nucleoprotein structures shown in . However, in contrast to the somatic core histones, a substantial amount of H2AL1/L2—and, to a lesser extent, TH2B—was also present in the soluble nuclear extract, suggesting that the decompacting buffer (Fig. S2 A, protocol 1) had partially disrupted the corresponding nucleoprotein structures (Fig. S2 B). We therefore set up another procedure using a less stringent decondensing buffer (Fig. S2 A, protocol 2), which prevented the partial disruption of histone variants (Fig. S2 C) but gave the same MNase-digestion pattern (Fig. S3 B and not depicted). To investigate the association of H2AL1/L2 within the nucleosomal or the smaller structure, the two structures (obtained using the less stringent protocol 2) were separated on a sucrose gradient, and the associated histones or histone variants were analyzed by Western blot (). As expected, the somatic-type histones H2A, H2B, H3, and H4 cofractionated with the larger fragment, confirming its nucleosomal nature (, lanes 4–7; and , stain [lane 2] and H3 and H4 [lane 3]). In contrast, the H2AL1/L2 variants specifically cofractionated with the small MNase-resistant DNA fragment, indicating that they could be major constituents of the new DNA packaging structure identified in condensing spermatids (, H2AL1/L2 [lanes 1–4]; and , V [lane 2]). Remarkably, neither H3 nor H4 could be detected in this structure, even when higher concentrations of proteins from these fractions were analyzed (, stain [lane 1] and H3 and H4 [lanes 2 and 4]). Interestingly, TH2B cofractionated with both DNA fragments, suggesting that, unlike H2B, it associates not only with nucleosomes but also with the new structures. To confirm the association between the DNA fragments and histones, further purifications of the nucleoprotein structures were performed using hydroxyapatite. This ion-exchange medium presents a high affinity for DNA, which allows the capture and purification by phosphate elution of nucleosomes or other nucleoprotein structures (). In these experiments, H2AL1/L2 not only were pulled down on hydroxyapatite but also perfectly coeluted with the small DNA fragment (Fig. S3, B and C), showing that the small DNA fragment and H2AL1/L2 are part of the same structure. Moreover, it was confirmed that TH2B, present within the nucleosomes, also associates with the new structure. The association of TH2B and the novel histone variants within an unknown DNA packaging structure prompted us to investigate their ability to reorganize the genome when ectopically expressed in somatic cells. We first monitored the ability of H2AL1 and H2AL2 to dimerize with either H2B or TH2B. HA-tagged H2B and TH2B were coexpressed with H2A, H2AL1, and H2AL2 fused to GFP. H2B and TH2B were then immunoprecipitated using an anti-HA antibody, and the presence of H2A, H2AL1, and H2AL2 was detected by Western blot (). The data show that H2A is able to form dimers with TH2B but less efficiently than with H2B. In contrast, H2AL2 displayed a strong preference for dimerization with TH2B compared with H2B (). H2AL1 also dimerized more efficiently with TH2B than with H2B, albeit with a lower affinity. These findings nicely correlate with the results obtained after the analysis of the chromatin of condensing spermatids (), showing the presence of TH2B and the absence of H2B in the histone H2AL1/L2–containing structures. An important question is whether the H2AL1/L2–TH2B dimers are capable of forming regular nucleosomes and/or could induce the formation of a different type of structure in somatic cells. To investigate this issue, the chromatin from cells expressing H2AL2/TH2B was digested by MNase, and the resulting oligonucleosomes were fractionated on a sucrose gradient. shows that all the H2AL2 and TH2B were efficiently incorporated into regular nucleosomes, and no smaller DNA packaging structure was found. We then analyzed the stability of the H2AL2–TH2B–containing nucleosomes compared with nucleosomes containing only somatic histones. Oligonucleosomes from cells expressing tagged H2A–H2B or tagged H2AL2–TH2B were immobilized on hydroxyapatite and eluted with increasing concentrations of salt. The results show that H2AL2–TH2B dimers are released at lower salt concentrations than H2A–H2B dimers, suggesting that H2AL2–TH2B–containing nucleosomes are less stable than nucleosomes containing somatic histones (). Altogether, these data show that TH2B and H2AL2 preferentially associate with each other, are able to form nucleosomes when expressed in somatic cells, and are found in new nucleoprotein structures specifically organizing pericentric heterochromatin in condensing spermatids. This study shows for the first time that pericentric heterochromatin exhibits very specific features at all stages of mouse spermiogenesis. Upon completion of meiosis, round spermatid chromatin retains somatic-type epigenetic marks, which then undergo reprogramming at the beginning of the elongation phase. A wave of histone acetylation affecting the whole genome marks the initiation of the chromocenter decompaction and spreading. In elongating spermatids, pericentric heterochromatin acquires a novel and unique organization where it is both marked by H4ac and H3K9me3 and loss of HP1. The observed general histone hyperacetylation and the removal of HP1 from pericentric regions are reminiscent of the reported disappearance of HP1 from chromocenter after prolonged trichostatin A treatment in somatic cells (). It is therefore very probable that at the beginning of spermatid elongation, the global histone acetylation triggers the disruption of HP1 binding as well as the induction of pericentric heterochromatin reprogramming. Interestingly, although HP1 is lost, the H3K9me3 mark remains when H4 acetylation invades the pericentric regions and could be involved in delaying the completion of histone acetylation within the major satellite region. Indeed, although both marks colocalize in pericentric heterochromatin, a more detailed analysis suggests that they are not present on the same subregions and, consequently, on the same nucleosomes. This implies the occurrence of an active removal of the H3K9me3 mark before or simultaneously with H4 acetylation. It could be either demethylated or widely exchanged with unmethylated histone H3 or H3 variants. Interestingly, the proteomic identification of the surviving histones isolated from condensing spermatids (steps 12–16) presented here showed the presence of the H3 variants H3.3 and H3t. Hence, the exchange of trimethylated H3K9 with H3.3 and H3t in the pericentric regions could also account for the progressive removal of the H3K9me3 mark. Later in spermiogenesis, the surviving nucleosomal organization of the pericentric regions could provide a basis for the preferential recruitment of new H2A variants identified here in condensing spermatids. Interestingly, our data point to TH2B as a key player in the incorporation of H2AL1/L2. First, TH2B is expressed much earlier than H2AL1/L2 during spermatogenesis but is still present at late spermiogenesis stages, when H2AL1/L2 accumulate. Second, using an ectopic expression approach, we show that H2AL1 and H2AL2 largely prefer TH2B to H2B, as a dimerization partner. Moreover, unlike H2B, TH2B was found with H2AL1/L2 associated to the small MNase-resistant DNA fragments, strongly suggesting that H2AL1 or H2AL2 also dimerize with TH2B in spermatogenic cells. Third, TH2B possesses the ability to induce nucleosome instability when incorporated in vitro into nucleosomes containing somatic-type histones (), and H2AL2–TH2B–containing nucleosomes were found here less stable than those containing H2A–H2B. Altogether, these observations support the hypothesis that TH2B-containing nucleosomes would be preferential sites for H2AL1/L2 incorporation through direct dimerization. These unstable nucleosomes would then become targets for important structural reorganization in condensing spermatids, leading to the formation of the new structures evidenced here by MNase digestion. Several testis-specific factors could potentially play a role in such a process. For example, the bromodomain-containing factor, TIF1-∂, is expressed in elongating spermatids (steps 9–11) concomitantly with the apparition of H2AL1/L2 and preferentially localizes on pericentric heterochromatin (). Although H2AL1/L2–TH2B dimers were incorporated into the nucleosomes in somatic cells, neither H3 nor H4 could be found in the H2AL1/L2–TH2B–containing structure in condensed spermatids. These observations are reminiscent of previous studies showing that a MNase digestion of chromatin from somatic cells could also produce subnucleosomal particles, some of which are composed of H2A–H2B dimers associated to a DNA fragment of ∼50 bp (; ; ). Whether the H2AL1/L2–TH2B are the testis-specific counterparts of these somatic subnucleosomal particles or whether they associate with other sperm basic proteins to form completely new DNA packaging structures is to be addressed in the future. In contrast to H2AL1/L2, the presence of TH2B is strongly reduced in epididymal spermatozoa (). However, longer Western blot exposures and MS analyses reveal that a small amount of TH2B remains (unpublished data), and an IF approach also suggested that hTSH2B (the human homologue of TH2B) could be detected in a subset of spermatozoa (). The exact nature of the H2AL1/L2-containing structures in sperm thus remains to be established. The possibility that H2BL1, which is synthesized later, partly replaces TH2B in the very late stages of spermiogenesis is an attractive hypothesis requiring further investigation. In the human genome, no H2A and H2B variants were found with substantial sequence similarity to the mouse H2AL1, H2AL2, and H2BL1. However, other testis-specific variants could act as functional homologues in human germ cells. For example, H2BFWT, a human testis–specific H2B variant present in mature sperm, is found uniquely in primates and, like H2BL1, shows a relatively low similarity with H2B and TH2B (). Interestingly, several human sperm H2B variants preferentially accumulate on telomeres (; ; ). Thus, although not conserved, specific histone variants could organize particular regions of the genome, like telomeres, centromeres, or pericentric heterochromatin, within the globally protamine-packaged genome. This sperm-specific packaging of pericentric heterochromatin could be important for postfertilization chromatin reorganization events. Indeed, after fertilization, a genome-wide epigenetic reprogramming occurs and establishes the totipotency of the zygote from the differentially organized paternal and maternal genomes, inherited from the highly specialized male and female gametes. Interestingly, recent data report important differences between the onset of paternal and maternal pericentric heterochromatin epigenetic marks in the first steps of preimplantation embryogenesis in the mouse (; ). Therefore, it is tempting to speculate that the presence of H2AL1/L2 in a novel organizational unit of major satellites in sperm would, after fertilization, act as a guide for epigenetic reprogramming of paternal pericentric heterochromatin. It is also likely that histone variants play a role in the establishment of other epigenetically inherited structures in the male genome. Indeed, in addition to a clear enrichment in pericentric heterochromatin, H2AL1/L2-containing structures are also dispersed elsewhere in the genome (Fig. S4 B). Characterization of these regions will constitute an additional, exciting challenge for the future, which should shed further light on both the nature and transmission of paternal epigenetic information. Antibodies against H2AL1, H2AL2, H3AL3, H2BL1, and H2BL2 were generated in rabbits by three injections of 200 μg of purified His-tagged recombinant proteins. Sera were diluted at 1:1,000 for Western blot and 1:250 for IF. Anti-H1t2 is described by . Anti-H3 was provided by Abcam. Anti-TP1 and anti-TP2, anti-protamine, and anti-acetylated antibodies were provided by S. Kistler (University of South Carolina, Columbia, SC), R. Balhorn (Lawrence Livermore National Laboratory, Livermore, CA), and M. Yoshida (RIKEN, Wako, Japan), respectively. Other anti-histones antibodies were provided by Upstate Biotechnology and used as advised by the supplier. Three different protocols were used, depending on the required quantity, purity degree, and maturation stage of male germ cells. Fractions enriched in spermatogenic cells at different stages of maturation were obtained by sedimentation on a BSA gradient as previously described (). Pure fractions of nuclei from step 12–16 spermatids were obtained by sonication of mice testis as described by . Pure fractions of epididymal sperm heads were obtained as follows. Epididymes were opened with a razor blade to free spermatozoa in a DME drop. The spermatozoa were then pelleted by centrifugation (4°C, 1,300 , 10 min), resuspended in 1.5 ml DME containing 1 mg/ml salmon sperm DNA, sonicated at 250 J to break flagella, and centrifuged (1,300 , 10 min, 4°C) on a discontinuous Percoll gradient (100%/70%/40%). The pellet, containing pure sperm heads, was washed once in DME/1 M NaCl and then in DME. The quality of each fraction or preparation was controlled by observation under a phase-contrast microscope. Microdissected tubules were prepared as described by . Testis imprint preparations were performed by gently pressing the testis (previously cut in two and frozen in liquid nitrogen) onto glass slides, air-drying, incubating in 90% ethanol for 3 min, and air-drying again. Permeabilization of cells was allowed in 0.5% saponine, 0.25% Triton X-100, and 1× PBS for 15 min at room temperature. Nonspecific binding was blocked with 5% dried milk, 0.2% Tween 20, and 1× PBS for 30 min at room temperature. Primary antibodies were diluted in 1% dried milk, 0.2% Tween 20, and 1× PBS (dilutions are indicated in Antibodies section). Incubations were performed overnight at 4°C in a humidified chamber. Slides were then washed three times for 5 min in the antibodies dilution buffer. Secondary antibodies (anti–rabbit cross-linked to Alexa 488 and/or anti–mouse cross-linked to Alexa 546 [Invitrogen]) were diluted at 1:500 in the same buffer as the primary antibodies and incubated for 30 min at 37°C. Washes were performed as for primary antibodies. DNA was counterstained by DAPI, and slides were mounted in MOWIOL. The protocol of IH experiments is described in detail by . e g e r m i n a l c e l l s f r o m o n e m o u s e t e s t i s w e r e o b t a i n e d b y d i l a c e r a t i n g t h e s e m i n i f e r o u s t u b u l e s i n ∼ 5 0 0 μ l D M E / H a m F 1 2 m e d i u m ( 1 : 1 ) , w a s h e d b y c e n t r i f u g a t i o n a t 1 , 0 0 0 r p m f o r 1 0 m i n , r e s u s p e n d e d , i n c u b a t e d f o r 1 0 m i n i n 1 % s o d i u m c i t r a t e a t r o o m t e m p e r a t u r e , a n d c e n t r i f u g e d a g a i n . T h e g e r m i n a l c e l l s o f t h e p e l l e t w e r e c a r e f u l l y d i s s o c i a t e d a n d f i x e d t w i c e i n m e t h a n o l / a c e t i c a c i d ( 3 : 1 ) f o r 1 0 m i n a t r o o m t e m p e r a t u r e a n d t h e n s p r e a d o n t o d r y , c l e a n s l i d e s . T h e s l i d e s w e r e a i r - d r i e d a n d k e p t a t r o o m t e m p e r a t u r e f o r u p t o 1 w k , u n t i l F I S H w a s d o n e . F o r t h e i m m u n o - F I S H e x p e r i m e n t s , I F w a s p e r f o r m e d a s u s u a l , t h e p o s i t i o n s o f t h e a c q u i r e d I F i m a g e s w e r e r e c o r d e d , a n d t h e s l i d e s w e r e w a s h e d i n 2 × S S C a t 3 7 ° C f o r 3 0 m i n , d e h y d r a t e d b y i m m e r s i n g i n a s e r i e s o f e t h a n o l ( 7 0 % / 9 0 % / 1 0 0 % ) , a n d a i r - d r i e d . The slides were denatured in 70% formamide/2× SSC for 12 min at 82°C (germinal cell preparation) or 1 min at 70°C (metaphase), dehydrated by passage through a cold ethanol series and air-dried (20× SSC: 175.3 g/l NaCl and 88.2 g/l sodium citrate in water, pH adjusted to 7 with NaOH). The DNA probes were labeled with either Biotin 11 dUTP or Digoxigenin dUTP by Nick Translation kit (Roche). A 10-μl sample of hybridization mix (50% formamide, 20% dextran sulfate, 1× SSC, and 1× SSPE; 20× SSPE: 174 g/l NaCl, 27.6 g/l NaHPOHO, and 7.2 g/l EDTA, pH adjusted to 7.4), containing 50–100 ng of each of the labeled probes, 10 μg sonicated salmon sperm DNA, and, when needed, 5–10 μg mouse cot DNA, was heated at 72°C for 10 min, preincubated at 37°C for 30 min, and applied to each slide. The preparations were then placed in the dark, under sealed coverslips, for hybridization during 24–48 h in a moist chamber at 37°C. The coverslips were then carefully removed, and the slides were washed in 2× SSC for 2 min at 70°C and preincubated for 15 min in phosphate-buffered detergent (PBD; Qbiogene) at room temperature. The digoxigenin-labeled and biotin-labeled probes were detected, respectively, by a 15-min incubation at 37°C with anti–digoxigenin-rhodamine (1:100; Boehringer) or streptavidin–Alexa 488 (1:200; Invitrogen), diluted in PBD, and washed three times for 5 min in PBD at room temperature. The preparations were finally counterstained with 250 ng/ml DAPI in Vectashield (Vector Laboratories). The quantification of the enrichment of each probe in major satellite sequences was performed on FISH slides as follows. The quantification was performed on mouse metaphases hybridized with the small fragment, the nucleosomal fragment, or control probes in codetection with major satellites. For each probe, the fluorescence signals on each whole chromosome and on its corresponding pericentric region (i.e., the area defined by major satellite hybridization) were quantified using MetaMorph software (Molecular Devices). The “pericentric/whole chromosome” signal ratio for the major satellite probe was considered 100% enrichment. For each of the other probes, this ratio (pericentric/whole chromosome) was also determined and normalized to the value obtained for a major satellite probe. A microscope (Axiophot; Carl Zeiss MicroImaging, Inc.) coupled to a −40°C chilled charge-coupled device camera (Hamamatsu) was used for 2D image acquisitions. Images were acquired at room temperature with a 63× objective (1.4 NA). Confocal images were taken using a confocal laser scanning microscope (Carl Zeiss MicroImaging, Inc.) and quantified using MetaMorph. The images presented in the figures were processed using Photoshop (Adobe). Step 12–16 spermatids obtained from 10 testes, or 10 NIH 3T3 cells used as control, were lysed (15 min, in ice) in 150 μl buffer 1 (50 mM Tris, pH 7.4, 150 mM NaCl, 1% NP-40, 0.5% DOC, and 0.1% SDS) or buffer 2 (50 mM Tris, pH 7.4, 300 mM NaCl, 0.1% NP-40, 0.1% DOC, 1 mM DTT, and antiprotease cocktail [Complete EDTA Free; Roche]). After centrifugation (20,000 , 4°C, 10 min), the pellet was resuspended in 150 μl of the initial buffer. In the case of buffer 2, a short sonication (80 J) was performed to allow the suspension of chromatin fragments of ∼5,000 bp, and unlysed spermatids were eliminated by an additional centrifugation. MNase digestion was performed on the buffer 1 or 2 lysates by the addition of 150 μl MNase buffer (10 mM Tris, pH 7.5, 10 mM KCl, and 1 mM CaCl) and 15 U of micrococcal nuclease S7 (Roche), and incubation at 37°C for the indicated times. Reaction was stopped by 5 mM EDTA (final concentration). MNase-digested fractions were separated by ultracentrifugation (80,000 , 20°C, 7 h) on a 10–30% linear sucrose gradient (in 1 mM phosphate buffer, pH 7.4, 80 mM NaCl, 0.2 mM EDTA, and antiprotease cocktail). DNA analyses were performed on 10 μl of MNase-digested fractions by treatment with proteinase K followed by electrophoresis on a 2% agarose gel. Proteins of the collected fractions were analyzed Western blots. After separation by SDS-PAGE, discrete bands were excised from the Coomassie blue–stained gel. The in-gel digestion was performed as previously described (). Gel pieces were then sequentially extracted with 5% (vol/vol) formic acid solution, 50% acetonitrile, 5% (vol/vol) formic acid, and acetonitrile. After drying, the tryptic peptides were resuspended in 0.5% aqueous trifluoroacetic acid. The samples were injected into a CapLC nanoLC system (Waters) and first preconcentrated on a 300 μm × 5 mm precolumn (PepMap C18; Dionex). The peptides were then eluted onto a C18 column (75 μm × 150 mm; Dionex). The chromatographic separation used a gradient from solution A (2% acetonitrile, 98% water, and 0.1% formic acid) to solution B (80% acetonitrile, 20% water, and 0.08% formic acid) for >35 or 60 min at a flow rate of 200 nl/min. The LC system was directly coupled to a mass spectrometer (QTOF Ultima; Waters). MS and MS/MS data were acquired and processed automatically using MassLynx 4.0 software. Database searching was performed using the MASCOT 2.1 program. Two databases were used: a homemade list of well-known contaminants (keratines and trypsin) and an updated compilation of SwissProt and Trembl databases with specifying Mus as the species. For searching the Mus database, the variable modifications allowed were as follows: acetyl-lysine, -ter acetylation, dimethyl-lysine, methyl-lysine, protein -acetylation, methionine oxidation, serine and threonine phosphorylation, methionine sulphone, and cysteic acid. Because of the potential high frequency of basic amino acid clusters (e.g., for histone proteins), four missed cleavages were also allowed. Proteins, which were identified with at least two peptides, both showing a score >40, were validated without any manual validation. For proteins identified by only one peptide having a score >40, the peptide sequence was checked manually. Peptides with scores >20 and <40 were systematically checked and/or interpreted manually to confirm or cancel the MASCOT suggestion. Reverse transcription reactions were performed with the StrataScript First- Strand Synthesis system (Stratagene) using random hexamer primers. qPCR reactions were performed using Brilliant SYBR Green qPCR MasterMix on an Mx3005p cycler (Stratagene). cDNA from total testis was used for the standard curve, and data were normalized using Brdt cDNA abundance, as Northern blot attests the constant level of this mRNA in meiotic and postmeiotic cells (). The coding sequences of H2AL1, H2AL2, H2A, H2B, and TH2B were inserted into pCDNA3.1 (Invitrogen) modified by insertion of HA or Flag tags or into peGFP-C (BD Biosciences). Plasmids were cotransfected in Cos7 cells by the Fugene transfection system (Roche), and cells were collected 24 h after transfection. Coimmunoprecipitations were performed as described by . Preparation of oligonucleosomes and capture on hydroxyapatite were performed as described by . Fig. S1 shows codetection and quantification of H4Ac and H3K9me3 in elongating spermatids by immunofluorescence. Fig. S2 shows protocols used for the MNase digestion of condensed spermatids and analysis of the resulting release of chromatin-associated proteins. Fig. S3 shows fractionation and capture on hydroxyapatite of the MNase-resistant structures of condensed spermatids. Fig. S4 demonstrates quantification of FISH experiments, showing an enrichment of the two MNase-resistant structures of condensed spermatids in major satellite sequences. Online supplemental material is available at .

Understanding how microtubules (MTs) reorganize during the cell cycle to assemble into a bipolar spindle is a classic problem of cell biology. Mitotic and meiotic spindles are highly dynamic structures, which assemble around chromosomes or sister chromatids and distribute them into each daughter cell. Errors in spindle assembly lead to severe DNA damage and aneuploidies, responsible for various forms of cancer. Therefore, it is essential that bipolar spindle assembly occurs correctly. Two pathways cooperate to assemble bipolar spindles. One pathway involves centrosomes, which generate astral MTs that continuously search for chromosomes. This is the “search-and-capture” model, which was postulated by . Accumulating evidence suggests that the small GTPase Ran is also a key player in the spatial control of spindle formation during the M phase (for reviews see ; ; ). Production of RanGTP depends on the activity of the regulator of chromosome condensation (RCC1), Ran's nucleotide exchange factor. RCC1 remains bound to chromosomes during the M phase. Thus, it was originally proposed that a high concentration of RanGTP around chromosomes acts as a local switch for spindle assembly (; ). This hypothesis has been validated for spindles assembled in vitro and for those assembled in somatic cells. In these systems, higher levels of RanGTP have been detected near chromosomes than in regions distant from chromatin, as indicated by fluorescence resonance energy transfer (FRET) and fluorescence lifetime imaging microscopy techniques (, ; ). Experiments in the cell-free system of initially demonstrated a central role for RanGTP in centrosome-dependent MT production and in chromatin-induced, centrosome- independent spindle assembly (; ; ; ). High levels of RanGTP stimulate the nucleating capacity of centrosomes but are not essential for basic centrosome nucleation activity. In contrast, chromatin-mediated MT formation depends entirely on the presence of RanGTP in the cell-free system (). More recently, siRNA experiments and microinjections in living cells of , , and humans demonstrated that the presence of RanGTP is required for essential functions in spindle formation in centrosome-containing systems (; ; ; ; ; ). Ran induces MT formation by releasing various spindle assembly factors, including NuMA, TPX2, and HURP, from the inhibitory effect of importins in the vicinity of chromosomes (; ; ; ; ; ). Many studies have focused on elucidating the function of RanGTP in spindle assembly in meiotic egg extracts. However, there is no in vivo evidence demonstrating the role of RanGTP in meiotic spindle formation in vertebrates. Meiotic spindle assembly in developing vertebrate oocytes occurs in the absence of centrioles (; ; ). During meiosis, two successive M phases occur without an intermediate S phase to produce haploid gametes. The first meiotic division is reductional with the segregation of homologous chromosomes. The second meiotic division is equational and resembles mitotic division. Cytostatic factor (CSF) then arrests vertebrate oocytes in metaphase II for many hours, until fertilization. In mouse and oocytes, MTs nucleate around condensing chromosomes and spindles self-organize in the presence of multiple MT organizing centers (MTOCs). In mouse oocytes, chromosomes gather quickly on a broad metaphase plate through interactions of their arms and MTs. Kinetochore–MT interactions are established at the end of the first meiotic M phase (MI) only (). Therefore, the oocyte model system is useful for the study of acentrosomal spindle assembly and for the assessment of the role of the Ran pathway in meiosis. We detected the accumulation of RanGTP around the chromosomes during all stages of mouse meiotic maturation with a previously described FRET-based probe for RanGTP-regulated release of importin β cargo molecules (). The overexpression of Ran mutants in mouse oocytes and the knock down of RCC1 in oocytes led to assembly of functional meiosis I spindles in the presence of excess or low RanGTP levels. In contrast, meiosis II spindle assembly was strictly dependent on RanGTP levels in both species. In mouse oocytes, we show that meiosis II progression also depended on RanGTP levels. We demonstrate that there is a mechanism that promotes spindle formation in the absence of both chromatin-induced MT production and centriole-based MTOCs. We followed spindle formation during meiotic maturation in mouse oocytes with high temporal resolution. Mouse oocytes are transparent and ideal for time-lapse microscopy. Meiosis resumes spontaneously in mouse oocytes when they are removed from the ovaries. We initially maintained the oocytes in prophase I at the germinal vesicle (GV) stage in dibutiryl cAMP (dbcAMP)–supplemented medium and injected them with RNA encoding GFP-tagged tubulin (tubulin-GFP; ). GV breakdown (GVBD) occurred ∼2 h after release from the dbcAMP-containing medium (). MTs were nucleated from asters that were distant from the chromosomes (). These randomly distributed asters reorganized around chromosomes into a bipolar barrel-shaped structure ∼2 h after GVBD. The MI spindle migrated along its long axis to the cortex of the oocyte () and then anaphase occurred, and the first polar body was extruded ∼9 h after GVBD. A metaphase II (MII) spindle rapidly assembled from remaining MTs around sister chromatids and remained stable during arrest (). We investigated the potential role of the Ran system during meiosis by assessing the levels of Ran and RCC1 by Western blot analysis. Endogenous Ran and RCC1 were present in immature (GV) and mature (MII) oocyte extracts from mouse and (). Although RCC1 levels were approximately two times higher in mouse MII oocyte extracts than in GV oocyte extracts (, right), this difference was much greater in oocytes. In these oocytes, the amount of Ran, RanGAP, and RanBP1 were similar in immature and mature eggs, but the amount of RCC1 increased during meiosis resumption (). We biochemically analyzed meiosis in the system and assayed the amounts of RCC1 during in vitro maturation of oocytes with high temporal resolution by quantitative Western blot analysis and a Cdk1 activity assay. Overall, RCC1 levels were seven times higher beginning with GVBD and reached maximal levels during MII (, right). A RanGTP gradient has been demonstrated to surround the chromatin in egg extracts and mitotic cells (, ; ). We used a mouse system to determine whether a RanGTP gradient was present in large living cells, for example, the mouse oocyte (∼100 μm in diameter). For this purpose, we used the previously described FRET probe, Rango, which has a high FRET signal when it is liberated from importin β by RanGTP (). An increase in the FRET signal (increase in I/I ratio) emitted by the Rango probe indirectly reports RanGTP levels: the presence of a free cargo gradient is indicative of a RanGTP gradient. The local increase in the I/I ratio () indicated the accumulation of free Rango and, hence, RanGTP near the chromosomes. The I/I ratio was consistently high in the same region, confirming the increase in FRET (unpublished data). The distribution of the RanGTP gradient in oocytes shows that RanGTP accumulated around chromosomes and decreased linearly with distance from the metaphase plate (). Thus, there was a broad (the size of an oocyte, ∼100 μm), shallow RanGTP gradient in mouse oocytes. We followed RanGTP-induced release of importin β in live oocytes and indirectly detected the accumulation of RanGTP around chromosomes during all steps of mouse meiotic maturation (). The kymograph shows that the local accumulation of RanGTP strictly followed the movement of chromosomes to the cortex during MI (). This is the first evidence of such a RanGTP gradient occurring during the time course of division in a living organism. We showed that Ran and RCC1 were present in mouse oocytes and that a broad gradient of RanGTP-induced release of importin β occurred during meiosis resumption. Next, we examined the role of Ran. For this purpose, we manipulated the levels of RanGTP by injecting two mutant forms of Ran into GV-stage oocytes. Ran has much weaker affinity for guanine nucleotide than does wild-type Ran, binds RCC1, and is a potent inhibitor of Ran's GDP–GTP exchange activity (). Thus, it inhibits the generation of RanGTP. Ran is unable to hydrolyze GTP and remains locked in the GTP-bound state (). We first checked that our mutants were expressed continuously and at similar levels during mouse meiotic maturation (Fig. S1, A, B, and C, available at ). We directly measured the efficiency of our mutants in ovo during meiotic maturation by injecting the Rango probe with each mutant form of Ran. The mutants blocked nuclear import in GV oocytes (), as described previously (). From GV to MII, there were lower levels of liberated Rango in oocytes receiving Ran than in controls, consistent with inhibition of RanGTP production (, compare 2 and 5 with 1 and 4). There were substantially higher levels of RanGTP in the cytoplasm of oocytes receiving Ran than in controls (, compare 3 and 6 with 1 and 4). In both cases, there was no detectable gradient of RanGTP in oocytes. We then injected RNA encoding tubulin-GFP with each form of Ran into mouse oocytes and followed spindle formation by video microscopy and by visualization of immunofluorescence after oocyte fixation. Meiosis resumption and first polar body extrusion had similar kinetics in control and injected oocytes (). This suggests that ectopic expression of the different forms of Ran did not affect the cell cycle per se. We observed the same succession of events with kinetics similar to those present in tubulin-GFP–injected oocytes () in oocytes injected with wild-type Ran (Ran; and Video 1, available at ). This shows that overexpression of Ran did not affect meiotic maturation and spindle assembly, which is consistent with a previous report (). Asters appeared in the cytoplasm after GVBD, not exclusively around the chromosomes in mouse oocytes injected with Ran, consistent with observations made in egg extracts. These asters fused to form a bipolar spindle that was abnormally long (; , A and B; and Videos 2 and 3, available at ). 7 h after GVBD, spindles with a similar size as those in control oocytes (47% were within the range of the standard deviation for control spindles) or longer (averaging 30% longer) were present in 53% of the oocytes (). The MI spindle that formed under these conditions migrated, and the pole closest to the cortex pushed into the cortex. Consistent with the formation of longer spindles in oocytes injected with Ran, 58% of the oocytes extruded a large first polar body (). Ran injected into mouse oocytes delayed the establishment of MI spindle bipolarity by ∼3 h in most oocytes (; , and C; and Video 4, available at ). During this time, MTs were unorganized around the chromosomes (). Approximately 5 h after GVBD, two poles formed and a spindle with very small poles assembled around chromosomes. Despite a 3-h delay in spindle formation, anaphase took place normally and the first polar body was extruded. We analyzed chromosome spreads from control and injected oocytes that had undergone MI to assess whether meiosis I spindles that formed in the presence of Ran or Ran were functional. MII-arrested oocytes had only 20 monovalent chromosomes in all cases (, compare 2, 3, and 4), showing that homologous chromosome segregation occurred normally. MII spindles formed after Ran and Ran expression in mouse oocytes were much more disorganized than MI spindles under these conditions (). The injection of Ran did not affect MII spindle organization ( and , A and B), as observed by live imaging. The injection of Ran and Ran induced similar spindle defects in MII. Oocytes either displayed an opened bipolar spindle connected to numerous cytoplasmic asters (50% of oocytes injected with Ran and 78% of Ran; , A [3 and 4] and B) or no organized spindle (30% of oocytes injected with Ran and 17% of oocytes injected with of Ran; , A [5 and 6] and B). The expression levels of our constructs were very similar in oocytes injected with Ran, Ran, and Ran (Fig. S1 D). We activated MII oocytes that had been injected with Ran and Ran with strontium, as described previously () to determine whether spindles assembled in the presence of Ran mutants were functional. Controls (noninjected or Ran-injected oocytes) and Ran-injected oocytes had similar levels of activation (). Nevertheless, MII to anaphase II transition was much slower (with a 2-h delay) in oocytes injected with Ran than in controls. This suggests that the spindle checkpoint was activated in these oocytes, potentially slowing the exit from CSF arrest. This observation is consistent with the observed MII spindle defects after injection of Ran. Furthermore, all oocytes injected with Ran had severe cytokinesis defects: they cleaved instead of extruding a second polar body ( and ), and membrane blebbing was often present in the cytokinetic furrow. We observed obvious segregation defects in 46% of these oocytes but did not see these defects in controls. Only one pronucleus was present in control oocytes, whereas there were lagging chromosomes in Ran-injected oocytes ( and ). Thus, overexpression of Ran strongly impaired the progression of meiosis II. Our findings in mouse oocytes suggest that inhibition of RanGTP production primarily perturbed spindle formation during meiosis II without completely inhibiting the establishment of meiosis I spindles and without affecting entry into the M phase. As shown in , RCC1 levels in immature and mature oocytes were very different in oocytes. Increasing RCC1 levels in oocytes throughout meiotic maturation suggest that RanGTP has an important function during MII. As RCC1 levels rise substantially during maturation of oocytes, preventing the synthesis of RCC1 during maturation kept RCC1 levels lower than in controls. Injection of antisense deoxyoligonucleotides prevented the rise in RCC1 levels, which occurs concomitantly with meiosis resumption (). The knock down of RCC1 did not modify the kinetics of meiosis resumption (not depicted) and meiosis I spindle formation (). In control and RCC1 knockdown samples, 75–90% of MI spindle structures looked normal (). Similar to our findings in mice, oocytes with low levels of RCC1 and, thus, low levels of RanGTP, progressed through MI, as observed by polar body extrusion (not depicted) and the drop in histone H1 kinase activity after GVBD in both control and RCC1-depleted oocytes (). Similar to meiosis in mouse oocytes, MII spindle formation was sensitive to perturbation of the GTP–GDP cycle of Ran. Although 67% of the spindles in sham-injected oocytes had normal bipolar organization and aligned chromosomes (76% in uninjected controls), only 14% looked normal after RCC1 knockdown, and 86% of the structures were aberrant (monopolar MT structures, spindle-like structures without a correct antiparallel array of MTs around the chromosomes and without chromosome alignment) or appeared as chromatin aggregates with no MTs (). These spindle defects were not caused by disruption of Cdk1 activities, as shown by histone H1 kinase assays (). Our findings suggest that oocyte maturation requires a large increase in RCC1 levels and, thus, RanGTP production. Inhibition of RCC1 activity led to defects in MII, as in mice. Thus, the mechanisms of meiotic spindle assembly in mice and have similar principles: local RanGTP production is essential during MII but much more weakly influences spindle function during the reductive first meiotic cell division. #text Oocytes were collected from 11-wk-old OF1 female mice, cultured, and microinjected as described previously (). Oocytes were maintained at the GV stage in M2 medium supplemented with 50 μg/ml dbcAMP (Sigma-Aldrich; ). In vitro synthesized RNAs were microinjected into the cytoplasm of GV oocytes. Injected oocytes were kept in M2 + BSA medium supplemented with dbcAMP for 3 h to allow overexpression of the corresponding protein. The resumption of meiotic maturation (GVBD) was triggered upon release of the oocytes into a drug-free medium. MII oocytes were released from CSF arrest using 10 mM strontium in Ca/free M2 medium as described by . Collagenase-treated stage VI oocytes were incubated in 5 μg/ml progesterone (Sigma-Aldrich) containing Barth buffer. Oocytes were checked every 10 min for GVBD spot appearance. Oocytes with appearing GVBD spots were collected and further incubated for the indicated times. Oocytes used for immunoblotting and H1-kinase assay were frozen in liquid nitrogen and for immunofluorescence staining fixed in 20% DMSO and 80% methanol. pQE32-cRan, pQE32-hRan, and pQE32-hRan were a gift from D. Görlich (Zentrum für Molekulare Biologie der Universität Heidelberg, Heidelberg, Germany). The coding sequences of canine Ran and human Ran and Ran were subcloned by PCR at EcoRI–NotI sites of the pRN3-myc2 vector. The pRN3-β5-tubulin-GFP plasmid has been described (). The Rango probe was subcloned into pRN3. The pRN3-histone RFP has been described (). The in vitro synthesis of capped RNA was performed as described previously (). Immunofluorescence of mouse oocytes was performed as described previously (). For MTs, we used a rat monoclonal antibody against tyrosinated α-tubulin (YL1/2). γ-Tubulin was visualized using a rabbit polyclonal anti–γ-tubulin antibody (1:1,000; Abcam). Samples were observed with a confocal microscope (TCS-SP; Leica) using a Plan APO 40×/1.25. Z series were performed with one image per micrometer, and a maximum projection of all Z planes is shown. oocytes were processed essentially as described previously (). In detail, DMSO/methanol permeabilized oocytes were fixed in methanol overnight at −20°C, rehydrated in PBS, and blocked with 5% BSA in PBS. Oocytes were then incubated with monoclonal anti–α-tubulin antibody (Clone DM 1A; Sigma-Aldrich), washed, and incubated with Cy3-labeled goat anti–mouse antibody (Jackson ImmunoResearch Laboratories). Oocytes were washed again with PBS, transferred into 0.5× TBS buffer (12.5 mM Tris, pH 7.2, and 60 mM NaCl), incubated with 5 μM Cytox green (Invitrogen), and washed with 0.5× TBS. Oocytes were dehydrated in methanol again and transferred into Murray's solution (benzylalcohole/benzylbenzoate, 1:2). After clarification, oocytes were cut within the GVBD spot and placed on depression microscope slides. Images were taken using a confocal microscope (TCS-NT; Leica). Immunoblotting of mouse oocytes was performed as described previously (). oocytes were lyzed in H1 kinase buffer and centrifuged at 12,000 for 15 min. Amounts corresponding to two thirds of the total lysate of one oocyte were used for immunoblotting. The supernatant was analyzed by immunoblotting. For mouse oocytes, Ran was recognized using a monoclonal antibody from BD Biosciences. For oocytes, rabbit polyclonal anti-human RCC1 and Ran were described previously (), polyclonal antibody against RanGAP was a gift from T. Walther and I. Mattaj (European Molecular Biology Laboratory, Heidelberg, Germany), and rabbit polyclonal antibody against RanBP1 was a gift from M. Dasso (National Institutes of Health, Bethesda, MD). A rabbit polyclonal antibody against PP1G was used as control in . It was generated using recombinant full-length PP1G as antigen in accordance with standard procedures. For detection and quantification of secondary antibody signals on Western blots, an Odyssey system (Li-Cor) was used. Chromosome preparations were performed as described previously (). Chromosomes were stained using 5 μg/ml propidium iodide in distilled water, and slides were mounted in Citifluor (UKC Chemlab). oocyte lysates prepared as described for immunoblotting were incubated with histone H1 (Sigma-Aldrich) and γ-[P]ATP as described previously (). Amounts corresponding to one third of the total lysate of one oocyte were used for H1 kinase assays. Antisense oligonucleotides were designed against the two most common alleles of RCC1 mRNA found in the EST database (Sanger Institute). An oligonucleotide pair against RCC1 mRNA regions −18 to +6, CTTTCATAGTGCCGTCTGTTCTACA, and −18 to +7, CTTTTCATAGTGCAGTCTGTTCTCA, was used to inhibit RCC1 expression. As a control, an oligonucleotide pair against RCC1 mRNA region −44 to −20 was used: GATTACAAAATAAACCGCGCTCGCC and GATTACAAAATTAACCGCGCTCAGC, which led only to a very mild reduction of RCC1 levels (RCC1 amount between 80 and 100% of untreated oocytes). 75 ng of an oligo pair in PBS (37.5 ng/oligo) were injected per oocyte before progesterone addition. Efficiency of RCC1 expression inhibition was routinely tested by immunoblotting. Fig. S1 shows that Ran mutants are stably expressed during meiotic maturation. Video 1 shows control mouse oocytes expressing tubulin-GFP together with Ran during meiotic maturation. Videos 2 and 3 show mouse oocytes expressing tubulin-GFP together with Ran during meiotic maturation. Video 4 shows mouse oocytes expressing tubulin-GFP together with Ran during meiotic maturation. Online supplemental material is available at .

Newly synthesized membrane proteins in are believed to be delivered by the signal recognition particle (SRP) to the SecYEG translocon cotranslationally (; for review see ). Thus, the SecYEG translocon is used both for the transport of secretory proteins from the cytosol to the periplasm () and for the integration of membrane proteins (for reviews see ; ). However, the role of SecYEG in the latter is only poorly understood. It is assumed that a signal-anchor sequence or a stop-transfer hydrophobic segment of a membrane protein is inserted into the translocon first and then moves laterally and is integrated stably into the lipid bilayer. In the processes of membrane protein integration/assembly, an integral membrane protein, YidC, also plays crucial roles (). Although integration of a class of membrane proteins is mediated directly by YidC, other membrane proteins are believed to be inserted first into the SecYEG channel and are transferred to YidC, which will facilitate final anchoring into the lipid phase () or folding into physiologic conformations (). The SecDF–YajC complex has large periplasmic domains and interacts with both SecYEG and YidC (; ). It is thought to mediate the interplay between SecYEG and YidC. According to the crystal structure of the SecYEβ translocon from an archaea (), an hourglass-shaped polypeptide-conducting channel is formed within the SecY subunit of the heterotrimer in which two halves of SecY are arranged in a pseudosymmetrical manner. It was proposed that the front side of the complex can be opened laterally for the exit of hydrophobic segments of newly synthesized proteins from the translocon to enter the hydrophobic environment. To maintain the impermeability of the membrane to ions, the conductance of the translocon is proposed to be blocked by a plug that covers the pore from the periplasmic side in the resting state and by a ring of hydrophobic residues in the pore that forms a seal around the translocating peptide chain. Despite this remarkable progress, several questions remain about translocon function in membrane protein integration. What are the pathways from the translocon into either the periplasm or the membrane for exported versus membrane proteins, respectively? What are the mechanisms by which movement of a membrane protein in the vertical direction is redirected to lateral transfer toward the lipid phase? Is the translocon sufficiently flexible to allow the reorientation of hydrophobic helices (; )? How and when do lipid interactions occur to generate the final topology ()? Whatever the mechanisms involved, it is likely that membrane protein assembly involves thermodynamic partitioning and is catalyzed by proteinaceous machinery (for review see ; ). Given the requirement for macromolecules, the system may be sensitive to environmental stresses and substrate overloading. Although cells have a regulatory mechanism to cope with defects in biogenesis, little is known about the system that deals with abnormalities in membrane proteins. possesses two extracytoplasmic stress response pathways called Cpx and σ, which up-regulate the expression of a class of proteins involved in folding and proteolysis (; ; ; ). Both systems were shown to sense abnormal proteins in either the outer membrane or the periplasmic space (). Additionally, we have shown that the Cpx/σ systems sense the accumulation of certain abnormal plasma membrane proteins. For example, the lack of FtsH, a membrane-bound protease, results in up-regulation of the Cpx stress response pathway, which is exaggerated by the overproduction of a membrane protein substrate of FtsH (). In this study, we describe a class of missense mutants that are defective in the correct assembly of membrane proteins. Notably, these mutations up-regulate the Cpx/σ pathways, as does the depletion of YidC. In vitro experiments reveal that membrane vesicles containing the mutated forms of SecY do not support correct folding of the lactose permease of (LacY), a polytopic membrane protein of known structure (; ). It is suggested that in addition to the known function of accepting newly synthesized membrane proteins, SecY also plays a role in their folding into the native conformation. We showed previously that the accumulation of certain classes of membrane proteins under the conditions of compromised activities of quality control proteases results in up-regulation of the Cpx extracytoplasmic stress response (). We reasoned that the stress response mechanism can sense some abnormal states of plasma membrane proteins. To further explore this notion, we examined whether extracytoplasmic stress response is induced when a cellular factor for membrane protein assembly/folding is functionally impaired. For this purpose, we used the arabinose-controlled expression system of YidC (). When arabinose was removed from the medium, YidC abundance in the engineered strain decreased with time (, lanes 5–8). In parallel to this decrease, Cpx pathway gene expression elevated up to about fourfold (, closed circles). Another assay using a reporter of the σ pathway showed that YidC-deficient conditions also up-regulate this stress response pathway (, closed circles). A previous study indicates that YidC depletion results in the defective folding of a multipath membrane protein, LacY (). It is also known that YidC is associated with the SecYEG translocon via the SecDF–YajC complex (; ). We found that the mutation (), which decreases the expression level of the operon (Chiba, K., personal communication), also induces the Cpx stress response markedly (, third bar). In parallel, we observed that this mutation considerably impaired the insertion of a model membrane protein, MalF–PSBT(J) (1.3 S subunit of biotin transcarboxylase (J); , lane 3; see the next section and Materials and methods for the significance of this indication of insertion defect). From these results, we surmise that the YidC and SecD deficiencies result in the generation of some aberrant forms of membrane proteins, which, in turn, activate the extracytoplasmic stress response mechanisms. The roles of YidC and SecD in membrane protein biogenesis are probably at stages after the initial targeting of newly synthesized membrane proteins. Defects at the targeting steps caused by an SRP mutation is known to induce the σ-dependent cytosolic stress response, presumably by producing uninserted membrane proteins in the cytosolic part of the cell (; ). To examine the effects of a defect in the SRP-targeting factor on the envelope stress response, we used the (Ts) mutation (). It was found that these mutant cells contained lower than normal levels of the Cpx-controlled LacZ activity (). Thus, this mutation in the targeting factor does not up-regulate the Cpx extracytoplasmic stress response. We also examined the consequence of a simple reduction of the cellular abundance of the translocon. The () mutation (), which decreases the expression level of SecY by ∼70% (, lane 4), did not activate the Cpx pathway at all (, fourth bar), although a model membrane protein, MalF-PSBT(J), exhibited a considerable insertion defect in this mutant (, lane 4). Also, the mutation, which decreases expression () and consequently leads to rapid degradation of the uncomplexed fraction of SecY (, lane 2; ), impaired insertion of the model membrane protein (, lane 2) but activated the stress response only insignificantly (, second column). These results indicate that the lack of insertion process itself does not generate a membrane stress. Collectively, we suggest that membrane protein folding/assembly defects that occur within the plasma membrane after the targeting event lead to the up-regulated envelope stress responses. Membrane protein biogenesis is accomplished by a series of sequential steps: translation, targeting, translocation via the translocon, lateral exit from the translocon, folding, assembly, and stable integration into the membrane. Although previous genetic and biochemical studies revealed cellular mechanisms that mediate these steps (for reviews see ; ), these studies primarily addressed whether a particular model membrane protein was inserted into the membrane with the proper topology of the extracytoplasmic domain. In most cases, defects in the pathway were detected only by the accumulation of uninserted molecules. In this study, we use the term type 1 defect to describe this phenomenon (Fig. S1, available at ). A defect in subsequent steps can occlude the translocon or interfere with earlier steps in the pathway (i.e., by backup). As the biogenesis pathway is constantly loaded with newly synthesized proteins, the accumulation of uninserted molecules can result not only when the primary defect is at the targeting step but also when the defect is at a later stage (Fig. S1). In other words, a defect in any step of the pathway can appear to be type 1. From the results presented in the preceding section, we suggest that the induction of envelope stress responses can be used as a specific indicator of postinsertion defects such as folding and assembly. We use the term type 2 defect for this phenomenon. Although a membrane stress response may suggest a postinsertion defect, a more direct demonstration of the folding/assembly failure is required. To demonstrate this point, we use the system of , ), in which in vitro translation of LacY is coupled to insertion into the membrane and folding into the tertiary structure, which is recognized by conformation-specific mAbs. We use the term type 3 defect to describe a failure that is detectable by this assay (Fig. S1). It should be noted that the in vitro system would suffer minimally from the backup phenomena because it uses very minute amounts of radiolabeled membrane protein. In the following analyses of mutations, we used the MalF-PSBT(J) fusion protein (see Materials and methods; ) and the SecY-PhoA C6 fusion protein (Fig. S2; ) to reveal the type 1 defect, induction of the Cpx/σ stress response to identify the type 2 defect, and acquisition of conformational epitopes in LacY to identify the type 3 defect. We examined a series of mutants that were described previously () as well as additional mutants isolated in this study (see Materials and methods and ) with respect to their defects in protein export and in membrane protein insertion (type 1 defect). These mutants were originally isolated as those up-regulating either a secretion monitor reporter () or a membrane stress reporter (see Materials and methods) at 37 or 30°C. Whereas many of them are either cold sensitive or temperature sensitive for growth, we assessed their export/integration phenotypes at the growth-allowing temperature (37 or 30°C), at which their lesion had been recognized by the original screening procedures. The export of OmpA, as examined by pulse labeling and immunoprecipitation, was evidently retarded in the , , , and mutants (, lanes 2–5) but was much less pronouncedly retarded in the , , , , , and mutants (, lanes 6, 7, and 9–12). The export of maltose-binding protein showed dependence similar to that of OmpA (unpublished data). We assessed membrane protein insertion phenotypes of the mutants using MalF-PSBT(J) and SecY-PhoA. These two model membrane proteins gave consistent results, and results with the latter are presented in Fig. S2 (A and B; available at ). All of the mutant strains were normal in biotinylating a control fusion protein (MalF-PSBT(K)) that had the PSBT domain on the cytosolic side (unpublished data). , , or cells (, top; first to third lanes), in which the fusion protein itself, as visualized by anti-MalF immunoblotting, accumulated normally (, bottom). Remarkably, the integration process appeared to have proceeded normally in the two strongly export-defective mutants, and . The remaining mutants examined all proved to be considerably defective in the integration process, as indicated by appreciable biotinylation of the fusion protein (, fourth to seventh lanes and ninth to twelfth lanes). The aforementioned results allow classification of the mutations into three groups. First, the and mutations are severely defective in export but are nearly normal in integration (see Fig. S2 B for the normal kinetics of SecY-PhoA processing in ). Second, , , , , , and are nearly normal in export but are considerably defective in integration. Finally, and are defective in both processes. We also examined OmpA export in the presence of simultaneously expressed MalF-PSBT. In this case, export was considerably retarded even in the and mutants (, bottom; lanes 6 and 7) but not in the wild-type cells (, lane 1). Overproduced MalF-PSBT seems to occlude even the relatively normal export pathway of the translocons having these alterations. As already discussed in the previous section, the accumulation of untranslocated MalF-PSBT observed in some of the mutants studied in may have been produced secondarily by the primary defect that lies in a later step of the biogenesis pathway. We then proceeded to screen the mutants by monitoring the type 2 defect. Thus, we combined each of the aforementioned mutations with a LacZ reporter, , of the Cpx stress response pathway (). LacZ (β-galactosidase) levels were found to be elevated appreciably in the , , , , and mutants (). The extent of this induction was approximately threefold for and , which was comparable with the extent of induction observed with the constitutive mutant (). In contrast, the Cpx pathway was not strongly up-regulated in the , , or mutant (not depicted for ) despite their integration defects. Stress response was not at all induced in the export integration mutant or (). The σ pathway was also up-regulated in but not in (unpublished data). Thus, a class of integration-impairing mutations activates the stress response mechanisms (). To study the type 3 defect, we used LacY, a polytopic membrane protein of known 3D structure (). Inverted membrane vesicles (IMVs) were prepared from cells in which the stress response induction was not markedly observed ( , , , and ) as well as from cells in which the stress response induction was evidently observed (, , and ). They were combined with the reaction system for in vitro transcription/translation of LacY to examine its insertion and folding. Initially, we observed that several mutant IMVs were far less active (Fig. S3, lanes 2–4; available at ) than wild-type IMVs (Fig. S3, lane 1). These IMVs were prepared without a urea wash. We then washed the IMVs with 4 M urea (see Materials and methods) before the assay. Remarkably, the urea wash activated the mutant IMVs, all of which exhibited LacY insertion activities that were comparable with wild-type IMVs treated in the same manner (Fig. S3, lanes 5–8). LacY synthesized in vitro remained associated with mutant IMVs even after a 5-M urea wash (Fig. S3) or alkali extraction (not depicted). It is likely that the urea wash cleared the mutated translocons by removing abnormal polypeptides so that the translocons were then capable of accepting LacY, making it possible to observe the initial integration of LacY into the mutant IMVs. To characterize LacY inserted in vitro, immunoprecipitation was performed with mAb 4B1, which binds specifically to a conformational epitope in periplasmic loop VII/VIII of LacY (). LacY synthesized in vitro was inserted into urea-washed IMVs, as judged by the lack of extraction by further treatment with urea ( [lanes 1–4, 9, 10, and 13–15] and B [lanes 1, 2, 5, and 6]). As shown in , mAb 4B1 binds to LacY inserted into the (, lane 6), (, lane 12), and (, lane 17) IMVs as well as LacY inserted into wild-type IMVs (, lanes 5, 11, and 16). In contrast, the mAb binds much less effectively to LacY inserted into , , or IMVs (, lanes 7, 8, and 18). These results strongly indicate that LacY inserted into the membrane of the , , or mutant does not efficiently fold into a normal conformation. Another mAb, 4B11, recognizes a discontinuous epitope that contains determinants from both loops VIII/IX and X/XI on the cytoplasmic surface of LacY (; ). As shown in (lanes 4 and 8), mAb 4B11 binds very weakly to LacY synthesized in vitro with and IMVs as compared with the mAb binding to LacY inserted into wild-type vesicles (, lanes 3 and 7). The IMVs also exhibited a lower than normal ability to support the LacY folding (, lane 12). Thus, LacY is misfolded in , , and membranes. There is an excellent correlation between the LacY folding defect and the stress response induction (). To address the in vivo significance of the aforementioned findings, we examined the stability of newly synthesized and membrane-integrated LacY-His by pulse-chase experiments in which membranes were isolated, washed with urea, and subjected to affinity isolation. , , , , , , and cells, which were labeled briefly with [S]methionine and chased with unlabeled methionine for 1, 2, and 4 min. As reported previously (), LacY was stably maintained in membranes of wild-type cells (, top). Its stability was not affected adversely in the , , or mutant (). In contrast, LacY was markedly destabilized in the mutant such that the radioactive band decreased to an almost undetectable level after a 4-min chase (). LacY synthesized in the and mutants was also unstable, although less so than in the mutant. These results are consistent with the notion that LacY molecules that were membrane integrated in the , , and mutants are degraded by a proteolytic mechanism of the cell, presumably as a result of a failure in folding. They agree well with the results obtained in the in vitro experiments. Collectively, we conclude that LacY, a paradigm for polytopic membrane proteins, is misfolded when membrane insertion is mediated by a translocon with the SecY129, SecY238, or SecY125 alteration. The results presented in this paper extend our previous conclusion that abnormal plasma membrane proteins can generate a stress that is sensed by the Cpx signal transduction mechanism (). Failure in membrane protein biogenesis can generate a stress that is sensed by the Cpx and the σ regulatory mechanisms that have generally been believed to sense protein abnormalities at the outer membrane–periplasmic regions of the cell. The Cpx pathway induction in mutants lacking phosphatidylethanolamine () may also be ascribed to the generation of some malfolded membrane proteins. The Cpx and σ regulatory systems contain both plasma membrane–integrated and periplasmic components (; ), and some of them could participate in this sensing. Molecular mechanisms of membrane stress recognition are emerging new subjects. In contrast to the YidC and SecDF defects, the SRP defects that occur at the targeting step do not elicit the envelope stress responses; instead, the σ-mediated cytosolic stress response is up-regulated under Ffh-compromised conditions (; ). It is noteworthy that the , , and mutants are defective in membrane protein insertion but are only weak or insignificant inducers of envelope stress responses. It is possible that some of these mutations primarily affect the targeting step of membrane protein assembly. Indeed, synthetic phenotypes with SRP alterations have been described previously for the mutation (; ) affecting a cytosolically exposed part of SecY (). Importantly, we have identified several amino acid substitutions in SecY that not only impair membrane protein biogenesis but also induce the extracytoplasmic stress responses. A formal possibility that the altered SecY protein itself is sensed as malfolded by the stress response mechanisms is unlikely for the following reasons. First, stress response induction is abolished by the introduction of wild-type in trans (unpublished data), indicating that a loss of function is responsible. Second, the , , , and mutants are nearly normal in the protein export ability, making it unlikely that the mutated SecY proteins have been considerably denatured. We propose that the mutations lead to membrane protein biogenesis failures, generating abnormally folded proteins within the plasma membrane. This stress response seems to be mechanistically distinct from the phage shock protein induction associated with some and defects (), as we confirmed the lack of PspA induction in some of the SecY alterations (unpublished data). Conclusive evidence that a class of alterations lead to inserted but abnormally folded membrane proteins came from our analyses of in vitro translation and integration of LacY. In vitro–synthesized LacY inserted competently into urea-washed membrane vesicles from the mutants in manners that were resistant to further treatment with urea, which removes LacY molecules only peripherally attached to the membrane (; ). Strikingly, however, there is a clear distinction between LacY molecules synthesized in the presence of the wild type, SecY39, SecY40, or SecY205 IMVs and those synthesized in the presence of the SecY125, SecY129, or SecY238 IMVs. Whereas the products of the former reactions react with the conformation-specific mAbs against LacY, the products of the latter reaction do so only weakly. These antibodies have been used to probe folding states of this integral membrane protein under a variety of conditions, including the presence or absence of YidC (; ). Thus, we conclude that a class of mutations alters the translocon in such a way that its function to effectively support the folding of membrane proteins, at least that of LacY, is compromised. Consistent with this notion, LacY is destabilized in these mutant cells. Unstable LacY is also produced in YidC-depleted cells (). Our results showing that such SecY defects are accompanied by the induction of membrane stresses suggest that the folding-assisting role of SecY is not limited toward LacY but is exerted toward several membrane proteins. Previously, showed that the class of mutations affects localization of the PhoA domain attached to a membrane protein. However, the significance of this observation with respect to the normal function of SecY in membrane protein assembly is obscure; for instance, this mode of PhoA translocation depends on SecB, which is considered to be unrelated to membrane protein integration. described that in yeast, some mutations affect the topological preference of a mutated transmembrane (TM) segment that naturally assumes dual orientations. Our present observations with SecY are distinct from the aforementioned precedents in that the mutations impair folding of a native, unaltered membrane protein, LacY. Different amino acid substitutions of SecY affected OmpA export and MalF-PSBT/SecY-PhoA insertion differentially (). Because the and alterations, which impair export but not integration, are suggested to affect aspects of SecY–SecA interactions (; , ), it is conceivable that the aforementioned differentiation comes from the differential utilization of SecA by exported versus membrane proteins (). However, it should be noted that insertion of both MalF-PSBT () and SecY-PhoA (unpublished data) depends on SecA. Intriguingly, we showed that export of a periplasmic protein, DsbA, exhibits a similar spectrum of dependence on different mutations as the membrane protein integration studied here (). As argue, higher hydrophobicity of the signal sequence of DsbA might direct it to the SRP-dependent pathway of translocation. Thus, membrane proteins and DsbA may use a targeting route different from that used by OmpA and maltose-binding protein, and the mode of targeting may determine the mode of translocon utilization. Recently, speculated on the basis of their electron microscopic data of SecYEG, which is in complex with the ribosome bearing a nascent membrane protein, that a working form of SecYEG is its front-to-front dimer. Furthermore, they proposed that the two SecYEG units are nonidentical such that one of them has a more hydrophobic channel than the other. Although this model should be examined critically by higher resolution structural studies, it offers a nice explanation for our finding of differential translocon utilization by different classes of exported or membrane proteins. We mapped positions affected by the five membrane protein folding–defective SecY alterations on the 3D structure of SecY () modeled on the basis of the published coordinates for the SecYEβ complex (). Although the mutations are located in diverse positions of SecY (Fig. S4, available at ), some comments are possible. First, Ala351 (the site of mutation) is located within the cytosolic loop 5 (C5), which is of implicated importance in the interaction with SecA or the ribosome. Second, Glu238 () and Cys385 () are located near the cytosolic end of TM6 (before C4) and TM9 (after C5), respectively, which face each other. Finally, and , impairing both export and integration, are mapped within regions of implicated importance in vertical and/or lateral gating of the translocon (). Ser76 (the site of the alteration) is located near the C-terminal end of the short plug helix called TM2a by . The bulky Phe residue here may compromise the proposed movement of this plug toward the periplasm to open the gate, which is in agreement with the late step translocation defect by (). Our results seem to suggest that this plug movement is also important for the effective clearance of the pore by the lateral release of a membrane protein. Pro84 () is located in the middle of TM2. The importance of Pro84 in membrane protein integration has been suggested from our reisolation of the same Pro84-Leu substitution by the new mutant screening designed for membrane stress induction (). The TM2, TM7, and TM3 regions of SecY were proposed to comprise the lateral gate, through which substrate signal peptide or a region destined to a TM configuration moves in or out (). Pro84 in TM2 could contribute to the gate-opening function possibly by perturbing the helix structure of this TM segment. How does SecY contribute to the correct folding of membrane proteins? One obvious function is to recognize a TM sequence and assist its movement out of the translocon. Additionally, it may have a role of reorienting a hydrophobic segment to generate its correct TM topology (; ). Finally, SecY may interact with SecDF and/or YidC to create a productive assembly line for a membrane protein to follow. Each of the membrane protein folding–defective mutations identified in this study may affect one of the aforementioned functions or still other functions of the SecY translocon, and their further analysis will be useful for our full understanding of the mechanisms of membrane protein integration and folding that occur in living cells. Minimal medium M9 () and complete nutrient medium L (containing 10 g bacto-tryptone, 5 g of yeast extract, and 5 g NaCl per liter; pH was adjusted to 7.2 by NaOH) were used. For growing plasmid-carrying strains, 50 μg/ml ampicillin, 20 μg/ml chloramphenicol, and/or 25 μg/ml tetracycline were added to the medium. The following K-12 strains were used in this study. MC4100 ( Δ[]; ) was used as a wild-type strain. ]), SP556 (MC4100, λRS88[′ ]), and NH192 (MC4100, [Cs]) were previously described by , , and ; note that NH192 was erroneously named as NH195 in this reference), respectively. AD202, an ∷ derivative of MC4100 (), was the parental strain for the following strains: TY0 ( ∷Tn), TY8 (∷Tn), TY24 (∷Tn), and AD208 (∷Tn; ; ; for TY strains). SKP1101 (MC4100, ∷/pSKPP10) and SKP1102 (MC4100, ∷/pSKPP11) carried (Ts) and the wild-type allele of on the plasmids, respectively (obtained from G. Phillips, Iowa State University, Ames, IA; ). strain JS71 were described previously (). MC4100 mutants having alleles, SH470 (), SH464 (), SH465 (), SH466 (), SH467 (), and SH468 (), were constructed by cotransduction of the respective mutations in the original mutants (; ) with ∷Tn. Note that the mutation () is identical to (). SH625 () was a ∷Tn transductant of MC4100 using THE521 () as a donor. control experiments. derived from pREP4 (QIAGEN). Plasmid pSTD643 carrying a transcriptional fusion was constructed by cloning the P3 promoter region of , which was amplified from the MC4100 chromosome using primers 5′-CCGATACTCTTTCCC TGCAATGGG-3′ and 5′-CGCCAACGAGGTTATCATTCACTG-3′ (EcoRI and BamHI recognition sequences are underlined), into pFZY1 () after digestion with these enzymes. Plasmid pKY4-3-3 carried the gene fusion under the promoter (). Plasmid pSH10 carried the ′- transcriptional fusion controlled by the Cpx regulatory pathway (). Plasmid pGJ78-J carried a PSBT domain fused to the second periplasmic domain of MalF under the promoter (). Localized mutagenesis of the chromosomal region around ∷Tn and screening for SecY-affected mutants were performed as described previously () except that SP556 (′- ) was used as a recipient in P1 transduction (). Blue-colored transductants were looked for on -agar palates containing 25 μg/ml tetracycline, 40 μg/ml X-gal (5-bromo-4-chloro-3-indolyl-β--galactopyranoside), and 0.5 mM phenylethyl-thiogalactoside at 30 or 37°C. They were purified and checked for growth phenotypes at 20, 30, 37, and 42°C. Plasmid pHMC5A ( ; ) was then introduced to examine whether the growth defect (if any) and color development on the X-gal plates were corrected. Those complemented were further tested for cotransduction of the phenotypes with ∷Tn when they were used as a donor in backcross P1 transduction into the original recipient. We established each mutant by saving a transductant with the mutant phenotypes at this stage, which was further subjected to sequence determination for the ORF. Mutants thus established are summarized below in a format of allele name given, nucleotide changes with a deduced amino acid change in parentheses, and growth phenotypes (Ts, Cs, and Ts/Cs indicate poor growth at 42°C, 20°C, and at both of these temperatures, respectively): , C251T (Pro84-Leu) and C3901T (silent), Cs; , G712A (Glu238-Lys), Cs; , G896 (Gly299-Glu), Ts; , G1051A (Ala351-Thr), Ts/Cs; and , G1207A (Gly403-Arg). The amino acid substitution by is identical to that by ; thus, data for were omitted. Established strains of the SP556 (′- ) background were named SH237 (), SH245 (), SH247 (), and SH253 (). We also used these mutants of the MC4100 background named SH469 (), SH472 (), and SH471 () as well as AD2397 (AD202; ). The MalF-PSBT(J) fusion protein contains a biotinylatable domain (PSBT) fused to the second periplasmic domain of MalF, which is biotinylated in vivo only when it resides in the cytosol for an extended period (). Thus, it reports an insertion defect by increased biotinylation. The control fusion protein MalF-PSBT(K) contains PSBT fused to a cytosolic domain of MalF and is readily biotinylated in any strain of having the normal biotinylating enzyme. Protein pulse labeling with [S]methionine, chase with unlabeled methionine, and immunoprecipitation were performed essentially as described previously (). Antisera against maltose-binding protein, OmpA, and the N-terminal region of SecY were described previously (). Anti-MalF and anti-YidC sera were provided by J. Beckwith (Harvard Medical School, Boston, MA) and R.E. Dalbey (Ohio State University, Columbus, OH), respectively. Radioactive proteins after SDS-PAGE were visualized and quantified by a phosphorimager (BAS1800; Fuji). Immunoblotting was performed as described previously () using SDS-solubilized cellular proteins equivalent to those of ∼10 cells. Biotinylated proteins were detected with streptavidin-conjugated HRP (GE Healthcare). Protein images were visualized and quantified by a lumino-imager (LAS1000; Fuji). β-galactosidase activity was assayed as described previously (); averaged values of at least three independent determinations were reported with error bars. IMVs for this assay were prepared as described previously () from cells cultured at 37°C. Where specified, they were washed with 4 M urea () before in vitro LacY insertion reactions. In vitro LacY transcription, translation, and insertion were performed at 30°C for 30 min in the presence of IMVs as described previously (). Subsequently, membranes were collected by centrifugation through a 50% sucrose cushion and washed with 4 M urea to obtain membrane-inserted LacY. The reactivity of LacY with conformation-specific mAb 4B1 or 4B11 was assessed by a previously described immunoprecipitation assay () with minor modifications: reaction mixtures were incubated at 4°C for 3 h with protein A beads, which were recovered after dilution with wash buffer and an additional incubation for 15 min, and washings were repeated five times. Cells transformed with both a LacY-His expression plasmid () and a plasmid (pSTD343) were precultured in LB-ampicillin–chloramphenicol (100 μg/ml and 34 μg/ml, respectively) medium at 37°C overnight, centrifuged down, washed with M9 minimal medium, and inoculated into 50 vol M9 medium supplemented with amino acids (except Met and Cys) and thiamine for further growth at 37°C. Cells were then induced for the transcription with 1 mM IPTG for 5 min at OD = 0.6–0.8 and were pulse labeled with [S]Tran (MP Biomedicals) for 1 min followed by chase with excess concentrations of unlabeled Met and Cys for indicated times. Samples were placed on ice, and cells were harvested by centrifugation. Membranes were prepared by sonification and subjected to a 5-M urea wash as described previously (; ). Radioactive LacY-His was isolated by metal affinity chromatography and visualized as described previously (). Fig. S1 provides graphical explanations of different types of membrane protein biogenesis defects. Fig. S2 shows the results of integration assays using the SecY-PhoA fusion protein. Fig. S3 shows LacY integration activities of IMVs prepared from mutants as well as their activation by urea washings. Fig. S4 shows locations of the membrane protein folding–impairing amino acid substitutions on the 3D structure of SecY. Online supplemental material is available at .

Cardiomyocytes are thought to terminally differentiate and withdraw from the cell cycle after birth. Therefore, cardiac injury causes permanent myocardial loss and results in cardiac dysfunction (; ). However, three research groups, including ours, have recently reported the isolation of cardiac stemlike cells based on the two distinct cell surface antigens, such as stem cell antigen 1 (Sca-1; ; ) and c-kit (). More recently, islet-1–positive cells have been reported to be a distinct population of cardiac progenitors in the postnatal heart, although the most of them do not contribute to the formation of the left ventricle and their existence in the adult heart is still unclear (; ). When these primitive cells were cultured under appropriate conditions, the cells expressed cardiac proteins (; ; ) and exhibited spontaneous beating (). When transplanted into injured hearts, the cells differentiated into cardiomyocytes (; ) and cardiac function was improved (). Although it is still unclear whether these primitive cells fit the precise definition of stem cells, e.g., self-renewal capacity and reconstitutive capability of the total organ, these findings suggest that the heart has intrinsic stemlike cells, which may participate in its regeneration. Side population (SP) cells are first identified as mouse hematopoietic stem cells with long-term multilineage reconstitution abilities based on their unique ability to efflux the DNA-binding dye Hoechst 33342 (, ). SP cells exist in a variety of organs, such as bone marrow, skeletal muscle, liver, brain, lung, skin, and heart (; ). reported that the ATP-binding cassette transporter, ABCG2 (also known as breast cancer resistance protein 1 [Bcrp1]), is a molecular determinant of this SP phenotype in hematopoietic stem cells. In mouse lung and rat liver, the SP phenotype has been reported to be largely determined by the expression of ABCG2 (; ). Among the tissue-derived SP cells, bone marrow and skeletal muscle SP cells have been well investigated. Bone marrow SP cells were first identified as a primitive population of hematopoietic stem cells (). The bone marrow–derived SP cells show long-term multilineage reconstitution in lethally irradiated recipients and form hematopoietic colonies in vitro (, ; ). have reported that bone marrow SP cells also differentiate into endothelial cells and cardiomyocytes in ischemic hearts. reported that transplantation of skeletal muscle SP cells into the irradiated mdx mouse results in the reconstitution of the hematopoietic compartment of the transplanted recipients and regeneration of donor-derived, dystrophin-positive muscle in the affected muscle. Skeletal muscle SP cells have the in vitro hematopoietic activity, and differentiate into skeletal myocytes when cocultured with satellite cell–derived myoblasts (). These results suggest that SP cells have features of somatic stem cells, and that cardiac SP cells (CSPs) may be a promising candidate for cardiac stem/progenitor cells. CSPs from postnatal hearts have been reported to differentiate into cardiomyocytes when cocultured with cardiomyocytes (; ; ). However, factors that induce differentiation of CSPs into cardiomyocytes have not been identified. Several growth or humoral factors have been reported to possess the ability to induce the differentiation of primitive cells into cardiomyocytes. During the development, bone morphogenetic proteins (BMPs) and fibroblast growth factors promote cardiogenesis in chick (; ). Both canonical and noncanonical Wnts play an important role in the cardiac differentiation (; ; ). Oxytocin (OT) and dynorphin B induce differentiation of embryonic stem cells and P19 embryonal carcinoma cells into cardiomyocytes (; ; ). Besides growth or humoral factors, chemical compounds such as DMSO and 5′-azacytidine have been reported to promote the cardiomyocyte differentiation of embryonic or somatic stem cells (; ). These findings suggest that both extracellular signals and epigenetic modification are capable of turning the fate of stem cells to cardiomyocytes. Recently, have reported that c-kit–, MDR-1–, or Sca-1–positive cardiac stem cells migrate and proliferate in response to hepatocyte growth factor and insulin-like growth factor-1, respectively. However, it is still elusive whether CSPs, by responding to the ischemia-induced factors, move to the injured area of the heart and differentiate into cardiomyocytes. We first report that CSPs from postnatal rat hearts differentiate into cardiomyocytes both in vitro and in vivo. Both OT and trichostatin A (TSA) induced postnatal CSPs to differentiate into beating cardiomyocytes. After intravenous transplantation of CSPs into normal adult rats, CSPs migrated and homed in the interstitial space of myocardium. When CSPs were intravenously transplanted into the cryoinjured heart, the number of CSPs was significantly larger in the border area than in the remote or infarct area after transplantation. Furthermore, CSPs differentiated into cardiomyocytes, endothelial cells, or smooth muscle cells in the border area. These findings suggest that CSPs are resident cardiac stem cells, which can migrate and regenerate myocardium in response to the ischemia-induced factors. Fluorescent sorting analysis revealed that there were two populations of cells in fetal, neonatal, and adult rat hearts referred to the SP and the main population (MP) cells in bone marrow (). When the cells were incubated with 50 μM verapamil, which is an inhibitor of multidrug resistance (MDR) and MDR-like proteins, there was no SP, suggesting that rat hearts contain SP cells. The proportion of CSP in the total cardiac-derived cells was ∼4.0%, ∼2.0%, and 1.2% in fetal, neonatal, and adult hearts, respectively. In neonatal CSPs, ∼14% of the cells expressed CD45, ∼59% expressed CD29, and ∼8% expressed CD31 (). The percentage of CD31-positive cells was 13.1 ± 4.0% under the fluorescent microscope (). To examine whether CSPs were in a noncycling quiescent state, cardiac cells were stained with both Hoechst 33342 and Pyronin Y (PY). The percentage of cells in PY-negative G0 stage was significantly higher in CSPs (74.3 ± 1.4%) than in cardiac MP cells (CMP; 34.0 ± 2.6%; ). A comparable result was obtained from the bone marrow SP and MP cells (PY-negative G0 stage of bone marrow SP, 79.8 ± 3.1%; bone marrow MP, 41.7 ± 5.4%; ). This suggests that CSPs represent a quiescent stem cell population in the heart. We next examined whether cardiac SP cells express Bcrp1, the molecular determinant of the SP phenotype. RT-PCR analysis showed that the Bcrp1 gene was expressed in freshly isolated SP cells from neonatal rat hearts, as well as in those from mouse bone marrow, but not in MP cells of hearts and of bone marrow (). Immunostaining with anti-Bcrp1 antibody revealed that Bcrp1 protein was detected on the cell surface of CSP, as well as bone marrow SP cells, but not in the MP cells (). In neonatal rat hearts, most Bcrp1-positive cells (∼95.4%) were CD31 (, a [coronary artery] and b [capillary]), but there were some CD31-negative/Bcrp1-positive cells (, arrowheads). Most of the CD31-negative/Bcrp1-positive cells (94.3 ± 9.8%) existed in the perivascular area (, arrowheads). There were also a few CD31-negative/Bcrp1-positive cells in the interstitial space (5.6% ± 9.8%; , arrowheads) between cardiomyocytes, which were stained with sarcomeric α-actinin (SA; , arrows) and distant from CD31-positive vessels (, arrows). There were no significant differences in the percentage of CD31-negative/Bcrp1-positive cells per total Bcrp1-positive cells among apex, mid, and base of left ventricles (), and also among chambers (i.e., atrium, left, and right ventricles; ). It has been reported that N-cadherin, CD29, and β1 integrin mediate the adhesion of stem cells to specialized mesenchymal cells and extracellular matrix in the niche environment (; ). Bcrp1-positive cells in the interstitial space coexpressed CD29 and N-cadherin around the surface of the cells (, arrowheads). At the junction of Bcrp1-positive cells and the neighboring cell, abundant coexpression of CD29 and N-cadherin was observed (, arrowheads). The perivascular Bcrp1-positive cells, which were localized adjacent to the smooth muscle cell actin (SMA)–positive cells, coexpressed CD29 (, arrowheads). These findings suggest that Bcrp1-positive cardiac stem or progenitor cells were localized in the specialized area of the myocardium, which may be similar to the stem cell niche in other organs, such as hematopoietic and gonad systems (; ; ). Isolated CSPs attached to the gelatin-coated dishes by 24 h with medium containing FBS. To induce differentiation into cardiomyocytes, we cultured CSPs with various growth factors, such as BMP2, BMP4, and OT, or on the feeder layers of the mesenchymal cells. Only treatment with OT was able to induce CSP into beating cardiomyocytes. After 2 d of treatment with OT, CSPs started to show various cell shapes (). 10 d after treatment, the attached cells started to proliferate, and elongated spindle-shaped cells became predominant (). 3 wk after treatment, some clusters of beating cells were recognized among flattened cells (, c and Video 1, available at ). Next, we examined whether methylation inhibitors or histone deacetylase inhibitors induced cardiac differentiation of CSPs. The treatment with TSA, but not with 5-azacytidine, induced CSPs into beating cardiomyocytes. The morphology of the CSPs treated with TSA was similar to that of CSPs treated with OT, although proliferation of the elongated spindle-shaped cells was less observed in the treatment with TSA compared with OT at ∼10 d (). 3 wk after the treatment, some clusters of beating cells were recognized among flattened cells (, and Video 2). There were no differences between OT- and TSA-induced cardiomyocytes in regard to the percentage of beating cells (OT, 0.27 ± 0.2%; TSA, 0.50 ± 0.21%; , shaded bar) and the percentage of SA-positive cells (OT, 3.8 ± 3.8%; TSA, 5.5 ± 3.9%; , open bar). Low magnification immunofluorescent images of SA and nuclear DNA of OT- and TSA-induced cardiomyocytes are shown in (a [OT] and b [TSA]). Fine striation was observed in OT- and TSA-induced cardiomyocytes (Fig. S2 A, a–d). Nontreated SP cells never exhibited spindle-shaped morphology or beating, and MP cells treated with OT or TSA detached from culture dishes within 1 wk. Next, we examined the gene expression of cardiac transcription factors and contractile proteins in CSPs by RT-PCR. Before treatment with OT or TSA, none of these cardiac genes were expressed (, P). Three weeks after treatment with OT or TSA, cardiac transcription factors, including , , and (), and contractile proteins, such as β–myosin heavy chain and (), were expressed (, OT and TSA). Treatment with 100 nM OT antagonist (OTA; [d(CH2)5-1,Tyr(Me)-2,Thr-4,Orn-8,Tyr-NH2-9] vasotocin) completely inhibited OT-induced expression of cardiac genes (, OT+OTA), indicating that OT induced cardiomyocyte differentiation through authentic OT receptors. Cardiac gene expression was not observed in cells cultured with vehicle (, V). To examine the expression of cardiac proteins, the CSPs treated with OT or TSA were stained with specific antibodies against cardiac proteins. The cells treated with OT or TSA expressed GATA4 (), atrial natriuretic factor (ANF; ), cardiac troponin T (cTnT; ), MLC2v (), and SA (). Notably, staining of each contractile protein showed a fine striated pattern, suggesting that treatments with OT or TSA induced differentiation of CSPs into mature cardiomyocytes. It has been reported that SP cells from skeletal muscle and bone marrow differentiate into various types of cells, such as adipocytes, endothelial cells, and skeletal muscle and cardiac myocytes (; ; ). To determine whether CSPs from the heart have multipotency of differentiation, we examined whether these cells could differentiate into cells other than cardiomyocytes. When CSPs were cultured with osteogenic inducers, including β-glycerophosphate, dexamethasone, and ascorbic acid-2 phosphate, some SP cells stained positive with alkaline phosphatase, which is one of the early markers of osteocytes (Fig. S1 A, a, available at ). RT-PCR analysis revealed that expression of alkaline phosphatase gene was induced in cardiac SP cells after treatment with osteogenic inducers (Fig. S1 A, b, lane O). On the other hand, cardiac SP cells treated with OT or TSA did not express alkaline phosphatase (Fig. S1 A, b, lanes OT and T). When cardiac SP cells were cultured in adipogenic induction with MDI-I mixture for 20 d, some SP cells showed cytoplasmic accumulation of oil droplets stained with Oil Red O, indicating that CSPs differentiated into adipocytes (Fig. S1 A, c). When GFP CSPs were transplanted into the normal rat via the tail vein, GFP CSPs were distributed over the various organs, such as lung (), spleen (), liver (), skeletal muscle (), bone marrow (), and heart (). In the lung and spleen, there were less GFP cells at 12 wk than at 1 wk after transplantation (12 wk/1 wk ratio; 0.19 for lung and 0.67 for spleen; ). On the contrary, in the liver, skeletal muscle, and heart, more GFP cells existed at 12 wk than at 1 wk after transplantation (12 wk/1 wk ratio; 1.63 for liver, 2.0 for skeletal muscle, and 3.0 for heart; ). 4 wk after transplantation, some GFP CSPs in the liver expressed albumin (). In skeletal muscle, GFP CSPs had multiple nuclei and expressed desmin (). However, there were no GFP CSPs positive for CD45 in the bone marrow (). GFP CSPs in the heart expressed CD29 () and were localized in the interstitial space of myocardium, which was delineated by collagen type IV (). In the normal heart, transplanted GFP CSPs did not express cTnT (not depicted). Next, we examined whether myocardial injury facilitates migration and homing of transplanted GFP CSP- and CMP-derived cells into the heart. 4 wk after transplantation of CMP, there were very few GFP CMPs in both normal and injured hearts (). In CSP-transplanted rat hearts, there were a few GFP CSPs, even in the normal myocardium (, arrowheads), whereas many GFP CSPs existed in the injured heart (, arrowheads). Many more GFP CSPs existed in the cryoinjured heart (15.0 ± 6.2 per 10 cells; = 3) in comparison with the normal heart (1.3 ± 1.0 per 10 cells; = 3; ) 4 wk after transplantation. There was no substantial difference in the number of GFP CMPs between normal (0.3 ± 0.6 per 10 cells; = 3) and cryoinjured heart (0.7 ± 0.6 per 10 cells; = 3; ). GFP CSPs were more abundant in the border zone of injured hearts (12.5 ± 2.5% of total cells) than in the normal (4.8 ± 1.4%) or injured (5.2 ± 1.7%) area (). Some GFP CSPs in the border and injured area expressed cTnT (), vimentin (), von Willebrand factor (vWF; ), and calponin (). The percentage of cTnT-positive GFP cells in the total GFP cells was 4.4%, vimentin-positive GFP cells 33%, vWF-positive GFP cells 6.7%, and calponin-positive GFP cells 29% (). The SA-actinin–positive GFP cells (8.6%; = 56) showed fine striated sarcomere structure. The majority of SA-positive CSPs were small cells without organized sarcomere structure (Fig. S2 B, a, arrowheads), suggesting that these cells remain in the stage of immature cardiomyocytes or cardiac precursor cells. There were some well-differentiated cardiomyocytes (Fig. S2 B, b, c, and d, arrowheads). Among the SA-positive GFP cells we examined, only a few small cells contained multi-nuclei without striation (Fig. S2 B, e and f, arrowheads). Small cell size and premature structure of this multinucleated cell suggest that the cell of multinuclei comes from mitosis rather than cell fusion. CSPs in the normal area of the injured heart did not express any of the aforementioned markers. These findings suggest that the injured myocardium recruits circulating CSPs, but not CMP, to the heart and stimulates the migration of CSPs toward the injured area. Furthermore, some environmental cues from the injured heart induce the differentiation of CSPs into cardiomyocytes, fibroblasts, endothelial cells, and smooth muscle cells. The novelty of our findings can be summarized as follows. First, we showed that a single factor, OT or TSA, can induce differentiation of CSPs into beating cardiomyocytes, which is quite different from the findings of previous studies (; ; ; ). Previous studies used coculture method to induce differentiation. Because OT is a physiological hormone, our findings may lead to identification of the intrinsic signals for cardiomyocyte differentiation. Second, we showed the precise location and distribution of CSPs in the heart. We distinguished CSPs from endothelial cells by immunohistochemical methods and clearly demonstrated the specific location of CSPs. Third, we demonstrated the expression of CD29 and N-cadherin on the cell surface of CSPs, suggesting that CSPs may be regulated in the niche in the heart. Fourth, we first demonstrated a sequential event of migration and homing of CSPs in the injured heart. There was no report concerning the in vivo dynamics of CSPs, and our findings suggest that the injured heart secrets some factors that recruit CSPs. Finally, we showed that transplanted CSPs followed the various steps of cardiomyogenesis, such as cardiac precursors and immature and mature cardiomyocytes. In addition, we showed that CSPs differentiate into multiple cell lineages other than cardiomyocytes, including fibroblasts, endothelial cells, and smooth muscle cells. Two groups have reported expression of cardiac proteins in CSPs when cocultured with primary cardiomyocytes () or with CMP (). Because both groups did not examine the contractile ability of SP-derived cells, it has remained unclear whether CSPs differentiate into mature cardiomyocytes. In addition, by the coculture method, it is difficult to distinguish if cardiomyocyte differentiation is accomplished by transdifferentiation or fusion. Recently, that CD31-negative CSPs also differentiate into functionally beating cardiomyocytes by coculture with adult rat cardiomyocytes. In this study, we first demonstrated that CSPs could differentiate into mature cardiomyocytes, which showed not only cardiac gene expression but also sarcomere formation and spontaneous beating, by single reagents such as OT and TSA. There were more CSPs in the rat heart of the early developmental stage. Fetal rat CSPs account for ∼4% of total isolated cells, 2% of neonatal rat CSPs, and 1.2% of adult rat CSPs. Our result of the developmental change of the CSP fraction is similar to the previously reported one in mouse hearts (). The percentage of CSPs from adult mouse was ∼0.24% in our experiments (unpublished data). The percentage varied from 0.02% to 2% in previous reports (; ; ; ). Although there may be a difference in the percentage of CSPs among the species, the cell surface markers of isolated CSPs were variable among the reports. and reported that a large portion of isolated CSPs from adult mouse are CD31 positive. In this study, CD31-positive cells were only 7.6% in isolated CSPs from neonatal rats. Our immunohistochemical analysis indicated that most Bcrp1-positive cells in the heart are CD31-positive endothelial cells. The reason for these variations may be attributed to distinct isolation techniques and to the fact that most endothelial cells were lost during the step of cell isolation discussed in this study. Considering the conclusion of that CD31-negative CSPs represent a distinct cardiac progenitor cell population, our CSPs isolated from neonatal rats are a condensed population of cardiac progenitors. The ability to induce CSPs into the mature cardiomyocytes is comparable between OT and TSA (). There were only a few studies showing quantitative analysis of the frequency of monocultured CSP-derived cardiomyocytes. reported that ∼10% of CD31/Sca-1/CSP expressed disorganized α-actinin and troponin I, but they did not show the characteristic sarcomeric organization and spontaneous beating, suggesting immature cardiomyocytes. have reported that when CSP-derived cardiosphere was dissociated and cultured, 0.28% of the total cells differentiated into cardiomyocytes, which were positive for α-actinin and sarcomeric myosin (). In this study, ∼5% of CSPs differentiated into cardiomyocytes with fine sarcomere structures and spontaneous beating (Fig. S2 A, a–f). Therefore, both OT and TSA possess more powerful cardiogenic activity against CSPs than the previously reported methods. OT, a hypothalamic neuropeptide, induces uterine contraction and milk ejection. In recent years, however, functional OT receptors have been found in various organs, such as kidney, ovary, testis, thymus, heart, vascular endothelium, osteoclasts, myoblasts, pancreatic islet cells, adipocytes, and several types of cancer cells (). OT receptors and OT biosynthesis are detected in atria and ventricles of the rat heart, and OT is thought to be involved in ANF release from cardiomyocytes (; ). CSPs are a heterogenous population of the cells, including cardiac stem/progenitor cells, endothelial progenitor cells, and other unknown cells. When CSPs are treated with OT or TSA, mesenchymal-like cells were observed near cardiomyocytes. Presently, we do not have the evidence to indicate that OT receptors are expressed in cardiac stem/progenitor cells, but not in other cells. It has recently been reported that elevated OT and OT receptor protein levels in growing fetal hearts and OT receptor immunostaining were predominantly detected in cardiomyocytes and endothelial cells (). These observations suggest that OT acts on cardiomyogenesis, but it remains to be determined whether OT has direct effects on cardiac stem cells. We have recently reported that OT induces differentiation of adult cardiac Sca-1 cells into mature cardiomyocytes (). Because of the lack of Sca-1 in rats and the unavailability of decent antibodies against rat c-kit, we could not determine the relationship between CSPs and other cardiac stem cells populations, such as Sca-1 or c-kit cells. The expression of cardiac transcription factors was absent in freshly isolated CSPs. We performed semiquantitative RT-PCR, showing the expression levels of in CSPs were negligible (Fig. S1C, a and b). Therefore, CSPs may be more primitive stem or progenitor cells in comparison with cardiac Sca-1 cells, in which faint but substantial expressions of cardiac transcription factors were observed. Our findings suggest that the OT-mediated signaling may play a pivotal role in the differentiation of various cardiac stem cells into cardiomyocytes. Histone deacetylases (HDAC) catalyze the deacetylation from conserved lysine residues in the N-terminal tails of histones (). Silencing of genes has been shown to be accomplished by histone deacetylation, and inhibition of HDAC reverses the silencing effect. HDAC are critically involved in cell cycle regulation, cell proliferation, cancer development, and cell differentiation (; ). Recently, HDAC inhibitors such as TSA, valproic acid, and butyric acid have been reported to modulate cell type–specific gene expression. The lymphoid lineage-determining factor Ikaros is repressed under the circumstances with hypoacetylation of core histones at promoter sites, and this repression is relieved by TSA (). reported that valproic acid induces neural differentiation of adult hippocampal neural progenitors through the induction of neuroD. In this study, TSA induced de novo expressions of , , and , suggesting that acetylation of chromatin activates specific master genes, products of which promote the expression of a series of cardiac transcription factors. It remains to be determined what genes are activated and involved in cardiomyocyte differentiation by the treatment of TSA. SP cells are thought to be a population of quiescent stem cells, which reside in the niche of the organs and contribute to life-long maintenance or repair of the tissue (; ). Quiescence of CSPs was confirmed by PY staining. Stem cell niches play a pivotal role in controlling the self-renewal and differentiation of stem cells (for review see ). Niches consist of stem cells, niche stromal cells, and extracellular matrix, and the interaction between stem cells and the cellular microenvironments through adhesion molecules is important, as are paracrine factors. Bcrp1-positive cells in the heart coexpressed CD29 and N-cadherin on their cell surface and were located in the interstitial space and perivascular area. Although the niche stroma cells for cardiac Bcrp1-positive cells were not specified in this study, the fact that most Bcrp1-positive cells existed in the perivascular area suggests that pericytes or adventitial mesenchymal cells may be a component of the stem cell niches. During the preparation of this manuscript, reported that c-kit–positive cardiac stem cells and lineage-committed cells are clustered together, forming their niches in adult mouse heart. In their paper, α4β1 integrin–mediated adhesion to laminin and fibronectin, as well as E- and N-cadherin–mediated cell–cell communications are supposed to be the fundamental structure of the cardiac stem cell niches. Some groups have reported that the frequency of cardiac stem cell clusters, including MDR1-positive cells, is inversely related to the hemodynamic load sustained by the anatomical regions of the heart; they accumulate in the atria and apex and are less numerous at the base and mid portion of the left ventricle (). However, the frequency of CD31-negative/Bcrp1-positive cells in neonatal hearts did not show significant difference in the anatomical regions in this study. The reason for this discordant result may be that the left ventricle of neonatal hearts is under less hemodynamic load than that of adult hearts. Intravenously transplanted CSPs were trapped in the lung and spleen, but redistributed in heart, liver, and skeletal muscle. CSPs in the heart were localized in the basal membrane between the myocardium and expressed CD29 on their cell surface, suggesting that CSPs penetrate the fenestrated endothelium, migrate into the basal lamina, and reside along with cardiomyocytes. Although some CSPs in liver and skeletal muscle expressed tissue-specific proteins such as albumin and desmin, respectively, transplanted CSPs in the normal heart did not express cardiac contractile proteins. It has been reported that transplanted bone marrow cells fuse with hepatocytes and skeletal muscle and regenerate the tissues (; ; ; ). Therefore, highly fusogenic hepatocytes and myotubes may fuse with transplanted CSPs and express differentiated marker proteins, whereas CSPs homing to the heart may not fuse with cardiomyocytes, and thus maintain stem or progenitor status. Tissue damage, such as total body irradiation or chemotherapy, leads to secretion of chemokines and cytokines and facilitates hematopoietic stem cell migration and repopulation (). reported that skeletal muscle–derived stem cells home and migrate to the perivascular space of a damaged muscle of mice after intravenous transplantation, and that the molecules involved in this process are L-selectin and mucosal addressin cell adhesion molecule-1. CSPs distributed in lung, spleen, liver, and skeletal muscle, but did not home specifically to the normal heart tissue. However, CSPs infused into rats with cryoinjured hearts homed in the heart, suggesting that the factors inducing migration and homing of stem cells may be released from injured heart. Further studies are necessary to understand the molecular mechanisms of differentiation, expansion, and migration of cardiac stem cells. Neonatal Wistar rats and wild-type mice (C57BL/6) were purchased from Takasugi Experimental Animal Supply Co. LTD. Neonatal and adult GFP transgenic rats were purchased from Japan SLC, Inc. (). All protocols were approved by the Institutional Animal Care and Use Committee of Chiba University. Cardiomyocytes of neonatal rats were prepared as previously described (). Rabbit anti–mouse Bcrp1 antibody was provided by S. Takeda (National Center of Neurology and Psychiatry, Tokyo, Japan; ). Other antibodies used in these studies are listed in . Other reagents that are not specified were obtained from Sigma-Aldrich. Cardiac cells were resuspended at the density of 1.0 × 10 cells/ml in PBS with 3% FBS. The cells were incubated in 1 μg/ml Hoechst 33342 dye for 60 min at 37°C in the dark, with or without 50 μM verapamil. After the incubation, cells were analyzed for Hoechst 33342 dye efflux by EPICS ALTRA flow cytometric analysis (Beckman Coulter). Before analysis, 2 μg/ml of propidium iodide was added to distinguish live cells from dead cells. Hoechst 33342 dye was excited at 350 nm using UV laser. Fluorescent emission was detected through 450-nm BP (Hoechst blue) and 675-nm LP (Hoechst red) filters, respectively. Propidium iodide in cells was excited at 488 nm, and fluorescence emission was detected through a 610-nm BP filter. For cell surface marker analysis, the cells were incubated with phycoerythrin (PE)-conjugated anti-CD45 antibody, PE-conjugated anti-CD31 antibody, or FITC-conjugated anti-CD29 antibody for 10 min on ice and washed with PBS supplemented with 3% FBS. The procedures for mouse bone marrow SP cells and PY staining were previously described (; ). CSPs were cultured on gelatin-coated dishes with Iscove's Modified Dulbecco's Medium supplemented with 10% FBS. 24 h after seeding, the cells were treated with 10 pg/ml TSA or 100 nM of OT (both Sigma-Aldrich) for 72 h. Male Wistar rats were anesthetized with 50 mg/kg ketamine i.p. and xylidine (10 mg/kg, i.p.) and a 6-mm aluminum rod, which was cooled to −190°C by immersion in liquid nitrogen, applied to the left ventricular free wall to produce cryoinjury, after the tail vein injection of 3 × 10 CSPs or CMPs derived from neonatal GFP transgenic into syngenic wild-type adult rats (CSP transplantation, = 3; CMP transplantation, = 3). As control groups, normal rats, which were subjected to the injection of 3 × 10 of CSPs ( = 3) or CMP ( = 3) were prepared. 4 wk after injection, rats were killed and lung, spleen, liver, skeletal muscle, and heart were fixed according to the periodate-lysine-paraformaldehyde fixative methods and snap-frozen in liquid nitrogen. 6-μm cryostat sections of fresh-frozen or fixed rat heart were prepared. Fresh-frozen sections were fixed with 1% formaldehyde for 15 min at room temperature. Blocking and staining procedures were performed according to the protocol described in the previous paragraph. Confocal images were acquired at room temperature using a microscope (Radiance 2000; Bio-Rad Laboratories) with Plan Apo 60×/1.40 NA oil immersion objective (Nikon) and Laser Sharp 2000 confocal software (Bio-Rad Laboratories). For , , Fig. S1 A (a and c), and Fig. S3 A (a–f), Axioscop 2 Plus (Carl Zeiss MicroImaging, Inc.) with Plan-NEOFLUAR 100×/1.30 NA oil immersion and 40×/0.75 NA objectives (Carl Zeiss MicroImaging, Inc.). SP cells were isolated from cardiac cells using EPICS ALTRA flow cytometric sorting. Total RNA was obtained from SP cells, TSA-treated SP cells, and the neonatal rat heart by RNA-Bee reagent (TEL-TEST). RT-PCR was performed using 0.1 mg of total RNA. For semiquantitative analysis, reverse transcribed products were pooled and fivefold serial dilutions were used for PCR. PCR was performed in a reaction volume of 20 μl with 200 nM deoxynucleoside triphosphates, 500 nM each of sense and antisense primers, and 2.5 U/100 μl polymerase (Roche). Every PCR condition was confirmed to be within the linear range and within semiquantitative range for these specific genes and primer pairs. The primers used in this study and the PCR conditions are described in . To confirm that the obtained bands were not derived from contaminated genomic DNA, a negative experiment was done for each sample without reverse transcriptase before PCR. Amplified samples were electrophoresed on 2% agarose gels and stained with ethidium bromide. For semiquantitative RT-PCR analysis, PCR was performed on undiluted cDNA and on fivefold serial dilutions of cDNA, and the intensity of the ethidium bromide–stained bands was quantified using the Image program (Wayne Rasband, National Institutes of Health). Diluted pools showing the same intensity for β-actin were used for further PCR and quantification of Nkx-2.5 gene expression. The protocol for osteocyte- and adipocyte-induction was previously described (). Alkaline phosphatase staining (leukocyte alkaline phosphatase assay kit) was used to examine the differentiation of osteocytes. For detection of accumulated oil droplets, Oil Red O staining was performed followed by nuclear hematoxylin counterstaining. The significance of differences among mean values was determined by test. P values were corrected for multiple comparisons by the Bonferoni correction. The accepted level of significance was P < 0.05. Fig. S1 shows the osteogenic and adipogenic differentiation of CSPs. Fig. S2 shows the fine sarcomeric patterns of OT- and TSA-induced CSP-derived cardiomyocytes. Live images of beating cells were taken with an inverted microscope (Carl Zeiss MicroImaging, Inc.) equipped with chilled charge-coupled device camera (Hamamatsu) using I-O DATA Videorecorder software. Online supplemental material is available at .

Tight junctions, adherens junctions (AJs), and desmosomes promote adhesion between epithelial cells, initiate the assembly of the mechanical cytoskeleton linkage, and facilitate the formation of a polarized epithelial monolayer (). AJs initiate these processes and are essential for morphogenesis, wound healing, and the retention of cell polarity and tissue integrity (). In epithelia, AJ formation is mediated by the calcium-dependent homophilic binding of E-cadherin molecules on neighboring cells (). These interactions link adjacent cells and promote the nucleation of a cytoplasmic protein complex consisting of p120-, β-, and α-catenins, which bridges E-cadherin clusters and the actin cytoskeleton. The biological necessity of AJ proteins has been underscored by a high correlation between the malfunctioning of AJ proteins, E-cadherin in particular, and tumor metastasis (). During tumor progression, the E-cadherin gene can be functionally silenced or inactivated by distinct mechanisms (). In addition to transcriptional repression by SIP-1, δEF-1, Snail/Slug, E12/47, and Twist (), posttranslational regulation of E-cadherin stability modulates its activity. Precisely tuned exocytic and endocytic pathways control the amount of E-cadherin residing on the plasma membrane (PM) and are important for modulation of E-cadherin function and AJ assembly (). Recent evidence suggests that Rab11 (), p120-catenin, ARF6, tyrosine phosphorylation, and ubiquitination () all control the trafficking and assembly of E-cadherin in mammalian cells. Additionally, transport of E-cadherin is regulated by the composition of the cadherin–catenin complex as well as the vesicular trafficking machinery (), where multiple adaptor and signaling proteins orchestrate trafficking specificity and efficiency. Clathrin adaptor protein (AP) complexes are important in the sorting of cargoes containing dileucine or tyrosine-based sorting motifs (). In epithelial cells, AP1B is the unique isoform that mediates basolateral transport (; ). Although AP1B is closely related to the more ubiquitously expressed form of AP1, AP1A, it targets to a distinct membrane compartment defined as the recycling endosome (; ). Recently, it has been shown that this compartment is an intermediary in transport from the Golgi to the PM () and also functions in the recycling of internalized basolateral membrane proteins (; ). Phosphoinositides are key mediators of membrane trafficking (). Membrane assembly and cargo binding of AP2 are both dependent on binding to phosphatidylinositol-4, 5-bisphosphate (PI4,5P) via its α and μ subunits (; ). There is evidence that other AP complexes are also modulated by phosphoinositide lipid messengers (). In addition, PI4,5P regulates actin polymerization, focal adhesion assembly, and several components of the vesicular trafficking machinery (). However, the mechanism by which PI4,5P generation is regulated to mediate these trafficking events has not been defined. Recent studies have unveiled that the spatial targeting and temporal regulation of type I phosphatidylinositol phosphate kinases (PIPKIs) is a critical mechanism for PI4,5P generation (). Here we show that in epithelial cells PIPKIγ targets to AJs by a direct interaction with the E-cadherin dimer. PIPKIγ regulates E-cadherin trafficking by acting as a scaffold between E-cadherin and AP complexes. We also demonstrate that localized generation of PI4,5P via these complexes is necessary for E-cadherin transport and AJ formation. Upon examination of the basolateral membrane in polarized epithelial cells, we found that PIPKIγ colocalized with E-cadherin () but not with occludin (not depicted). PIPKIγ also presented in a cytosolic vesicular compartment and partially colocalized with E-cadherin at this site (, arrows). These regions of colocalizion were confirmed by constructing vertical sections of z-series images shown in , suggesting an interaction between PIPKIγ and a component of AJs. To examine this possibility, E-cadherin and PIPKIγ were immunoprecipitated. As shown in , PIPKIγ and E-cadherin associate in vivo, along with other cadherin-associated proteins, demonstrating that PIPKIγ associates with E-cadherin complexes. N-cadherin and VE-cadherin also associate with PIPKIγ (), suggesting that PIPKIγ associates with the classical cadherin complexes. PIPKIγ is predominantly expressed as two distinct splice variants, PIPKIγ635 and PIPKIγ661, which differ by a 26–amino acid C-terminal extension. HA-tagged PIPKIγ splice variants were expressed in human embryonic kidney 293 (HEK293) cells, and their association with the endogenous N-cadherin complex was analyzed. PIPKIγ635 and PIPKIγ661 both coimmunoprecipitated with N-cadherin indistinguishably (Fig. S1 A, available at ), indicating that this association does not depend on the PIPKIγ661 C-terminal extension. However, the endogenous PIPKIγ associated with E-cadherin was indistinguishable in apparent molecular weight from PIPKIγ661, which is the predominant splice variant expressed in these cells. To ascertain whether this association was direct, in vitro GST pull-down assays were performed using recombinant GST-tagged PIPKIγ and His-tagged E-cadherin cytoplasmic domain (ECDT). ECDT showed specific binding to both GST-PIPKIγ635 and GST-PIPKIγ661, but not to GST alone () or GST-PIPKIα (Fig. S1 B). E-cadherin molecules form lateral homodimers in vivo, and oligomer formation is critical for AJ assembly and stability (). Consequently, we examined if dimerization was important for association with PIPKIγ. A parallel dimeric ECDT was constructed by inserting a heptad repeat (HR) peptide sequence between the His tag and ECDT (; Fig. S1, C and D). The dimeric construct had a greater binding affinity for PIPKIγ compared with the monomeric protein (). This was not caused by the HR tag because an HR-fused integrin cytoplasmic domain did not bind PIPKIγ (; Fig. S1 E). Moreover, when expressed in cells, the Myc-tagged dimeric ECDT (Myc-HR-ECDT) selectively bound PIPKIγ and p120-catenin with ∼10-fold greater affinity compared with the monomer ( and Fig. S1 F). Consistent with a previous paper (), β-catenin bound the monomeric and dimeric E-cadherin C terminus with the same affinity. To further determine whether PIPKIγ binding to E-cadherin involves other AJ components, wild-type or mutated E-cadherin was expressed in HEK293 cells and assessed for endogenous PIPKIγ association (). Elimination of either the p120-catenin (ECDΔp120; EED to AAA) or β-catenin (ECDΔβctn; ECD847, deletion of the last 35 amino acids) binding sites had no effect on PIPKIγ association. A chimera of truncated E-cadherin (deletion of the last 70 amino acids) fused to a truncated α-catenin that lacks the β-catenin binding site () abrogated both β-catenin (not depicted) and PIPKIγ binding (). These results indicate that PIPKIγ binding to E-cadherin is independent of α-, β-, or p120-catenin and narrowed the PIPKIγ interaction region on E-cadherin to residues 837–847. To confirm this putative PIPKIγ binding site, the last 45 amino acids of E-cadherin were truncated (ECD836). This truncation resulted in ablation of both β-catenin and PIPKIγ binding (). These combined data demonstrate that PIPKIγ directly interacts within a region including amino acids 837–847 of E-cadherin that is a highly conserved domain in the type I classical cadherins (). To examine whether PIPKIγ modulates E-cadherin function through this direct interaction, we introduced wild-type or p120/β-catenin binding site–deleted Myc-HR-ECDT into MDCK cells to compete with endogenous E-cadherin for PIPKIγ binding (Fig. S2 A, available at ). Expression of dimeric ECDT that specifically binds PIPKIγ resulted in a loss of AJs identified by E-cadherin staining and a cytosolic accumulation of PIPKIγ in cells, indicating that this phenotype is likely caused by sequestration of PIPKIγ (Fig. S2 B). Overexpression of PIPKIγ661 was sufficient to rescue the loss-of-AJ phenotype induced by wild-type Myc-HR-ECDT expression (Fig. S2 C). These data establish that a specific interaction between PIPKIγ and E-cadherin plays a key role in E-cadherin function and appears to be a limiting factor in AJ formation. To further determine the functional role of PIPKIγ at AJs, we knocked down endogenous PIPKIγ expression using siRNAs. Although the cellular E-cadherin content was not changed (), loss of PIPKIγ caused a striking loss of E-cadherin from the PM with an apparent accumulation in a cytoplasmic compartment (). Upon loss of PM E-cadherin, the cells spread () and underwent a morphological transition from a polarized epithelial to a more migratory mesenchymal- like phenotype (Fig. S2 D), supporting the requirement for PIPKIγ in E-cadherin–mediated AJs assembly. In addition, MDCK stable cell lines were generated that inducibly express HA-tagged wild-type (PIPKIγ661WT) or kinase-dead PIPKIγ661 (PIPKIγ661KD), the major endogenous PIPKIγ isoform associated with cadherins (Fig. S1 A). Expression of PIPKIγ661WT or PIPKIγ661KD was induced by removing doxycyclin from the growth media (). As shown in Fig. S3 A (available at ), upon PIPKIγ661KD expression, both E-cadherin PM targeting and AJ assembly appeared defective compared with parental or PIPKIγ661WT-expressing cells (Fig. S3 A). These cells formed E-cadherin–mediated cell–cell contacts much more slowly then parental cells when maintained at confluence (Fig. S3 A, 72 vs. 16 h). These observations are consistent with a dominant-negative effect for PIPKIγ661KD and also establish a requirement for PI4,5P generation in AJ assembly. Our combined results demonstrate a highly specific role for PIPKIγ in the assembly of E-cadherin–based AJs, possibly by modulating the trafficking of E-cadherin. It has been shown that depletion of extracellular calcium by EGTA results in a loss of E-cadherin homoligation, internalization of E-cadherin, disassembly of AJs, and cell scattering (). To explore possible modulation of E-cadherin trafficking by PIPKIγ, the exocytic and endocytic trafficking of E-cadherin was quantified in parental, PIPKIγ661WT-, and PIPKIγ661KD-expressing MDCK cells. To quantify internalization, cell surface E-cadherin was biotinylated, followed by extracellular calcium removal to induce E-cadherin internalization. Shown in , PIPKIγ661WT expression considerably enhanced, whereas PIPKIγ661KD expression inhibited, E-cadherin internalization when compared with parental cells. These data indicate that E-cadherin endocytosis is dependent on PIPKIγ661 kinase activity and PI4,5P generation. E-cadherin internalization can be reversed upon replenishment of calcium, providing a method to assess the role of PIPKIγ in recycling E-cadherin back to the PM. Compared with parental cells, recycling of E-cadherin was accelerated when PIPKIγ661WT was expressed, whereas PIPKIγ661KD blocked PM deposition of E-cadherin (). Immunofluorescent staining experiments supported these results, as cells overexpressing PIPKIγ661WT had a dramatically faster rate of E-cadherin internalization and also showed an accumulation of both E-cadherin and PIPKIγ in an intracellular compartment (Fig. S3 B). In contrast, overexpression of PIPKIγ661KD significantly slowed the internalization of E-cadherin (Fig. S3 B). PI4,5P regulates multiple events, including actin reorganization, that could affect E-cadherin assembly and AJ formation. Because changes in PIPKIγ expression levels may alter global PI4,5P levels and induce nonspecific responses, cellular PI4,5P was quantified by HPLC analysis (Fig. S4, A and B, available at ). No considerable changes in global cellular PI4,5P levels were observed when PIPKIγ expression levels or activity was altered. The overall structure of the actin cytoskeleton also showed no substantial change between PIPKIγ661WT- or KD-overexpressing cells and control cells (unpublished data). Upon depletion of PIPKIγ, cells exhibited an increase in actin stress fibers and prominent membrane ruffles, indicating a morphological transition from a polarized epithelial to migratory phenotype (unpublished data). Further effort was made to observe localized changes in PI4,5P levels via a bead-based adhesion assay using latex beads coated with recombinant E-cadherin ectodomain (hE/Fc). As shown in Fig. S4 C, both E-cadherin and PIPKIγ661 assembled on the surface of the hE/Fc-coated beads but not the poly-lysine–coated control beads, demonstrating that PIPKIγ661 was recruited to the nascent AJs. Although our previous data indicate that PI4,5P is necessary for E-cadherin assembly, the level of PI4,5P detected by the pleckstrin homology domain along the hE/Fc-coated bead surface was similar to the surrounding PM signal (Fig. S4 D), supporting the hypothesis that local PI4,5P levels are sufficient to regulate E-cadherin assembly. Multiple components of the trafficking machinery, including the AP complexes, bind to and are regulated by PI4,5P (; ; ). A yeast two-hybrid screen using the C terminus of PIPKIγ661 as bait identified interactions with the μ subunits of both AP1B (μ1β; amino acids 135–423) and AP2 (μ2; full length). This was an exciting observation as the μ subunits are key regulatory subunits of the AP complexes (; ). The direct interaction of PIPKIγ661 with μ1β was confirmed by direct binding of purified components (, left), and the in vivo association was established by coimmunoprecipitation (, right). PIPKIγ635 did not interact with either μ subunit (; ), indicating that the last 26 residues of PIPKIγ661 are required. In addition, in vitro binding of μ1β-adaptin stimulated the kinase activity of PIPKIγ661 (), whereas binding of the soluble ECDT (His-HR-ECDT) had no effect on PIPKIγ661 activity under these conditions (not depicted). Epithelial cells typically express two variants of the AP1 complex, AP1A and AP1B. These AP1 complexes differ only in their μ subunits, μ1α and μ1β, which are almost 80% identical but target to distinct membrane compartments and have distinct functions (). To further define the interaction between PIPKIγ and the AP1 complexes, we examined the interaction between PIPKIγ and both μ1α and μ1β in parallel experiments. As shown in (left), PIPKIγ661 binds to μ1α and μ1β indistinguishably in vitro. The primary binding site of the μ subunits is in the last 26 amino acids of PIPKIγ661 because the C terminus of PIPKIγ661 but not PIPKIγ635 bound the μ subunits (). A weak interaction between full-length PIPKIγ635 and the μ subunits was observed under less rigorous GST pull-down conditions, where detergent and BSA concentrations were decreased. This may be consistent with the observations by , which indicate that there may be a secondary μ subunit binding site in the kinase domain of the PIPKIs. However, in vivo the interaction is specific for PIPKIγ661 and only the interaction between μ1β subunit (AP1B) and PIPKIγ661 was detected (, right). The in vivo specificity of PIPKIγ661 for μ1β may be regulated by targeting to AP1B-positive membrane compartments via an interaction with other trafficking components, or the PIPKIγ661–AP1B interaction maybe specifically regulated by other mechanisms. Endogenous E-cadherin associates with the PIPKIγ–AP1 complex (). This association was disrupted by internalization of E-cadherin triggered by extracellular calcium depletion (). When calcium was restored and E-cadherin recycling to the PM was triggered, E-cadherin reassembled into the PIPKIγ–AP1 complex (). To further examine the interactions between E-cadherin, PIPKIγ661, and AP1, GST pull-down assays were performed. Although there is no direct interaction between the μ1β subunit of AP1 and HR-ECDT, PIPKIγ661 was sufficient to link HR-ECDT to μ1β in a GST pulldown experiment (). PIPKIγ661 contains a Yxxφ sorting motif (YSPL; ; ). The substitution of the tyrosine with phenylalanine in the sorting motif was reported to reduce binding to the μ subunits (; ; ). Concurrent with these results, the Y644F mutation diminished PIPKIγ661 binding to μ1β ( and ), and consequently the amount of E-cadherin C terminus pulled down by μ1β was considerably reduced (), indicating that the interaction between PIPKIγ661 and μ1β-adaptin is necessary and sufficient to link E-cadherin to the AP1B complex. Because AP complexes play an important role in protein transport, our data suggest that PIPKIγ661 regulates E-cadherin trafficking via a direct interaction with and regulation of AP complexes. Such a model would require AP1B for E-cadherin transport to the PM. To address this hypothesis, we used LLC-PK1 cells, which do not express μ1β (i.e., are AP1B deficient). In LLC-PK1 cells, many basolateral proteins are mistargeted and cell polarity is disrupted (; ). To assess the role of μ1β in E-cadherin transport, GFP–E-cadherin was expressed (). A small fraction of GFP–E-cadherin was able to translocate to the PM; however, the majority was observed in a perinuclear compartment, indicating inefficient transport of E-cadherin to the PM. In cells expressing GFP–E-cadherin, there was a greatly enhanced recruitment of endogenous PIPKIγ to GFP–E-cadherin–containing compartments, which is consistent with the association between PIPKIγ and E-cadherin. Upon expression of μ1B in the LLC-PK1 cells, GFP–E-cadherin was efficiently targeted to sites of cell–cell adhesion, and endogenous PIPKIγ colocalized with E-cadherin at AJs (). The expression of μ1α, however, did not rescue E-cadherin trafficking to the PM (unpublished data). These data support a model were PIPKIγ associates with E-cadherin and this interaction is required for functional recruitment of AP1B to PIPKIγ via its interaction with the YSPL motif in the PIPKIγ661 C terminus. The functional relationship between PIPKIγ and AP1B is reinforced by the observation that endogenous PIPKIγ and AP1 colocalized in vesicle compartments (). Both E-cadherin and PIPKIγ partially colocalized with γ-adaptin in cytoplasmic compartments after removal of calcium (, arrows), suggesting a functional link between E-cadherin, PIPKIγ661, and AP1 in E-cadherin trafficking. Interestingly, when E-cadherin recycling was triggered by replenishing calcium, we observed that overexpression of PIPKIγ661 enhanced the recruitment of AP1B to PM. In parental MDCK cells, AP1 showed typical perinuclear localization with a small fraction targeting to the PM. When PIPKIγ661 was expressed, AP1 targeted to the basolateral membrane where it colocalized with E-cadherin and PIPKIγ661 (). However, when PIPKIγ635 was expressed, AP1 organization was strikingly distinct, as it was concentrated in a central perinuclear compartment with no localization near the PM and little colocalization with E-cadherin (). In these cells, E-cadherin was largely trapped in the cytosol and was not efficiently targeted to the PM. In PIPKIγ661KD-expressing cells, AP1 weakly localized beneath the PM or showed strong colocalization of both E-cadherin and PIPKIγ661KD in a large perinuclear compartment, but there was little detectable PM E-cadherin (). These data again support a model where both the PIPKIγ–AP1 interaction and PIPKIγ kinase activity are necessary for recruitment of AP1 to the PM and the efficient trafficking of E-cadherin to the PM. To explore this hypothesis, the internalization and recycling of E-cadherin in cells ectopically expressing PIPKIγ635 was first determined. As shown in , when internalization and recycling of E-cadherin was measured by surface biotinylation, overexpression of PIPKIγ635 had a dominant-negative effect and inhibited E-cadherin trafficking to and from the PM compared with parental cells. Again, these results support a functional role for the PIPKIγ661–AP1B interaction in modulation of E-cadherin trafficking. Consistent with this conclusion, both ectopically expressed GFP–E-cadherin () and endogenous E-cadherin () were sequestered in a cytosolic compartment in PIPKIγ635-overexpressing cells, displaying a phenotype similar to that observed when endogenous PIPKIγ was knocked down. To further characterize these E-cadherin–containing vesicles in PIPKIγ635-expressing cells, we induced accumulation of transferrin receptor (TfnR) in the recycling endosome using an established approach (). As shown in (top), internalized E-cadherin in parental MDCK cells showed partial colocalization with both endogenous PIPKIγ and internalized TfnR, representing the recycling endosome, and colocalization among these three proteins was also observed (, arrows). Interestingly, overexpression of PIPKIγ635, which interacts with E-cadherin but not AP1B, blocked E-cadherin colocalization with the TfnR compartment, but E-cadherin did colocalize with PIPKIγ635 (, bottom). These data suggest that PIPKIγ661 mediates the transport of E-cadherin from the trans-Golgi network to the recycling endosome, which has been argued to serve as an intermediate between the trans-Golgi network and the basolateral PM (; ). Further, this data establishes that trafficking of E-cadherin to this compartment requires a functional interaction between PIPKIγ661 and AP1B. If PIPKIγ serves as an adaptor between E-cadherin and AP complexes, one would expect that an E-cadherin mutant lacking or with diminished PIPKIγ binding would not be transported efficiently to the PM. A V832M germline mutation was identified in hereditary diffuse gastric cancer (), which lacks the ability to mediate cell–cell adhesion or suppress invasion (). In these patients, the wild-type E-cadherin gene is repressed, and only the mutant is expressed in the carcinomas (). Interestingly, the V832M mutation lies in the PIPKIγ binding region. To determine whether this mutation impacts PIPKIγ binding, E-cadherin V832M was introduced into HEK293 cells and its association with PIPKIγ was quantified. This mutant showed a substantially reduced ability to bind PIPKIγ (). Consistent with published data (), β-catenin binding was normal (not depicted). The basolateral transport of this V832M mutation was also explored in both LLC-PK1∷μ1β (unpublished data) and MDCK cells using GFP-fused E-cadherinV832M. As shown in , although the V832M mutant was visualized on the PM as reported by others (), a large accumulation of this E-cadherin mutant was observed in a cytosolic compartment. This phenotype was similar to that of wild-type E-cadherin observed in the PIPKIγ635-overexpressing cells () or the LLC-PK1 cells deficient in μ1β (). Wild-type E-cadherin in LLC-PK1∷μ1β () and MDCK () cells was transported efficiently to the basolateral membrane and little was visualized in the cytosol. This result is consistent with a requirement for an interaction between E-cadherin and PIPKIγ661 for normal trafficking of E-cadherin. Proteins that interact with the ECDT not only mediate the elemental functions of E-cadherin, such as AJ assembly, actin organization, and cell proliferation, but also regulate E-cadherin by modulating its expression and trafficking. Here we have shown that PIPKIγ directly binds to both E-cadherin and AP complexes. This dual interaction supports a mechanism for the highly regulated generation of PI4,5P to spatially drive the assembly of the trafficking machinery and to specifically control E-cadherin trafficking. These results reveal a novel mechanism where PIPKIγ661 functions as both scaffolding and signaling molecule during E-cadherin trafficking (). In this model, the AP complex interacts indirectly with the E-cadherin cargo via the PIPKIγ661 scaffold, which directly binds to AP complexes via a Yxxφ sorting motif in its C terminus. This represents a novel paradigm in which PIPKIγ661 serves as a cargo adaptor for AP complexes. Although PIPKIγ661 binds to both AP1A and AP1B indistinguishably in vitro, we found that PIPKIγ661 preferentially interacts with AP1B in vivo, and regulation of E-cadherin trafficking to the basolateral PM appears to be specific for AP1B. AP1A and AP1B both use Yxxφ sorting motifs for cargo recognition. However, despite the substantial sequence and structural homology of the μ1 subunits, the AP1 complexes are targeted to distinct compartments by an unknown mechanism (; ). PIPKIγ661 might be specifically recruited to AP1B-containing membrane domains in vivo via an interaction with one or more other proteins. PIPKIγ661 may also have additional lower affinity binding contacts with AP1B, as recently reported that multiple PIPKIs bind to the μ2 subunit of AP2 complex via the kinase domain. In our hands, the YSPL motif of PIPKIγ661 was the preferential binding site for the μ subunits of AP1 and AP2 (), but PIPKIγ635 did bind μ1- and μ2-adaptin subunits under less rigorous conditions (unpublished data), suggesting additional lower affinity interacting sites between the AP complexes and PIPKIγ. For E-cadherin trafficking, the YSPL motif of PIPKIγ661 is the key interaction with AP complexes and subsequent interactions may regulate kinase activity. As the kinase domains of the PIPKI isoforms are highly homologous, other isoforms of PIPKIs (e.g., PIPKIα) may interact with and regulate some AP complex–dependent trafficking events (). These putative interactions could be through the conserved kinase domains or, like PIPKIγ661, could be mediated by specific binding partners via interaction with variable regions of the PIPKI isoforms. Nevertheless, because PI4,5P is a key moderator of the recruitment and assembly of trafficking machinery (; ), the localized generation of PI4,5P at sites where E-cadherin and other cargoes are assembled into the trafficking machinery is an indispensable step in this process. This finding suggests that any association between a PIPK and the trafficking machinery must be spatially and temporally regulated. The interaction between the PIPKIγ661 and the AP1B complex fits this criterion, as this association is detected when E-cadherin is recycled back to the PM but not when E-cadherin is being internalized (). This observation supports the concept of cellular signals coordinating the interactions between PIPKIγ661 and the AP complexes. A dileucine motif in the juxtamembrane region of the ECDT is required for basolateral sorting (), and this motif was proposed to be a cargo signal recognized by the β subunit of the AP1 complex (). There is no solid evidence supporting this interaction at present. However, if this is true, the E-cadherin–PIPKIγ661–AP1B complex could be further stabilized via the interaction of the E-cadherin dileucine motif with the β subunit of AP1B. Alternatively, other trafficking components may recognize this motif and cooperate with PIPKIγ and AP1B to provide specificity. The E-cadherin–PIPKIγ661–AP1B interaction serves as a foundational signal for exocytic targeting and basolateral sorting of E-cadherin. Indeed, internalized E-cadherin accumulated at the recycling endosome, as indicated by internalized TfnR. This compartment contained, in addition, endogenous PIPKIγ supporting a role for PIPKIγ in trafficking though this compartment. Consistent with this observation, the overexpression of PIPKIγ635, which binds E-cadherin but not μ1β (AP1B), blocked E-cadherin trafficking to the TfnR-positive compartment of the recycling endosome. E-cadherin did colocalize with PIPKIγ635, indicating that PIPKIγ interacts with E-cadherin in this compartment. The combined results demonstrate that the association of PIPKIγ661 with AP1B is require for E-cadherin trafficking through this compartment. The recycling endosome is a major site of AP1B, and this further supports our hypothesis that PIPKIγ functions as an adaptor in E-cadherin trafficking and facilitates E-cadherin transport to and from the recycling endosome via binding to AP1B and generation of PI4,5P. E-cadherin endocytosis can occur in a clathrin-dependent (; ) or independent manner (). Calcium removal stimulates E-cadherin endocytosis by the clathrin-dependent pathway (). As there is no known Yxxφ sorting motif in the ECDT, the interaction between PIPKIγ661 and E-cadherin may recruit AP2 for clathrin-dependent E-cadherin endocytosis. Additionally, Arf6 promotes E-cadherin internalization () and has been shown to associate with and stimulate the activity of PIPKIγ (, ). Arf6, in cooperation with PI4,5P, was also shown to directly interact with and promote the recruitment of AP2 to the PM (; ). These combined results suggest that PIPKIγ661, E-cadherin, AP2, and Arf6 may cooperate to regulate E-cadherin internalization in epithelial cells. This would position PIPKIγ661 as a nexus between AP complexes and E-cadherin in endocytic recycling. Nevertheless, there is no direct evidence demonstrating that AP2 mediates the internalization of E-cadherin, and further investigation is needed to characterize the role of PIPKIγ in E-cadherin endocytosis. Our data demonstrates that a loss of PIPKIγ in cultured epithelial cells results in the severe mistargeting of E-cadherin, suggesting a strong functional connection between PIPKIγ and E-cadherin. Interestingly, the PIPKIγ knockout mouse does not share the same phenotype as the E-cadherin knockout mouse (; ). This is not surprising, as the knockout phenotypes of other modifiers of E-cadherin function, such as p120-catenin, differ from that of the E-cadherin knockout as well (; ). Considering the existence of multiple pathways for E-cadherin trafficking, the roles of these E-cadherin modifiers may only become apparent during the development of specific tissues in later stages of animal development. Dimerization is an essential property of E-cadherin assembly driving AJ formation (). The association of both PIPKIγ and p120 catenin with the E-cadherin dimer may be a mechanism to functionally regulate E-cadherin assembly and could promote AJ formation by stimulating E-cadherin clustering. Because PIPKIγ specifically binds to E-cadherin dimers, the in situ PI4,5P generation resulting from this interaction may drive other local complementary cellular events, such as actin reorganization (). Actin assembly is important not only in AJ assembly but also for E-cadherin internalization/exocytosis (). The association of PIPKIγ with E-cadherin may be crucial for downstream signaling, as Rac and phosphoinositide-3 kinase are activated by E-cadherin and both regulate the stability of AJs by modulating actin assembly (; ; ). Phosphoinositide-3 kinase requires PI4,5P for signaling, and Rho family small G proteins regulate some PIPKI isoforms (). As a result, PIPKIγ may also regulate AJ assembly through local cooperation with phosphoinositide-3 kinase and small G protein signaling. The generation of phosphoinositide messengers upon assembly of AJs has implications beyond simple control of E-cadherin trafficking. Because E-cadherin is a major suppressor of invasion of epithelial tumors, the cell biological data suggest that PIPKIγ may play a similar role. In exploring this possibility, we have discovered that a loss of E-cadherin correlates with a loss of PIPKIγ in human breast cancers (unpublished data). This finding supports a physiological role for PIPKIγ in assembly of E-cadherin junctions and potentially a role in progression of epithelial tumors. The C terminus of E-cadherin was amplified by PCR and constructed into normal or modified () pET28 to generate the His-tagged E-cadherin tail or HR–E-cadherin tail, which were then subcloned into pCMV-Myc vector (CLONTECH Laboratories, Inc.). Wild-type E-cadherin, E-cadherinΔp120ctn, E-cadherinΔβctn, and E-cadherin/αctn were provided by B. Gumbiner (University of Virginia, Charlottesville, VA). E-cadherin836 was amplified by PCR, and E-cadherinV832M was generated using the QuikChange II Site-Directed Mutagenesis kit (Stratagene) according to the manufacturer's instructions. Both μ1α and μ1β constructs were provided by I. Mellman (Yale University, New Haven, CT). cDNAs encoding N-terminal truncated μ1α and μ1β (1–135 aa truncated) were amplified by PCR and subcloned into pET42. All of the PIPKI constructs were created as described previously (, ; ). Duplexes of siRNA oligos (for both human and mouse: aagttctatgggctgtactgc, aaggacctggacttcatgcag; for canine: gaaggctcttgttcacgat) were synthesized by Dharmacon. Monoclonal antibodies for E-cadherin (recognizing the C terminus), N-cadherin, human VE-cadherin, p120catenin, β-catenin, γ-adaptin, and FITC-conjugated anti–E-cadherin were purchased from Transduction Laboratories. The H68.4 monoclonal anti-CD71 (TfnR) antibody was purchased from BioGenex Inc. Polyclonal PIPKIγ antibody was generated as described previously (). Regular mouse and rabbit IgG and secondary antibodies were obtained from Jackson ImmunoResearch Laboratories. Anti-HA antibody was purchased from Covance. Anti-Myc and FITC-conjugated anti-Myc were obtained from Upstate Biotechnology. HRP-conjugated anti-GST antibodies were purchased from GE Healthcare. Anti–E-cadherin antibodies recognizing the ectodomain were purchased from Zymed Laboratories (monoclonal, for immunoblotting) and Sigma-Aldrich (rat monoclonal, for immunofluorescence). MDCK-TetOff cells (CLONTECH Laboratories, Inc.) and HEK293 cells were cultured in Dulbecco's modified eagle medium (Mediatech, Inc.) supplemented with 10% FBS (Invitrogen). Lysates of human umbilical vein endothelial cells were obtained from E. Bresnick (University of Wisconsin-Madison, Madison, WI). MDCK cells were transfected using FuGENE 6 (Roche) for 48 h, and then 100 μg/ml hygromycin B was added to the medium to select stable clones and 10 mg/ml doxycycline was used to suppress PIPKIγ expression. To induce expression, doxycycline was removed for 72 h. HEK293 cells were transfected using the calcium phosphate–DNA coprecipitation method for 48 h. For siRNA knockdown, MDCK cells in a 6-well plate were transfected twice at 48-h intervals with 5 pmol/well siRNA using the calcium phosphate–DNA coprecipitation method. Cells were analyzed 48 h after the second transfection. LLC-PK1 cell lines were provided by I. Mellman and cultured as described previously (). Indirect immunofluorescence and confocal microscopy were performed as described previously (). For triple labeling, double labeled samples were blocked by 0.5 mg/ml of normal mouse IgG in 3% BSA/PBS at 37°C for 30 min, rinsed in PBS twice, and incubated with FITC-conjugated anti–E-cadherin or anti-Myc antibodies in 3% BSA/PBS at 37°C for 1 h. Confocal images were acquired using photomultiplier tubes through LaserSharp2000 (Carl Zeiss MicroImaging, Inc.) with a PlanApo 100× oil objective (NA 1.4) on an inverted microscope (Eclipse TE2000; Nikon) with Radiance 2100 MP Rainbow (Bio-Rad Laboratories). Z series were created by sequentially scanning FITC, Texas red, or Cy5 channel at 0.3-μm steps. Single sections were exported to Photoshop CS2 (Adobe) for final image processing. Fluorescence intensity was quantified using ImageJ 1.62 (National Institutes of Health) and plotted using SigmaPlot 8.0. Cells were lysed in lysis buffer (50 mM Tris⋅HCl, pH 7.5, 150 mM NaCl, 0.5% NP-40, 1 mM EDTA, 1 mM PMSF, and 10% glycerol), and then used for immunoprecipitation (). The immunocomplexes were separated by SDS-PAGE and analyzed. Unless stated, immunoprecipitations were performed using 800 μl of cell lysate from one confluent 60-mm dish. One fourth of each precipitate and 20 μl of each lysate were analyzed. GST-tagged PIPKIα, PIPKIγ635, and PIPKIγ661, His-tagged E-cadherin, or HR-E-cadherin tail were purified from , and GST pull-down assays were performed (). One fourth of the GST beads for each pull down were loaded on the gel and 20 μl of each purified protein was loaded as an input control. Images were scanned and exported to Photoshop CS2 for final processing. Intensity of bands was quantified using ImageJ 1.62 and plotted using SigmaPlot 8.0. Cells were allowed to grow on coverslips for 72 h to reach confluence and were then incubated with 2 mM EGTA for 20 min before performing indirect immunoflurescence. Confluent MDCK cells grown in 24-mm-diam Transwells (Costar) were biotinylated by 1 mg/ml sulfo-NHS-SS-biotin (Pierce Chemical Co.) and analyzed as described previously (). Cells were lysed in 500 μl of lysis buffer and one third of the precipitates were analyzed. Internalization of E-cadherin was induced by 0.5 mM EGTA at 18°C. To measure the recycling of E-cadherin, MDCK cells were treated with 2 mM EGTA for 40 min at 37°C and chased in normal medium, and then surface biotinylation was performed. Activity of 10 μg of purified recombinant PIPKIγ proteins was assayed against 20 μg of Folsch Brain Extract III as previously described (). Kinase activity was quantified using Storm 840 (Molecular Dynamics) and plotted using SigmaPlot 8.0. Fig. S1 shows that PIPKIγ directly binds E-cadherin. Fig. S2 shows that the direct interaction with PIPKIγ is important for E-cadherin assembly. Fig. S3 shows that functional PIPKIγ is required for E-cadherin assembly. Fig. S4 shows that a modification of PIPKIγ activity has no effect on global PI4,5P level. Online supplemental material, including Figs. S1−S4 and supplemental Materials and methods, is available at .

Fibrillin microfibrils of the ECM, which associate with elastic fibers, are implicated in the regulation of TGFβ in large latent complexes (LLCs; for review see ; ). Fibrillin-1 is a multidomain cysteine-rich glycoprotein containing 43 calcium-binding EGF (cbEGF)–like domains and 78 cysteine-containing TB motifs (). Fibrillin-1 mutations cause the heritable disorder Marfan syndrome (MFS) with severe cardiovascular, skeletal, ocular, and lung manifestations (for review see ). Enhanced TGFβ signaling is a major contributor to the pathology of MFS. A model has been proposed in which fibrillin-1 mutations perturb the normal microfibril regulation of latent TGFβ and, thereby, contribute to MFS pathogenesis (for review see ). The clinically overlapping conditions, Loeys-Dietz aortic aneurysm syndrome, familial thoracic aortic aneurysms and dissections, and marfanoid craniosynostoses are also caused by enhanced TGFβ signaling but, in these cases, are caused by cytoplasmic kinase mutations in TGFβ receptor (TGFβR) I or II (; , ; ; ; ). Mouse MFS models have revealed that enhanced TGFβ activity in fibrillin-1 haploinsufficient mice leads to primary developmental failures, including distal alveolar septation (), and, in heterozygous mutant mice, leads to mitral valve defects (). Haploinsufficiency triggers secondary cellular events that result in intimal hyperplasia and adventitial inflammation with TGFβ involvement as well as aortic failure (for review see ). Losartan, an angiotensin II blocker that lowers blood pressure and leads to the clinically relevant attenuation of TGFβ signaling, prevented aortic aneurysm in a mouse MFS model (). Tight-skin mice have enhanced TGFβ activity and sclerosis as a result of an internal fibrillin-1 duplication and a larger than normal secreted protein (; ). TGFβ is secreted from cells as a dimeric small latent complex (SLC) comprising noncovalently associated latency-associated propeptide (LAP) and active TGFβ and/or as a large LLC comprising SLC bound covalently to a latent TGFβ-binding protein (LTBP) through a TB motif (for reviews see ; ). Only LTBP-1 and -3 bind TGFβ strongly. It has been proposed that by interacting with LLC, fibrillin microfibrils may act as a growth factor highway in tissues (for review see ). LTBPs are structurally related to fibrillins (for review see ). LTBP-1 but not LTBP-3 can bind in vitro to fibrillin-1 (). This interaction involves three C-terminal domains of LTBP-1 and four N-terminal domains of fibrillin-1. LTBP-1 is an associated but not an integral component of microfibrils (; ), and it colocalizes with fibrillin microfibrils in some tissues (; ). The prodomain of another TGFβ superfamily member, BMP-7, can bind an N-terminal fibrillin-1 fragment in vitro (). Activation of TGFβ, a potent growth factor that regulates cell proliferation, migration, differentiation, and survival, is normally tightly regulated. However, physiological activation mechanisms leading to receptor signaling are incompletely understood. They may involve LTBP-1–mediated proteolytic release, thrombospondin-1 (TSP-1) competition with SLC, integrin presentation, pH changes, and reactive oxygen species (for reviews see ; ; ; ; ). Autoantibodies to a fibrillin-1 proline-rich region induce fibroblast activation possibly by releasing sequestered TGFβ1 from microfibrils (). BMP-1 also controls TGFβ1 activation by cleaving LTBP-1 (). Once activated, TGFβ binding to TGFβRI and II heterodimers leads to the phosphorylation of TGFβRI, which, in turn, phosphorylates signaling proteins Smad2 and Smad3 (for reviews see ; ). Smad2 and Smad3 phosphorylation allows association with Smad4, nuclear translocation, and specific gene activation or repression. We have discovered that in the presence of cells, a specific fibrillin-1 sequence encoded by exons 44–49 regulates the bioavailability of endogenous TGFβ1, thereby stimulating Smad2 signaling. Fibrillin-1–mediated TGFβ release from ECM does not require intact cells, proteolysis, or changes in the expression of TGFβ or its receptors. A fibrillin-1 fragment containing the TGFβ- regulating sequence specifically binds deposited fibrillin-1 in the insoluble cell layer through a strong interaction with the fibrillin-1 N-terminal region. This interaction, which directly inhibits the association of C-terminal LTBP-1 with fibrillin-1, can thus release LLC from microfibrils. This novel mechanism is likely to contribute to TGFβ dysregulation in MFS and related diseases and in acquired fibrotic disorders. Our first step was to determine whether fibrillin-1 could stimulate the Smad2 pathway. Recombinant fragments encompassing full-length human fibrillin-1 () were tested for their ability to induce Smad2 phosphorylation in human dermal fibroblasts (HDFs) that were cultured in serum-free conditions. Overlapping fragments PF10 and PF11 but not overlapping fragments PF8, PF9, PF12, and PF14 were found to stimulate Smad2 signaling (). No other fibrillin-1 fragments or human plasma fibronectin stimulated Smad2 phosphorylation (). Thus, the Smad2-stimulating effect was mapped to a specific fibrillin-1 sequence of six contiguous cbEGF-like domains that are encoded by exons 44–49 (, asterisk). Similar results were obtained using the mouse osteoblast cell line 2T3 (unpublished data). ELISA assays revealed that purified PF10 alone contained no active TGFβ (R 0.9988), and repeated mass spectrometry failed to detect any trace of LAP or TGFβ tryptic peptides in purified PF10 preparations (; unpublished data). TGFβ1 signals through a heteromeric complex of TGFβRI and II, which have serine/threonine kinase activity (for reviews see ; ). We investigated whether the Smad2 signaling effects of fibrillin-1 fragments PF10 or PF11 were exerted through these receptors (). First, an antibody that blocks TGFβRII was used in cell signaling inhibition assays. In the presence of the inhibitory TGFβRII antibody, there was no Smad2 stimulation by PF10 () or PF11 (not depicted). The TGFβRII-inhibiting antibody also blocked TGFβ1-induced Smad2 phosphorylation (). A chemical inhibitor, [3-(pyridin-2-yl)-4-(4-quinonyl)]-1H-pyrazole, which is an ATP-competitive inhibitor of TGFβRI kinase (), was then used in Smad2 signaling inhibition assays to ascertain whether TGFβRI was also involved in the PF10-mediated stimulation of Smad2 signaling. No Smad2 signal in response to PF10 (), PF11 (not depicted), or TGFβ1 () was detected when TGFβRI was neutralized by this inhibitor. Thus, PF10 and PF11 exert their effects on Smad2 signaling through TGFβRI and II. Using an antibody that specifically inhibits active TGFβ1, the Smad2 signal was markedly reduced upon stimulation with PF10 () or PF11 (not depicted). In control experiments with supplemented TGFβ1, the inhibitory TGFβ1 antibody also blocked Smad2 phosphorylation (). Thus, Smad2 phosphorylation by PF10 or PF11 requires active TGFβ1, and fibrillin-1 does not directly activate these receptors. Using ELISA assays, we found that the supplementation of HDF cultures with PF10 or PF11 increased active TGFβ1 in HDF serum-free medium (). After supplementing 1 μM HDF cultures for 90 min, PF10 treatment had enhanced active TGFβ1 to 23.7 pM and PF11 to 18.5 pM compared with medium from untreated HDF cultures, which contained only trace levels of TGFβ1. Positive control experiments with added recombinant active TGFβ1 contained high levels of TGFβ1 as expected. Using human plasma fibronectin, there was no increase in active TGFβ1 (R 0.9988; ). PF10-treated cultures had slightly more total than active TGFβ1 (). PF10, which lacks the N-terminal three domains of PF11, also consistently generated a stronger Smad2 phosphorylation signal than PF11 at equal concentrations (0.15 μM; ). However, both fragments showed a similar time-dependent Smad2 signaling response in which a marked increase in phosphorylated Smad2 from 5 to 20 min was seen with PF10 () and PF11 (not depicted). The sequence within PF10 and PF11 that regulates active TGFβ1 levels and Smad2 signaling was localized to six cbEGF-like domains (, asterisk). We investigated whether its ability to enhance levels of active TGFβ was conformation dependent. After the preincubation of PF10 or PF11 with the calcium chelator EDTA at a concentration of 100 mM, increased Smad2 phosphorylation was detected in the EDTA-treated samples but not in the untreated or EDTA-only controls (unpublished data). No EDTA-induced increase in Smad2 phosphorylation was detected in control HDFs supplemented with TGFβ1 that had been preincubated with EDTA. PF10 treatment with 0.2 mg/ml elastase, which degrades PF10 (), fibrillin molecules, and microfibrils (), also enhanced PF10-induced Smad2 signaling (). Full-length fibrillin-1 molecules that were purified from HDF culture medium stimulated Smad2 phosphorylation, but not as strongly as PF10 (). However, Smad2 signaling activity was barely detectable after supplementing cultures with microfibrils purified from bovine ciliary zonules (), possibly as a result of masking of the TGFβ regulatory sequence. ELISA assays revealed that regulation of active TGFβ levels by PF10 or PF11 or by fibrillin molecules purified from HDF culture medium in the HDF cultures for 90 min was dose dependent (0.0625–2 μM). Linear regression analysis showed that the slope of the regression line for PF10 was greater than PF11, although it was not statistically significant. However, PF10 did show a statistical increase in active TGFβ1 when compared with intact fibrillin-1 molecules (R 0.9993; ). TSP-1 activates TGFβ1 by interacting with SLC (). The active TGFβ1 sequence RKPK associates with the LAP sequence LSKL; SLC interactions with TSP-1 sequences KRFK and WSXW result in the release of active TGFβ1. These TSP-1 sequences are not present within PF10. A comparison of the effects of human TSP-1 and fibrillin-1 fragment PF10 on TGFβ1 showed that at equimolar concentrations (15 nM), PF10 treatment increased 1.1 pM of active TGFβ1 more than TSP-1 (R 0.9989; ). Having shown that PF10 treatment increases active TGFβ1 in HDF cultures supplemented with serum-free medium, we used ELISA assays to determine whether this effect requires intact cells, cell layer ECM, or HDF-conditioned medium. 1.5 μM PF10 strongly enhanced active TGFβ1 when incubated with cell layers in freshly added serum-free medium (). In contrast, when PF10 was added to conditioned medium alone, it induced a very small but significant increase in active TGFβ1 at 15 and 60 min (4–7% of active TGFβ1 levels induced by cell layers; R 0.9985; ). We also compared total and active TGFβ1 levels in cell layers before and after cell lysis (). PF10 treatment of lysed cell layers led to release into serum-free medium of 83% of the levels of both total and active TGFβ1 that were released using unlysed cultures (), with no statistical difference between active and total TGFβ1 levels released from the lysed cell layers. Thus, the deposited cell layer ECM is the main requirement for the PF10-mediated increase in active TGFβ1. The cell lysis experiments indicated that most of the TGFβ1 regulatory effect of PF10 resided within lysed cell layers. Nevertheless, we decided to further study whether cell surface receptors influenced the PF10-mediated increase in Smad2 signaling because integrins have previously been implicated in TGFβ activation (). The addition of integrin function–blocking antibodies to β1 or αv had no significant effect on PF10- or PF11-mediated Smad2 signaling (). The blocking antibody (mAb 16) to α5 did have a small but significant enhancing effect on TGFβ activation (); an α5-integrin–blocking antibody has previously been shown to activate TGFβ in cultures (). However, when HDFs were coincubated with PF10 in the presence of 1.5 mM EDTA, which chelates divalent cations and inhibits integrins (), there was no effect on the PF10-mediated increase in Smad2 signaling (unpublished data). Syndecan-4–null mouse embryonic fibroblasts significantly increased Smad2 signaling in response to PF10, as did the wild-type control fibroblasts (). Thus, PF10-mediated TGFβ regulation occurs in the absence of syndecan-4. Activation of TGFβ from the SLC complex can involve pericellular proteolysis (for review see ). To investigate whether proteases are involved in the fibrillin-1–mediated increase in Smad2 signaling, HDFs were preincubated for 30 min with inhibitors of serine (aprotinin and leupeptin), cysteine (leupeptin), and/or metalloproteinases (4-Abz-Gly-Pro--Leu--Ala-NH-OH). Quantitative analysis of densitometric data that was normalized against β-actin confirmed that none of these protease inhibitors had any substantial effect on PF10-stimulated Smad2 signaling (unpublished data). PF10 and PF11 induction of TGFβ signaling could be caused by rapid changes in the gene expression of TGFβ and its receptors. mRNA samples from HDFs supplemented for 30 min with PF10 or PF11, with TGFβ1 as a positive control, or with no ligand as a negative control were used in semiquantitative RT-PCR experiments. There were no detectable differences in the expression levels of TGFβ1 and TGFβRI/II/III during the time frame of fibrillin-1–mediated enhanced TGFβ signaling (unpublished data). #text As we previously reported, it was not possible to coat BIAcore chips with the fibrillin-1 fragments (; ). However, solid-phase binding assays of overlapping fibrillin-1 fragments () revealed that PF10 strongly and specifically interacted with the N-terminal region of fibrillin-1 (fragment PF1) with relatively high affinity (dissociation constant [K] = 90 ± 14 nM; ). Thus, this interaction mediates the association of PF10 with full-length fibrillin-1. We also examined the effects of two MFS disease–causing mutations in the N-terminal region (PF1) on interactions with PF10. Both MFS mutant forms of PF1 showed altered affinities for PF10. Mutant PF1 exhibited increased affinity (K = 52 ± 13 nM), whereas mutant PF1 bound very weakly. These altered affinities may affect PF10-stimulated Smad2 signaling and possibly MFS phenotype. #text xref sup #text xref sub #text xref #text Recent studies have shown that a major functional relationship exists between fibrillin-1 and TGFβ activity (for reviews see ; ). Fibrillin-1 is postulated to regulate TGFβ through the association of LLC with fibrillin-rich microfibrils, although it is not clear how this regulation occurs. We have discovered that a specific fibrillin-1 sequence encoded by exons 44–49 (in recombinant fragments PF10 and PF11) enhances endogenous active TGFβ1 and Smad2 signaling. This sequence, which is present within a pepsin-resistant microfibril proteolytic fragment (), contains no TB motif such as those in LTBP-1 and -3 that bind LAP through disulphide linkage (for review see ). Thus, fibrillin-1 enhances active TGFβ1 by a novel mechanism and may contribute directly to the lung, skeletal, and vascular pathologies of MFS and related diseases. We excluded the idea that purified PF10 or PF11 contained traces of latent or active TGFβ by mass spectrometry and immunoblotting, and we did not detect any TGFβ1 activity in our purified PF10 or PF11 preparations. The smaller fibrillin-1 sequences tested had greater ability to stimulate Smad2 signaling. PF10 induced slightly greater levels of active TGFβ and Smad2 signaling than PF11, which comprises PF10 plus three additional upstream domains, and both fragments induced greater levels of active TGFβ and Smad2 signaling than intact fibrillin. Small-angle x-ray analysis and single-particle transmission electron microscopy of the solution structure of fibrillin-1 recently revealed that the region spanning TB4 to TB6 (PF11) is relatively compact, with PF10 being the most linear region within PF11 (). The additional three-domain globular region of PF11 and other domains in full-length fibrillin-1 may exert conformational effects that reduce the availability of the sequence encoded by exons 44–49. We previously showed that elastase effectively degrades microfibrils and fibrillin molecules (), and, here, we have found that the elastase degradation of PF10 enhances Smad2 signaling. In tissues, such proteolytic fragments may potently stimulate TGFβ-induced signaling. We found only trace levels of Smad2 signaling induced by tissue-purified microfibrils. The active PF10 sequence may be masked by molecular folding and/or by associated molecules. We previously mapped this fibrillin-1 region to the microfibril interbead (). TSP-1 has previously been identified as a physiological activator of TGFβ (). ELISA dose-response curves have revealed that PF10 was more effective at activating endogenous TGFβ1 than TSP-1. The active TGFβ sequence RKPK associates with the LAP sequence LSKL in the SLC (); the TSP-1 activation of TGFβ involves competitive binding of a TSP-1 sequence (KRFK) that releases TGFβ from SLC. Sequence analysis of the six cbEGF domains in PF10 that regulate TGFβ bioavailability revealed no similar motifs, so PF10-mediated TGFβ1 activation involves a different mechanism. Latent TGFβ can also be activated by integrin αvβ6 or proteolysis (; ). However, PF10-mediated regulation of TGFβ1 did not involve cell surface β1- or αv-integrin receptors, syndecan-4, or pericellular proteolysis. Moreover, changes in TGFβ-induced signaling could not be accounted for by the enhanced expression of TGFβ or its receptors. To determine how fibrillin-1 enhances active TGFβ and stimulates Smad2 signaling, we first investigated what PF10 interacts with in HDF cultures. Mass spectrometry revealed that PF10 bound specifically to full-length fibrillin-1 in the microfibril-rich insoluble fibroblast layer, which is a proposed repository of LLC (for review see ), and also bound with high affinity to the N-terminal fibrillin-1 fragment PF1 in solid-phase binding assays. The PF1 sequence localizes adjacent to microfibril beads (; ). We found no evidence for SLC interactions with either PF10 or PF1 or for LTBPs interacting directly with PF10. Crucially, however, preformed PF10–PF1 complexes reduced PF10-induced Smad2 signaling, confirming a key role for this interaction in regulating active TGFβ1. Moreover, MFS mutant PF1 fragments that had increased or decreased affinity for PF10 showed reduced or unchanged Smad2 signaling, respectively. Next, we showed that the PF10 interaction with PF1 directly inhibits C-terminal LTBP-1 binding to the fibrillin-1 N terminus so that, at appropriate concentrations, it will displace LLC from microfibrils. Finally, we confirmed that PF10 has no effect on Smad2 signaling in UMR-106 cell cultures, which do not constitutively express fibrillin-1 or LTBP-1 (). Thus, we have delineated a novel mechanism that regulates TGFβ bioavailability () in which PF10, by binding microfibrils close to the beads through interactions with the fibrillin-1 N-terminal sequence, can displace LTBP-1 and LLC from microfibrils. One possible mechanism of subsequent TGFβ activation may be the BMP-1 cleavage of LTBP-1 (). Alternatively, TGFβ may become activated during the release of LLC from microfibrils through conformational changes because fibrillin-1– and SLC-binding sites are within the same C-terminal region of LTBP-1. The LTBP-1 N terminus can be transglutaminase linked to ECM (for review see ), but release of the LTBP-1 C terminus from microfibrils may be sufficient for TGFβ activation. Further experiments confirmed that PF10 releases TGFβ1 mainly from lysed cell layers as expected because fibrillin-1 is a major deposited ECM component. The small increase in TGFβ1 levels when cells are intact may be caused by additional microfibrils assembling at the cell surface. Low levels of active TGFβ1 released by PF10 from conditioned medium probably reflect the known presence of some secreted fibrillin-1 molecules and aggregates in medium (; unpublished data). Enhanced PF10-mediated Smad2 signaling after EDTA treatment indicates that calcium-dependent conformation of the cbEGF-like domain array influences activation, perhaps by altering the PF1–PF10 interaction. We have also found that supplementing cultures with heparin enhances PF10-dependent TGFβ activation, but we have excluded that this heparin effect is caused by direct heparin–PF10 interactions (unpublished data). Heparin strongly binds PF1 in a conformation-dependent manner (; unpublished data), so we speculate that it may enhance LTBP-1 displacement from PF1 by PF10. Pathological fibrillin-1–mediated regulation of TGFβ bioavailability may be induced by microfibril degradation products. Progressive proteolytic damage and aortic degeneration are hallmarks of classic MFS. Disease-causing amino acid substitutions are spread throughout the molecule (for review see ), but some mutations occur within PF10 that may directly alter TGFβ activation. They include classic MFS causing amino acid substitutions in exons 44 and 46, exon 47/48 domain interface, exons 47 and 48, and deletions of exons 44, 44–46, 46, and 49 (). Furthermore, mutations in any region of fibrillin-1 that disrupt domain and molecular conformations can increase proteolytic susceptibility to inflammatory enzymes (; ; ; ; ), leading to microfibril proteolysis and release of TGFβ-regulating fragments. Microfibrils from unaffected individuals are also highly susceptible to degradation by matrix proteases such as elastase (). Thus, microfibril proteolysis could be a common mechanism for the release of active TGFβ1 from ECM in heritable and acquired fibrillinopathies. In summary, we have shown that a specific fibrillin-1 sequence regulates the bioavailability of TGFβ1. We are currently investigating whether fibrillin-1 similarly regulates levels of other TGFβ isoforms and whether other fibrillins can regulate TGFβ. Tissue culture reagents were purchased from Life Technologies or Mediatech. 293-EBNA cells were purchased from the American Type Tissue Culture Collection and were routinely maintained in DME with 10% FBS, 2 mM -glutamine, 100 U/ml penicillin/streptomycin, and 250 μg/ml G418. HDFs were purchased from Cascade Biologics, Inc. and maintained in low serum growth supplement from the same supplier. UMR-106 rat osteosarcoma cells were originally obtained from T.J. Martin (St Vincent Institute of Medical Research, Fitzroy, Victoria, Australia). 2T3 cells were a gift from S. Harris (University of Texas Health Science Center, San Antonio, TX; ). Recombinant fibrillin-1 fragments encompassing full-length human fibrillin-1 were expressed in 293-EBNA cells using a modified pCEP-His vector and were purified as previously described (; ; ). Secreted fibrillin molecules and multimers were purified from confluent HDF culture medium by cesium chloride density gradient centrifugation and size fractionation using a Sephacryl 200 column equilibrated in 0.1 M NaCl, 1 mM CaCl, and 50 mM Tris, pH 8.0. Identity and purity were confirmed by immunoblotting using an anti–fibrillin-1 mAb raised to the N terminus (amino acids 45–450; mAb 2502; Chemicon Europe) and by mass spectrometry (provided by B. Raynal, University of Manchester, Manchester, UK). Microfibrils were purified from adult bovine ciliary zonules as previously described (). The presence of microfibrils was confirmed using atomic force microscopy (). A C-terminal fragment of human LTBP-1 (amino acids 1,008–1,394) was generated by PCR amplification using Vent DNA polymerase (New England Biolabs, Inc.), a high fidelity DNA polymerase, according to the manufacturer's instructions. The template was human LTBP-1 cDNA in the vector pSV7d (a gift from K. Miyazono, University of Tokyo, Tokyo, Japan). A 10-histidine epitope tag was engineered into the primers at the C terminus of the recombinant LTBP-1 fragments. The PCR products were ligated into pCEP-Pu expression vector (a gift from E. Kohfeldt, Max Planck Institute of Biochemistry, Martinsried, Germany) in frame with the BM40 signal sequence. Insert sequences were confirmed by automated sequencing (MWG). Constructs were transfected into 293-BNA cells using LipofectAMINE 2000 (Invitrogen). Transfected cells were selected in 1 μg/ml puromycin, and resistant cells were expanded into triple-layer flasks. Recombinant fragments were purified using a nickel-NTA agarose column (QIAGEN) according to the manufacturer's instructions. Bound protein was eluted with low pH or with 100–300 mM imidazole. The protein was further purified using a mono-Q ion exchange column in conjunction with a protein purification system (BioCad 700E; Applied Biosystems). Bound protein was eluted with a linear 0–1-M NaCl gradient. Coomassie blue staining was used to visualize the purity of the fragment, and mass spectrometry/peptide mass mapping was used to validate the recombinant LTBP-1 fragment. Confluent HDFs were incubated for 24 h using serum-free DME supplemented with 4.5 g/L glucose and -glutamine (Cascade Biologics, Inc.). The cells were incubated in 0.5 ml of fresh serum-free DME containing 0.15 μM of recombinant fibrillin-1 fragments, 0.15 μM of medium-purified fibrillin-1 molecules, or 0.15 μM of tissue-purified microfibrils for 15 min at 37°C. 4 nM of recombinant human TGFβ1 (Sigma-Aldrich) was used as a positive control. Human plasma fibronectin was used as an additional control (FC010; Chemicon Europe). Cells were washed twice with PBS, incubated with NET buffer supplemented with fresh proteinase inhibitors (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% NP-40, 2.5 mM EDTA, 100 μM NaVO, 1% aprotinin, 1 mM PMSF, and 1% leupeptin) for 30 min, and scraped from the tissue culture flask. Cell lysates were electrophoresed, and Western blots were undertaken using a Smad 2 antibody (AB3849; Chemicon Europe). Western blots were developed using electrochemiluminescence (GE Healthcare). film (BioMax; Kodak) was used to visualize positive bands. Each Western blot was stripped after use and reprobed with β-actin to ensure equal loadings of total protein (AC-15; Sigma-Aldrich). In some experiments, the effects of pretreating fibrillin-1 fragments with EDTA, elastase, or heparin were determined. Protein fragments were preincubated with 100 mM EDTA, pH 7.4, 100 μg/ml heparin (3,000 kD; Sigma-Aldrich), or 0.2 mg/ml porcine pancreatic elastase (Sigma-Aldrich) for 15 min before SDS-PAGE and Western blot analysis of Smad2 signaling. Signaling assays were also performed using the mouse osteoblast cell line 2T3 (), syndecan-4–null and wild-type mouse embryonic fibroblast cell lines (gift from M.J. Humphries, University of Manchester, Manchester, UK), and UMR-106 rat osteosarcoma cells (gift from T.J. Martin). Quantitative analysis was performed by densitometry with data normalized against β-actin. The densitometry values are plotted as a ratio of Smad2 signaling against corresponding β-actin. Data are represented as the mean of three repeated experiments and were statistically analyzed using unpaired tests (Prism 2.0 software; GraphPad). Error bars represent the SD of the three experiments. Results are statistically significant when the p-value is <0.05 (*, P < 0.05; **, P < 0.001; ***, P < 0.0001). HDFs were incubated with inhibitory antibodies or chemical inhibitors for 30 min at 37°C in 0.5 ml of serum-free DME before lysis and signaling assays (as described in the previous section). An anti-TGFβ1 mAb (mAb 240; R&D Systems) and an anti–human TGFβRII antibody (AF-241-NA; R&D Systems), which was designated RII in , were used at concentrations of 15 μg/ml. A chemical inhibitor of TGFβRI, [3-(pyridin-2-yl)-4- (4-quinonyl)]-1H-pyrazole (Merck Biosciences), which is designated as RI in , was used at a concentration of 20 μg/ml. The inhibitory integrin antibodies αv (17E6; Merck Biosciences), α5 (mAb 16), and β1 (mAb 13; gifts from M.J. Humphries) were used at concentrations of 20 μg/ml. Freshly prepared protease inhibitors were used at neutral pH at the following concentrations: aprotinin (serine) at 100 μM, leupeptin (cysteine; Sigma-Aldrich) at 100 μM, and a matrix metalloproteinase inhibitor (4-Abz-Gly-Pro--Leu--Ala-NH-OH; inhibits matrix metalloproteinases 1, 3, 8, and 9; Merck Biosciences) at 150 μM. Quantitative analysis was performed by densitometry with data normalized against β-actin. The densitometry values are plotted as a ratio of Smad2 signaling against corresponding β-actin. The amounts of active TGFβ1 present in HDF medium were determined using the TGFβ1 EMax Immunoassay kit (Promega). Recombinant fragments were added to HDFs in 0.5 ml of serum-free DME for 90 min at 37°C. The media were collected, and 200 μl was used in the EMax immunoassay, which was performed according to the manufacturer's instructions. For measurement of total (active + latent) TGF, the samples were acidified using HCl and were reneutralized before measurement using NaOH according to the ELISA manufacturer's instructions (Promega; ). TGFβ standard curves were undertaken for every assay. The standard curve is linear between 15.6 and 1,000 pg/ml of the TGFβ1 standard. All experiments were performed in triplicate and on the same microtitre plate. The data are represented as the mean values of one experiment. In some cases, other statistical methods were used: linear regression analysis was undertaken using SPSS 12.0 software (SPSS), and two-way analysis of variance (ANOVA) was performed followed by a posthoc multiple comparisons test using Tukey's test (SPSS 12.0 software). Furthermore, a protected two-tailed test was performed in conjunction with ANOVA in some cases. Recombinant proteins were added to HDFs in 0.5 ml of serum-free DME for 90 min at 37°C. Total RNA was isolated using the SV Total RNA Isolation kit (Promega). RNA was quantitated using an RNA/DNA calculator (GeneQuant Pro; GE Healthcare). cDNA was synthesized from the extracted RNA using RT-PCR, and the products were resolved using 2.5% ultrapure agarose gels (Invitrogen). Oligonucleotide primers for PCR were designed using Primer3 software. 0.5 mg of the fibrillin-1 fragment PF10 was bound to a nickel chelate affinity chromatography column using a chromatography system (AKTAprime; GE Healthcare). HDF cell layers that had been lysed with NET buffer containing fresh protease inhibitors (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% NP-40, 2.5 mM EDTA, 100 μM NaVO, 1% aprotinin, 1 mM PMSF, and 1% leupeptin) were passed over the column followed by a wash using 150 mM NaCl, 50 mM Tris-HCl, and 1 mM CaCl, pH 7.4. The bound proteins were subsequently eluted using a gradient of 1 M NaCl, 50 mM Tris-HCl, and 1 mM CaCl, pH 7.4. The procedure was repeated using the insoluble cell layer after treatment with 0.5 mg/ml collagenase in the presence of protease inhibitors (2 mM PMSF and 5 mM -ethylmaleimide) in 150 mM NaCl, 50 mM Tris-HCl, and 1 mM CaCl, pH 7.4, for 24 h. After eluting bound molecules, the affinity column was subjected to a final elution step using 500 mM imidazole, 150 mM NaCl, 50 mM Tris-HCl, pH 7.4, and 0.5 mM CaCl. All fractions were desalted using a HiTrap desalting column (GE Healthcare). All samples were reduced and alkylated as previously described (). To identify proteins bound to PF10, samples were analyzed using a mass spectrometer (Micro-Q-TOF; Waters) and the Mascot search engine (Matrix Science). The Mascot protein score is derived from the sum of the ion scores for each peptide detected from that protein. The ion score of a peptide, which reflects the probability of the observed peptide mass matching the mass of the peptide in the database, is expressed as a value, log(P), where P is the probability (). The SwissProt database was used. Peptide tolerance and mass spectrometry/mass spectrometry tolerance were set to ±0.3 D. In the final imidazole elution, PF10 was the only ECM sequence, which confirmed the affinity protocol. Solid-phase binding was performed as previously described (). In brief, 0–200 nM of soluble ligands were biotinylated, and flat-bottomed microtitre plates (Thermo Labsystems) were coated with the N-terminal fibrillin-1 fragment (PF1) at 200 nM in TBS (50 mM Tris-HCl, pH 7.4, and 0.1 M NaCl) overnight at 4°C. BSA blocking, washing, binding, and detection steps were subsequently performed. Soluble biotinylated protein dilutions of 0–200 nM for binding curves were used. All assays were performed in triplicate and were repeated at least twice to confirm the observed results. K values for dose-dependent interactions were calculated using nonlinear regression with one-site binding (hyperbola). All data are shown as mean values ± SEM. 1 μg of the fibrillin-1 fragment PF10 was incubated with 1 μg of recombinant latent TGFβ1 for 2 h at 37°C. The cross-linking agent BS (bis[sulfosuccinimidyl] suberate; Pierce Chemical Co.) was added at a concentration of 0.25 mM and incubated for 15 min at 4°C. The proteins were electrophoresed, and potential bands of interest were analyzed using mass spectrometry as outlined above (see Affinity chromatography and mass spectrometry). Blot overlays were performed essentially as previously described (). 25 μg of the fibrillin-1 fragment PF10 was electrophoresed using SDS-PAGE and was transferred onto nitrocellulose as described in the Smad2 signaling assays section. The membranes were blocked and incubated with 50 μg/ml of latent TGFβ1 (299-LT/CF; R&D Systems) at 4°C overnight. A primary antibody to latent TGFβ1 (AF-246-NA; R&D Systems) followed by an enzyme-conjugated secondary antibody was used to detect bound ligand. Blots were developed using enhanced chemiluminescence as described in the Smad2 signaling assays section. Kinetic binding analysis of latent TGFβ1 with fibrillin-1 was undertaken by surface plasmon resonance using a biosensor (BIAcore 3000; BIAcore). To investigate possible interactions between fibrillin-1 and the SLC, 1.8 μg/ml of latent TGFβ1 (299-LT/CF; R&D Systems) was immobilized onto a CM5 sensor chip in 10 mM acetic acid, pH 5.5. All subsequent binding experiments were performed in 10 mM Hepes, pH 7.4, 0.1 M NaCl, 1 mM CaCl, and 0.005% surfactant P20. 200 nM of fibrillin-1 fragments were applied to the sensor chip at a flow rate of 30 μl/min for 3 min. After 2.5-min dissociation, the chip was regenerated using 50 mM acetic acid for 30 s. The response value for each injection was calculated using the binding assay result wizard (BIAcore control software 3.2; BIAcore). As a positive control, an SLC antibody (AF-246-NA; R&D Systems) was passed over the chip. To analyze the binding of fibrillin-1 fragments to LTBP-1, a C-terminal fragment of LTBP-1 (designated CT LTBP-1; residues 1,008–1,394) was immobilized onto a CM5 sensor chip at 25 μg/ml in 50 mM sodium acetate, pH 5.2. 0–150 nM of the fibrillin-1 fragments PF1 and PF10 were applied to the sensor chip (15 μl/min) for 6 min and were left to dissociate for 10 min. Regeneration was performed in 10 mM Hepes, pH 7.4, 0.4 M NaCl, 1 mM CaCl, and 0.005% surfactant P20. The K for the PF1 interaction was calculated by plotting a saturation binding curve using the equilibrium response value at the top of the curve as described previously (). The PF1 interaction was performed three times, and the final K was calculated from a mean of these values. Increasing concentrations of PF10 (0–30 μM) were preincubated with 50 nM PF1 for 15 min before being applied to the sensor chip for 3 min (30 μl/min) and were left to dissociate for 10 min. CT LTBP-1 on the sensor surface was then regenerated. The maximum response was plotted against concentration using Prism 2.0 software (GraphPad). No binding response occurred between PF10 and CT LTBP-1, so it was possible to determine whether PF10 inhibits the interaction between PF1 and CT LTBP-1. The IC was calculated using nonlinear regression analysis (sigmoidal dose response; variable slope).

Mitofusins are mitochondrial outer membrane GTPases required for mitochondrial fusion (). Although yeast have one mitofusin, Fzo1 (), mammals contain two mitofusins, Mfn1 and Mfn2 (; ; ). In the absence of either Mfn1 or Mfn2, cells have greatly reduced levels of mitochondrial fusion, and the imbalance of fusion and fission events leads to mitochondrial fragmentation (). In the absence of both Mfn1 and Mfn2, no mitochondrial fusion can occur, leading to severe mitochondrial and cellular dysfunction (). Moreover, mitochondrial dynamics play an important role in apoptosis (), and maintenance of mitochondrial fusion has been linked to protection against apoptosis (; ; ). Mutations in Mfn2 cause Charcot-Marie-Tooth disease (CMT) type 2A, an autosomal dominant peripheral neuropathy (). Most types of CMT disease involve Schwann cell dysfunction, resulting in the demyelination of peripheral nerves. However, CMT2A is an axonal form in which the axons of the longest sensory and motor nerves are selectively affected (). There is currently no effective treatment for this disease. Interestingly, another neurodegenerative disease, dominant optic atrophy, is caused by mutations in OPA1 (; ), a mitochondrial intermembrane space protein that is also necessary for mitochondrial fusion. The sensitivity of neurons to mutations in Mfn2 and OPA1 suggests that such cells are particularly dependent on mitochondrial dynamics, which likely impacts the recruitment of mitochondria to extended neuronal processes (). Indeed, the disruption of mitochondrial dynamics has been experimentally linked to neuronal dysfunction (; ; ; ). Several issues regarding mitofusin function and its relation to neurodegenerative disease remain poorly understood. First, it is unclear to what extent there is functional interplay between Mfn1 and Mfn2 during mitochondrial fusion. In experiments with Mfn1- or Mfn2-null cells, either mitofusin can functionally replace the other, indicating functional redundancy (, ). However, some studies suggest distinct pathways or mechanisms for Mfn1 versus Mfn2 (; ). OPA1 action has been reported to depend on Mfn1 but not Mfn2 (). An in vitro study indicates that overexpressed Mfn1 is more effective than Mfn2 in tethering mitochondria, an effect that is correlated with a higher rate of GTP hydrolysis for Mfn1 (). Second, Mfn1 and Mfn2 have been shown to physically associate with each other (; ), but the functional significance of such heterooligomeric complexes is poorly understood. Finally, it is unknown why Mfn2 mutations in CMT2A cause such highly cell type–specific defects. Patients with CMT2A show deficits in the longest sensory and motor peripheral nerves, with a subset showing additional degeneration in the optic nerve (). The length-dependent degeneration of peripheral nerves likely reflects an inherent challenge of neurons to supply functional mitochondria to the nerve terminals, but it remains unclear why primarily the peripheral and optic nerves are affected. In this study, we have analyzed Mfn2 disease alleles that cause CMT2A. We find that most of these mutants are not functional for fusion when allowed to form only homotypic complexes. However, these Mfn2 mutants can be complemented through the formation of heterotypic complexes with wild-type Mfn1. These results emphasize the close interplay between Mfn1 and Mfn2 in the mitochondrial fusion reaction, demonstrate the functional importance of Mfn1–Mfn2 heterooligomeric complexes, and provide insights into the pathogenesis of Mfn2-dependent neuropathy. We have previously generated mouse embryonic fibroblast (MEF) cell lines with null mutations in both Mfn1 and Mfn2 (double Mfn-null cells; ; ). These cell lines enable straightforward structure-function analysis of mouse mitofusins. Human and mouse Mfn2 are 95% identical, and all of the residues that were found mutated in the original CMT2A study () are conserved in mouse Mfn2. In the present study, we introduced nine of the originally reported point mutations into mouse Mfn2; these include mutations occurring immediately before the GTPase domain (V69F and L76P), within the GTPase domain (R94Q, R94W, T105M, P251A, R274Q, and R280H), and in a C-terminal heptad repeat region (W740S; ). By expressing these disease alleles in wild-type and double Mfn-null MEFs, we could assess their subcellular localization, effects on mitochondrial morphology, and ability to mediate mitochondrial fusion. Mock-infected wild-type MEFs have a range of mitochondrial profiles, but the vast majority of cells show considerable amounts of tubular mitochondria ( and Fig. S1 A, available at ). The expression of wild-type Mfn2 or the GTPase mutant Mfn2 by retroviral transduction did not affect mitochondrial morphology. We found that all of the CMT2A mutants properly localized to mitochondria as determined by immunofluorescence. However, seven of the nine CMT2A alleles (all except Mfn2 and Mfn2) caused substantial mitochondrial aggregation when cells were infected at a high multiplicity of infection (Fig. S2). At low infection rates, most infected cells have only one proviral copy and express about fourfold Mfn2 compared with endogenous Mfn2 in wild-type cells (Fig. S3). Under these conditions, only Mfn2, Mfn2, and Mfn2 caused high levels of mitochondrial aggregation (). In contrast, such mitochondrial aggregation was not found in cells expressing wild-type Mfn2 and was found only in a few cells expressing Mfn2. This mitochondrial aggregation phenotype may reflect the aberration of Mfn2 function by CMT2A mutations; however, the effect is clearly dosage dependent and is not observed at physiological expression levels (see ). Therefore, its relevance to CMT2A disease remains to be determined. To evaluate the Mfn2 CMT2A alleles for mitochondrial fusion activity, we expressed them in double Mfn-null cells. Cells lacking mitofusins are fully deficient for mitochondrial fusion and show completely fragmented mitochondrial morphology (). The expression of wild-type Mfn2 restored mitochondrial fusion, resulting in tubular mitochondrial morphology (, B and C; and Fig. S1 B). In contrast, the GTPase mutant Mfn2 behaved as a complete loss of function allele, showing no ability to restore mitochondrial tubules. The CMT2A mutants Mfn2, Mfn2, Mfn2, Mfn2, and Mfn2 are similarly unable to promote mitochondrial tubules in double Mfn-null cells. In contrast, cells expressing Mfn2, Mfn2, Mfn2, or Mfn2 showed a considerable restoration of mitochondrial tubules. Therefore, more than half of the CMT2A mutants are nonfunctional. To definitively evaluate the fusion activity of CMT2A alleles, we tested them in a polyethylene glycol (PEG) mitochondrial fusion assay. In this assay, double Mfn-null cells containing either mitochondrially targeted EGFP or mito-DsRed were each infected with retrovirus expressing a CMT2A allele. Hybrids between the two cell lines were scored for mitochondrial fusion. Cell hybrids that formed between double Mfn-null cells or cells expressing Mfn2 never showed mitochondrial fusion (). In contrast, the expression of wild-type Mfn2 resulted in extensive mitochondrial fusion: 75% of the cell hybrids exhibited a complete overlay of EGFP and DsRed (scored as full fusion) or a nearly complete overlay with some singly labeled mitochondria remaining (scored as extensive fusion). 20% of these hybrids had no colabeled mitochondria (scored as no fusion) and invariably had fragmented mitochondria. These hybrids likely arose from uninfected cells. When clonal infected cell lines were used (), all cell hybrids showed extensive mitochondrial fusion. We found excellent agreement between the ability of a CMT2A allele to restore mitochondrial tubules to double Mfn-null cells and their fusion activity in the PEG assay. The Mfn2 CMT2A alleles Mfn2, Mfn2, Mfn2, and Mfn2 induced fluorophore mixing as efficiently as wild-type Mfn2, indicating that they are highly functional. In contrast, mutants Mfn2, Mfn2, Mfn2, Mfn2, and Mfn2 were all completely deficient for mitochondrial fusion. Interestingly, the five nonfunctional alleles are all in positions that are conserved between Mfn1 and Mfn2. Three of the four functional alleles are in nonconserved positions. Our PEG fusion assays showed that Mfn2, along with four other CMT2A alleles, has no mitochondrial fusion activity in double Mfn-null cells. This allele is particularly interesting because position 94 is the most commonly mutated residue found in CMT2A. Multiple clinical studies have found familial or de novo mutations of residue 94 to either Q or W (, ; ; ; ). To definitively study the in vivo properties of this allele, we used homologous recombination to place the R94Q mutation into the endogenous mouse Mfn2 locus in embryonic stem (ES) cells (). For positive selection, the targeting construct contained a neomycin expression cassette flanked by sites. After the generation of mice containing the knockin allele, Cre-mediated recombination was used to excise the neomycin cassette in vivo, resulting in an locus containing the R94Q mutation and a short scar located in the adjacent intron (). We mated mice heterozygous for the Mfn2 allele, and homozygous embryos were used to derive Mfn2 homozygous MEF cell lines. Our molecular analyses indicate that these cell lines express no wild-type Mfn2 while expressing endogenous levels of Mfn2 (). To confirm the expression of Mfn2 in these cell lines, we used RT-PCR to analyze Mfn2 RNA transcripts. We amplified exon 5 (which encodes residue 94) and the adjoining sequences of Mfn2 cDNA by PCR. The presence of the R94Q mutation within the amplified cDNA fragment was diagnosed by digestion with the restriction enzyme MspA1I, which cuts uniquely at a site introduced by the R94Q mutation. As expected, the cDNA fragment was amplified from cDNA of wild-type and Mfn2 homozygous cells but not Mfn2-null cells (). The cDNA from wild-type cells is completely resistant to MspA1I digestion, whereas the cDNA from Mfn2 homozygous cells was completely digested by MspA1I, demonstrating that all Mfn2 transcripts contain the R94Q mutation. Having confirmed mRNA expression of the mutant allele, we next confirmed protein expression. Immunoblot analysis indicated that endogenous levels of Mfn1 and Mfn2 are present in wild-type and Mfn2 homozygous cell lines (). Given that Mfn2 has no fusion activity in double Mfn-null cells ( and ), we expected Mfn2 homozygous cells to have fragmented mitochondria similar to those found in Mfn2-null cells (, ). Surprisingly, the scoring of mitochondrial profiles indicated that most Mfn2 homozygous cells have predominantly tubular mitochondria; this is in striking contrast to Mfn2-null cells, which have extensive mitochondrial fragmentation (). In addition, we did not find any mitochondrial aggregation in the Mfn2 homozygous cell line. Therefore, although Mfn2 behaves as a null allele when expressed in double Mfn-null cells, it is clearly highly functional in our homozygous knockin cells. In evaluating these results, it is important to consider the total complement of mitofusins in each cellular context because Mfn1 and Mfn2 can form both homooligomeric (Mfn1–Mfn1 or Mfn2–Mfn2) and heterooligomeric (Mfn1–Mfn2) complexes (; ). When Mfn2 is expressed in double Mfn-null cells, only Mfn2–Mfn2 homooligomeric complexes can be formed, and such complexes are clearly inactive for mitochondrial fusion. In Mfn2 homozygous knockin cells, endogenous Mfn1 is still present (). Therefore, three possible complexes can be formed: Mfn1–Mfn1, Mfn1–Mfn2, and Mfn2–Mfn2 (). The phenotype of Mfn2-null cells (which contain only Mfn1–Mfn1 homooligomeric complexes) indicates that endogenous levels of Mfn1–Mfn1 complexes alone are not sufficient to promote tubular mitochondria. Given that Mfn2–Mfn2 complexes are nonfunctional (), these results strongly suggest that Mfn2 can cooperate with Mfn1 to form Mfn1–Mfn2 complexes capable of promoting fusion. If this model of complementation is correct, Mfn2 should be able to physically associate with wild-type Mfn1. We tested whether the Mfn2 CMT2A mutants could coimmunoprecipitate with wild-type Mfn1 and Mfn2. In MEFs, all of the Mfn2 CMT2A mutants associated with Mfn1 at normal levels with the exception of Mfn2, which showed lower levels (). Similarly, the Mfn2 CMT2A mutants associated with Mfn2, although at slightly reduced levels compared with wild-type Mfn2. Again, Mfn2 had low binding. It should be noted that when analogous immunoprecipitation experiments were performed in transfected 293T cells, the reduction in Mfn2 binding was subtle (unpublished data). Therefore, although Mfn2 has reduced binding to wild-type Mfn1 and Mfn2, this defect is not observed at high expression levels. The engineered GTPase mutant Mfn2, which interacted strongly with Mfn1, interacted poorly with Mfn2. These results suggest that the mutant Mfn2 molecules can interact with wild-type Mfn1 and Mfn2 and can potentially participate in or modify the fusion reaction. To learn more about the complementation of Mfn1 and Mfn2 and whether this is a unique property of the Mfn2 allele, we tested all of the nonfunctional Mfn2 CMT2A alleles for complementation with wild-type Mfn1 and Mfn2. We expressed alleles Mfn2, Mfn2, Mfn2, Mfn2, and Mfn2 in either Mfn2- or Mfn1-null cells and scored mitochondrial profiles. Most Mfn2-null cells have fragmented mitochondrial morphology, with only ∼13% of the cells having short mitochondrial tubules. The expression of wild-type Mfn2 in these cells restores normal tubular mitochondrial morphology (). Remarkably, the expression of each of the five CMT2A alleles into Mfn2-null cells resulted in extensive mitochondrial tubulation. The GTPase mutant Mfn2 was also able to induce mitochondrial tubulation, although its effect was considerably weaker than that of the CMT2A alleles. Because Mfn2-null cells contain Mfn1, the expression of CMT2A alleles in Mfn2-null cells results in the formation of three possible complexes: Mfn1–Mfn1, Mfn1–Mfn2, and Mfn2–Mfn2 (). These results strongly support and generalize our interpretation of the Mfn2 homozygous knockin cell line: Mfn2 disease alleles can cooperate with Mfn1 to promote fusion activity. This activity is likely mediated by Mfn1–Mfn2 heterooligomers. Because even mutant Mfn2 shows a low level of complementation with Mfn1, Mfn2 need not have GTPase activity to cooperate with Mfn1. In contrast, the expression of mutants Mfn2, Mfn2, Mfn2, Mfn2, and Mfn2 in Mfn1-null cells did not induce tubulation (). Mfn1-null cells expressing these alleles had extensively fragmented mitochondria. In this experiment, only Mfn2 complexes can be formed: Mfn2–Mfn2, Mfn2–Mfn2, and Mfn2–Mfn2 (). Therefore, in contrast to Mfn1–Mfn2 complexes, Mfn2–Mfn2 complexes do not appear to be competent for fusion. The aforementioned experiments demonstrate that Mfn1 can complement Mfn2 CMT2A alleles. By the nature of the experiment, it is impossible to know whether the complementation is occurring on the same mitochondria (in cis), between adjacent mitochondria (in trans), or both. To test whether the nonfunctional Mfn2 mutants can support fusion with wild-type mitochondria in trans, we returned to the PEG cell hybrid assay for mitochondrial fusion. In this assay, mitochondria from double Mfn-null cells cannot fuse with mitochondria from wild-type cells, indicating a requirement for mitofusins on adjacent mitochondria (; ). We expressed Mfn2 alleles in double Mfn-null cells and assessed mitochondrial fusion in cell hybrids with wild-type cells. In this experimental scheme, Mfn2–Mfn2 complexes present on one set of mitochondria are tested for fusion with mitochondria containing a full complement of wild-type mitofusin complexes (Mfn1–Mfn1, Mfn2–Mfn2, and Mfn1–Mfn2 complexes). As expected, when double Mfn-null cells expressing wild-type Mfn2 were fused with wild-type cells, we found extensive colabeling of mitochondria (). Moreover, the Mfn2 CMT2A alleles Mfn2, Mfn2, Mfn2, and Mfn2 induce readily detectable but moderate levels of fusion that are lower than those of wild-type Mfn2 but are much more than those of Mfn2 (). However, the Mfn2 allele allows essentially no mitochondrial fusion. These results indicate that most Mfn2 CMT2A mutants can function in trans with wild-type mitofusin complexes. To determine whether this complementation is caused by interactions with wild-type Mfn1–Mfn1 or Mfn2–Mfn2 complexes, we next tested the Mfn2 CMT2A alleles in mitochondrial fusion assays with Mfn2-null and Mfn1-null cells. Mfn2 mutants Mfn2, Mfn2, Mfn2, and Mfn2 promoted moderate levels of mitochondrial fusion in cell hybrids with Mfn2-null cells (). In contrast, the same mutants induced no mitochondrial fusion in hybrids with Mfn1-null cells (). These results demonstrate that most Mfn2 alleles can promote fusion when exposed to membranes containing Mfn1 but not Mfn2. As expected from its failure to promote fusion with wild-type mitochondria, Mfn2 showed no fusion activity with either the Mfn2- or Mfn1-null cells. Previous immunoprecipitation studies indicated that Mfn1 and Mfn2 form heterooligomeric complexes (; ). However, most functional studies have focused on Mfn1 or Mfn2 in isolation, and, therefore, we have little information on the functional importance of the heterooligomeric complex. The only direct demonstration that this complex is functional comes from the observation that cell hybrids between Mfn1- and Mfn2-null cells show low levels of mitochondrial fusion, suggesting that Mfn1–Mfn2 heterotypic complexes formed in trans have fusion activities that are roughly comparable with homooligomeric Mfn1 or Mfn2 complexes alone (). Our current study of Mfn2 disease alleles reveals an intimate interplay between Mfn1 and Mfn2 in mediating mitochondrial fusion. A subset of Mfn2 disease alleles lack mitochondrial fusion activity in isolation but show substantial fusion activity in the presence of Mfn1. In addition, PEG fusion assays () indicate that this cooperation between Mfn1 and mutant Mfn2 at least partially occurs through interactions in trans. Such close physical and functional interactions between Mfn1 and Mfn2 support the view that they have similar biochemical functions during mitochondrial membrane fusion. These results highlight the importance of heterooligomeric Mfn1–Mfn2 complexes in the control of mitochondrial dynamics. Our study greatly extends a different type of complementation demonstrated in the yeast mitofusin Fzo1p. Fzo1p demonstrates strong complementation between specific pairs of null alleles, resulting in the restoration of mitochondrial tubules (). For example, an mutant containing a GTPase mutation can cooperate with an mutant containing a heptad repeat mutation to promote mitochondrial fusion. Such complementation reflects the oligomeric nature of mitofusin complexes and indicates that each subunit of the oligomer need not be fully functional to provide function to the complex. However, this previous study () was limited to Fzo1 homooligomeric complexes, unlike the heterooligomeric complexes studied here. Indeed, we have not been able to demonstrate a similar type of complementation in Mfn1 or Mfn2 homooligomeric complexes (unpublished data). Our results reveal some functional heterogeneity in Mfn2 mutants that underlie CMT2A disease. The Mfn2 allele behaved somewhat differently from the other nonfunctional alleles. Mfn2, like the other nonfunctional alleles, could be complemented by wild-type Mfn1. However, it showed reduced physical interactions with Mfn1 and did not show complementation with Mfn1 in trans. Presumably, Mfn2 can be complemented by Mfn1 in cis but not in trans. Our results have important implications for understanding the pathogenesis of CMT2A, especially because four of the five nonfunctional mutant alleles described in this study (Mfn2, Mfn2, Mfn2, and Mfn2) are among the most commonly identified Mfn2 mutations (, ; ; ; ; ). In contrast to the broad expression pattern of Mfn2, one of the remarkable features of CMT2A disease is its apparent cell type specificity. In most patients, the clinical features are restricted to the motor and sensory neurons of the peripheral nervous system. In a subset of patients (designated as hereditary motor and sensory neuropathy type VI), the optic nerve is additionally affected (). A recent study has suggested possible involvement of the central nervous system (). This clinical picture suggests that most cells in CMT2A patients likely have only mild perturbations in mitochondrial dynamics. Moreover, in a typical patient, only the longest peripheral sensory and motor neurons are affected. This length dependence suggests that even in the peripheral nervous system, the defects in mitochondrial dynamics are not catastrophic because only the neurons with the highest demands for precise control of mitochondrial fusion are damaged. Our studies of the Mfn2 knockin mice are ongoing, but initial observations support the conclusion that CMT2A disease results from a mild perturbation in mitochondrial dynamics. Thus far, we have not observed a neurological phenotype in the heterozygous knockin mice. The lack of an obvious peripheral neuropathy in these mice may reflect the fact that motor neurons in mice are much shorter than in humans, where their extreme length likely places more stringent requirements on the precise regulation of mitochondrial fusion. Although Mfn2-null animals die in utero, Mfn2 homozygous animals are born live and die at ∼3 wk of age. The much milder phenotype of Mfn2 homozygous animals compared with Mfn2-null animals further supports our conclusion that Mfn2 can be partially complemented by endogenous Mfn1. Mfn2 homozygous animals have severe movement defects (unpublished data), and we are currently analyzing the basis for this phenotype. In considering the effects of Mfn2 mutations, our results indicate that the full complement of mitofusins in any given cell type is the most relevant parameter in determining the dysfunction of mitochondrial fusion. This concept is clearly illustrated in our analysis of Mfn2 CMT2A mutations in MEFs. When Mfn2 is expressed in double Mfn-null cells, it is completely deficient for fusion activity, indicating that homooligomeric Mfn2 complexes are nonfunctional. In contrast, MEFs containing homozygous Mfn2 knockin mutations show only mild defects in mitochondrial morphology, a phenotype that is quite different from the extensive mitochondrial fragmentation observed in Mfn2-null MEFs. This observation indicates that in the presence of endogenous wild-type Mfn1, Mfn2 is actually highly functional. By expressing Mfn2 and other mutant alleles in Mfn1-null versus Mfn2-null cells, we found that wild-type Mfn1 but not Mfn2 can cooperate with mutant Mfn2 to promote mitochondrial fusion. These results suggest that the widespread expression pattern of Mfn1 (; ) protects mitochondrial dynamics in most cells in CMT2A patients carrying nonfunctional alleles of Mfn2. CMT2A is an autosomal dominant disease, with patients carrying one mutant and one wild-type allele of Mfn2. In cell types that express Mfn1, Mfn1 homooligomeric complexes would be normal, and Mfn1–Mfn2 heterooligomeric complexes would also be largely normal as a result of the cooperation between Mfn1 and mutant Mfn2 (). For Mfn2 homooligomeric complexes, Mfn2–Mfn2 complexes would be functional, whereas Mfn2–Mfn2 and Mfn2–Mfn2 complexes would be nonfunctional. Therefore, of the three classes of mitofusin complexes, only a subset of one class is nonfunctional, resulting in mild mitochondrial fusion defects in most cells. In cell types with low or no Mfn1 expression, the full complement of mitofusin complexes consists primarily of Mfn2 homotypic complexes. In relative terms, such cells would experience a severe loss of mitochondrial fusion because the majority of mitofusin complexes (Mfn2–Mfn2 and Mfn2–Mfn2) lack mitochondrial fusion activity. Therefore, we propose that in CMT2A disease, the widespread expression pattern of Mfn1 serves to protect mitochondrial fusion in most cells through heterooligomeric complex formation with mutant Mfn2. Peripheral nerves may contain little or no Mfn1 expression to compensate for mutant Mfn2. The resulting defects in mitochondrial dynamics coupled with the extreme length of these neurons lead to neuronal dysfunction and axon degeneration. The mitofusin 7xMyc and 3xHA constructs were described previously (). The CMT2A point mutations were introduced to Mfn2-7xMyc in pcDNA3.1 by PCR with primers encoding the mutations. After cloning, the entire amplified region was verified by sequencing. The mutant cDNAs were then cloned into the retroviral construct pCLBW, and viral supernatant was produced and collected as described previously (). Immunofluorescence against Mfn2-7xMyc was performed as described previously (). In brief, cells were grown on poly--lysine–treated coverslips, fixed in formalin, permeabilized with 0.1% Triton X-100 in PBS, and blocked with 5% bovine calf serum in PBS. The 9E10 primary antibody was detected with a Cy3-labeled secondary antibody. Coverslips were mounted with GelMount and imaged with a plan NeoFluar 63× NA 1.25 oil immersion objective (Carl Zeiss MicroImaging, Inc.) on a laser-scanning confocal microscope (model 410; Carl Zeiss MicroImaging, Inc.). Images were acquired with LSM software (version 1; Carl Zeiss MicroImaging, Inc.) and pseudocolored in Photoshop CS (Adobe). Mitochondria were visualized by mitochondrially targeted GFP or DsRed as previously described (). In other cases, mitochondria were stained using 150 nM MitoTracker red CMXRos (Invitrogen) and postfixed in acetone. PEG fusion assays were performed in the presence of cycloheximide as described previously (, ). Cell hybrids were fixed 7 h after PEG treatment. The mitochondrial GFP signal was enhanced by incubation with an anti-GFP antibody conjugated to AlexaFluor488 (Invitrogen). The two arms of the targeting construct were derived from Mfn2 genomic sequence (129/SvJ background) and subcloned into the targeting vector pPGKneobpAlox2PGKDTA. Before subcloning of the left arm, the R94Q mutation was engineered into exon 5 by PCR. The targeting construct was verified by DNA sequencing. The linearized targeting construct was electroporated into low-passage 129/SvEv ES cells as described previously (). Correctly targeted ES clones were identified by PCR using the primer sets A and B depicted in . Chimeric mice were generated by the injection of ES cells into C57BL/6 blastocysts. After confirmation of germline transmission, the floxed neomycin cassette was removed by mating the knockin mice with the EIIA-cre deletor line (). Heterozygous knockin animals were mated, and MEFs were derived from day 10.5 embryos as described previously (). Homozygous embryos were identified by PCR genotyping of extraembryonic membranes. Wild-type, Mfn1-null, Mfn2-null, and Mfn2–Mfn2 MEFs were cultured in DME containing 10% bovine calf serum, 1 mM -glutamine, and penicillin/streptomycin. Double Mfn-null MEFs were cultured with 10% FCS in place of bovine calf serum. MEFs were resuspended directly in 800 μl STAT-60 (IsoTex Diagnostics, Inc.), and RNA was isolated according to the manufacturer's instructions. cDNA was generated by first-strand synthesis on total RNA using oligo(dT) and Superscript II RT (Invitrogen). A cDNA fragment containing exon 5 was subsequently amplified (primers 5′-GGGGCCTACATCCAAGAGAG-3′ and 5′-GCAGAACTTTGTCCCAGAGC-3′). This product was digested overnight at 37°C with MspA1I. MEF cell lysates were prepared from confluent 6-cm plates. For protein lysates, cells were washed once with PBS and resuspended in 400 μl lysis buffer (150 mM NaCl, 50 mM Tris, pH 8.0, 4 mM MgCl, 1% Triton X-100, and protease inhibitor cocktail [Roche]). Nuclei were removed by centrifugation, and postnuclear lysates were quantified with a protein assay (Bio-Rad Laboratories). 12 μg of each sample was separated by an 8% SDS-PAGE and immunoblotted with an anti-Mfn2 antibody (Sigma-Aldrich), an anti-Mfn1 antibody (), or anti–β-actin as a loading control. Mitofusin antibodies (diluted 1:1,000) were detected by HRP-conjugated secondary antibodies and ECL detection reagents (GE Healthcare). Double Mfn-null cells were infected with retrovirus encoding Mfn1-3xHA or Mfn2-3xHA. Infected cells were selected by culture in media containing bovine calf serum, which does not support uninfected double Mfn-null cells. Each cell line was subsequently infected with virus encoding Mfn2-7xMyc constructs or Drp1-7xMyc. Postnuclear lysates were generated as described above for MEFs (5–6 d after infection) and were immunoprecipitated with 9E10 antibody coupled to protein A–Sepharose beads. HA.11 (Covance) and 9E10 antibodies were used for immunoblotting. Fig. S1 shows the mitochondrial profiles of MEFs expressing Mfn2 CMT2A alleles; this data is summarized in C. Fig. S2 shows mitochondrial aggregation in MEFs highly overexpressing Mfn2 CMT2A alleles. Fig. S3 shows that at low infection rates, recombinant Mfn2 is present at approximately fourfold the level of endogenous Mfn2. Online supplemental material is available at .

UV radiation (UVR) is a potent carcinogen that acts directly on DNA. Accumulated lifetime exposure to UVR is the key environmental risk factor for development of nonmelanoma skin cancers (NMSCs), such as basal and squamous cell carcinomas (). The cellular response to DNA damage is centered on p53, a transcription factor that exerts its tumor-suppressive function by inducing cell cycle arrest, cell senescence, or apoptosis (). The importance of p53 in the prevention of UVR-induced skin cancer is underscored by the observation that after chronic UV irradiation, p53-deficient mice exhibit a vastly increased incidence and reduced latency of NMSC compared with wild-type (wt) animals (). Programmed cell death initiated by UVR is required to remove precancerous keratinocytes, yielding so-called sunburn cells (SBCs). Their formation appears to represent a crucial tumor-suppressive response because they arise from the cell type of origin for NMSC and their development requires functional p53 (; ). Two distinct signaling pathways activate the caspases that mediate apoptosis (). The extrinsic pathway is initiated by “death receptors” (several members of the TNF-R family) and proceeds via caspase-8 and its adaptor FADD (Fas-associated death domain), whereas the intrinsic or mitochondrial pathway is regulated by the interacting pro- and antiapoptotic members of the Bcl-2 protein family and leads, after mitochondrial outer membrane permeabilization, to caspase-9 activation. Although UVR-induced apoptosis clearly involves the downstream effector caspases (), the relative roles of the extrinsic and intrinsic pathways are controversial. The extrinsic pathway is favored by evidence that membrane localization of the death receptors Fas (also called APO-1 or CD95) and TRAIL-R is up-regulated in a p53-dependent manner after UVR exposure () and that UV-irradiated (FasL-deficient) mice exhibit reduced SBC formation (). On the other hand, UV-irradiated mice overexpressing Bcl-2 in keratinocytes exhibited fewer SBCs and more skin tumors than control animals (). In the intrinsic path to cell death, the key initiators are the BH3-only members of the Bcl-2 family (). Different death stimuli activate distinct subsets of these death ligands. For example, Noxa and Puma are up-regulated during p53-mediated cell killing, and their genes are direct p53 targets (; ; ). Gene-targeting experiments in mice have demonstrated that Puma plays a major and Noxa a more restricted role in p53-mediated apoptosis (; ; ). lymphoid and myeloid cells were remarkably resistant to genotoxic damage (; ; ). The role of Noxa has been less clear, as its loss gave MEFs only slight, albeit significant, resistance against etoposide and did not affect any apoptotic responses in lymphoid cells (; ). Here, we have sought to delineate the pathways to cell death elicited by UV irradiation of primary MEFs, the MEFs rendered more sensitive to genotoxic damage by transformation with the adenovirus E1A and oncogenes () and keratinocytes within whole mouse skin. We demonstrate that the Bcl-2 family regulates not only the death induced by p53 but also a p53-independent pathway in response to UVR. By exploiting MEFs that lack different BH3-only proteins, we show that in primary cells, both Noxa and Puma contribute to UVR-induced apoptosis. Unexpectedly, only Noxa plays a major role in the transformed MEFs and keratinocytes, where the , but not the , gene proved to be induced. To examine whether UV-irradiated MEFs die by apoptosis, several well-established parameters were assessed in both primary and E1A/ transformed MEFs ( and Fig. S1, available at ). Morphological and flow cytometric examination indicated that relatively few primary MEFs died when exposed to low-dose UVR (5–50 J/m). Although they exhibited some morphological signs of apoptosis at these doses, at higher doses (e.g., 200 J/m), they appeared to die predominantly by a nonapoptotic mechanism, as indicated by the prevalence of large, vacuolated cells at 6 h after irradiation (). In contrast, at both low and high doses, the MEFs sensitized to genotoxic damage by oncogenic transformation () exhibited classical morphological signs of apoptosis, such as chromatin condensation and membrane blebbing (), as well as hallmark biochemical features. Notably, caspase activity was evident from the generation of the cleaved p12 and p85 fragments of the canonical caspase substrates ICAD (inhibitor of caspase-activated DNase) and PARP (poly ADP ribose polymerase), respectively (Fig. S1, A and B). The resulting release of the active DNase CAD from its inhibitor ICAD presumably accounts for the characteristic DNA fragmentation (Fig. S1 C) and the cell population with DNA content of <2C (Fig. S1 D). Consistent with the different fates of primary and E1A/ transformed MEFs, active caspase-3 increased in E1A/ MEFs but not in primary MEFs (). Thus, although E1A/ transformed MEFs are highly sensitive to UVR-induced apoptosis, primary MEFs are more resistant and die at high doses, predominantly by a nonapoptotic mechanism. To investigate whether the extrinsic or intrinsic signaling pathway drives UVR-induced apoptosis, two independently generated E1A/ras transformed wt MEF lines were transfected with expression vectors that encode Bcl-2 or proteins that inhibit the death receptor pathway: a well-characterized dominant-negative mutant of FADD (FADD-DN; ) or the caspase-8 inhibitor CrmA (). Both inhibitors were expressed () and functional, as they blocked the death of MEFs stimulated with FasL (not depicted). Nevertheless, neither FADD-DN nor CrmA notably inhibited UVR-induced apoptosis (). In striking contrast, Bcl-2 overexpression () almost completely ablated UVR-induced cell death (). We conclude that the UVR-induced apoptosis of the transformed MEFs proceeds through the Bcl-2–regulated pathway. The strong protection conveyed by Bcl-2 () prompted us to investigate which of its BH3-only antagonists drives this process. Primary MEFs derived from wt mice or those lacking Bim, Bad, Noxa, or Puma were exposed to graded doses of UVR, and cell viability was monitored 24 h later. Wt primary MEFs exhibited dose-dependent rates of death, which was just as extensive in MEFs lacking Bad or Bim (Fig. S2 A, available at ). In contrast, the cells lacking p53, Noxa, or Puma all displayed greater viability than wt controls at doses up to 50 J/m (). Because many BH3-only proteins have partially redundant functions (), we also generated primary MEFs from mice lacking both Noxa and Puma, or Noxa and Bim. MEFs were significantly more resistant than cells at doses of 100 and 200 J/m (; P < 0.025), although the slight increases at lower doses were not significant. MEFs were at least as refractory to UVR as the MEFs (), indicating that Noxa and Puma account for all p53-mediated killing of UV-irradiated primary MEFs. primary MEFs were no more refractory than those lacking Noxa alone (Fig. S2 B). Thus, at doses up to 50 J/m, the UVR-induced apoptosis of primary MEFs is driven mainly through p53 and its targets Noxa and Puma, whereas Bad and Bim are dispensable. , , , and MEFs () probably reflects nonapoptotic cell death. , , , and E1A/ lines 1–3). Because the E1A oncoprotein sensitizes MEFs to apoptosis by inactivating the RB tumor suppressor, we verified by intracellular FACS analysis that all transformed lines of every genotype expressed similar levels of E1A (Fig. S3 A, available at ). When the cell lines were subjected to UVR at 10–50 J/m and viability analyzed 24 h later, the p53-deficient E1A/ MEFs were consistently the most resistant, and Noxa-deficient MEFs demonstrated a marked, albeit lower, level of resistance ( and Fig. S3). In contrast, loss of Bim conveyed no protection (). primary MEFs (), their transformed derivatives behaved like wt MEFs except for slightly greater viability at 10 and 20 J/m (; P < 0.025). To determine whether these protective effects persisted, viability was also analyzed at 24, 48, and 72 h after an intermediate dose of UVR (15 J/m). and cells died as rapidly as wt cells (). Indeed, even most of the cells deficient in p53 or Noxa had succumbed by 72 h after irradiation; only 15% of them remained viable, although this still greatly exceeded the 1% viability of the wt cells ( ii). Thus, Noxa is the major initiator of UVR-induced apoptosis downstream of p53 in E1A/ transformed MEFs. transformed MEFs were more sensitive to UVR than the corresponding or Bcl-2 overexpressing cells suggested that Noxa is not the sole initiator of UVR-induced apoptosis. and embryos. transformed cells appeared to survive slightly better than MEFs at all doses tested ( i) and had a small but significant advantage at 24 h after stimulation with 15 J/m ( ii; P < 0.05). transformed cells were no more resistant to UVR than counterparts deficient in Noxa alone (, iii and iv). To clarify why loss of either Noxa or Puma protected the primary MEFs () but only loss of Noxa substantially protected the transformed MEFs (), we quantified and transcripts by PCR at various times after treatment with UVR, or etoposide as a positive control, and normalized the values to unstimulated controls. In transformed MEFs, UVR induced mRNA 2.3-fold but, surprisingly, slightly reduced the mRNA level (). MEFs (, top). In contrast, etoposide up-regulated both and , albeit to a greater extent (∼11- vs. 4.5-fold). The absence of mRNA at 6 h after etoposide treatment indicates that this transcript is rapidly induced but then degraded (). In primary wt MEFs, UVR, like etoposide, induced both and mRNA (). Interestingly, mRNA was up-regulated slightly more by a high dose (100 J/m) than a low dose (25 J/m) of UVR (, top right; 2.5- vs. 1.8-fold maximum at 2 h), whereas mRNA exhibited the reciprocal pattern (, top left; 1.4- vs. 2.2-fold maximum at 4 h). These results show that UVR increases mRNA levels in primary but not in E1A/ transformed MEFs and that UVR dose affects the relative abundance of and transcripts in primary cells. In hemopoietic progenitor cells, DNA damage induces the transcriptional repressor Slug, which in turn ablates the activation of by p53 (). To explore whether Slug or its close relative Snail might account for the absence of mRNA in UV-irradiated E1A/ MEFs, we analyzed their activation profile by quantitative PCR. Snail mRNA levels were not affected by treatment with UVR or etoposide, in both the primary and transformed MEFs (, left). As we hypothesized, however, mRNA remained unchanged after irradiation of primary MEFs but was induced approximately threefold by UVR in wt E1A/ MEFs (). E1A/ MEFs (). E1A/ MEFs (). Thus, Slug may well be responsible for the absence of induction in UV-irradiated transformed MEFs (see Discussion). , , or mice (). Remarkably, Bcl-2 overexpression not only enhanced resistance to UVR in wt MEFs (, right) but also in the MEFs lacking Noxa, Puma, or even p53 (). MEFs exposed to 50 J/m UVR, Bcl-2 overexpression increased survival from <5 to 70% and in or cells, it saved ∼95% of the cells (). MEFs were transfected with constructs encoding FADD-DN or CrmA, but neither of these inhibitors augmented resistance to UVR (). These results demonstrate that UVR-induced apoptosis must proceed via a Bcl-2–inhibitable pathway in addition to that orchestrated by p53 and its death effectors Noxa and Puma. To extend these results to an in vivo context, we analyzed the induction of apoptosis in the skin of UV-irradiated mice of different genotypes, first by histology and enumeration of SBCs, identified by their eosinophilic cytoplasm, pyknotic nuclei, and detachment from surrounding cells (, arrowheads), and then by TUNEL staining (Fig. S4, available at ), to reveal cells undergoing DNA fragmentation. Unstimulated control skin from all genotypes (, left) exhibited well-ordered stratification of keratinocytes, an intact stratum corneum, and the absence of SBCs. UV-irradiated skin from p53- or Noxa-deficient animals exhibited a relatively preserved morphology and small numbers of SBCs (). mice were hallmarked by large vacuoles, reduced epidermal thickness, loss of the stratum corneum, and an increased number of SBCs (). At 72 h after UVR, wt skin exhibited substantial parakeratosis, hyperkeratosis, keratinocyte loss, and dysplasia (Fig. S5). and skin at this time point, and these regions typically overlaid viable, well-ordered, and stratified keratinocytes (Fig. S5). Moreover, unstimulated control skin of each genotype displayed very few TUNEL-positive cells (Fig. S4). skin at 24 h, whereas considerably fewer arose in UV-irradiated skin from the , , or animals (Fig. S4). These data indicate that, although loss of Puma provided no protection, the loss of p53 or Noxa significantly reduced the number of apoptotic keratinocytes within the epidermal layer (; P < 0.025). Quantitative PCR analysis revealed that the mRNA expression profiles of and observed in cultured MEFs held for irradiated whole mouse skin. In wt skin, mRNA was up-regulated approximately sevenfold and mRNA approximately twofold at 24 h after exposure to 1,000 J/m UVR (). mice (). To determine whether cell surface–bound death receptors, such as Fas (Apo-1/CD95) or TNF-R1, are required for UVR-induced apoptosis, potent inhibitors of the extrinsic pathway, namely, FADD-DN and the caspase-8 inhibitor CrmA, were expressed in the transformed MEFs. As neither blocked the UVR-induced apoptosis, the death receptor pathway appears to be dispensable in these cells. Other reports, however, have implicated Fas signaling in cellular responses to UVR, such as immune suppression (). Furthermore, cell surface expression of Fas was shown to be up-regulated by p53 after UVR exposure (). However, in agreement with our findings, UVR-induced apoptosis of primary MEFs and keratinocytes did not require the critical death receptor mediator caspase-8 (; ). How can the dispensability of death receptor signaling for UVR-induced apoptosis of keratinocytes be reconciled with the suppression of SBC formation in the skin of UV-irradiated FasL- deficient (C3H/HeJ ) mice ()? One possible explanation is that death receptor signaling amplifies rather than initiates UVR-induced apoptosis. Alternatively, FasL-Fas signaling in the skin may represent a non–cell autonomous process, whereby FasL on leukocytes recruited by inflammatory cytokines engages Fas on UVR-damaged keratinocytes. The strong inhibition of UVR-induced apoptosis by Bcl-2 overexpression implicated proapoptotic members of this family in initiating the response. Indeed, both Noxa and Puma, BH3-only proteins previously associated with commitment to γ-irradiation–induced apoptosis, proved to play critical roles. Primary MEFs lacking either Noxa or Puma exhibited substantial resistance to doses up to 50 J/m, although little protection was evident at higher doses, where nonapoptotic death appeared to predominate. Significantly, primary MEFs lacking both Noxa and Puma proved as refractory as those lacking p53 at all doses and times studied. This finding suggests that, in these cells, Noxa and Puma are the essential mediators of all p53-induced death after UV irradiation. MEFs, which exhibited a small but statistically significant survival advantage over transformed MEFs. Because DNA damage is not believed to induce mRNA and because UVR also causes cytoplasmic damage, cytoskeletal alterations provoking Bim release () may occur independent of the DNA damage response. In any case, the function of Bim in the UVR response probably overlaps that of BH3-only proteins that play a more prominent role, such as Noxa. Noxa alone is a weak inducer of apoptosis, because it predominantly neutralizes Mcl-1 and robust cell killing requires additional BH3-only proteins, such as Bim or Puma, which can neutralize other prosurvival members (). cells died at a rate intermediate between and wt counterparts. primary MEFs survived high-dose UVR (100 or 200 J/m) considerably better than MEFs, the transformed MEFs showed no greater resistance to UVR than counterparts, at any dose or time examined. Examination of expression revealed that transformation ablated mRNA induction specifically in response to UVR. Because etoposide induced expression comparably in the primary and transformed MEFs, the suppression is probably mediated by a UVR-induced signal acting parallel to p53 rather than by the p53 pathway itself. A precedent for failure of induction in the face of DNA damage and active p53 emerged with the discovery that in γ-irradiated hematopoietic progenitors, expression is silenced downstream of p53 by the transcriptional repressor Slug (). Indeed, we found that in E1A/ transformed MEFs, but not in primary MEFs, UVR induces transcription of Slug, which may well be responsible for the failure of p53 to induce expression in these cells. However, in contrast to the hematopoietic progenitors, was up-regulated in the transformed MEFs irrespective of whether p53 was present. Although we show that Noxa is the principal initiator of UVR-induced apoptosis, Puma is the principal initiator of apoptosis downstream of p53 activated by other genotoxic stimuli (; ). cells were markedly resistant (; ). This difference probably reflects the fact that UVR produces pyrimidine dimers, whereas γ-irradiation and etoposide produce double-strand breaks in DNA. Although all genotoxic damage leads to p53 stabilization, the activity of p53 can be altered by the different kinases (e.g., ATM and ATR) selectively activated by these two types of DNA damage (; ). A plausible model is that the levels and/or activity of Noxa and Puma are modulated selectively by different posttranslationally modified forms of p53 and possibly also by p53-independent signals (, X or Y) that are determined by the nature of the DNA damage. and, remarkably, even cells still died after irradiation. The failure of FADD-DN or CrmA to inhibit the death of Noxa-deficient cells demonstrates that their UVR-induced apoptosis does not require death receptor signaling. In contrast, the substantial protection conveyed by Bcl-2 overexpression, even in the absence of p53, demonstrates that UV irradiation can trigger at least two distinct pathways to apoptosis. One of these pathways leads via p53 to induction of and , whereas the other activates a Bcl-2–inhibitable apoptosis inducer via a p53-independent route (, Z). What molecular mechanisms might mediate this p53-independent but Bcl-2–inhibitable pathway to UVR-induced cell death? Transformation by E1A may well have activated this pathway. By antagonizing RB, the E1A in the transformed MEFs deregulates E2F activity, including that of E2F1 (), which can enhance expression of several BH3-only proteins () and repress that of Mcl-1 (). Another interesting candidate is the JNK signaling pathway. In response to UVR, the JNKs phosphorylate diverse nuclear and cytoplasmic substrates, including c-Jun, a constituent of the transcription factor AP-1 that is rapidly induced upon UVR exposure (). MEFs () and c-Jun–deficient fibroblasts () are abnormally resistant to UVR. , , , and mice. animals exhibited significantly fewer SBCs than wt controls at 24 h after irradiation, and the extent of SBC formation in skin was comparable to that in skin. skin was indistinguishable from wt controls. and skin, demonstrating that loss of p53 or Noxa provided longer term protection and not only a delay in apoptosis. The greater role of Noxa in the skin may be related to our finding that was increased approximately sevenfold and only approximately twofold at 24 h after irradiation. p53 protein is strongly up-regulated in UV-irradiated skin from 2 to 24 h after exposure (), and accordingly, the robust induction of Noxa at 24 h required p53. In summary, our analyses of two different cell line systems and the skin of intact mice identify Noxa as the principal mediator of UVR-induced apoptosis. Furthermore, we provide evidence of a pathway that collaborates with the p53 pathway to activate , plus the activation of a p53- and Noxa-independent pathway to apoptosis that can be blocked by Bcl-2. The delineation of these pathways will provide insight into how the stimulus specificity of BH3-only protein activation is conferred downstream of p53. All experiments with animals were conducted according to the guidelines of the Melbourne Research Directorate Animals Ethics Committee. (), (), (), (), and mice () has been described. and mice were generated on an inbred C57BL/6 background using C57BL/6-derived embryonic stem cells, whereas , , and mice were produced on a mixed C57BL/6 × 129Sv background using 129Sv-derived embryonic stem cells and were backcrossed with C57BL/6 mice for >10 generations. MEFs were derived from day 14.5 embryos and were cultured to ∼80% confluence before passage 1 (P1). The Phoenix packaging line was used to produce high-titre, replication-incompetent retrovirus using the Fugene 6 Transfection method (Roche). In brief, Phoenix cells were transfected with plasmid DNA (pWZLH.12S [E1A] and pBabePuro.H-Ras; gifts from M. Schuler, Johannes Gutenberg University, Mainz, Germany; D. Green, St. Jude Children's Research Hospital, Memphis, TN; and S. Lowe, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). Primary MEFs were infected by centrifugation for 45 min at 32°C in the presence of viral supernatant. This process was performed on 2 consecutive days, and transfected cells were selected by incubation with 3 μg/ml puromycin and 100 μg/ml hygromycin B (Invitrogen) for at least 1 wk. FLAG–FADD-DN and FLAG-CrmA DNA sequences were subcloned into the MSCV-IRES-GFP vector backbone to generate retroviral expression constructs (). GFP-labeled cells were sorted using a FACStar cell sorter (Becton Dickinson) to obtain an 85–90% GFP population. Early passage (P2) primary MEFs and E1A/ transformed MEFs were maintained in a high-glucose DME supplemented with 10% fetal calf serum, 10 M asparagine, and 50 μM 2-mercaptoethanol. Cells were seeded at 2.5 × 10 per well (6-well tissue culture plate) and cultured to ∼60–70% confluency before irradiation with UVC (UV lamp; Sankyo Denki). UVR output was quantified with a Spectroline Shortwave (254 nm) Ultraviolet meter. Cells were stimulated for 48 h with 100 ng/ml recombinant human FasL (FLAG-tagged; Qbiogene) cross-linked with 1 μg/ml anti-FLAG M2 monoclonal antibody (Sigma-Aldrich). Cells were washed twice with ice-cold PBS before lysis with Onyx lysis buffer (20 mM Tris-HCl, pH 7.4, 135 mM NaCl, 1.5 mM MgCl, 1 mM EDTA, 1% Triton X-100, 10% glycerol, 2 mM sodium orthovanadate, 50 mM sodium fluoride, 1 μg/ml pepstatin, 1 μg/ml aprotinin, and 1 μg/ml leupeptin). 20 μg of total protein per sample were separated by gel electrophoresis (Tris-glycine Novex Pre-cast gels; Invitrogen) and transferred to nitrocellulose membranes (Hybond-C extra; GE Healthcare). Membranes were probed with antibodies to poly ADP ribose polymerase (Qbiogene), β-actin (Sigma-Aldrich), HSP70 (a gift from R. Anderson, Peter MacCallum Cancer Centre, East Melbourne, Australia), FLAG epitope tag (a gift from L. O'Reilly and D. Huang, Walter and Eliza Hall Institute of Medical Research, Melbourne, Australia), ICAD (BD Biosciences), active caspase-3 (a gift from Y. Lazebnik, Cold Spring Harbor Laboratory), and Bim (clone 3C5; Qbiogene) and visualized using the ECL Western detection kit (GE Healthcare). Cytospin preparations (1 × 10 and 1 × 10 cells) were fixed in ice-cold methanol and stained with hematoxylin and eosin. Total RNA was extracted from ∼1 × 10 cells or ∼100 mg of whole mouse skin using Trizol (Invitrogen) and treated with DNase I (Promega). First-strand cDNA synthesis was performed using 5 μg total RNA, SuperScript II reverse transcriptase (Invitrogen), and oligo(dT) primer (Promega) according to the manufacturer's instructions. One tenth of the reverse-transcription reaction was subjected to quantitative PCR using Quantitect SYBR Green PCR Master Mix (QIAGEN) in 10-μl reaction volumes and the ABI PRISM system (Applied Biosystems). β-Actin was used as an internal control, and basal transcript levels were estimated from cDNA samples from and cells. For statistical comparison, a test was used, with P values <0.05 considered significant. The supplemental text discusses DNA fragmentation assay, cytochrome release assay, TUNEL staining, and confocal microscopy. Fig. S1 shows hallmarks of apoptosis in E1A/ transformed MEFs. Fig. , , and primary MEFs. In Fig. S3, intracellular immunofluorescent staining demonstrates that all E1A/ transformed cell lines express E1A comparably and that cyctochrome release after UV irradiation is abrogated by the loss of Noxa and p53. In Fig. S4, TUNEL analysis confirms the induction of apoptotic cell death in UV-irradiated epidermal keratinocytes. In Fig. S5, analysis of skin at 72 h after UV irradiation demonstrates the long-term protection afforded by the loss of Noxa or p53. Online supplemental material is available at .

The granzymes, a family of structurally related serine proteases expressed in cytotoxic lymphocytes, cooperatively bring about the death of transformed and virus-infected cells after their cosecretion with perforin (). Perforin is critical for permitting the access of granzymes and other granule-bound toxins to their substrates within the target cell (; ), and as a result, all granzyme-dependent cell death pathways are halted in its absence, leading to marked immunodeficiency in humans (; ) and mice (). By comparison, deficiency of an individual granzyme is better tolerated by gene-targeted mice. The absence of granzyme B (GrB), which cleaves target cell proteins adjacent to aspartate residues, results in delayed target cell DNA fragmentation during apoptosis of most, but not all, cell types (; ; ). GrA induces a caspase-independent form of cell death that involves the induction of single-stranded DNA nicks after cleavage and activation of constituents of the SET complex (). Proapoptotic function has also more recently been described for GrC () and GrM (; ). All of the granzymes and the closely related myeloid serine proteases, such as cathepsin G (CatG), are members of the chymotrypsin superfamily and, like chymotrypsin, are synthesized as preproenzymes (; ). The signal peptide is cleaved by a signal peptidase in the endoplasmic reticulum, before transport to the granules (; ). The proteases are then stored within the granules as active enzymes after further limited proteolysis that removes a two-amino-acid activation peptide at their N terminus (). This enables the protease to assume the correct conformation for access and hydrolysis of substrate (). A similar process is used in the activation of the mast cell proteases, which are also stored as active enzymes in the lysosomal compartment (; ). Inhibition of a specific dipeptidyl peptidase, CatC (dipeptidylpeptidase I) can prevent this processing and activation (). Transfection of mammalian COS-7 (; ) or yeast () cells with granzyme cDNA constructs that retained the activation dipeptide resulted in the expression of inactive protease. However, the activity could be rescued by exogenous treatment of cell lysates or purified granzyme protein with CatC (; ), by coexpression of progranzyme and active CatC in the same cells or by deleting the dipeptide sequence before transfection (; ). Further evidence for a critical role of CatC in the processing of granzymes and myeloid cell serine proteases came from gene-knockout studies, in that a CatC-deficient mouse was generated by homologous recombination with an inactive CatC gene (). Several studies have examined the effect of CatC deletion on the activity both of the granzymes and the structurally related mast cell and neutrophil proteases (; ; ). Effector lymphocytes generated from these mice were reported to be as deficient as perforin-null mice in their ability to induce apoptosis of target cells, although only DNA fragmentation was assayed in these studies (; ). No tryptase and only minimal ASPase activity could be detected in the granules isolated from CatC-null lymphocytes (; ). Thus, it appeared that CatC deficiency effectively recapitulated the GrAB-null phenotype in cytotoxic lymphocytes and would also result in minimal activity from the other granzymes. Given the proposed generic role for CatC in granzyme activation, we reasoned that cytotoxic T lymphocyte (CTL) from CatC mice might allow us to study the physiological role of perforin in a setting where all of the killer cell granzymes were not expressed. Also, as perforin is present in CTLs of both GrAB and CatC mice, comparing these CTLs would also allow us to determine the physiological role of the remaining “orphan” granzymes expressed in the GrAB mice. Surprisingly, we found low but clearly measurable residual GrB activity in the absence of CatC expression, and that effector lymphocytes from CatC-null mice could induce apoptosis of target cells by a mechanism that was indistinguishable from classic apoptosis. Unlike GrAB mice, which are exquisitely sensitive to infection with ectromelia virus (ECTV; mouse pox), CatC-deficient effector lymphocytes also mediated resistance to ECTV in vivo, so that infected mice were as resistant to the virus as wild-type C57BL/6 (B6) mice. It has previously been reported that CatC gene-targeted mice lack GrA and -B activities and therefore have severe defects of granule-mediated lymphocytotoxicity (; ). To explore this issue further, we generated allo-reactive CTLs in standard, one-way (H-2 anti–H-2), mixed lymphocyte reactions and assessed the capacity of B6.CatC effector cells to kill mouse MS9II (H-2) target cells. We had previously used time-lapse microscopy at 37°C to characterize in real time the morphological and molecular events accompanying the death of MS9II target cells in response to allo-reactive CTLs raised in B6 or B6.GrAB mice (). This methodology allowed us to examine changes in cell morphology and the appearance of classic markers of apoptosis in response to CTL attack. By adding soluble annexin V (AV) and propidium iodide (PI) to the medium, we were able to track phosphatidylserine exteriorization (a relatively early marker of apoptosis) and loss of plasma membrane integrity (a late marker), respectively, as a function of time. Both B6 and B6.GrAB CTLs induced classic apoptotic changes, such as cell shrinkage, marked membrane blebbing, and nuclear collapse with indistinguishable kinetics (). However, a major point of difference was that although B6 CTLs induced early and marked target cell AV binding, cells killed by GrAB CTLs did not stain with AV until very late, simultaneously with loss of plasma membrane integrity (). Given that, like GrAB CTLs, CatC CTLs had also been found to lack GrA or -B activity (), we were surprised to find that MS9II cells exposed to CatC CTLs underwent morphological and molecular events of apoptosis that were indistinguishable from B6 CTLs, both in their amplitude and timing. B6 or B6.CatC effector cells were added to equal numbers of the much larger and adherent MS9II target cells. The time course of death in a single representative target cell is shown (). In this cell, rounding occurred at ∼1 h 10 min, followed by AV binding at 1 h 45 min and PI uptake at 2 h 25 min. and Video 1, available at , illustrate the sequence of events for the same cell, including conjugate formation with the CTL (t = 0.5 h), cell rounding and plasma membrane blebbing (t = 1.5 h), AV binding (t = 2.0 h), and PI uptake (t = 2.5 h). By making similar observations on a large number of dying target cells, it was possible to compare both the morphology and kinetics of death induced by CatC-deficient and -sufficient CTLs () with those raised in GrAB mice (). Taking the time of rounding as a common reference point (i.e., t = 0), no difference was observed in the kinetics of cell death (morphological changes; AV or PI staining) over populations of cells killed by B6 CTLs ( = 20) or B6.CatC CTLs ( = 25). The killing observed in these assays was mediated though the granule pathway in that CTLs raised simultaneously in perforin-deficient mice were unable to kill the same target cells, as demonstrated both by time-lapse microscopy (not depicted) or 4-h Cr release assays, whereas target cells released Cr in response to both wild-type and CatC CTLs (). Cell death by perforin-induced lysis was excluded, as the kinetics and morphology of this form of cell death are quite easily distinguished from apoptosis (). We also demonstrated by Western blot that the CatC mice had not up-regulated their level of perforin or GrB protein expression compared with B6 CTLs (). In total, these observations strongly suggested that the CatC-deficient effector cells induce apoptotic death similar to wild-type B6 killer cells, but differing from cell death induced by CTLs raised from mice with structural disruption of both their GrA and -B genes. To exclude the possibility that the aforementioned findings were peculiar to MS9II target cells or the specific strain combination used to raise allo-reactive CTLs, we exposed Cr-labeled P815 mouse mastocytoma tumor cells (H-2) to B6 CTLs stimulated with irradiated BALB/c splenocytes. Unlike MS9II cells (which have a relatively long doubling time), the DNA of rapidly dividing P815 cells labels well with I-UdR, so that cells doubly labeled with Cr and I can be used to simultaneously evaluate plasma membrane permeability and DNA fragmentation. Both day 7 primary () and day 3 secondary () CTLs of B6 mice produced strong Cr and I-DNA release that was only slightly diminished when CTLs from B6.CatC mice were used. This killing was once again mediated through the granule pathway, as doubly stimulated perforin-deficient CTLs induced neither Cr or I release, whereas the cell death induced by day 7 primary B6 CTLs was totally inhibited when free calcium in the reaction medium was complexed with EGTA ( and Fig. S1, available at ). Additional effector cell types were examined to determine whether the cytotoxic activity observed in allo-stimulated CatC-null CTLs was also seen in other effector lymphocytes. Unfractionated splenic natural killer (NK) cells or purified DX5CD3 cells activated with IL-2 each induced Cr release from Yac-1 target cells in a perforin-dependent manner, but this did not vary whether CatC was expressed or not (unpublished data). Although perforin and granzymes synergistically bring about target cell apoptosis, it is well recognized that purified perforin is unable to induce DNA fragmentation, but only causes cell lysis (). When delivered by a CTL or NK cell, GrA and -B activate separate signaling pathways, resulting in DNA nicking or oligonucleosomal fragmentation, respectively, in a perforin-dependent manner (; ). Rapid fragmentation of target cell DNA, a hallmark of apoptosis induced through the granule-exocytosis mechanism, in most cell types is largely due to the action of active GrB (; ,; ). Consistent with these previous findings, we found that DNA fragmentation was abolished in P815 target cells exposed to GrAB CTLs, whereas appreciable Cr release was still observed, as previously reported (; ; ). In contrast, CatC-null CTLs were efficiently able to induce DNA fragmentation of P815 cells (). Although an earlier study failed to find evidence of GrA or -B activity in the lymphocytes of CatC-null mice (), the similar kinetics of the apoptotic events detailed in CatC-sufficient and -deficient CTLs () suggested that some serine protease (i.e., granzyme) activities might be generated in allo-stimulated, CatC-deficient CTLs. In addition to activating granzymes, CatC has been reported to be critical for activation of serine proteases in monocytes/neutrophils () and mast cells (). To show that inactivation of the CatC gene abolished the activity of certain specific serine proteases in myeloid cells and to reconfirm the myeloid cell phenotype of the CatC mice, CatG and neutrophil elastase (NE) were assayed in bone marrow neutrophils purified from mice administered granulocyte colony-stimulating factor, as described previously (). The hydrolysis of NE- and CatG-specific substrates was measured in a chromogenic assay using lysates of bone marrow cells from wild-type B6 and B6.CatC mice. As expected (), the activity of both enzymes was virtually abolished (reduced by >98%) in the absence of CatC expression (Fig. S2, available at ). To test the possibility that the CTLs of CatC mice were still capable of some residual granzyme activation, lysates prepared from allo-stimulated primary and secondary CD8 CTLs (>90% pure) were assayed for their capacity to hydrolyze the GrB-specific substrate Ala-Ala-Asp-S-benzyl (cleavage after asp or ASPase activity), N-CBZ-L-lysine thiobenzyl ester (BLT; to measure trypsin-like GrA [tryptase] activity) and Phe-Leu-Phe-S-Benzyl (chymotrypsin-like or chymase activity; ). It became clear that although GrB activity was markedly reduced, it was not abolished in the CatC cells. In kinetic assays, ASPase activity was reduced but still clearly detectable in the lysate of day 7 primary () or day 3 secondary () allo-stimulated CatC-null CTLs, albeit with some delay compared with CatC-sufficient cells. Not surprisingly, there was greater ASPase activity in twice-stimulated CD8 cells; however, maximum substrate hydrolysis was reached by 10 min in both cases. The ASPase substrate, Ala-Ala-Asp-SBzl is cleaved only by GrB: it is not cleaved by caspases, proapoptotic cysteine proteases that also cleave after specific aspartate residues. We also performed an additional important control to examine the reliance of substrate turnover on the presence of bona fide granzyme activity: we demonstrated that no ASPase activity whatsoever was present in the CTLs of GrAB-null cells, in which the GrB gene is structurally disrupted (). This observation ruled out any incidental cleavage by proteases other than GrB as a cause of substrate hydrolysis in the CatC-null cells. The presence of ASPase activity in the CatC-null lysate was again confirmed when lysates containing equal amounts of protein from one of the experiments shown in were progressively diluted and the assay was performed for 15 min. At each lysate dilution, the level of ASPase activity was about one third of that seen in B6 lysates (). As expected, normal activity was noted in the B6.pfp lysates. As the substrate used in these assays is specific for GrB, these results indicate that a proportion of the GrB precursor expressed in the CatC-null effectors had been activated. In contrast to ASPase activity, no measurable tryptase (GrA) activity was detected in either the CatC-null or GrAB lysates, indicating that GrA (and the second tryptase, GrK) were not activated in the CatC CTLs (). As expected, the turnover of the chymase substrate in the same lysates was not reduced in any of the lysates tested (). This indicated that CatC is not essential for the processing of at least some of the granzymes with chymase activity, presumably GrC–F (). It was previously demonstrated that normally resistant B6 mice lacking both GrA and -B or perforin are incapable of controlling primary infection by the natural mouse pathogen ECTV, an orthopoxvirus (,). These mice are unable to control ECTV replication, and viral load in their spleen is several orders of magnitude higher than B6 mice (). We therefore estimated viral titers in the spleens and livers of CatC-null and -sufficient B6 mice after ECTV infection into the footpad (). Viral titers in spleen and liver were estimated at 2, 4, 6, and 8 d after infection and showed no measurable difference at any time point. In contrast, deficiency of both GrA and -B has been shown to increase viral titer in the liver and spleen ∼10,000-fold by days 6 and 8, compared with granzyme-expressing mice (). In addition, we found that the CatC-null mice had only a slightly higher mortality than wild-type mice, but far lower than that seen in perforin or GrAB-deficient mice (unpublished data). Overall, our data indicated that the phenotype of CatC-null mice with respect to ECTV infection in vivo is distinct from that of perforin or GrAB-deficient mice on the same genetic background; namely, perforin- and GrAB gene-disrupted mice are far more susceptible. The lymphocyte and myeloid cell serine proteases are unusual amongst proteolytic enzymes in that they are stored in secretory lysosomes as active enzymes (). Activation is a two-step process and involves removal of the leader sequence followed by cleavage of two final residues at the N terminus at the time of granule packaging (; ; ). Studies using protease inhibitors or in which purified CatC was added to recombinant granzymes (; ; ; ; ) or, finally, with CatC-knockout mice () indicated that CatC is the principal protease that removes the activation dipeptide. Our current study clearly confirms that CatC is responsible for virtually all of the GrA and most of the GrB activity in CTLs. However, our results also clearly indicate that not all of the GrB activity is eliminated in CatC mice and that the extent of the immune defect of CatC-null mice may have been overestimated in the past. Surprisingly, we have found that although CatC seems to be critical for activating GrA, it is not essential for GrB activity. As a result, CatC-deficient mice retain substantial CTL cytotoxicity, inducing apoptosis in vivo and in vitro. The reduced GrB activity of CatC CTLs was still sufficient to induce DNA fragmentation in target cells that was comparable to that induced by wild-type effector cells. This indicated that the quantity of GrB released by wild-type CTLs is well in excess of the amount required to trigger DNA fragmentation in P815 cells. Much of the DNA fragmentation induced by GrB is normally reliant on caspase activation (; ), so the sufficiency of CatC CTLs in inducing DNA fragmentation indicates that the quantity of GrB delivered by these cells is adequate to effectively initiate caspase processing. Similarly, live-cell imaging demonstrated that target cells exposed to CatC CTLs died by classic apoptosis: both the morphology and the acquisition of apoptotic markers (phosphatidylserine exposure, PI uptake) as a function of time were indistinguishable between cells incubated with CatC-deficient or -sufficient killer cells. In vivo, CatC-null mice were far less susceptible to both GrAB and pfp mice to ECTV, and viral titers generated in their livers and spleens were comparable to those in wild-type B6 mice. Indeed, the kinetics of virus replication in CatC mice matched that seen with GrA mice (,), which is consistent with our findings that the CatC lacked GrA expression but retained measurable GrB activity. Overall, the current study makes it clear that CatC-null mice are considerably less immunologically compromised than either GrA/B-cluster or perforin-deficient mice and that alternative mechanisms exist for GrB activation in the absence of CatC. A lack of neutrophil serine protease activity has been observed in CatC mice (), as was confirmed in our study. A role for these proteases in defense against bacterial and fungal pathogens has been suggested (; ), but the only study that addressed this issue, surprisingly, showed that the absence of CatC contributed to increased survival from sepsis (). A deficiency of cell-mediated immunity is not a prominent feature of CatC deficiency in humans. Most CatC-deficient patients present with Papillon-Lefevre syndrome (PLS), a rare autosomal recessive disorder (; ) marked by severe, early onset periodontal disease, with subsequent premature loss of both primary and secondary dentition, and palmoplantar hyperkeratosis (; ). Although the critical CatC substrates responsible for PLS have not been defined, it has been suggested that CatC may play a role in maintaining the structure and integrity of the epidermis surrounding the teeth (; ). An increased susceptibility to bacterial infections both in the mouth and within organs such as the liver has been reported and may reflect defects in neutrophil activity, manifested as an absence of CatG, NE, and protease 3 activity (; ; ). Recent reports have produced somewhat conflicting results with regard to lymphocyte function in PLS patients. Three of the seven patients described in the aforementioned studies also had a slight reduction in GrA protein levels, but considerable GrA and -B activities were detected (), and IL-2–activated T cells generated from these patients were able to efficiently kill K562 target cells. However, naive NK cells isolated from a further 20 PLS patients (representing seven families) displayed an ∼50% reduction in activity against K562, compared with a pool of normal healthy controls (). The profile of effector molecules expressed by these two cell types is distinct, perhaps accounting for the apparently discordant results (). In particular, resting NK cells have very low levels of GrB but a very high constitutive level of GrH, whereas GrB expression increases markedly with IL-2 stimulation (). In another study, it was shown that the NK cells of a PLS patient exhibited GrB activity after stimulation with IL-2 (). We have found similar results in CatC-null NK cells (unpublished data). CatC, an amino-dipeptidase, belongs to a family of lysosomal papain-like cysteine proteases (). Unlike related proteases, which function as monomeric endopeptidases, CatC is an oligomeric exopeptidase with a general preference for acidic N-terminal sequences (see below) until it encounters a “stop sequence,” usually a charged residue, such as arginine, lysine, or proline (). However, CatC is also well recognized for its ability to cleave the dipeptide at the N terminus of serine proteases found in lymphocytes and neutrophils. Exposure of the invariant N-terminal isoleucine in all granzymes allows it to form an ion pair with the aspartate within the catalytic pocket, which leads to the formation of a functional catalytic center (). It is unclear whether further trimming is prevented, as the isoleucine is seen as a stop sequence, or whether the ion pair formation is instantaneous, which makes the N terminus unavailable for further processing. In the absence of CatC, other less specific lysosomal enzymes could potentially remove the activation dipeptide. However, this degree of redundancy is limited, as we showed that GrB activity remains markedly compromised, whereas GrA activity remains almost completely absent. Interestingly, an analogous study by showed that tryptases but not chymases are still activated in CatC mast cells. Although were unable to find evidence for dipeptide-deleted GrA and -B, they did, surprisingly, find that 50% of the GrC they were able to purify from CatC-deficient CTLs had had its activation dipeptide removed, and they suggested that an alternative protease, DPPIV, might be responsible. This seems unlikely to us, as substrate analysis has suggested that DPPIV has a strict preference for proline in the P1 position, although alanine was tolerated in two identified substrates (). Comparison of the dipeptide sequence in serine proteases found that all of the granzymes, with the exception of GrA, have an acidic residue (Glu) in the P1 position (; ,; ), as do the mast cell chymases (; ), carboxypeptidase A (), and NE (). In contrast, GrA has arginine (), CatG has lysine (), and the mast cell tryptases have glycine (). Therefore, in terms of the specificity of an alternative dipeptidase acting in the absence of CatC, there is not a consistent preference for a specific residue in the P1 position. We are currently attempting to identify the proteases responsible for partially compensating for CatC's absence, using biochemical and genetic approaches. Inbred B6 (H-2), BALB/c (H-2), and C3H/J (H-2) mice were purchased from the Walter and Eliza Hall Institute of Medical Research. GrA and GrB cluster–deficient (GrAB) mice () were obtained from M. Simon (Max-Planck-Institut für Immunbiologie, Freiburg, Germany) and were maintained at the Peter MacCallum Cancer Centre. Mice were genotyped using the PCR screening protocols previously described (). Originally, the GrA mice () were derived on the B6 background and the GrB cluster–deficient mice () on the129 background. After appropriate matings and selection of GrAB progeny, the mice were backcrossed for eight generations to B6. The CatC (dipeptidyl peptidase 1) knockout (B6.CatC) mice () were provided by C. Pham (Washington University School of Medicine, St. Louis, MO) and maintained at the Peter MacCallum Cancer Centre. The B6.CatC mice had previously been backcrossed to B6 for 11 generations. We used a PCR screening protocol provided by C. Pham to confirm disruption of the CatC gene and acquisition of the Lac Z cassette in knockout animals (). We also recapitulated the myeloid phenotype of the mice by confirming the absence of active CatG and NE from neutrophils (Fig. S2; see the following paragraph). Mice 5–10 wk of age were used in all experiments, and the studies conformed to Peter MacCallum Cancer Centre and the John Curtin School of Medical Research animal experimental ethics committee guidelines. Whole cell lysates of CTLs generated in mixed lymphocyte cultures were normalized for protein content and analyzed for granule serine protease activity by the hydrolysis of synthetic peptide thiobenzylester substrates: for ASPase activity, Boc-Ala-Ala-Asp-S-Bzl (a gift from J. Powers, Georgia Institute of Technology, Atlanta, GA); for tryptase activity, BLT (Sigma-Aldrich); and for chymotrypsin-like (chymase) activity, Suc-Phe-Leu-Phe-SBzl (Enzyme Systems Products), as described previously (; ). The failure of CatC mice to activate the bone marrow–derived serine proteases NE and CatG was confirmed by the lack of cleavage of specific peptide paranitroanilide substrates in chromogenic assays as previously described (; ). Absorbance readings (at 405 nM) were taken at 15 min. The data points shown represent the means ± SEM. P815 (mouse mastocytoma; H-2) and MS9II (mouse fibroblast; H-2) cells were maintained in DME (JRH Bioscience) supplemented with 10% (vol/vol) FCS (JRH Bioscience), 2 mM glutamine (JRH Bioscience), 100 U/ml penicillin, 100 μg/ml streptomycin (Invitrogen), and 1 mM sodium pyruvate (Invitrogen). Primary mouse CTLs were maintained in supplemented RPMI (Invitrogen) also containing 100 μM nonessential amino acids and 0.1 mM 2-mercapto-ethanol. Western blot analysis () was used to confirm perforin expression using the rat PI-8 antibody (Kamiya Biomedical; ) and GrB expression using the rat anti-GrB antibody (eBioscience). Allogenic CTLs capable of recognizing H-2 expressed on P815 cells or H-2 on MS9II cells were generated in one way in vitro mixed lymphocyte reactions. Splenocytes isolated from B6 and the gene-targeted mouse strains listed above were cultured for 7 d in RPMI medium with lethally irradiated BALB/c (H-2) or C3H/J (H-2) splenocytes. The stimulator/responder cell ratio was 1:1, and responder cells were seeded at 2 e6 cells per milliliter of culture medium in 24-well plates. In some experiments, the cells were restimulated in a similar fashion for a further 3 d. Cell death induced by CTL populations was assessed by Cr and I-DNA release, as described previously (). Time-lapse confocal microscopy was performed essentially as previously described (). In brief, MS9II cells were plated and allowed to adhere to 96-well culture plates and incubated overnight at 37°C in a humidified CO incubator. The plates were transferred to a temperature-controlled stage (Prior Proscan) maintained at 37°C on a microscope (IX-81; Olympus). PI (Sigma-Aldrich) was added to the cultures, in tissue culture medium (DME) at 50 ng/ml and AV-FLUOS (Roche) at 2 μg/ml. Cells were exposed to an equal number of activated CD8 CTLs and viewed for the times indicated. Images were captured at specified intervals using a charge-coupled device camera (ORCA-ER; Hamamatsu) controlled by MetaMorph software (Universal Imaging Corp.). The images were viewed with an LCPlanFl objective lens at 20× (NA 0.4). As the fluorescence intensities of AV-FLUOS and PI varied, we plotted the fluorescence reading for each frame relative to the maximum for that fluorochrome over the entire time course (using MetaMorph and Excel [Microsoft]), after subtraction of background fluorescence at each time point. For each fluorophore, the maximum fluorescence plotted was therefore defined as 1.0, and baseline fluorescence was 0. The kinetics of ECTV replication were estimated as previously described (). Mice were infected with 10 pfu ECTV (Moscow strain) into the footpad. Three mice were killed on days 2, 4, 6, and 8 after infection, and liver, spleen, and blood were harvested. The viral titers in spleen and liver were determined as described previously (). Fig. S1 demonstrates a chromium release assay showing that apoptosis induced by CD8 CTLs is granule mediated. Fig. S2 demonstrates a chromogenic substrate cleavage assay showing CatG and NE activities in CatC-null and -expressing neutrophils. Video 1 is a time-lapse video of CatC-null CTL-induced death of an MS9II target cell. Online supplemental material is available at .

Granzyme B (GzmB) is an aspartic acid–directed protease that is contained within the specialized secretory granules of cytotoxic T lymphocytes (CTLs) and natural killer (NK) cells (; ; ). Entry of GzmB, as well as other granzymes, into virally infected or transformed cells is facilitated by perforin, which is a pore-forming protein that is also contained within CTL/NK granules (; ; ). Upon entry into target cells, GzmB can initiate apoptosis in the target through restricted proteolysis of substrate proteins (; ; ; ; ; ). To date, several substrates for GzmB have been identified and, among these, the BH3-only protein BID has been implicated as playing a particularly important role (; ; ; ; ). BID has been proposed to be the preferred substrate for GzmB (; ), and BID proteolysis is thought to initiate apoptosis through activating Bax and/or Bak and promoting their oligomerization within the mitochondrial outer membrane. Oligomerization of Bax/Bak results in permeabilization of mitochondria and facilitates escape of intermembrane space proteins such as cytochrome (). Release of cytochrome into the cytosol is especially relevant, as this protein acts to initiate the assembly of a complex between Apaf-1 and caspase-9, which results in the activation of the latter and precipitates a series of further caspase activation events culminating in the death of the cell (; ; ). However, because GzmB can also directly process and activate caspases, such as caspase-3 and -8 (; ; ; ; ), this provides a more direct route to caspase- dependent cell death. Several other substrates for GzmB have been reported. Among these, ICAD (DFF45) is also thought to play an important role in cytotoxic lymphocyte killing, as this protein acts as an inhibitor of a DNase (CAD/DFF40) that becomes activated during many forms of apoptosis (; ). Thus, it has been suggested that GzmB-mediated proteolysis of ICAD during CTL/NK-mediated killing can lead directly to CAD activation and DNA degradation (; ). Several studies have debated the relative importance of BID versus caspases and other GzmB substrates during CTL/NK-mediated killing (; ; ; ; ; ). One model proposes that BID proteolysis is the critical event in CTL/NK-mediated killing because BID (of human origin) appears to be more susceptible to GzmB-dependent proteolysis than caspases such as caspase-3 (). However, other studies have reported that BID proteolysis occurs secondary to caspase-3 activation in certain models (). One possible reason for these discrepant findings is that groups working in this area use GzmB preparations from either murine or human sources. Thus, some conclusions have been based on experiments where human GzmB has been used in conjunction with mouse cell lines or vice versa. This approach is predicated upon the assumption that human and murine GzmB have essentially identical substrate preferences, but this has not been established. However, this assumption does not take into account the divergence between human and mouse GzmB at residues that may affect substrate recognition or catalysis. Although the similarity between human and mouse GzmB is substantial, these are distinct proteases and are likely to display subtle, but important, differences in substrate selection. Similar considerations probably apply to many other conserved proteases, such as caspases. We provide evidence that human and murine GzmB are distinct enzymes. Although both of these enzymes can cleave a similar cohort of protein substrates, we have found clear differences in their ability to cleave important substrate proteins such as BID, ICAD caspase-8, and several other substrates. These observations suggest that experiments that use GzmB (and possibly other granzymes) from one species cannot be readily generalized to other species. These observations have broad implications for our current understanding of the role of GzmB, as well as other granzymes, in CTL/NK-mediated killing. GzmB is a highly conserved protease, as the murine and human forms of this granzyme exhibit 69% identity (). However, although the amino acids that make up the catalytic triad are perfectly conserved between rat, mouse, and human GzmB (, residues highlighted in red), several of the residues that are predicted to influence substrate recognition exhibit significant divergence (, residues highlighted in blue). Because of the high degree of homology between human and murine GzmB, it is generally assumed that these enzymes cleave essentially the same cohort of substrates upon entry into target cells. Because of this, many investigators use these enzymes in cross-species experiments where mouse GzmB is used to kill human target cells or vice versa. However, the divergence seen in residues involved in substrate recognition suggests that human and mouse GzmB may not behave in an identical manner toward the same protein substrates. To explore this issue, we generated recombinant human and mouse GzmB in and active site–titrated these enzymes using the synthetic substrate peptide Ac-IETD-AFC and the viral serpin CrmA (). Using this assay, we chose concentrations of the human and murine enzymes that displayed equivalent rates of IETD-AFC hydrolysis and used these concentrations of GzmB for subsequent experiments (). We then compared hGzmB and mGzmB for their ability to process and activate effector caspases within cell-free extracts prepared from either human (Jurkat) or mouse (J774) cell lines. To permit direct comparison, cell-free extract preparations were adjusted to the same final protein concentration (10 mg/ml). Effector caspase activation was assessed by measuring the hydrolysis of Ac-DEVD-AFC, which is a peptide that is cleaved efficiently by caspase-3 and -7, but not by GzmB (). As illustrated in , addition of 100 nM of either human or mouse GzmB to cell-free extracts of human origin provoked similar rates of DEVD-AFC hydrolysis, suggesting that these enzymes activated effector caspases within the extracts with equal efficiency. Note that neither granzyme hydrolyzed DEVD-AFC when cell-free extract was omitted from the assay (, left). Essentially identical results were observed when cell-free extracts generated from mouse J774 cells were used in the same assay (, right). These data suggest that human and mouse GzmB activate effector caspases (whether of human or murine origin) with a similar efficiency. As outlined earlier, it has been suggested that BID is the preferred substrate for GzmB (; ; ). Proteolysis of BID by GzmB results in the exposure of a myristoylation signal sequence that, upon myristoylation at this site, promotes translocation of BID to mitochondria to induce oligomerization of Bax and Bak on the mitochondrial outer membrane (). Because cleaved BID is a potent trigger for mitochondrial cytochrome release, proteolysis of this BH3-only protein may well be sufficient to ensure death of a target cell upon CTL attack. Thus, we were interested in determining whether the human and mouse forms of GzmB cleaved BID with similar efficiency. We also compared hGzmB and mGzmB in terms of their ability to proteolytically process caspase-3, -7, and -8, all of which are well established as substrates for hGzmB (). As a source of human BID and other human substrate proteins, we used cell-free extracts derived from Jurkat cells (; ). As a source of murine substrates, similar cell-free extracts were generated from the mouse macrophage line, J774. To exclude complex outcomes resulting from activation of caspases within the extracts caused by the addition of exogenous GzmB, we also included 10 μM zVADfmk in all reactions to suppress endogenous caspase activity. Note that this polycaspase inhibitor does not inhibit GzmB activity (; ). As illustrates, addition of 100 nM hGzmB to the cell-free extracts led to efficient proteolysis of BID of human and mouse origin, and to slightly less efficient proteolysis of caspase-3 and -7 in the same extracts. In contrast, although hGzmB was found to proteolytically process human caspase-8 (), it failed to cleave the mouse counterpart of this caspase (). mCaspase-8 was processed upon addition of cytochrome and dATP to J774 extracts to activate the apoptosome pathway to caspase activation (Fig. S1, ), confirming that the antibody used to detect mcaspase-8 was, indeed, capable of detecting the processed form of this enzyme. Thus, mcaspase-8 does not appear to be a substrate for hGzmB. Rather surprisingly, mGzmB failed to cleave BID of either species under conditions where hGzmB clearly did so (). Importantly, mGzmB cleaved both human and mouse caspase-3 and -7 very efficiently under the same conditions (), thereby precluding the possibility that mGzmB was not active in these experiments. In addition to the differences observed in respect to BID proteolysis, it can also be seen that mGzmB failed to cleave caspase-8 from either species (). We also performed titrations of human and mouse GzmB in Jurkat and J774 cell-free extracts and, again, observed marked differences in the ability of these enzymes to cleave BID and caspase-8 (). We also assessed proteolysis of nucleophosmin/B23 in these assays, as this protein has recently been reported to be a substrate for hGzmB (). However, whereas hGzmB efficiently cleaved nucleophosmin of human origin, mGzmB cleaved this protein very inefficiently under the same conditions (). Moreover, neither human nor mouse GzmB cleaved nucleophosmin within J774 cell- free extracts (). Thus, although human and mouse GzmB can proteolytically process caspase-3 and -7 (whether of mouse or human origin), these granzymes exhibited distinct differences in respect to their ability to cleave BID, caspase-8, and nucleophosmin. To exclude the possibility that the results we observed could be indirect effects, e.g., caused by activation of caspases within the cell-free extracts, we repeated these experiments using purified recombinant human and mouse BID. As illustrates, both human and mouse BID were cleaved efficiently by hGzmB. As a control, we also compared the efficiency of proteolysis of recombinant pro–caspase-7, and this was cleaved by hGzmB with comparable efficiency to that of BID (). However, in complete agreement with the results generated using cell-free extracts, although mGzmB cleaved caspase-7 with an essentially identical efficiency to hGzmB, it failed to exert any detectable proteolytic activity toward BID from either species (). These data provide robust evidence that human and mouse GzmB display divergent activity toward BID, but similar activity toward caspase-7. We then extended our analysis to several other proteins that have been reported to be substrates for GzmB. ICAD, the inhibitory subunit for the CAD deoxyribonuclease that plays a role in apoptosis-associated DNA hydrolysis, has also been reported to be a direct substrate for GzmB (; ). Consistent with this, hGzmB cleaved ICAD efficiently in Jurkat cell-free extracts (). However, mGzmB failed to cleave human ICAD over a similar concentration range, which again underscores the differences between the human and mouse granzymes in terms of their substrate preferences (). We conducted the same experiment in J774 (mouse) cell-free extracts and, similar to our earlier observations made with mcaspase-8 (), neither GzmB species cleaved mouse ICAD under the conditions where human ICAD was readily cleaved by hGzmB (). We also assessed several additional GzmB substrates in these assays (cochaperone p23, CD2-associated protein, and α-tubulin), and in each case found substantial differences between the activity of human and mouse GzmB toward these substrates (). From the preceding experiments, it was apparent that human and mouse BID are poor substrates for mGzmB, whereas both human and murine caspases -3 and -7 are readily cleaved by this granzyme ( and ). In contrast, human GzmB cleaved BID of either species very efficiently ( and ). An important implication of these observations is that murine GzmB may be unable to activate the BID pathway to kill target cells and may rely predominantly on direct activation of caspases. This suggests that inhibition of caspases may clonogenically rescue cells from mouse, but not human, GzmB. To test this possibility, we loaded human CEM cells with either human or mouse GzmB using the pore-forming protein streptolysin O (SLO). As shown in , loading of CEM cells with human or mouse GzmB led to very rapid death (within 6 h) with all of the typical features of apoptosis. Inclusion of the caspase inhibitor zVADfmk in these assays transiently delayed apoptosis induced by hGzmB (). However, within 48–72 h, cells treated with hGzmB had essentially all died, irrespective of the presence of the caspase-inhibitory peptide (). In contrast, when the same cells were loaded with mGzmB at concentrations that achieved ∼95% cell death within 6 h in the absence of zVADfmk (), these cells were substantially protected when the caspase inhibitor was included in the assay. Moreover, the protection afforded by zVADfmk was long-term, as these cells proliferated upon subsequent culture (). Essentially identical results were also observed using human Jurkat cells in the same assay (), and these cells could also establish colonies when plated in soft agar (). Although caspase inhibition selectively maintained the clonogenic potential of two different human cell lines treated with mGzmB, it remained possible that this granzyme could kill mouse cells in a caspase-independent manner. Thus, we performed the reciprocal experiment using mouse J774 cells, and in this case we used a different polycaspase inhibitor, Q-VD-Oph (). As shows, whereas J774 cells were readily killed by both GzmB species, inhibition of caspases only restored clonogenic potential in cells that were treated with mGzmB. Furthermore, under conditions where mGzmB efficiently killed J774 cells, no detectable processing of BID was observed, whereas caspase-3 processing was readily detected (). In contrast, hGzmB cleaved both BID and caspase-3 in the same cells (). Using MCF-7 cells deficient in caspase-3, we also explored whether the absence of this caspase was sufficient, on its own, to confer protection against mGzmB. For comparison, we also used MCF-7 cells reconstituted with caspase-3 by stable transfection of a expression plasmid (). However, although the absence of caspase-3 did delay the onset of GzmB-initiated cell death, no long-term protection was seen in the absence of this caspase (Fig. S2, available at ). This suggests that GzmB-mediated processing of other caspases, such as caspase-7 ( and ), is probably sufficient to kill cells in the absence of caspase-3, and that inhibition of both proteases is necessary to achieve clonogenic protection from mGzmB. We have shown that human and mouse GzmB behave as distinct proteases and share only partial overlap in their substrate range. In particular, whereas hGzmB efficiently cleaved the BH3-only protein BID, mGzmB failed to do so. Similarly, hGzmB promoted efficient proteolysis of the human forms of ICAD and caspase-8, whereas mGzmB failed to cleave these proteins. Interestingly, neither human nor mouse GzmB cleaved murine ICAD or murine caspase-8 under the conditions used in this study. Several other examples of divergent substrate preferences between human and murine GzmB were found. Collectively, these data demonstrate that human and mouse GzmB exhibit clear differences in the cohort of proteins that these proteases target upon entry into target cells. These observations are also supported by a study from that was published during revision of our study, as these authors also found similar functional divergence between human and mouse GzmB. There has been much debate concerning the relative importance of the substrates cleaved by GzmB (; ; ; ). Although some have suggested that BID may be the most relevant GzmB substrate during CTL attack, others have suggested that BID proteolysis is indirect and is mediated by caspase-3, which is also activated by GzmB (). The contention that BID is a preferential substrate for GzmB is supported by studies that have shown that Bcl-2 can block GzmB-initiated apoptosis, whereas caspase inhibitors fail to do so (). The latter observations would support an upstream role for BID, independent of caspase activity, in promoting mitochondrial outer membrane permeabilization as a result of Bax/Bak activation. In this scenario, inhibition of caspase activity would fail to block mitochondrial outer membrane permeabilization, and cells would die in a caspase-independent manner. Bcl-2 would be expected to block death in this context through antagonizing the effects of BID on mitochondria. The observation that BID is a poor substrate for mGzmB suggests that this protease does not efficiently activate the BID pathway to cell death in the mouse. Consistent with this, although inhibition of caspase activity failed to clonogenically rescue cells from hGzmB, clonogenic rescue was readily seen when mGzmB was used in the same context. This was true irrespective of whether human or mouse target cells were used. These observations have important implications for studies where murine granzymes were evaluated in human cell lines and vice versa. For example, recent studies have used BID-null mouse embryonic fibroblasts in conjunction with hGzmB to conclude that BID is an essential target for GzmB in this system (). Our observations suggest that mGzmB kills in a predominantly caspase-dependent and BID-independent manner, whereas the reverse could be true for hGzmB. However, an alternative interpretation is that BID is not important for GzmB-mediated cell death in either organism. Thus, conclusions based on mixing human enzymes with mouse substrates (or the reverse) should be interpreted with caution. The implications of our findings also extend to other related proteases, such as caspases, where it is often assumed that human and murine orthologues perform essentially identical functions. However, the sequence identity between human and murine caspase orthologues is typically much lower (on average 40%) than seen with GzmB. This makes it even more likely that functionally important differences in the substrate range of human and mouse caspase orthologues will emerge upon examination of this issue. Indeed, there is already one clear example of sequence divergence between murine and human caspases that has major functional consequences. Whereas murine caspase-12 appears to be a functional enzyme (), human caspase-12 has acquired substitutions that render this caspase inactive (). Although this is clearly an extreme example, it is likely that less obvious examples of functional divergence between caspase orthologues will emerge, as we have seen for GzmB, when this is explored. In summary, our data demonstrate that human and murine GzmB behave as distinct enzymes with divergent substrate preferences. These observations illustrate how subtle differences in enzyme structure can radically affect substrate selection. The following antibodies were used: anti–caspase-3, anti–human caspase-7, anti-XIAP, and anti–human BID (BD Biosciences); anti–mouse caspase-7 (Millipore); anti–mouse BID (R&D Systems); anti–human caspase-8 (Oncogene); anti–mouse caspase-8, (Alexis); anti–α-tubulin and anti-actin (ICN, UK); anti-p23 (Affinity BioReagents); anti-nucleophosmin (Invitrogen); anti–CD2-associated protein (Santa Cruz Biotechnology, Inc.); anti-DFF45/ICAD (Calbiochem and Biosource). The peptides zVADfmk and Ac-DEVD-AFC were purchased from Bachem, IETD-AFC was purchased from Alexis, Noble agar was purchased from BD Biosciences, and SLO was purchased from Aalto Bio Reagents. Vivaspin 0.5 ml microcentrifuge concentrators were purchased from Sartorius. Unless otherwise indicated, all other reagents were purchased from Sigma-Aldrich. The cDNA encoding the mature form of mouse GzmB was cloned into the expression vector pPIC6α (Invitrogen). X-33 cells were transformed with pPIC6α.mGzmB via electroporation and were put through two rounds of selection in 900 μg/ml blasticidin, resulting in X-33 clones stably harboring mouse GzmB expression plasmid. yeast clones stably harboring a human GzmB expression plasmid were provided by W. Wels (Chemotherapeutisches Forschungsinstitut Georg-Speyer-Haus, Frankfurt, Germany). Yeast clones expressing hGzmB were grown in 1-liter cultures for 3 d at 25°C, whereas clones expressing mGzmB were grown in 2-liter cultures for 4 d at 25°C. In both cases, GzmB was purified from culture supernatants using nickel affinity chromatography, followed by extensive washing in buffer containing 5 mM imidazole. Purified GzmB was eluted with 500 mM imidazole and washed extensively in PBS, pH 7.2, followed by concentration in microconcentrator units (Sartorius). The pET15b expression plasmid encoding full-length human BID was provided by X.-M. Yin (University of Pittsburgh School of Medicine, Pittsburgh, PA), the pET23dwHis expression plasmid containing full-length mouse BID was provided by X. Wang (University of Texas Southwestern Medical Center, Dallas, TX), and G.S. Salvesen (The Burnham Institute of Medical Research, Pittsburgh, PA) provided the pET23b expression plasmid containing full-length human caspase-7. Expression plasmids were transformed into the bacterial strain BL21 (DE3) and were expressed and affinity purified as previously described (), with the exception that caspase-7 was induced for 40 min to minimize the production of the processed form of this protease. Full-length CrmA was cloned into the expression plasmid pGEX4TK2 (GE Healthcare) in frame with the GST coding sequence at the N terminus. GST-CrmA was expressed in DH5α and affinity purified using glutathione–Sepharose as previously described (). 1 μM GST-CrmA was activated by incubation with 500 μM dithiothreitol for 10 min at 37°C. Active CrmA was then incubated with human or mouse GzmB for 10 min at 37°C. Residual GzmB activity was then determined by diluting samples into PBS containing 500 μM Ac-IETD-AFC and measuring hydrolysis of the latter peptide by fluorimetery, as described in Fluorometric assays. Typically, 3–4 μg of recombinant human or mouse BID was incubated for 2–3 h at 37°C with varying concentrations of human or mouse GzmB in 10 μl reaction volumes. Human recombinant pro–caspase-7 was treated in an identical manner, but with 10 μM zVADfmk added to inhibit caspase autoprocessing. Reaction products were separated by SDS-PAGE and visualized by Coomassie blue staining. Densitometry was performed using ImageJ software (National Institutes of Health; ) and catalytic rates were calculated using these data. Cell-free cytosolic S-15 extracts were generated from Jurkat and J774 cells, as previously described (; ). Typically, human and murine extracts were normalized to 10 mg/ml total protein content, and then diluted twofold in WCEB (20 mM Hepes, pH 7.5, 10 mM KCl, 1.5 mM MgCl, 1 mM EDTA, 1 mM EGTA, 1 mM DTT, 100 μM PMSF, 10 μg/ml leupeptin, and 2 μg/ml aprotinin). GzmB and zVADfmk were added to the reactions at final concentrations of 6–100 nM and 10 μM, respectively. Reactions were incubated for 1–2 h at 37°C to facilitate GzmB-mediated proteolysis. Cell-free reactions were then assessed by immunoblotting or fluorimetric assay. Reactions containing Jurkat or J774 cell-free extracts (normalized for total protein content) were typically assembled in a final volume of 10 μl. After a 30-min incubation at 37°C, 2.5-μl samples were diluted to a final volume of 100 μl in WCEB containing 50 μM Ac-DEVD-AFC. To determine the activity of recombinant GzmB, protein was diluted in 30 μl PBS containing 500 μM Ac-IETD-AFC. All samples were measured in an automated fluorimeter (Spectrafluor Plus; TECAN) at wavelengths of 430 (excitation) and 535 nm (emission). For clonogenic assays on Jurkat and J774 cells, a 3-ml bottom agar layer was poured on 6-cm culture dishes containing RPMI with 20% fetal calf serum, 1% -Glutamine, 2 mg/ml sodium bicarbonate, 0.001 mg/ml folic acid, 10% cell-conditioned medium, and 0.5% Noble agar, and this was allowed to set for 30 min. A top layer, containing a 100-μl aliquot of cells, was added to 4 ml of the same medium containing 0.33% Noble agar (prewarmed to 37°C), and this was poured over the bottom layer and allowed to set. The agar layer was then overlaid with 2.5 ml of the medium (without agar) and incubated at 37°C for 3–4 wk. Plates were washed briefly in PBS, pH 7.2, and stained for 30 min with a solution of 0.05% crystal violet in PBS, pH 7.2, containing 20% methanol. For clonogenic assays using MCF-7 cells, cells were plated on 6-cm culture dishes with RPMI, 20% fetal calf serum, 1% -Glutamine, and 10% cell-conditioned medium and incubated at 37°C for 1 wk. Plates were then washed briefly in PBS, pH 7.2, and stained for 30 min with a solution of 0.05% crystal violet in PBS, pH 7.2, containing 3% paraformaldehyde. Typically, SLO and GzmB were added to cells (at concentrations indicated in the figure legends) at 10 cells/ml. For clonogenic assays, 30,000 cells were treated in a final volume of 30 μl in 96-well U-bottomed plates, followed by incubation at 37°C for 45 min. SLO was deactivated by diluting cell cultures 1:10 with RPMI containing 10% FCS, and cells were transferred to 96-well flat-bottomed plates for further observation. Cell death was subsequently assessed by flow cytometry using annexin V/propidium iodide (PI) staining (). The cell images presented in , , , and Fig. S2 A were taken on an inverted microscope (IX71; Olympus) using a 40× objective, and images were captured using a Colorview II camera (Soft Imaging System) equipped with Analysis image acquisition software. All images were acquired at room temperature. Images of crystal violet–stained colony formation assays presented in , , and Fig. S2 D were captured on a digital camera (CoolPix 5000; Nikon) and imported into Photoshop (Adobe) for uniform brightness and contrast adjustment. Densitometric analysis of the Coomassie blue–stained gels presented in was performed using ImageJ software. All Western blot images were scanned from x-ray film using a scanner (Epson Perfection) and Photoshop software. All figures were prepared using PageMaker 7.0 software (Adobe). Fig. S1 shows the cytochrome /dATP–induced processing of caspase-8 in mouse J774 cell-free extracts. Fig. S2 shows that caspase-3 deficiency is not sufficient to afford clonogenic rescue from human or mouse GzmB. Online supplemental material is available at .

Notch signaling regulates a diverse array of cell fates and cellular processes during embryonic development and contributes to adult homeostasis. Notch is a cell surface receptor that not only functions in ligand binding but is also the downstream signal transducer through a process of regulated intramembrane proteolysis (RIP; ). This mode of signaling depends on prior furin-mediated proteolysis to form an intramolecular, heterodimeric Notch (hNotch) receptor (; ; ). The extracellular and membrane-bound intracellular furin-cleavage fragments of hNotch are held together through noncovalent interactions that prevent receptor activation in the absence of ligand (; ). Binding of DSL (Delta/Serrate/Lag-2) ligands to hNotch activates signaling by inducing additional proteolysis, first within the Notch extracellular domain (NECD) via a disintegrin and metalloprotease (ADAM), which facilitates γ-secretase proteolysis within the membrane-spanning region to release the Notch intracellular domain (NICD; ; ). Trans location of NICD to the nucleus allows it to interact with the DNA-binding protein CSL (CBF1, SuH, LAG-1) and recruit coactivators to activate transcription of Notch target genes (). Although activating proteases have been identified, the mechanism by which ligand binding leads to Notch proteolysis is still not well understood. DSL ligands, like Notch, are type 1 transmembrane proteins, and, accordingly, activation of Notch signaling requires direct cell–cell contact. Interestingly, endocytosis in the ligand cell is required to induce a signal in the Notch cell, suggesting additional roles beyond ligand presentation (; ; ). Studies in first suggested that ligand endocytosis of bound Notch promotes ADAM cleavage, leading to receptor dissociation and signaling (). The exclusive uptake of the Notch ectodomain by Delta cells imaged in flies (; ) is consistent with the idea that NECD sequences prevent receptor activation and must be removed before Notch can be proteolytically activated. Indeed, truncation of NECD sequences yields forms of Notch that are constitutively cleaved in the absence of ligand (; ; ; ). Furthermore, dissociation of mammalian hNotch via calcium chelators () or mutations within the heterodimerization domain mimics signaling induced by DSL ligands (). That activating heterodimerization mutations are responsible for aberrant Notch signaling in T-cell acute lymphoblastic leukemia () provides additional support for Notch dissociation in receptor activation. The NECD transendocytosis model for ligand activation of Notch is appealing; however, the ubiquitous expression of Notch makes it difficult to be certain that NECD imaged in Delta cells was actually donated by the neighboring Notch cell. Interpretation of such immunolocalization studies is further complicated by the exchange of full-length Delta and Notch between interacting cells (; ; ). Thus, the staining patterns could also represent internalization of cell surface Notch with Delta within the same cell, rather than transfer between cells. To determine if activation of mammalian Notch signaling involves NECD transendocytosis, and to dissect the relative roles of endocytosis versus ADAM cleavage in receptor dissociation, we established a coculture system in which epitope-tagged forms of Notch permit direct visual confirmation for transfer of either NECD alone and/or intact Notch to DSL ligand cells. Our data reveal that after interaction with Notch cells, DSL ligand cells preferentially take up NECD, whereas the bulk of the NICD remains in the receiving cell, where it undergoes proteolytic activation. In addition, both receptor dissociation and internalization of NECD by DSL ligand cells require hNotch formation by furin processing, yet occur independent of ADAM proteolysis. Based on our findings, we conclude that hNotch must first dissociate before activating Notch proteolysis can occur, and we propose that DSL ligand endocytosis participates in the physical dissociation of Notch, rather than promoting enzymatic dissociation as previously suggested (). To unambiguously determine if Notch signaling in mammalian cells also involves transendocytosis of NECD by interacting ligand cells, and to circumvent the problems associated with ubiquitous expression of Notch, we engineered epitope-tagged forms of Notch1 (N1) and stably expressed them in C2C12 cells for use in coculture studies. We have previously reported that L cells stably expressing DSL ligands (Delta-like1 [Dll1] or Jagged1 [J1]) activate Notch signaling when cocultured with N1-expressing cells (; ; ). Based on these studies, Dll1 cells were cocultured with cells expressing an N-terminal HA-tagged N1 (HA-N1) in the presence of rabbit polyclonal antibodies to the Dll1 extracellular domain (148G), as outlined in . After incubation, the cells were fixed, permeabilized, and stained with both a fluorescently conjugated mouse monoclonal HA antibody, to identify HA-N1, and a fluorescently conjugated anti–rabbit antibody, to detect surface and internalized Dll1. Confocal imaging revealed HA puncta polarized toward Dll1 cells at sites of cell–cell contact, producing a nonuniform punctate expression pattern for N1 within cell membranes (, arrows). Importantly, this ligand-induced receptor clustering and nonuniform punctate expression pattern was not detected for HA-N1 cells in direct contact with parental L cells (). Moreover, the acquired images suggest that HA and Dll1 signals colocalize within vesicular structures of Dll1 cells (, arrowheads). Although confocal sectioning suggested that the HA-Dll1–positive vesicles were intracellular (Fig. S1, available at ), we designed a staining protocol () to distinguish HA staining at the surface from that located inside Dll1 cells. In support of this staining protocol, cells expressing a C-terminal HA-tagged Dll1 showed a signal only after permeabilization, whereas an extracellular HA-tagged Dll1 was detected by staining with HA antibodies both before and after permeabilization (). Given this validation, cocultures stained in this manner revealed signals for Dll1 and HA-N1 at the cell surface (, arrows), whereas permeabilization and staining with the second HA antibody identified a vesicular signal within Dll1 cells (arrowheads), which is consistent with transfer of HA-N1 to Dll1 cells. Similar staining patterns were also observed for HA-N1 cells cocultured with J1 cells (), suggesting that Notch transendocytosis is a general phenomenon of DSL ligands (), as previously reported for the two ligands, Delta and Serrate (). Dissociation of Notch by DSL ligand cells in has been linked to signaling, but its requirement in activating proteolysis has not been directly demonstrated. To address this question, cocultures were stained with both an antibody specific for the γ-secretase–cleaved active NICD (Val1744), as well as for HA antibodies to mark the N1 N-terminal tag. To enhance detection of active NICD, proteosome inhibitors were used to prevent its rapid turnover (). As discussed in the previous section, ligand cells in direct contact with HA-N1 cells typically induced a polarized punctate HA signal at sites of cell–cell contact (, arrows), which is indicative of ligand-induced receptor clustering. Such interacting cells were then scored for HA puncta associated with ligand cells (, arrowheads), which is suggestive of HA-N1 transfer, as well as a signal for active NICD in the nucleus of HA-N1 cells, which is indicative of Notch signaling. This analysis revealed that 60–70% of these interacting cells displayed both a punctate HA signal associated with ligand cells and a signal for active NICD in the nucleus of HA-N1 cells (). Receptor clustering and transendocytosis by DSL ligand cells, as well as a signal for active NICD in the nuclei of HA-N1 cells, all correlated with activation of a Notch reporter in coculture assays (). In contrast, HA-N1 cells in direct contact with parental L cells did not display polarized HA-positive puncta or a signal for active NICD, and the interacting L cells were negative for HA staining (). In addition, cells expressing an N1 protein tagged at its C terminus with EGFP displayed signals for both EGFP and active NICD in the nucleus that were specific for interactions with Dll1 cells (Fig. S2, available at ). Together, our findings suggest that DSL ligand cells interact with Notch cells to induce receptor clustering and transendocytosis, and that these events promote Notch proteolysis and downstream signaling. Importantly, this coculture system has provided direct support for N1 transendocytosis by DSL ligand cells, and has allowed us to investigate the requirements for transfer of Notch to ligand cells, as well as the relevance of such transfer to signaling. The intracellular HA-positive vesicular signal detected for ligand cells in direct contact with HA-N1 cells ( and ) could be caused by the transfer of the NECD alone and/or the entire N1 protein. In transendocytosis of full-length Delta and Notch, as well as a multipass membrane protein, BOSS (), has been reported. In mammalian systems, transendocytosis of full-length ephrins and EPH receptors is important for biological effects mediated by this signaling system (; ). However, because we find that transfer of N1 to ligand cells correlates with a signal for active NICD in interacting N1 cells, transfer of the N-terminal subunit exclusive of its membrane-bound C-terminal subunit must occur. To directly address the relative levels of NECD versus full-length N1 transfer, we examined cells expressing an N-terminal HA and C-terminal EGFP-tagged N1 protein (HA-N1-EGFP). Signals for both tags colocalized to large puncta at the N1 cell periphery and within membrane extensions polarized toward ligand cells (, arrows and overlay). In contrast, vesicular structures associated with ligand cells were primarily positive for the N-terminal tag (, arrowheads and overlay), which is indicative of NECD in the absence of the C-terminal subunit. To quantitate the transfer of NECD versus full-length N1 for both ligand and N1 cells, we determined the dissociation ratio, which is a measure of the HA–N-terminal tag relative to that detected for the EGFP–C-terminal tag (Fig. S3, available at ; see Materials and methods). We reasoned that if NECD was being removed from the N1 cell and transferred to the ligand cell then the dissociation ratio calculated for ligand cells should be greater than that determined for N1 cells, where similar levels of both tags would be detected if the receptor remained intact. The dissociation ratio was higher for ligand cells compared with N1 cells (), suggesting that the majority of the transendocytosed signal represents transfer of NECD independent of NICD. Transendocytosis of NECD alone by ligand cells may depend on ADAM cleavage of Notch. Alternatively, ADAM proteolysis may require receptor dissociation and NECD removal. To distinguish between these possibilities, we examined the effect of inhibiting metalloproteases with BB94 on HA-N1 transendocytosis and receptor dissociation. Specific staining to identify intracellular HA signals identified signals within Dll1 cells for both BB94 and DMSO control cocultures (; arrowheads). Examination of cells expressing a double-tagged HA-N1-EGFP identified Dll1 cells with HA signals in the absence of EGFP signals (; arrowheads), which was not the case for Notch cells. Consistent with this, the dissociation ratio determined for Dll1 cells compared with Notch cells was considerably higher (). Importantly, BB94 treatment did not alter the dissociation ratios, suggesting that metalloprotease inhibition does not perturb the separation and removal of the N-terminal tag from its C-terminal tag on HA-N1-EGFP. Despite detection of NECD transendocytosis, NICD generation () and Notch reporter activity () were reduced by the addition of BB94, underscoring the importance of ADAM proteolysis in Notch activation. Although ADAM proteolysis is required for NICD generation, our findings provide the first evidence that ADAM proteolysis does not function in receptor dissociation. Not surprisingly, the γ-secretase inhibitor DAPT prevented a signal for activated NICD () and reporter activity (); however, NECD transendocytosis () and separation of the N and C termini () were unaffected. Together, our findings suggest that NECD release and transendocytosis mediated by Dll1 cells precede and facilitate Notch proteolysis, rather than being the results of ADAM cleavage, as previously suggested. Because ADAM proteolysis was not required for NECD transendocytosis, we reasoned that the physical dissociation of NECD to Dll1 cells might rely on hNotch1 formation that requires furin processing (; ). To investigate this, we engineered N- and C-terminal–tagged versions of a previously described N1 mutant defective in furin processing and hNotch1 formation (). Importantly, this mutational approach should specifically block NECD transendocytosis without affecting ligand endocytosis, and determine if specific transendocytosis of NECD is required for signaling. Western blot analysis of HA-N1-EGFP–expressing cells identified cleaved forms of both endogenous and ectopic N1 proteins, the latter having decreased mobility because of its C-terminal EGFP tag (). Although HA-N1ΔFC-EGFP cells express cleaved endogenous N1, a cleaved ectopic N1 was not detected. Importantly, surface labeling with biotin identified both uncleaved and cleaved forms of HA-N1-EGFP, whereas only an uncleaved form was detected on the surface of HA-N1ΔFC-EGFP cells (). This protein analysis confirms that the mutant HA-N1ΔFC-EGFP protein is not proteolytically processed into a heterodimeric form, but instead exists at the cell surface as an uncleaved receptor. Imaging of cells stably expressing HA-N1ΔFC cocultured with Dll1 cells revealed clusters of HA-positive puncta at borders of interacting cells (, arrow), indicative of ligand–receptor interactions. Despite this ligand-induced receptor clustering, detection of the HA–N-terminal signal within interacting Dll1 cells was significantly reduced, but not completely eliminated (). However, given that this particular analysis detects only the N-terminal tag, it is possible that the signal represents internalization of intact HA-N1ΔFC. To investigate this, we examined cells expressing a double-tagged HA-N1ΔFC-EGFP, and found puncta positive for both tags within N1 cell membranes (, arrows and overlay), as well as associated with Dll1 cells (arrowheads), which is suggestive of the transfer of intact receptor. Consistent with this, the HA-N1ΔFC-EGFP dissociation ratio calculated for Dll1 cells was close to 1.0, reflecting an equivalent transfer of both tags (). EGFP fluorescence calculated for the entire mutant-expressing cell was decreased compared with the N-terminal tag (Fig. S4, available at ), resulting in a dissociation ratio >1.0; however, similar dissociation ratios were calculated for wild-type– and mutant-tagged proteins within interacting cell membranes (), and both tags were congruent at the cell surface, where interactions with ligand cells would occur (compare and , arrows). It is possible that conformational changes induced by the deletion alter detection of the C-terminal tag on proteins within the cell. The losses in dissociation calculated for HA-N1ΔFC-EGFP cells correlate with significant losses in active NICD and reporter activity (). The signaling detected with HA-N1ΔFC-EGFP cells is likely caused by activation of endogenous N1 (). Together, our analysis of the furin cleavage mutant indicates that losses in hNotch formation are associated with significant losses in receptor dissociation and signaling, indicating that heterodimer formation is required for NECD transendocytosis by ligand cells, and that this event is a prerequisite for Notch activation. Our data indicate that NECD transendocytosis occurs independent of ADAM proteolysis, but requires a furin-processed heterodimer. Given the obligatory role of dynamin-dependent endocytosis in Notch signaling (), we next asked if Dll1 endocytosis might contribute to heterodimer dissociation. For these studies, we generated Dll1 mutants lacking most (Dll1ΔICD), or all (Dll10CD), intracellular sequences because these sequences are required for Delta endocytosis (). Removal of Dll1 intracellular sequences prevented uptake of Dll1 extracellular antibodies, as well as a soluble form of N1 (N1Fc; ). However, internalization of transferrin by Dll1ΔICD cells was unaffected (), indicating that the block in endocytosis is Dll1 specific. Endocytosis-defective Dll1ΔICD and Dll10CD cells induced clustering of tagged forms of N1 at sites of cell–cell contact (, arrows); however, both NECD transendocytosis () and the ligand cell dissociation ratio were significantly reduced compared with full-length Dll1 (), indicating that ligand-mediated endocytosis of Notch is required to separate the N- and C-terminal subunits. Importantly, losses in NECD transendocytosis associated with Dll1ΔICD and Dll10CD cells correlate with a decrease in NICD-positive nuclei (, G and H; not depicted) and reporter activity (; not depicted), highlighting a critical requirement for Dll1-specific endocytosis in Notch signaling, as well as a specific role for ligand activity beyond receptor binding and clustering. To directly address a requirement for ligand cell endocytosis in NECD transendocytosis, we transiently expressed mutant forms of dynamin (dynk44A-EGFP; ) or Eps15 (EGFP-Eps15DIII; ) in Dll1 cells because these constructs are known to inhibit endocytosis. Expression of either dynk44A-EGFP or EGFP-Eps15DIII suppressed transfer of HA-N1 to Dll1 cells () and reduced ligand-induced Notch signaling (). Consistent with the losses in NECD transfer, when either dynamin or Eps15 activities were inhibited, NECD internalized by Dll1 cells overlapped with transferrin, a known cargo of clathrin-mediated endocytosis that requires both dynamin and Eps15 (). Given that perturbation of either dynamin or Eps15 activity did not decrease Dll1 cell surface expression or N1Fc binding (Fig. S5, available at ; not depicted), the losses in reporter activity () indicate that NECD transendocytosis by Dll1 cells is required to generate a signal in interacting Notch1 cells. If hNotch dissociation is driven by ligand endocytosis, and if both events are required for Notch proteolysis then this could explain, in part, the requirement for DSL ligand endocytosis in Notch signaling. Our findings suggest a two-step model in which ligand endocytosis induces nonenzymatic dissociation of bound Notch to allow uptake of NECD by ligand cells (, step 1). The remaining, membrane-bound hNotch subunit (NTM/S1) would undergo constitutive cleavage by ADAM and γ-secretase to produce NEXT/S2 and NICD/S3, respectively (, step 2). In support of this model, we found that a recombinant NTM/S1 form was constitutively active in a reporter assay, and that BB94 reduced the level of signaling (), indicating that ADAM cleavage is required for maximal activity. Moreover, losses in active NICD/S3 (detected by the Val1744 antibody) produced by BB94 and DAPT provide additional evidence for constitutive cleavage of recombinant S1 by both ADAM and γ-secretase, respectively (). Nonetheless, to directly demonstrate that S1 undergoes ADAM cleavage to produce the transient S2 form, it was necessary to truncate and myc tag recombinant S1 to resolve the S1, S2, and S3 forms by SDS-PAGE. After expression in COS cells, both S2 and S3 cleavage products were detected (); however, ADAM inhibition decreased the appearance of S2 and S3, whereas DAPT blocked the appearance of S3 and lead to the accumulation of S2, indicating that S1 is an ADAM substrate (; ). Together these findings are consistent with our model that after hNotch1 dissociation, membrane-bound S1 is constitutively cleaved first by ADAM, and then by γ-secretase. The phenomenon of Notch transendocytosis by ligand cells has been described in and linked to Notch signaling (), yet the mechanism and relevance of Notch uptake by ligand cells to Notch activation remains controversial (). In fact, Notch transendocytosis via interacting ligand cells has been suggested to simply reflect clearance of the ADAM-shed Notch ectodomain, rather than directly affecting Notch proteolysis (). To explore the idea that NECD transendocytosis represents a critical step in Notch activation, we developed a system that employs dual epitope-tagged proteins to directly image the transfer of NECD to interacting ligand cells. Our results both confirm and extend the studies in , where distinct cellular staining patterns for the Notch extracellular and intracellular domains were interpreted to represent dissociation of the Notch receptor after ligand activation (; ; ). We found that mammalian N1 receptors were clustered by, transferred to, and taken up by ligand cells, and that these events correlate with detection of cleaved, active NICD in the nucleus of contacted Notch cells. We conclude that ligand activation of Notch involves separation of the heterodimer, and that this allows translocation of the dissociated NECD to the ligand cell and of the cleaved NICD to the nucleus of the Notch cell. The dependence of NECD transendocytosis on functional ligand endocytosis first suggested that ligand internalization of bound Notch imparts a molecular strain to induce conformational changes that allow ADAM cleavage within the NECD (). However, other studies have suggested that ADAM cleavage is independent of ligand endocytosis, and that activating γ-secretase cleavage occurs only after the ADAM-cleaved NECD is removed by endocytosis (). We initially favored the idea that removal of NECD from the intact Notch receptor would require ADAM proteolysis, but we found that metalloprotease inhibition did not impair heterodimer dissociation or NECD transendocytosis. Although we cannot rule out that proteases other than metalloproteases might cleave NECD, our findings indicate that Notch dissociation and transendocytosis occur independently of, and before, ADAM cleavage, rather than arise as a consequence of proteolysis. Additional support for this idea comes from our demonstration that the secreted extracellular matrix proteins microfibril-associated glycoprotein 1 and 2 activate Notch signaling through hN1 dissociation, independent of ADAM proteolysis (). Our data indicate that ADAM activity is not required for NECD transendocytosis, but we also find that ADAM activity is absolutely necessary for Notch signaling induced by DSL ligands. Furthermore, we found that forms of N1 defective in heterodimer formation were reduced in both NECD-specific transendocytosis and separation of the N- and C-terminal tags after contact with Dll1 cells. Moreover, losses in furin-generated hNotch1 correlated with losses in NICD generation and Notch reporter activation, providing a direct link between heterodimer dissociation, NECD transendocytosis, and Notch activation. Based on our findings, we suggest that receptor dissociation by ligand cells serves to convert hNotch1 from an ADAM-insensitive substrate into one that is readily cleaved. In this regard, the heterodimeric nature of Notch is unique among ADAM substrates, whereas the membrane-bound subunit is similar in structure to conventional ADAM substrates (). In fact, we show that the constitutive signaling activity intrinsic to a recombinant S1 form is dependent on ADAM cleavage, underscoring the role of the NECD in protecting the membrane-spanning subunit from proteolytic activation in the absence of ligand. Studies have demonstrated that ADAM cleavage is a prerequisite for γ-secretase processing (; ), and an inverse relationship between the size of the ectodomain and the efficiency of γ-secretase cleavage has been proposed (). Therefore, we conclude that the ADAM requirement in Notch signaling reflects a role for this protease in trimming the extracellular sequences of the membrane-bound subunit, which would allow efficient γ-secretase cleavage to produce appropriate levels of NICD for biological responses. We found that endocytic-defective ligands were unable to efficiently dissociate hNotch1 and activate Notch signaling. Interestingly, losses in ligand endocytosis that perturb signaling did not diminish receptor binding or clustering, which is consistent with our previous report that ligand binding is not sufficient to induce Notch signaling (). We demonstrate that in addition to ligand binding, productive Notch signaling requires receptor dissociation that is promoted by ligand endocytosis. Although a low level of signaling was detected when ligand endocytosis was defective, this activity may reflect cell detachment and/or Notch endocytosis, which could produce sufficient force for receptor dissociation. In fact, prefixed Delta-expressing S2 cells (that are presumably endocytosis defective) have also been reported to activate Notch target genes (). Moreover, soluble DSL ligands are not as potent as cell-associated ligands () and must be preclustered (; ; ) and/or immobilized (). Although it is unknown whether endocytosis, by itself, could provide such a physical force, both the actin cytoskeleton and dynamin have been implicated in inducing membrane constriction and tension during the process of endocytosis (; ). In summary, this study has defined a direct role for NECD transendocytosis by DSL ligand cells in the activation of Notch signaling, and established a system to study the underlying mechanisms. In contrast to current models, we found that NECD transendocytosis occurs independent of ADAM proteolysis and requires separation of the hNotch1 subunits, which are driven not by ligand binding, but by ligand endocytosis. Previous studies have proposed that endocytosis of ligand–receptor complexes produces a change in Notch receptors that promotes proteolytic activation (). Our findings suggest that physical separation of the Notch heterodimer is the structural change induced by ligand endocytosis, and we further speculate that ligand endocytosis acts via a mechanical force to first disrupt the noncovalently attached hNotch subunits, and that dissociation is required for proteolytic activation. Our model proposes that a critical event in Notch signaling is nonenzymatic dissociation of the Notch receptor, bringing Notch activation closer to the realm of mechanotransduction than previously proposed proteolytic cleavage models. Stable Dll1, J1, and N1 cell lines were generated and cultured as previously described (; ; ; ). The cDNA constructs used were as follows: pBos-rDll1, pBos-rDll1ΔICD (rDll1 truncated at D573), pBos-rDll10CD (rDll1 truncated at V560), pBos-HA-rDll1 (three tandem HA epitopes inserted between P536 and W537 of rDll1), pBos-rDll1-HA (three tandem HA epitopes inserted after V714), pBos-HA-rN1 (three tandem HA epitopes inserted between R23 and C24 of rN1), pBos-HA-rN1ΔFC (36 aa deletion from R1628 to H1663), pBos-HA-rN1-EGFP (EGFP inserted after K2561 of HA-rN1), pBos-HA-rN1ΔFC-EGFP (EGFP inserted after K2525 of HA-rN1ΔFC), pcDNA3.1-Dyn1K44A, pEGFP(N1)-Dyn1K44A, pEGFP(C2)-Eps15DIII (gifts from S. Schmid, The Scripps Research Institute, La Jolla, CA), and pEGFP(N2; CLONETCH Laboratories, Inc.). pCDNA3-p120 (a gift from J. Aster, Brigham and Women's Hospital, Boston, MA) mimics the furin cleavage fragment of human N1 (M1-R23 joined by a BamHI linker to E1666-K2555). Potential translational initiation codons (AUG, CUG, and GUG) found upstream of the S3/γ-secretase site were modified via a PCR strategy to produce pCDNA3-p120mis (mutated initiation sites). Wobble mutations were introduced into CUG and GUG codons to maintain the amino acid sequence; however, AUG methionine residues were changed to valine. The amino acid positions affected, as well as the respective altered codon sequences, are as follows: Leu1667(CTT), Met1670Val(GTA), Val1700(GTA), Val1722(GTT), Val1727(GTA), Leu1735(CTT), Met1738Val(GTA), Val1740(GTA), Val1746(GTT), Leu1748(CTA), Val1751(GTA), Val1755(GTT), Leu1756(CTT), and Leu1757(CTA). Sequencing confirmed that only intended substitutions were introduced. For p120misΔmyc, the C terminus of p120mis from M2094-K2555 was replaced with six tandem myc epitope tags. Cloning details are available upon request. C2C12 stable cell lines were cell surface biotinylated with Sulfo-NHS-Biotin. Biotinylated proteins were isolated with SAV-immobilized beads after quantification and equilibration. The SAV precipitates were analyzed by Western blotting with 93–4 serum or OPA1 antibodies (BD Biosciences). COS7 cells were transfected with p120mis or p120misΔmyc using Lipofectamine (Invitrogen). Immediately after transfection, the cells were treated either with 3 μM BB94, 10–20 μM DAPT, or DMSO for 22 h, with inclusion of 2–10 μM MG132 for the last 5 h. Cell lysates were either collected directly into 2xSB + 100 μM DTT for Western blotting with anti-myc (9E10), anti–α-tubulin, and anti-Val1744 antibodies or immunoprecipitated with antibodies to N1 intracellular domain (PCR12), followed by Western blotting with antibodies to N1 intracellular domain (93–4) or antibodies to activated NICD (Val 1744). The primary antibodies used were as follows: rabbit polyclonal anti–extracellular domain Dll1 (148G), 1:500; rabbit polyclonal anti–extracellular domain J1 (PCR8), 1:500; mouse anti-HA (262K), 1:1,000; rabbit anti-cleaved N1 (Val1744), 1:1,000 (Cell Signaling Technology); mouse anti-HA conjugated to Alexa Fluor 488 (16B12), 1:1,000; mouse anti-GFP (3E6), 1:1,000; rabbit anti-GFP conjugated to Alexa Fluor 488, 1:1,000; (Invitrogen). The secondary antibodies used were as follows: goat anti–rabbit conjugated to Alexa Fluor 633; goat anti–rabbit conjugated to Alexa Fluor 568; goat anti–rabbit conjugated to Alexa Fluor 488; goat anti–mouse conjugated to Alexa Fluor-568; and goat anti–mouse conjugated to Alexa Fluor 488, all at 1:1,000 (Invitrogen). Cells were fixed in PBS containing 4% formaldehyde and 4% sucrose for 10 min; aldehydes were quenched with 50 mM ammonium chloride in PBS for 15 min at room temperature, permeabilized with 0.1% Triton X-100 for 5 min, blocked in staining buffer (10% goat serum, 1% BSA, and 0.05% sodium azide in PBS) for 1 h, and incubated with antibodies diluted in staining buffer for 2 h or overnight. Samples were mounted in ProLong Gold (Invitrogen), and images were acquired at room temperature on an LSM 5 PASCAL laser-scanning microscope equipped with a Plan-Neofluar 100×/1.3 NA objective (both from Carl Zeiss MicroImaging, Inc.). Within each experiment, instrument settings (laser intensity, gain) were kept constant. Images were acquired and analyzed using LSM PASCAL software. Where appropriate, images are projections of several confocal sections. For all image quantification, data collected from three independent experiments were used to calculate the mean, SEM, and P value from unpaired tests. Dll1Fc binding to HEK 293T cells transfected with HA-N1 or HA-N1ΔFC was analyzed by flow cytometry (FACSCalibur; Becton Dickinson), as previously described (; ). Dll1 cells were incubated with soluble N1Fc (R&D Systems) preclustered with goat anti–human Fc (1:150; Jackson ImmunoResearch Laboratories) conjugated to FITC or Texas red and coincubated with transferrin-polylysine-FITC conjugate (Sigma-Aldrich) at 37°C before fixation and analysis. For antibody uptake, cells were incubated with 148G (1:2,000) before fixation, permeabilization, and detection with goat anti–rabbit secondary antibodies. Cell-surface staining of HA was performed on HA-N1-EGFP or HA-N1ΔFC-EGFP C2C12 cells on ice with anti-HA antibody (16B12) and goat anti–mouse conjugated to APC (Invitrogen) and analyzed by flow cytometry. Ligand-induced signaling measured by the CSL-luciferase Notch-responsive reporter were assayed using either parental L cells or stable cell lines expressing Dll1, Dll1ΔICD, or HEK293T cells transfected with Dll1 plus EGFP, Dll1 plus dynK44A-EGFP, or Dll1 plus EGFP-Eps15DIII, in the presence or absence of 5 μM BB94, 50 μM DAPT, or DMSO assayed using the Dual-Luciferase Reporter Assay System (Promega), as previously described (; ; ). Ligand-independent signaling was measured in COS7 cells transfected with Lipofectamine (Invitrogen), as described by . The lipofection mix was removed after 5 h and replaced by DME supplemented with 1% FBS and 3 μM BB94, 10 μM DAPT, or DMSO. Cells were collected after 4–5 h for luciferase assays that were performed using the Dual-Luciferase Reporter Assay System. The assay was performed twice in triplicate. Fig. S1 shows coendocytosis of Dll1 and Notch1 N terminus by Dll1 cells. Fig. S2 displays ligand-induced Notch cell activation and NICD nuclear translocation. Fig. S3 exemplifies the methodology for calculation of Notch1 dissociation ratio. Fig. S4 illustrates that deletion of the furin cleavage site in Notch1 permits surface expression and ligand binding. Fig S5 shows that inhibition of Dll1-specific, or general endocytosis does not reduce Dll1 surface expression or Notch1 binding. Online supplemental material is available at .

The location and movement of intracellular vesiculotubular structures is controlled by microtubule-dependent kinesin and dynein motor proteins, as well as actin-dependent myosin motor proteins. Microtubule-based vesicle motility usually occurs in a bidirectional, stop-and-go manner because of the alternating activities of kinesin motors for plus-end movement and dynein motors for minus-end movement toward the microtubule organizing center (MTOC; ; ; ). How motor proteins are targeted to individual vesicles, how they dock on specific receptors, and how motor activity is controlled in a spatial and temporal manner are all processes that are poorly understood. Cytoplasmic dynein is an ∼1.2-MD multisubunit protein complex, and it is the major motor for centripetal transport of membranous cargoes along microtubules (). Dynactin, which is also an ∼1.2-MD multisubunit complex, is a critical component of most, if not all, of the cytoplasmic dynein–driven activities. Dynactin participates in motor binding to microtubules (), increases motor processivity (; ), and acts as a multifunctional adaptor connecting cargo and dynein motor (; ). At least 15 subunits of the dynein–dynactin motor are identified. The 1-MD dynein heavy chain dimer and the 300-kD p150 dimer of the projecting arm of dynactin contact microtubules (). p150 is connected to the dynein heavy chain via the dynein intermediate chains () and increases dynein motor processivity (; ). The actin-related protein 1 (Arp1) subunit forms a short filament at the base of dynactin and can bind membrane-associated βIII spectrin, which probably acts as the membrane receptor for the dynein–dynactin motor complex (; ). αIβIII spectrin is located on the cytosolic side of late endocytic compartments (LEs), Golgi, and other subcellular compartments (), implying that compartment-selective dynein motor recruitment cannot be controlled by βIII spectrin itself. Small GTPases of the Rab family are present on specific subcellular compartments to regulate vesicle transport and fusion. They are ideal candidates for orchestrating the spatiotemporal regulation of motor-driven vesicle trafficking. Several Rab GTPases have been shown to interact directly or indirectly with motor proteins. These include members of the kinesin motor family (Rab4, Rab5, and Rab6), the dynein motor (Rab6 and Rab7), and the myosin motors (Rab8, Rab11, and Rab27a; ). Rab6, which regulates Golgi transport, requires the effector bicaudal-D1 and -D2 (BicD1/2) to interact with the p50 subunit of dynactin (; ) or a third protein, egalitarian (Egl), which directly interacts with the dynein light chain in (). An activation state–dependent interaction of Rab6 with p150 has also been observed in a directed two-hybrid analysis (). We have studied another Rab protein, Rab7, which, through its effector Rab7-interacting lysosomal protein (RILP), recruits the dynein–dynactin motor to LEs, resulting in minus end–driven vesicular transport to the MTOC (). The Rab7–RILP–dynein motor cascade has been shown to act on many Rab7-containing compartments, including -containing phagosomes (; ), early melanosomes (), major histocompatibility complex class II–containing compartments (MIICs; ), and cytolytic granules (). The crystal structure of Rab7 in complex with a C-terminal domain of RILP revealed the details of this interaction. RILP forms a coiled-coil homodimer with two symmetric surfaces that bind two separate Rab7–GTP molecules to form a tetrameric complex (), which has been confirmed in biochemical experiments (; ). The recent finding that a member of the oxysterol-binding protein–related protein (ORP) family, ORP1L, also interacts with Rab7 and induces clustering of LEs () complicated a simple interpretation of Rab7–RILP–controlled dynein motor recruitment. ORPs have been implicated in diverse aspects of cellular processes, including sterol and phospholipid metabolism, vesicle transport, and cell signaling (). The mechanisms by which ORP proteins contribute to these processes have, however, remained largely unknown. We recently showed that ORP1L localizes to LEs and interacts via its ankyrin repeat region with the small GTPase Rab7 (). ORP1L was shown to stabilize the GTP-bound active form of Rab7 on LEs and to affect the subcellular distribution of these organelles, analogously, to RILP (). A third Rab7 effector, Rabring7 (), clusters LEs, much like the other effectors, and induces lysosomal acidification, but dynein motor recruitment has not been shown. Surprisingly, no obvious sequence similarity is found between the three Rab7 effectors. Apparently, multiple effectors interact with Rab7. They could be mutually exclusive, but they may also interact simultaneously with this Rab GTPase. How the Rab7 effector complexes recruit the dynein motor complex is also unclear. We have studied the interaction of RILP and ORP1L with the Rab7 GTPase, as well as their interactions with dynein–dynactin motor subunits. We show that Rab7 is part of a tripartite complex binding RILP and ORP1L simultaneously. RILP is essential for dynein motor recruitment through a direct interaction with the C-terminal portion of the p150 subunit of the dynein–dynactin motor. ORP1L recruits this complex to βIII spectrin domains, which appears to be critical for dynein motor activation and minus-end transport of LEs. Rab7, thus, recruits two proteins with diverse functions in the control of dynein motor–driven transport: RILP for motor binding and ORP1L for transport to the membrane-associated late endocytic βIII spectrin receptor for motor activation. The dynein–dynactin motor, thus, requires two receptors before actively transporting LEs to the microtubule minus end. Its projecting arm, p150, is recruited by RILP bound by active Rab7 in LE membranes. The other Rab7 effector, ORP1L, then transfers the Rab7–RILP–p150–dynein motor complex to βIII spectrin interacting with the base of the dynactin complex Arp1. Only after completion of this “mass protein action” does the dynein motor transport LEs to the microtubule minus end. Both RILP and ORP1L have been reported to localize on LEs, and their overexpression induces juxtanuclear clustering of these compartments, suggesting involvement in microtubule-dependent LE motility (; ). We studied the distribution of GFP-tagged RILP and Xpress-tagged ORP1L expressed in HeLa cells, and we visualized the proteins by confocal laser-scanning fluorescence microscopy. The proteins showed extensive colocalization on compact juxtanuclear organelle clusters (). We studied the contribution of their various domains to this phenotype (). The N-terminal portion of ORP1L, consisting of the ankyrin repeat region (ANK; aa 1–237; ) or the ANK and the PH domain regions (ANK + PH; aa 1–408; not depicted), displayed colocalization with RILP similar to that of full-length ORP1L (), demonstrating that the N-terminal Rab7-interacting ANK region of ORP1L suffices to specify this localization. We have previously shown that the ANK and ANK + PH domain regions of ORP1L induce clustering of LEs in the absence of ectopically expressed RILP (, ). We tested whether a truncated RILP that fails to recruit the dynein–dynactin motor, but still binds to the switch and interswitch regions of small GTPase Rab7 (ΔN-RILP; ; ), affects the LE-clustering phenotype induced by the ORP1L ANK domain (). Overexpression of ΔN-RILP inhibited clustering by ANK, although the two proteins still colocalized on the scattered LEs. To determine whether ORP1L and RILP might interact physically, HeLa cells were transected with Xpress-tagged RILP and subjected to immunoprecipitation with the Xpress mAb or irrelevant mouse IgG. Western blotting of the isolates with anti-ORP1L antibody revealed coprecipitation of endogenous ORP1L with Xpress-RILP (). To study the interaction in an endogenous setting, HeLa cell lysates were incubated with anti-RILP antibody or irrelevant rabbit IgG. Endogenous ORP1L was detected in the immunoprecipitates by Western blotting (), suggesting that ORP1L and RILP not only colocalize on LEs but are also part of a physical complex in cells. We then applied fluorescence resonance energy transfer (FRET) techniques to test whether a RILP–Rab7–ORP1L interaction could be visualized in living cells using fluorescently labeled proteins. When two fluorophores are in close proximity (<8 nm) and the fluorophores show spectral overlap, FRET can occur (). FRET can be detected by sensitized emission (when the acceptor emits light at the cost of donor fluorescence) or fluorescence lifetime imaging microscopy (FLIM). FLIM detects the time between photon absorbance by the donor fluorophore and its emission (in nanoseconds), which decreases when energy is transferred to acceptor fluorophores. FLIM, which, in principle, is more quantitative than sensitized emission for detecting FRET and FRET efficiencies (), was applied to study interactions between Rab7, RILP, and ORP1L in living HeLa cells. HeLa cells transfected with GFP–RILP and monomeric red fluorescent protein (mRFP)–Rab7, GFP–ORP1L and mRFP–Rab7, or GFP–ORP1L and mRFP–RILP, were analyzed by FLIM, and the lifetime of the GFP fluorophore was measured. The cells were cocultured with Mel JuSo cells stably expressing histone 2B (H2B)–GFP, which were used as an internal null FRET control. Because RILP and Rab7 have been previously shown to interact (; ) and cocrystallize (), these proteins constituted a positive control for the experimental setup. When GFP–RILP and mRFP–Rab7 were coexpressed, discrete perinuclear clusters were formed. A substantial decrease in GFP fluorescence lifetime on the GFP–RILP–positive structures was observed in the presence of mRFP–Rab7. The measured lifetime was 2.29 ± 0.06 ns (), when the lifetime for H2B–GFP in control cells (indicated by an asterisk) was at 2.56 ± 0.03 ns (). The calculated donor FRET efficiency, or E (see Materials and methods), between GFP–RILP and mRFP–Rab7 was 11.2 ± 2.5% (), indicating efficient FRET and close spacing of RILP and Rab7 in living cells. Measuring FLIM between GFP–ORP1L and mRFP–Rab7 resulted in a comparable reduced lifetime of the GFP fluorophore, 2.23 ± 0.03 ns (), corresponding to an E of 12.5 ± 1.85% (). To determine whether ORP1L and RILP are in close proximity not only to Rab7 but also to each other, FLIM was performed between GFP–ORP1L and mRFP–RILP. The decrease of fluorescence lifetime was somewhat less pronounced, but still significant (2.35 ± 0.04 ns; E = 7.0 ± 1.8%; ). Similar results were obtained when measuring FRET by sensitized emission (unpublished data). These data suggest that ORP1L is part of the same complex as RILP and Rab7. The absolute distances between the proteins cannot, however, be determined from these data because FRET efficiency is not only determined by the Förster distance (distance between the fluorescent groups) but also by the orientation factor and flexibility of the fluorophores (). Having established that ORP1L and RILP are part of a physical complex (), we set out to study the interaction between the two proteins by a series of pull-down experiments. Endogenous RILP was pulled down from HeLa cell lysate using purified, matrix-immobilized GST–ORP1L (). To study whether this interaction between ORP1L and RILP is direct, we used purified His-tagged RILP or the constitutively active GTP-loaded Rab7 mutant Q67L, which is produced in , to pull down purified GST–ORP1L. These experiments revealed that, although His–Rab7Q67L efficiently pulled down GST–ORP1L in accordance with our previous results (), no interaction was detected between His–RILP and GST–ORP1L (). Because both RILP and ORP1L bind to Rab7, we next tested, using purified proteins, whether Rab7 is able to bridge the two effectors and thereby form a tripartite complex. Because soluble full-length recombinant GST–ORP1L is produced in limiting amounts, GST–ANK, which suffices to bind Rab7 and can be efficiently produced, was used to perform the binding assay. GST fusion proteins of the Rab7-interacting ORP1L ANK fragment (GST–ANK) were incubated with His–RILP, GTP-loaded His–Rab7Q67L, or a mixture of both His-tagged proteins. As expected, GST–ANK pulled down His–Rab7Q67L, but not His–RILP. However, His–RILP was pulled down by GST–ANK when the incubation was performed in the presence of GTP-loaded His–Rab7Q67L (). This indicated that Rab7 is required to bridge RILP and ORP1L, and that the two effectors do not compete for the same binding site on Rab7. To test whether ORP1L and RILP interacted cooperatively with Rab7, we immobilized GTP-loaded GST–Rab7 and pulled down in vitro–translated and radiolabeled ORP1L or RILP. This experiment was performed in the presence of a step gradient of increasing amounts of purified RILP, ΔN-RILP, or ORP1L fusion proteins, respectively. GTP-loaded GST–Rab7 pulled down S-labeled ORP1L to a significant extent in the absence of RILP, but further addition of His–RILP to the reaction mixture increased ORP1L binding to Rab7 in a dose-dependent manner. S-labeled ORP1L binding to immobilized Rab7 increased up to approximately fourfold in the presence of His–RILP (). Addition of His–RILP had no effect on the binding of ORP1L ANK (not depicted) or ANK + PHD fragments to Rab7–GTP (). To further map the region of ORP1L involved in the observed stabilization, we extended the ANK + PHD domain to contain a region upstream of the oxysterol-binding protein–related domain (ORD) displaying a high probability of forming a coiled coil (ΔORD; ; ; ). Also this construct failed to produce the effect observed with full-length ORP1L (), suggesting that the ORD of ORP1L is involved in the stabilization effect observed in the presence of His–RILP. To determine if the N-terminal region of RILP, which is required for dynein–dynactin motor recruitment to LEs (), is necessary for facilitating the ORP1L–Rab7 interaction, the pull-down experiment with GTP-loaded Rab7 was repeated with in vitro–translated S-labeled ORP1L in the presence of His–ΔN-RILP. This truncated form of RILP failed to increase ORP1L binding to Rab7–GTP (), demonstrating that the same domain of RILP essential in dynein–dynactin motor recruitment to LEs () is also involved in the stabilization of the Rab7–ORP1L interaction. To test if the observed cooperativity of ORP1L binding to Rab7 in the presence of RILP is reciprocal, we pulled down in vitro–translated S-labeled RILP with immobilized His–Rab7–GTP, followed by incubation with increasing amounts of GST–ORP1L. RILP was observed to bind to immobilized Rab7, but the addition of GST–ORP1L had no effect on RILP binding (). These experiments suggest that RILP might stabilize the Rab7–ORP1L interaction in a unidirectional manner, and again indicate that binding of the two effectors to Rab7 is not mutually exclusive. GTP-loaded Rab7 collects RILP and ORP1L into a complex, and then recruits the dynein–dynactin motor protein complex to LE membranes. The dynein–dynactin motor complex interacts via the Arp1 filament with the β chain of membrane-associated αIβIII spectrin (), which thereby acts as a membrane receptor for the motor. Fluorescence microscopy analyses performed in HeLa cells demonstrated that the dynein–dynactin motor subunits Arp1, p50 (not depicted), and p150 were selectively recruited to LEs by myc-tagged RILP (). ΔN-RILP, in contrast, failed to recruit these dynein–dynactin motor subunits (). To identify the dynein–dynactin motor subunit required for the RILP-mediated targeting to LEs, we tested whether ectopically expressed RILP or ORP1L could selectively stabilize particular dynein–dynactin motor subunits on LE membranes when the dynein–dynactin motor complex is dissociated. Overexpression of (GFP–)p50 displaces the projecting arm of dynactin from the rest of the dynactin complex, leaving Arp1 membrane associated, whereas p150 and other more distal dynein–dynactin motor subunits are dispersed in the cytosol (; ; ). GFP–p50 was overexpressed in HeLa cells in the absence or presence of HA-RILP or Xpress–ORP1L. Cells were fixed and stained for endogenous p150 (). Expression of RILP alone, but not ORP1L (or ANK; unpublished data), sufficed to rescue p150 localization on LEs when the dynein motor complex was dissociated by p50 overexpression. It is noteworthy that the dynein–dynactin motor complex is no longer assembled into a functional motor when p50 is overexpressed, resulting in scattered vesicles still containing RILP and p150. To further test the involvement of endogenous RILP or ORP1L in motor recruitment and minus end–directed transport, the proteins were down-regulated by specific siRNAs, and the localizations of RILP (ectopically expressed), ORP1L (ectopically expressed), late endocytic markers, and p150 were determined (). Down-regulation of endogenous RILP or ORP1L induced scattering of LEs. p150 was recruited to the scattered vesicles only when RILP was overexpressed (). Importantly, p150 recruitment by RILP in the absence of ORP1L did not suffice to drive minus-end transport of LEs into a characteristic perinuclear cluster around the MTOC (). This suggests that RILP, but not ORP1L, interacts with the p150 subunit of the dynein–dynactin motor complex. However, interactions of RILP or ORP1L with other stabilizing motor subunits, or the involvement of additional proteins linking p150 and RILP, cannot be excluded by these experiments. To identify the potential interaction between RILP and p150, we first determined the p150 region recruited by RILP. ΔC, the GFP-tagged N-terminal 95-kD fragment of p150 (aa 1–876) was coexpressed with mRFP–RILP (). Although decorating microtubules and the MTOC, this fragment failed to be recruited to the RILP cluster. This suggested that RILP interacts (directly or indirectly) with the C-terminal 55-kD portion of p150. The coiled coil–containing region in p150 CC2 (aa 887–1,063) also failed to be recruited by RILP (not depicted), whereas GFP–p150 C25 (aa 1,049–1,278), representing the most C-terminal 25-kD region after the coiled coil, was recruited to the mRFP–RILP cluster (). This construct did not colocalize with mRFP–ΔN-RILP (), in agreement with the observation that this dominant-negative RILP variant inhibits p150 recruitment to LEs (; ). These data suggest that the N-terminal half of RILP interacts with the most C-terminal region of p150 to recruit the dynein–dynactin motor complex to LEs. To test whether this p150 fragment also inhibited minus-end transport by acting as a dominant-negative dynein motor fragment, analogously to overexpression of p50 (), the GFP-tagged versions of both proteins were overexpressed in HeLa cells, and late endocytic structures were visualized by labeling with anti-CD63 antibodies (). Overexpression of the C-terminal fragment of p150 induced a mild, but reproducible, scattering of LEs, whereas the effect of p50 overexpression was considerably more pronounced. This suggests that overexpression of the (presumably) monomeric version of the p150 C-terminal fragment competes weakly with the endogenous p150 homodimer for binding to the RILP homodimer. A direct interaction between RILP and the most C-terminal region of p150 can only be studied using purified proteins. GST–Rab7, GST–ORP1L, maltose-binding protein (MBP)–p150 C25, His–RILP, and His–ΔN-RILP were expressed in and purified. GST–Rab7 was immobilized on beads and loaded with GTPγS. To reconstitute the tripartite complex in vitro, the beads were subsequently loaded with His–RILP or His–ΔN-RILP in the presence or absence of ORP1L. The beads were washed, and equal amounts of purified MBP–p150 C25 were added to all reactions before another washing step. To determine specific binding, which is binding-dependent on the active GTP-bound conformation of Rab7, GTPγS was eluted by EDTA and replaced by excess GDP. The eluates were then incubated with amylose resin to capture the MBP–p150 C25 fusions before extensive washing. The bound fractions were analyzed by SDS-PAGE and Western blotting (). Only RILP, but not ΔN-RILP or ORP1L, was recovered from the amylose-bound MBP–p150 C25 fractions, in line with the observations above. This interaction was also reproduced in the absence of immobilized Rab7, suggesting that the GTPase is not required for binding in vitro of p150 to RILP (unpublished data). These results support a model in which the C-terminal domain of RILP interacts with active GTP-loaded Rab7 and the N-terminal domain of RILP (which is absent in ΔN-RILP) interacts with the C-terminal 25-kD fragment of the dynein–dynactin motor subunit p150. If the Rab7–RILP complex recruits the dynein motor through the p150 stalk, it could act as a second dynein motor receptor in parallel with βIII spectrin on the membrane of LEs. βIII spectrin has an N-terminal actin-binding domain that interacts with the base of the dynein–dynactin motor Arp1 () and is required for anchoring this motor to the membrane and for motor activity (). αIβIII spectrin is likely to be the only spectrin family member present on LEs (; ). To test whether αIβIII spectrin is located on LEs, and whether RILP or the dominant-negative form ΔN-RILP affects spectrin localization, we introduced GFP-tagged RILP or ΔN-RILP in Mel JuSo cells and stained these with antibodies for αI or βIII spectrin (). αI and βIII spectrin localized to various subcellular structures, including RILP- and ΔN-RILP–containing vesicles. Although full-length RILP was required for p150 recruitment, αIβIII spectrin remained membrane associated, even when ΔN-RILP was expressed. The membrane association of αIβIII spectrin is, thus, independent of RILP, and probably occurs through association with various (unidentified) late endocytic transmembrane proteins and phospholipids (). If RILP and βIII spectrin are receptors for the same motor protein, do they act in concert or independently? HeLa cells were transfected with GFP–RILP and a pSUPER construct that coexpresses mRFP and a control or βIII spectrin short hairpin RNA (shRNA). The cells were fixed and stained for endogenous p150. Although (GFP–)RILP recruited p150, down-regulation of βIII spectrin prevented transport of the RILP–p150–positive vesicles to the microtubule minus ends, which is similar to the situation observed after down-regulation of ORP1L ( and ). We subsequently tested whether the clustering of LEs imposed by ORP1L was also dependent on βIII spectrin. HeLa cells were transfected with GFP–ORP1L– and pSUPER vector–coexpressing mRFP and control or βIII spectrin shRNA. Cells were fixed and stained for p150 (). GFP–ORP1L clustered vesicles without recruiting detectable p150, which is in accordance with the finding that RILP is responsible for recruiting p150. Silencing of βIII spectrin expression resulted in scattering of ORP1L-positive vesicles, implying that minus-end transport of ORP1L and RILP-positive vesicles requires βIII spectrin (). Unlike RILP, endogenous ORP1L was detectable with the available antibodies. We tested whether down-regulation of βIII spectrin also induced scattering of organelles positive for endogenous ORP1L and other late endocytic markers. HeLa cells were transfected with the pSUPER vector coexpressing mRFP and a shRNA for βIII spectrin before labeling (). Scattering of the ORP1L-, Lamp-1–, and CD63-positive LEs was only detected in the mRFP-expressing transfected cells. These results suggest that βIII spectrin is required for minus-end transport of LEs labeled with endogenous or ectopically expressed ORP1L (). Notably, silencing of βIII spectrin (), as well as ORP1L (), prevents LEs labeled with Rab7–RILP–p150 complexes to be actively transported to the minus end of microtubules. Most intracellular organelles are transported along microtubules in a bidirectional manner, driven by at least two oppositely directed motor proteins. Late endosomes, like MIICs, move bidirectionally by the alternating activities of the dynein–dynactin motor (for minus-end transport) and the kinesin motor (for plus-end transport; ). The function of motor proteins must be spatially and temporally coordinated to explain this complex pattern of organelle motility. We show how minus-end transport of LEs by the dynein–dynactin motor is regulated by the tripartite complex ORP1L–Rab7–RILP. We show that RILP directly interacts with the 25-kD C-terminal region of p150. This explains how activation of Rab7, via RILP, recruits the dynein–dynactin motor to LEs. However, we also show that this motor recruitment does not suffice for minus-end microtubular transport of LEs. ORP1L is presumably essential for translocation of Rab7–RILP–p150 to βIII spectrin, which acts as the general dynein receptor on LEs and other organelles (). Only then does dynein motor–driven minus-end transport ensue. Rab7–RILP is the late endocytic–specific dynein motor receptor that cooperates with the general dynein receptor βIII spectrin for active dynein motor activity on LEs. This p150–RILP–Rab7–ORP1L–βIII spectrin cascade explains one critical step in the spatiotemporal control of late endocytic motility. βIII spectrin pairs with αI spectrin and the αIβIII spectrin heterodimer localizes to the cytosolic side of LEs, Golgi, and other organelles (; ). However, its presence does not suffice to recruit the dynein motor. Instead, activation of Rab7 specifies the target membrane for RILP- and ORP1L-mediated dynein–dynactin motor activities rather than the more ubiquitous spectrin receptor. Several Rab proteins have recently been found to control motor proteins either via a direct interaction or via effector proteins (). In all cases, it is unclear whether additional membrane motor receptors on specific compartments are also involved in RabGTPase-controlled, motor-driven transport. We define a novel interaction where Rab7 recruits RILP to interact with the C terminus of p150. This may stabilize the projecting motor arm on the dynein–dynactin base comprised of Arp1, p50, and other subunits (). The C-terminal fragment of p150 also interacts with dynein intermediate chain and Arp1 () in a manner that still leaves space for RILP interactions. That RILP interacts with a fragment near the dynein–dynactin motor base is not unexpected because the Rab7–RILP complex is membrane embedded and not likely to protrude far from the membrane. We have previously failed to detect this by cryoelectronmicroscopy (), probably because of the inefficiency of immunolabeling. In this study, we have used a series of techniques to study the interaction between Rab7 and two of its effectors, RILP and ORP1L. The FRET results indicate that Rab7, RILP, and ORP1L are in close proximity in living cells, and our in vitro experiments indicate that RILP and ORP1L can bind simultaneously to the same active GTP-loaded Rab7 molecule. In fact, the binding of ORP1L to Rab7 is stabilized by RILP. The N-terminal ANK region of ORP1L specifies the Rab7 interaction (). We now show that the ANK region binds directly to active GTP-loaded Rab7, but additional determinants in the C-terminal half of ORP1L stabilize the RILP–Rab7–ORP1L tripartite complex. Because the crystal structure of the Rab7–RILP complex revealed that two GTP-loaded Rab7 molecules bind on opposite sides of a RILP homodimer (), the RILP–Rab7–ORP1L complex is probably a heterotrimeric dimer with ORP1L positioned at the boundaries of the complex (). How does the RILP–Rab7–ORP1L complex recruit the dynein motor to LEs? Rab7 associates with LEs after GTP binding and recruits the effectors RILP and/or ORP1L, thus forming a tripartite complex. In theory, both effectors could mediate the recruitment of the motor. When we overexpressed p50 to dissociate the dynein–dynactin motor complex such that the p150 subunit becomes solubilized, RILP, but not ORP1L, was able to recruit soluble p150 to LEs. p150 is in fact the most membrane-distal subunit that becomes detached from the Arp1 filament at the base of dynactin after p50 overexpression (; ; ; ). Using p150 deletion constructs, we identified the RILP-interacting determinant as the most C-terminal 25-kD part of p150. In vitro reconstitution experiments confirmed that RILP, unlike ΔN-RILP or ORP1L, directly interacts with this C-terminal region of p150. Thus, the C-terminal half of RILP interacts with GTP–Rab7 and the N-terminal half interacts with the C-terminus of p150, and probably with the C-terminus of ORP1L (). The N-terminal half of RILP could not be successfully produced in because of aggregation of the expressed protein and was not recruited to LEs. Thus, a direct interaction of this domain with p150 could not be tested. Because both RILP and p150 are homodimers, these two proteins may bind as dimer–dimer, but this has not been experimentally verified. If the Rab7–RILP complex recruits the dynein–dynactin motor complex by directly interacting with p150, the function of ORP1L remains elusive. RILP mainly contains coiled-coil regions, whereas ORP1L is composed of multiple defined domains, including three ankyrin repeats, a PH domain, and an oxysterol-binding domain. Both βIII spectrin and ORP1L contain a PH domain with similar specificity (; ). It is possible that ORP1L is required for Rab7–RILP–dynein motor targeting to specific microdomains on LEs to deliver the dynein motor to its spectrin receptor. This is likely to be the 622-kD αIβIII spectrin heterodimer (; ), which remains bound to late endosomes when RILP is inactivated (). The 272-kD βIII spectrin protein contains two Arp1-binding sites (), repeat regions involved in several protein–protein interactions and a C-terminal PH domain (). Via its various molecular interactions, spectrin is strongly associated with multiple membrane compartments, including LEs. Using RNAi, we show that βIII spectrin is critical for minus-end dynein-mediated transport of LEs. Surprisingly, inactivation of ORP1L or βIII spectrin prevented minus-end dynein-mediated transport of LEs, even when Rab7–RILP recruited the p150 subunit of dynactin onto the membranes of LEs. The similarity of phenotypes suggests that ORP1L and spectrin are functionally connected. ORP1L could interact directly with αIβIII spectrin, perhaps via its ankyrin repeats. Because the PH domains of ORP1L and βIII spectrin have overlapping specificities (; ), ORP1L may also transfer the Rab7–RILP–p150 complex to the correct microdomains for docking the dynein motor on βIII spectrin. This transfer could be required for dynein motor activation by an unknown mechanism. In addition, such microdomains may contain a GTPase-activating protein to inactivate GTP–Rab7 and dissociate the RILP–dynein motor complex from LEs. This may be essential because a kinesin and/or dynein motor should be cyclically recruited or activated to drive the bidirectional transport of LEs in a stop-and-go fashion (). How the kinesin motor is recruited and whether this requires dissociation of the Rab7–RILP–dynein motor complex is unclear. Kinesin II can associate with p150 (; ), and the bidirectional movement may be driven by alternating activities of the kinesin and dynein motors on the ORP1L–Rab7–RILP-recruited–p150 complex docked on βIII spectrin, which acts as a receptor for both motor activities. Based on the data presented here, we propose the following model for Rab7-controlled recruitment of the dynein motor complex to its receptor αIβIII spectrin on LEs (). After GTP binding, Rab7 is recruited to late endocytic membranes. Active Rab7 binds two proteins with different functions. The RILP homodimer binds to the switch and interswitch regions of Rab7, arresting Rab7 in the active membrane-bound state (; ; ). ORP1L binding to Rab7 is stabilized by RILP to form a 360-kD complex. Within this complex, the RILP dimer associates with the base of p150 to facilitate the recruitment of the dynein–dynactin motor. ORP1L then targets the Rab7–RILP–dynein motor complex to the spectrin receptor on LEs. ORP1L and βIII spectrin are both required for a functional dynein motor, suggesting that RILP-mediated recruitment of the dynein motor to LEs does not suffice for correct docking on the LE–dynein receptor βIII spectrin. The dynein motor is apparently only activated with full cooperation of Rab7, RILP, and ORP1L. The complicated bidirectional motility of LEs is regulated by networks of macromolecular complexes controlled by the small GTPase Rab7. ORP1L, RILP, ΔN-RILP, and Rab7 cDNA constructs have been previously described (; ; , ; ). The mammalian expression vectors used were pcDNA4HisMax (Invitrogen), pcDNA3.1 (Invitrogen), and pEGFP-C (BD Biosciences). The mRFP–RILP and mRFP–Rab7 fusion constructs were generated by amplification of full-length mRFP by PCR (template plasmid provided by R.Y. Tsien, University of California, San Diego, La Jolla, CA) using forward primer 5′-CCCAGCTAGCACCACCATGGCCTCCTCCGAGGACGTCAT-3′ and reverse primer 5′-GAAGATCTGGCGCCGGTGGAGTG-3′. The mRFP fragment was ligated into the NheI and BglII sites in vector pEGFP-C1, from which the GFP moiety had been removed. The GFP–Arp1 construct was a gift from C. Hoogenraad (Erasmus Medical Centre, Rotterdam, Netherlands). The C-terminal fragments of p150 encoding aa 887–1,278 and 1,049–1,278 were generated by PCR and ligated into the EcoRI and BamHI sites of vector pEGFP-C2 (CLONTECH Laboratories, Inc.). For production of MBP–p150 fusion proteins, the same fragments were cloned into the EcoRI and BamHI sites of vector pMAL-c2X (New England Biolabs). A pSUPER vector coexpressing mRFP () was used to express short hairpins to down-regulate human βIII spectrin (the sequence targeted by the expressed RNAi is 5′-CGTGGCACGGCTCTGGGAC-3′). siRNA for human RILP was obtained from Dharmacon (ON-TARGETplus SMARTpool for accession no. ) and cotransfected with a vector expressing GFP–ORP1L (), using DharmaFECT 1 transfection reagent (Dharmacon). Human ORP1L siRNA oligos with sequence 5′-UrGrCrCrArGUrGrCrCrGrGrAUUrCUrGrATT -3′ were obtained from Proligo and were cotransfected with a vector expressing GFP–RILP () using Lipofectamine 2000 (Invitrogen). For production of hexahistidine (His)-tagged proteins, full-length Rab7, or Rab7Q67L, cDNAs were subcloned into the BamHI and XhoI sites of pET-28a (Novagen). A cDNA fragment encoding full-length RILP was subcloned into the NcoI–HindIII sites of vector pETM-11 (a gift from G. Stier, European Molecular Biology Laboratory, Heidelberg, Germany) for production of His-tagged RILP. ΔN-RILP was subcloned as a BglII fragment into a BamHI-digested vector pRSET-C (Invitrogen). Vector pRP265, which is a derivative of pGEX-2T (GE Healthcare) with a modified multiple cloning site, was used to generate GST–Rab7 fusion protein. Full-length ORP1L, ORP1S, and ANK fragment inserts were subcloned into the BamHI site of vector pGEX-1λT (GE Healthcare) for production of GST fusion proteins. Constructs are depicted in . GST–ORP1 and GST–RILP fusion proteins were used for generation of rabbit polyclonal antibodies (; , ). The other antibodies used were chicken anti-Rab7 (a gift from A. Wandinger-Ness, University of New Mexico, Albuquerque, NM), rabbit anti-Rab7 (Santa Cruz Biotechnology, Inc.), rabbit anti–human αI spectrin and rabbit anti–human βIII spectrin (Santa Cruz Biotechnology, Inc.), mouse anti-Xpress (Invitrogen), mouse anti-myc (Santa Cruz Biotechnology, Inc.), mouse anti-p150 (BD Biosciences), HRP-conjugated monoclonal anti-MBP (New England Biolabs), and polyclonal goat anti-GST (GE Healthcare). His-tagged RILP was produced in the strain Rosetta (DE3) pLysS (Novagen) in autoinduction high-density shaking cultures () at 24°C for 20 h. Cells were collected and resuspended in buffer A (25 mM Hepes, pH 7.5, 300 mM NaCl, Complete EDTA-free Protease Inhibitor Cocktail [Roche], 1 mM PMSF, 5 mM β-mercaptoethanol, 10 mM imidazole, and 0.05% [vol/vol] Triton X-100), and then lysed by sonication on ice. The cleared lysate was incubated with preequilibrated Talon Co resin (CLONTECH Laboratories, Inc.) for 1 h. The resin was packed into a column and washed with buffer A containing 20 mM imidazole, and His-RILP was eluted by a step-gradient of imidazole in buffer A. The eluted protein was concentrated in 25 mM Hepes, pH 7.5, 300 mM NaCl, and 10% (vol/vol) glycerol in 10-kD cut-off concentrators (Vivaspin-2; Sartorius). His-tagged Rab7 and His-tagged Rab7Q67L were expressed in BL21(DE3) (Stratagene) by induction with 0.5 mM IPTG for 5 h at 30°C. After harvesting, the cells were resuspended in 25 mM Hepes, pH 8.0, 300 mM NaCl, 5 mM MgCl, Complete EDTA-free Protease Inhibitor Cocktail, 1 mM PMSF, 5 mM β-mercaptoethanol, and 10 mM imidazole, and lysed by sonication on ice. The clarified lysates were passed through a HiTrap Chelating HP column (GE Healthcare) charged with Co and further purified using a HiTrap SP HP column (GE Healthcare) in 20 mM Mes, pH 6.0, 100 mM NaCl, 5 mM MgCl, and 5 mM β-mercaptoethanol. BL21(DE3) cells harboring the GST–Rab7 construct were induced with 0.5 mM IPTG for 5 h at 30°C. Cells were harvested, resuspended in a buffer containing 50 mM Tris-HCl, pH 7.5, 200 mM NaCl, 1 mM PMSF, Complete EDTA-Free Protease Inhibitor Cocktail, 5 mM MgCl, and 1 mM DTT, and lysed by sonication on ice. The cleared lysate was loaded on a preequilibrated GSTrap FF column (GE Healthcare) and washed extensively with the same buffer. GST–Rab7 was eluted with 50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 5 mM MgCl, 1 mM DTT, and 20 mM reduced glutathione. The MBP–p150 fusion proteins were expressed in strain Rosetta (DE3) pLysS (Novagen) and affinity purified using an amylose resin column (New England Biolabs) according to the manufacturer's instructions. All fusion proteins made were analyzed by SDS-PAGE and Coomassie staining; they are depicted in . Transfected HeLa cells (∼2 × 10) were washed with ice-cold PBS and scraped into 400 μl of lysis buffer (20 mM Hepes, pH 7.6, 150 mM NaCl, 2 mM MgCl, 10% glycerol, 0.5% Triton X-100, and 1 mM DTT) with Complete EDTA-free Protease Inhibitor Cocktail. Cells were kept on ice for 15 min and centrifuged for 15 min at 16,000 at 4°C, and the supernatant was preadsorbed at 4°C for 30 min with 30 μl of protein G–Sepharose 4 Fast Flow (GE Healthcare). The recovered supernatant was incubated with Xpress or irrelevant control antibodies at 4°C overnight. The lysate–antibody mixture was incubated at 4°C with protein G–Sepharose for 4 h, followed by washing with lysis buffer. For immunoprecipitation of endogenous ORP1L from HeLa lysates, cells (∼10) were lysed and preadsorbed as described for supernatant. The recovered supernatant was incubated with RILP or irrelevant control antibodies at 4°C for 2 h. The lysate–antibody mixture was incubated at 4°C with protein G–Sepharose for 2 h, followed by washing with lysis buffer. The immunoprecipitates were analyzed by SDS-PAGE and Western blotting. 100 μg GST–ORP1L fusion protein and an approximately equimolar amount of 25 μg GST were coupled to 30 μl glutathione–Sepharose 4B beads in coupling buffer (PBS, 2 mM MgCl, and 1 mM DTT) for 2 h at 4°C. Beads were washed with coupling buffer and equilibrated in lysis buffer (20 mM Hepes, pH 7.6, 150 mM NaCl, 2 mM MgCl, 10% glycerol, 0.5% Triton X-100, and 1 mM DTT). HeLa cells were lysed as described in the immunoprecipitations section. After centrifugation of the lysates for 15 min at 16,000 at 4°C, the supernatant was added to the beads and incubated for 2 h at 4°C. Beads were washed extensively with lysis buffer, and bound proteins were eluted with 20 mM glutathione. The eluted proteins were analyzed by SDS-PAGE and Western blotting. S-labeled full-length and truncated proteins were generated by in vitro transcription/translation using the TnT coupled reticulocyte system (Promega) according to the manufacturer's instructions. 50 μg wild-type GST–Rab7 fusion protein was coupled to 20 μl glutathione–Sepharose 4B beads in coupling buffer (PBS, 2 mM MgCl, and 1 mM DTT) for 2 h at 4°C. Beads were washed with coupling buffer and equilibrated in 20 mM Hepes, pH 7.5, 100 mM KAc, 0.5 mM MgCl, 1 mM DTT, 2 mM EDTA, and 10 mg/ml albumin. The beads were incubated with 10 μM GTP for 10 min at room temperature, after which MgCl was added to a final concentration of 10 mM and the incubation was continued for an additional 30 min. Purified His–RILP, His–ΔN-RILP, or GST–ORP1L (0 ng, 100 ng, 1 μg, and 5 μg) and in vitro–translated proteins (22 μl) were added to the beads and incubated in a total volume of 0.5 ml for 2 h at 4°C. Beads were washed extensively with wash buffer (20 mM Hepes, pH 7.5, 100 mM KAc, 5 mM MgCl, and 1 mM DTT), and proteins were eluted with 20 mM glutathione in wash buffer and resolved by SDS-polyacrylamide gels, which were exposed on x-ray film (Kodak) or analyzed by phosphor imaging (FLA-3000; Fujifilm). 50 μg His–RILP and 50 μg His–Rab7Q67L were coupled to 20 μl Talon Co resin (CLONTECH Laboratories, Inc.) in coupling buffer (20 mM Hepes, pH 7.5, 150 mM NaCl, 2 mM MgCl, 10% glycerol, and 5 mM β-mercaptoethanol) for 1 h at 4°C. Beads were washed and equilibrated in coupling buffer with 10 mM imidazole. 2.3 μg purified GST–ORP1L or 5 μg plain GST was added to the beads and incubated for 2 h at 4°C. Beads were washed extensively with binding buffer (20 mM Hepes, pH 7.5, 150 mM NaCl, 2 mM MgCl, 10% glycerol, 5 mM β-mercaptoethanol, and 10 mM imidazole) and eluted with 750 mM imidazole in 20 mM Hepes, pH 7.5, 150 mM NaCl, 10% glycerol, and 5 mM β-mercaptoethanol. Samples were analyzed by SDS-PAGE and Western blotting. 100 μg GST–ANK was coupled to glutathione–Sepharose 4B beads in PBS for 2 h at 4°C, after which the beads were washed and equilibrated in binding buffer (20 mM Hepes, pH 7.5, 150 mM NaCl, 2 mM MgCl, 10% glycerol, 0.5% Triton X-100, and 1 mM DTT). 10 μg purified His–RILP and/or 20 μg His–Rab7Q67L were added to the beads and incubated for 4 h at 4°C. Beads were washed with binding buffer and eluted with 20 mM glutathione (30 μl) in binding buffer. Samples were analyzed by SDS-PAGE and Western blotting. 30 μg GST–Rab7 was coupled to 15 μl glutathione–Sepharose 4B beads and preloaded with GTPγS as described for the pull-down of in vitro–translated fragments. Beads were washed to remove unbound proteins, and recombinant purified His–RILP (3 μg), His–ΔN-RILP (2 μg), or GST–ORP1L (8 μg) was added to the beads and incubated in a total volume of 0.4 ml for 2 h at 4°C. Beads were washed extensively with detergent/salt buffer (20 mM Hepes, pH 7.5, 300 mM NaCl, 1 mM DTT, 2 mM MgCl, and 0.1% [vol/vol] Triton X-100), and 4 μg of either MBP–p150 (aa 887–1,278) or MBP–p150 (aa 1,049–1,278) was added to the beads. Proteins were incubated further in a total volume of 0.4 ml of detergent/salt buffer for 30 min at 22°C. Beads were washed and proteins bound to GTP–Rab7 were eluted by supplementing the 0.4 ml of detergent/salt reaction buffer with EDTA to a final concentration of 20 mM. To precipitate the MBP–p150 C25 fusion (aa 1,049–1,278), the elution fractions were combined with 20 μl of amylose resin and incubated for 1 h at 4°C. The resin was washed extensively with detergent/salt buffer before analysis by SDS-PAGE and Western blotting. Transfected cells were fixed either with 4% formaldehyde in PBS for 30 min and permeabilized for 5 min with 0.05% Triton X-100 in PBS or with methanol (−20°C) for 5 min. Nonspecific binding of antibodies was blocked by 10% FBS/PBS for 30 min, after which cells were incubated with primary antibody in 5% FBS/PBS for 30 min at 37°C. Bound primary antibodies were visualized with Alexa Fluor secondary antibody conjugates (Invitrogen). Cells were mounted in Mowiol (Calbiochem) containing 50 mg/ml 1,4-diazocyclo-octane (Sigma-Aldrich) or in Vectashield mounting medium (Vector Laboratories). The specimens were analyzed with confocal laser scanning microscopes (TCS SP1 or TCS SP2) equipped with HCX PL APO and HCX PL APO lbd.bl 40×/NA 1.32 objective lenses (all Leica). The acquisition software used was Leica LCS. FLIM experiments were performed at 37°C in a 5% CO culture hood on an inverted microscope (DM-IRE2; Leica) fitted with a TCS SP2 scanhead and HCX PL APO lbd.bl 40×/NA 1.32 objective lenses and equipped with Lambert Instruments frequency domain lifetime attachment, controlled by EZflim software (Lambert Instruments). Cells were cultured in Delta T dishes (Bioptechs) in CBS medium (140 mM NaCl, 5 mM KCl, 2 mM MgCl, 1 mM CaCl, 23 mM NaHCO, 10 mM [D-]glucose, and 10 mM Hepes, pH 7.3, under 5% CO condition). GFP was excited with ∼4 mW of 488-nm light from a LED modulated at 40 MHz, and emission was collected at 490−550 nm using an intensified charge-coupled device camera (CoolSNAP HQ; Roper Scientific). To calculate the GFP lifetime, the intensities from 40 phase-shifted images (modulation depth ∼70%) were fitted with a sinus function, and lifetimes were derived from the phase shift between excitation and emission. For internal control, cells were cocultured with Mel JuSo cells expressing H2B–GFP only. Lifetimes were referenced to a 1-μM solution of rhodamine-G6 in saline that was set at a 4.11-ns lifetime. The donor FRET efficiency was calculated as = 1 − (measured lifetime/GFP lifetime in control cells) ().

Hydrocephalus describes the lethal accumulation of cerebrospinal fluid (CSF) in the brain. The incidence is 0.12–2.5 per 1,000 live and stillbirths, and the majority of human hydrocephalus is of genetic origin (). A number of genetic mouse models of congenital hydrocephalus have been described. These include mice; recently, the affected gene, , was identified using an insertional mutant (). The corresponding chromosomal region in the human genome has also been implicated in hydrocephalus (). is a conserved gene present in the genomes of various protists and metazoans and encoding a large protein of >500 kD. In the neonatal brain of the mouse, is expressed in the ciliated epithelial cells lining the lateral, third, and fourth ventricles (). is also expressed in the ciliated epithelia of bronchi and oviduct and in developing spermatocytes. This expression pattern suggests a ciliary/flagellar function for hydin. Indeed, mutations in several other genes encoding ciliary proteins cause hydrocephalus in mice. These include , which encodes an axonemal dynein heavy chain; , a homologue of central pair (CP) component PF16; and , which encodes the intraflagellar transport protein IFT88/Polaris (; ; ). Comparative genomics revealed that homologues of are present in species with cilia/flagella and absent in species lacking these organelles, such as and (; ; ). Hydin was identified in the flagellar proteomes of () and (). In the proteomic analysis of fractionated flagella (), hydin was identified by 69 different peptides, primarily from axonemal fractions, suggesting that it is abundant and tightly associated with the axoneme. Finally, in , hydin ablation by RNAi resulted in reduced motility (). However, despite the evidence indicating a function in flagella, neither hydin's location within the organelle nor its role in flagellar assembly and motility are known. In this study, we use a combination of biochemical, reverse genetic, and structural approaches to examine the function and location of hydin in the flagellum. hydin is a polypeptide of ∼540 kD encoded by a single copy gene, , spanning ∼17,700 bp on linkage group I. To generate an antibody to hydin, a fragment (, fragment A), representing exon 3, was amplified from genomic DNA by PCR and cloned into the bacterial expression vector pMAL-cRI v.2 encoding maltose-binding protein (MBP). The MBP-hydin peptide was expressed in , purified, and used for antibody production. A BamHI–SalI piece of fragment A was cloned into pGEX-6P-1, and the purified GST-hydin peptide was used for affinity purification of the antibody. The purified anti- hydin antibody stained a single band of ∼540 kD in Western blots of axonemes of control cells (, control lane). Similar to other flagellar proteins, expression of is strongly induced after deflagellation by pH shock (; ; Fig. S1, available at ). Western blot analysis of cell bodies and isolated flagella revealed that hydin was highly enriched in the flagella () and attached to the axonemes ( and see ). Immunofluorescence microscopy combining anti-hydin with anti-acetylated tubulin confirmed that hydin is present in the flagella of wild-type cells (). When isolated flagella were extracted with detergent, anti-hydin staining remained associated with the axonemes and overlapped with the anti-tubulin signal over the entire length of the axonemes (). Therefore, hydin is an axonemal protein and one of the largest proteins so far identified in the flagellum. To determine whether hydin is associated with the outer doublets or the CP of microtubules, we used a modified CP extrusion assay. When isolated flagella were treated with 1 mM ATP, trypsin, and 1% NP-40 for 2 min, the CP was partially expelled from the axonemes (). Double labeling with anti-hydin and anti-acetylated tubulin revealed that hydin was concentrated on the CP, which was visible as a thin microtubular structure projecting from the distal end of the axoneme (); accordingly, anti-hydin staining was absent from the proximal region of the axoneme that had been vacated by the CP. Treatment with an additional 1 mM of ATP after 2 and 4 min often resulted in the complete extrusion of the CP from the axonemes (). Hydin colocalized with the extruded CP but was absent from the axonemes that were now devoid of the CP (). These data show that hydin is located exclusively in the CP apparatus. In whole mount immunogold EM of extruded CPs (), gold complexes representing hydin decorated the projecting CP microtubules (). Labeling was present but sparse on well-preserved CPs and increased as the CPs disintegrated, suggesting that parts of hydin targeted by the antibody are not readily accessible in the native CP. The majority of the gold particles (76%; = 275) was present on the microtubule located on the concave side of the arc formed by the CP; for this analysis, only gold complexes on CPs with well-preserved, intact microtubules were scored. Previous studies have shown that this is the C2 microtubule (see for a schematic drawing of the CP; ; ). As a control, we carried out a similar immunolocalization using antibodies to CPC1 () and PF6 (), both components of the C1 microtubule (; ; ). As expected, these antibodies decorated the CP microtubule on the convex side of the arc (PF6, 94% and = 56; CPC1, 91% and = 23). Thus, that portion of hydin encoded by exon 3 is closely associated with the C2 microtubule. In , the mutants and fail to assemble the CP (, ; ). We examined the flagella of these mutants to confirm that hydin is a component of the CP. In immunofluorescence microscopy, the hydin signal was strong in detached flagella of wild type (CC124; ) but greatly reduced in detached flagella of and (). In the flagella of these mutants, the CP is replaced by an amorphous core (, ; ), which may account for the residual hydin signal. This amorphous core is lost from most (80–90%) of the isolated axonemes of the mutants (; ), and indeed, Western blot analysis of isolated axonemes from and revealed only traces of hydin in both mutants (). These results provide strong independent evidence that hydin is a component of the CP apparatus. Mutations of in have not yet been identified. To investigate the function of hydin, an RNAi vector targeting was constructed by cloning fragment S in sense and fragment A in antisense orientation downstream of the promoter and upstream of a triple HA tag and the terminator (). The resulting plasmid, pKL3-hyAS, was cotransformed into cells of strain CC3395. 39 of 204 independently derived transformants scored from three experiments showed a severe motility phenotype, with the majority of the cells resting at the bottom of the culture vessels. message levels were tested by quantitative RT-PCR in 11 of the nonmotile transformants. In 10 of the strains, was down-regulated by up to 87% (). The remaining strain had normal message levels and, in contrast to the others (see the following section), completely lacked flagella, suggesting that its phenotype was due to an insertion of the RNAi vector into a gene necessary for flagellar assembly. In Western blots of isolated axonemes of selected hydin RNAi strains, hydin was strongly reduced (). In immunofluorescence microscopy using the anti-hydin antibody, labeling of the flagella in the RNAi strains was much weaker than in wild type, although the strength of the residual staining varied somewhat between RNAi strains and individual cells (). Thus, hydin depletion was confirmed by both Western blotting and immunofluorescence microscopy. Similar to earlier observations for RNAi of other genes in (), hydin knockdown was not stable over an extended period of time. All strains returned to wild-type amounts of hydin and wild-type phenotype within 2–6 mo after transformation (unpublished data). The CP apparatus is thought to regulate dynein arm activity, allowing coordinated flagellar movement (). Light microscopic examination of living cells from 10 independent hydin RNAi strains revealed that the flagella were mostly paralyzed and only occasionally formed bends. Flagella also were shorter than those of wild-type cells, with some cells lacking flagella completely ( and ). Strikingly, the flagella of approximately half of the cells were arrested with one flagellum in the “hands-up” position, i.e., pointing away from the cell body, and one flagellum in the “hands-down” position, i.e., lying alongside the cell body (). The remaining cells either had both flagella arrested in the hands-down position (hyN3, 30%; hyS1, 26%) or both flagella in the hands-up position (hyN3, 14%; hyS1, 27%). This is distinctly different from the phenotype of other CP mutants: the CP mutants , , , and were paralyzed with both flagella in the hands-up position (95–100%; see ; , , and were too motile to analyze). The relative position of the eyespot was analyzed in 222 hydin RNAi cells showing asymmetric arrest of the two flagella (living or rapidly fixed as described by ); 45% of these had the trans flagellum and 55% had the cis flagellum, which is closer to the eyespot, in the hands-up position. Therefore, the tendency to arrest in the hands-up or hands-down position does not correlate with the cis or trans flagellum. Although the majority (90–99%) of the hydin RNAi cells were paralyzed, some cells showed residual flagellar movements resulting in twitching or spinning. We noticed that the flagellum in the hands-up position was the more active one, which typically would undergo a quick power stroke, a resting period of variable length, and a recovery stroke, after which it stopped again in the hands-up position ( and Video 3, available at ). The flagellum in the hands-down position was either completely paralyzed or beat at a lower frequency than the hands-up flagellum (Videos 1 and 2). We further observed that residual hydin in the cells examined by immunofluorescence microscopy tended to be located in just one of the two flagella (). Because hydin is required for flagellar motility, it is possible that the asymmetrical motility of flagella observed in some RNAi strains is related to the unequal distribution of residual hydin in the two flagella. To determine whether flagellar paralysis was accompanied by ultrastructural defects of the axoneme, four strains (hyD2, hyN3, hyN4, and hyS2) were processed for transmission EM, and flagellar cross sections were analyzed. In all four strains, similar ultrastructural defects were observed (). Most notably, the C2b projection and parts of the C2c projection of the C2 microtubule were missing from almost all 9 + 2 axonemes (∼90%; = 34 for strain hyS2 and 14 for hyN4) of the hydin RNAi strains (). Image averages generated from digitized images clearly revealed the differences in the CP apparatus of 9 + 2 axonemes of wild-type and RNAi cells (). In 5–12% of the cross sections, one CP microtubule was missing; an additional 3–5% of cross sections lacked both CP microtubules, which were replaced by a core of amorphous material similar to that described in other mutants lacking the CP (; , ; ). Finally, ∼2% of the cross-sectioned axonemes lacked one to four B-tubules or up to three entire doublet microtubules (). Previous ultrastructural analysis of KLP1-depleted axonemes suggested that KLP1 is a component of the C2c projection (). A part of this projection was largely missing in the axonemes of hydin RNAi cells, raising the question of whether hydin depletion affected the presence of KLP1 and other known CP proteins. Western blotting showed no difference in the amounts of the C1 proteins PF6, FAP101 (), or CPC1 in axonemes isolated from wild-type or hydin RNAi cells (). In contrast, probing with anti-KLP1 () showed that KLP1 was strongly reduced in the hydin RNAi axonemes. These data indicate that localization of KLP1 to the C2 tubule depends on hydin. To obtain biochemical evidence for the location of hydin within the CP, we took advantage of the fact that the C2 microtubule can be selectively depolymerized by extraction of axonemes with high salt, e.g., 0.6 M KCl (). These conditions solubilized most of the KLP1 and the outer dynein arm intermediate chain IC2 (, lane KCl-S), whereas most of the hydin and the C1 proteins PF6 and CPC1 stayed in the insoluble fraction (lane KCl-P), in good agreement with data from the proteomic study of flagella (). This result, indicating that hydin is associated with the C1 microtubule when the C2 microtubule is solubilized, was surprising given the effect of hydin depletion on the C2b and C2c projections, as well as on the C2 protein KLP1, and the immunogold localization of hydin to the C2 microtubule. However, ultrastructural analysis has shown that the C1b and C2b projections physically interact and has revealed that axonemes isolated from the mutant , which lack the C1b projection, also frequently lack the C2b projection (). Therefore, if hydin is a component of the C2b projection, one would predict that hydin would be missing or destabilized in the mutant. To test this, flagella were isolated from and wild-type cells, and the detergent-soluble membrane-plus-matrix fraction, axonemes, and KCl-extracted axonemes were compared by Western blotting (). As shown above, hydin fractionated with the axonemes in wild-type cells and still remained with the axonemes after high-salt extraction. In contrast, in , some of the hydin was released into the detergent-soluble membrane-plus-matrix fraction, and little remained in the axonemes after high-salt extraction. The majority of PF6, a component of the C1a projection, remained attached to the wild-type and axonemes after high-salt extraction. These data show that hydin interacts with the CPC1 complex of the C1 microtubule. shows a model for the location of hydin within the CP that is consistent with the immunolocalization, ultrastructural, and biochemical data. #text strains used in the work include CC3395 , , −, CC124 , , , −), and (CC-3706; , +), all from the Genetics Center. was obtained from R.P. Levine (Harvard University, Cambridge, MA), whereas , , , and were R. Lewin isolates originally obtained from the Culture Collection of Algae and Protozoa (Cambridge, UK); all have been maintained in this laboratory since 1974. Cells were grown in M medium I with 2.2 mM KHPO and 1.71 mM KHPO () or TAP () at 23°C with aeration and a light/dark-cycle of 14/10 h (), or shaken in constant light. A 1.3-kb fragment (fragment A) corresponding to exon 3 of in the U.S. Department of Energy's Joint Genome Institute's genome v. 2 was amplified by PCR from genomic DNA of CC3395 using primer pair hyf2a and hyr2a (see Tables S1 and S2, available at , for primers and PCR conditions). The PCR product was digested with HindIII and BamHI and ligated into pMAL-cR1 v. 2 digested with the same enzymes (New England Biolabs, Inc.). The construct was transformed into XL1 blue, and, for expression of the maltose-binding∷hydin fusion protein, into BL21. The fusion protein was purified by amylose affinity chromatography, and antibodies were produced in rabbits (CRP, Inc.) using the company's standard protocol. Fragment A was digested with BamHI and SalI and ligated into pGEX-6P-1 (GE Healthcare) restricted with the same enzymes. After transformation into XL1blue, the GST∷hydin-fusion protein was purified by affinity chromatography, subjected to SDS-PAGE (7.5%), transferred onto polyvinylidene difluoride membrane, stained with Ponceau S, and excised from the membrane using a razor blade. The immobilized fusion protein was used for affinity purification of anti-hydin from bleeds 1, 2, and 4. We constructed a novel expression vector based on the gene; this is an intronless gene that encodes the dynein light chain LC8 and is inducible by deflagellation. The upstream region of , including the first and second codon, was amplified from genomic DNA by PCR using primers LC8f1-3 and LC8r1-2 and GoTaq Flexi DNA polymerase (Promega). The downstream region of was amplified using primers LC8r2_2 and LC8f1-3, and the products were restricted with HindIII–BamHI and BamHI–EcoRI, respectively. Both products were then ligated into pUC119 digested with HindIII and EcoRI. The resulting plasmid (pLC8S) was digested with BamHI, and a triple HA tag, amplified from p3xHA () using primers HAf2tm68 and Har2 and restricted with BamHI and BglII, was inserted, resulting in plasmid pKL3. Fragment S covering the first three exons and introns of was amplified by PCR using primers hyf1s and hyr1s. Fragments A (see above) and S were restricted with BamHI–HindIII and HindIII–XbaI, respectively, and ligated into pKL3 digested with NheI and BamHI. The resulting plasmid, pKL3-hyAS, was linearized with EcoRI or DraI. Cotransformation was performed using the glass bead method () and as a selectable marker. Flagella were isolated from as described previously by . For Western blot analysis, flagella were extracted with 1% NP-40 for 5 min on ice, axonemes were collected by centrifugation (30,000 , 20 min, 4°C), and the detergent-soluble membrane-plus-matrix and axoneme fractions were prepared for SDS-PAGE. For further fractionation, axonemes were extracted with 0.6 M KCl for 30 min on ice, pelleted, and extracted with 0.5 M KI for 20 min on ice. Extrusion of the CP was induced as previously described (; ) with the following modifications: ATP (1 mM final concentration) and trypsin (5 μg/ml final concentration; Invitrogen) were added to isolated flagella in HMDEK-PEG (30 mM Hepes, 5 mM MgSO, 1 mM DTT, 0.5 mM EGTA, 25 mM KCl, and 0.5% PEG 20,000) at room temperature. Flagella were then demembranated by addition of 1% Nonidet NP-40 (Calbiochem). After 2 min, aliquots were fixed with formaldehyde (2% final concentration), allowed to settle onto poly--lysine–coated multiwell slides for 2–8 min, and submerged into −20°C methanol for 6–10 min. For complete extrusion of CP microtubules, more ATP (1 mM) was added 2 and 4 min after flagellar demembranation, and samples were processed as described after a total of 6 min of incubation. Methanol-fixed specimens were processed for immunofluorescence microscopy. DNA was isolated from using PlantDNAzol reagent (Invitrogen) following the instructions of the manufacturer. TRIzol LS reagent (Invitrogen) was used for RNA isolation, and cDNA synthesis was performed using PowerScript Reverse Transcriptase (CLONTECH Laboratories, Inc.) using the manufacturer's protocol, except that cDNA synthesis was performed for 50 min at 42°C, 20 min at 48°C, and 20 min at 55°C. The relative amount of cDNA representing message in samples was determined by real-time PCR using primers C_410060F and C_410060R that spanned intron 4; QuantiTect SYBR green PCR master mix (QIAGEN) was used to monitor amplification. The relative amount of the G protein β subunit (), which is constantly expressed under various conditions, was measured in each trial and used to correct for slight differences in amount of cDNA in each sample. Up to three independent sets of RNA were isolated and analyzed. Cells were fixed in glutaraldehyde for EM () and processed as described previously (). For whole mount immunogold EM, CPs were extruded as described above but omitting trypsin. After 22–40 min, the suspension was applied to carbon/formvar-coated grids and fixed with 2.5–3% formaldehyde. Immunostaining was performed as described by , and specimens were analyzed using electron microscopes (CM10 or -12; Philips). Cells were fixed and stained for immunofluorescence microscopy as described by with the following modifications: 1% polyethylenimine was used to immobilize strains with a cell wall, and primary antibodies were applied overnight at 4°C. See Table S3, available at , for antibodies and dilutions used in this study. After washing, specimens were mounted with ProLong Gold (Invitrogen). Images were acquired at room temperature using AxioVision software and a camera (AxioCam MRm) on a microscope (Axioskop 2 Plus) equipped with a 100×/1.4 oil differential interference contrast (DIC) Plan-Apochromat objective (Carl Zeiss Microimaging, Inc.) and epifluorescence. Image brightness and contrast were adjusted using Photoshop 5.0 and 6.0 (Adobe). Figures for publication were assembled using Illustrator 8.0 (Adobe). Capture times and adjustments were similar for images mounted together. Averaged images were prepared in Photoshop by setting the original images to 16% opacity and aligning them manually. Fig. S1 shows the induction of hydin transcripts after deflagellation, providing additional evidence that hydin is a flagellar protein. Video 1 shows a hydin RNAi cell of strain hyN3 in which the flagellum in the hands-up position beats more frequently than the other flagellum. Video 2 shows a hydin RNAi cell of strain hyN4 with the more active flagellum in the hands-up position and the other in the hands-down position. Video 3 shows a hydin RNAi cell of strain hyS1, in which the more active flagellum rests mostly in the hands-up position and the other flagellum in the hands-down position. Tables S1, S2, and S3 show PCR primers, PCR conditions, and antibodies, respectively, used in this work. Online supplemental material is available at .

The midbody is a transient structure formed during the final stage of cell division (; ). Its main cytoplasmic components are the constricted actomyosin-based contractile ring and microtubule bundles derived from the central spindle (; ). Often in association with these two scaffolds, numerous proteins involved in cytoskeletal organization, cytokinesis, cell cycle regulation, and signaling are concentrated at the midbody (; ; ; ). Two principal membrane structures exist at the midbody—the plasma membrane, which corresponds to that of the cleavage furrow, and cytoplasmic membrane vesicles of biosynthetic and endocytic origin, which fuse with the plasma membrane for abscission, the terminal step of cytokinesis (; ; ; ; ; ). Beyond abscission, little is known about membrane traffic events involving the midbody. As for the midbody, microtubule bundles are key cytoskeletal elements of primary cilia (; ; ). However, in contrast to the midbody, which forms during M phase, primary cilia are plasma membrane protrusions of interphase and postmitotic cells (; ). Primary cilia have emerged as important structures that function like an antenna and have a key role in signaling to the cell interior, including the regulation of cell cycle progression (; ; ; ; ; ; ; ). Considerable progress has been made with regard to transport of proteins within the cilium (; ). However, the dynamics of the ciliary plasma membrane, specifically in the context of the disappearance of the primary cilium before, and its reformation after, M phase, have not been resolved. Neuroepithelial (NE) cells are the primary progenitor cells of the mammalian central nervous system. They are polarized along their apical–basal axis, and the orientation of the cleavage plane relative to this axis determines whether division is symmetric or asymmetric (). NE cells initially increase in number by symmetric divisions, in which cleavage occurs precisely along their apical–basal axis, bisecting the apical plasma membrane domain (hereafter referred to as the apical membrane) of the dividing cell and thus distributing it equally to both daughters (; ; ). At the onset of neurogenesis, NE cells switch to asymmetric divisions, in which the orientation of the cleavage plane deviates from the apical–basal axis, resulting in the apical membrane being bypassed by the cleavage furrow and hence being inherited by only one of the daughter cells, which thus remains neuroepithelial, in contrast to its sister, which adopts a neuronal fate (; ). Concomitant with this switch, NE cells reduce the size of their apical membrane (). It is unknown whether this reduction is solely achieved by down-regulation of apical biosynthetic transport () or involves additional dynamics of the apical membrane. Extracellular membrane vesicles and their origin in eukaryotic cells have received increasing attention. We recently reported the existence in the lumen of the neural tube of two novel classes of extracellular membrane particles that bear a marker of the apical membrane of NE cells, prominin-1 (prom1; ), which will be collectively referred to as prom1 particles herein. One class consists of small (50–80 nm) electron-translucent vesicles, referred to as P4 particles. The other class comprises relatively large (0.5–1 μm) electron-dense particles, referred to as P2 particles, on which prom1 often appears to be distributed in a ring-like fashion. Importantly, the P4 particles were found to be distinct from the similar-size exosomes (), the internal vesicles of multivesicular bodies that are released into the extracellular space by exocytosis (; ; ). Likewise, the P2 particles appear to be morphologically distinct from the recently reported 0.3–5-μm nodal vesicular parcels implicated in left–right determination (). Given the presence of apical membrane constituents in the P2 and P4 particles (), their release would be a means of reducing the size of the apical membrane of NE cells as these cells switch from symmetric to asymmetric division during brain development (). Consistent with this possibility, P2 and P4 particles accumulate in the neural tube fluid before (P2) and during (P4) the onset of neurogenesis, and both types of particles decrease as neurogenesis progresses (). It is therefore important to identify the subcellular sites from which P2 and P4 particles originate. Although the origin of P2 particles has been enigmatic, microvilli have been considered the likely source of P4 particles (), given that microvilli may give rise to small extracellular membrane vesicles (; ) and that the P4 particle constituent prom1 is concentrated on microvilli of NE cells (). However, at the onset of neurogenesis, when the P4 particles accumulate in the neural tube fluid, most NE cells bear only few, if any, microvilli (), raising the possibility that NE cell structures other than microvilli give rise to P4 particles. Prom1, the characteristic membrane constituent of the P2 and P4 particles, has intriguing features. First, prom1, also called CD133 (), is not only found on NE cells but is widely expressed by many somatic stem cells (; ; ; ; ). Second, being a pentaspan membrane protein, prom1 is the defining constituent of a specific, cholesterol-based membrane microdomain that is characteristic of various types of plasmalemmal protrusions exhibiting substantial membrane curvature (; ; ). Collectively, these findings imply that the release of the prom1 particles into the neural tube fluid would be a means of not only reducing the size of the apical membrane of NE cells but also modifying its composition by depleting a stem cell–characteristic membrane microdomain. Here, we show that prom1 is concentrated at apical midbodies of symmetrically dividing NE cells and, after the onset of neurogenesis, on primary cilia, and that these NE cell surface structures are the sites of origin of extracellular prom1 particles. We focused on the neuroepithelium as a source of prom1 particles because the neural tube fluid of the mouse embryo has previously been shown to contain both P2 and P4 membrane particles () and NE cells express prom1 on their lumenal surface (). To investigate the formation of the prom1 particles, a mouse prom1-GFP fusion protein was expressed in the embryonic chick spinal cord neuroepithelium, which is structurally very similar to the mouse neuroepithelium, but a simpler experimental system (). Prom1-GFP was expressed at Hamburger and Hamilton (HH) stage 10–11 and analyzed 24 h later (HH17), i.e., at the onset of neurogenesis. Remarkably, prom1-GFP was not only concentrated at the apical surface of the transfected side of the neuroepithelium (, open arrow) but also detected as punctate structures at the surface of the contralateral, untransfected side of the neuroepithelium (, arrowheads). Higher magnification revealed that the punctate structures had a ring-like appearance (, insets), as previously reported for the P2 particles enriched in prom1 (). In contrast to prom1-GFP, monomeric red fluorescent protein (mRFP), which was cotransfected with prom1-GFP, remained confined to the transfected side of the neuroepithelium (). These observations suggested that the prom1-GFP particles associated with the contralateral neuroepithelium originated from the transfected side by some transfer of membrane. To explore this possibility, prom1-GFP was expressed in the chick spinal cord (HH10–11), and slice cultures prepared 24 h later were analyzed by time-lapse confocal imaging. This revealed the presence of pleiomorphic prom1-GFP particles in the lumen of the neural tube that rapidly passed through the field of observation (Video 1, available at ). Despite the dynamic movement of these particles, their generation from the apical surface of the transfected NE cells could be captured in some cases with the present experimental setup (). In these instances, prom1-GFP appeared to concentrate within the apical membrane, followed by release of a particle from the surface of the transfected NE cell (; and Videos 1 and 2). After its release, the prom1-GFP particle quickly disappeared from the imaged field (; and Videos 1 and 2). Interestingly, a released particle appeared to be able to enter the contralateral neuroepithelium (). We conclude that prom1 particles are released from the apical surface of NE cells into the neural tube lumen. To identify the sites of origin of the prom1 particles, we explored the possibility that these particles may originate from microtubule-based plasma membrane protrusions. The rationale for this was that prom1 is known to be concentrated in plasma membrane protrusions (; ), but the majority of the prom1 particles in the neural tube fluid lack actin (). Indeed, upon differential centrifugation of embryonic day (E) 11.5 mouse neural tube fluid followed by immunoblotting, the P2 pellet, in which most of the large prom1 particles are known to be recovered (), showed a striking enrichment of α-tubulin (). Consistent with this, double immunofluorescence of the prom1 particles in the lumen of the telencephalic ventricles revealed that a substantial fraction of them (57 ± 11%) contained α-tubulin, which appeared to be concentrated in the center of the ring-like prom1 staining (). Double immunogold EM of the P2 pellet corroborated this conclusion () and revealed, for the E10.5 telencephalon, α-tubulin inside electron-dense structures that showed prom1 labeling at their surface and that appeared to be detached (or at least emerging) from the apical surface of the neuroepithelium (). In light of these observations, we considered as putative donor membranes for the lumenal prom1 particles two apical membrane structures of NE cells that are known to be associated with clusters of microtubules, the cilium (; ) and the midbody, a cytoplasmic bridge connecting nascent daughter cells until the completion of cytokinesis (; ). EM of serial plastic sections of apical midbodies of dividing NE cells in the mouse E10.5 telencephalon yielded three principal results. First, we observed membrane buds of various sizes emerging from midbodies, as well as vesicular profiles in their immediate vicinity (, arrowheads; and Fig. S4, A and B, available at ), consistent with extracellular membrane vesicles arising from midbodies. Second, although certain midbodies showed the typical ordered array of microtubules and could be identified as such in single sections (), others showed the same morphological appearance as the previously observed pleiomorphic protuberances () and could only be identified as a midbody upon serial sectioning (; and Fig. S4 B). In the case of the latter midbodies, an ordered array of microtubules was less obvious (), and the texture of the cytoplasm in the core region appeared more heterogeneous, showing more electron-dense areas (usually in the cortical region) as well as less condensed patches (). Moreover, these midbodies were usually observed on NE daughter cells that appeared to have entered G1 by morphological criteria, such as the presence of a cilium, which is known to be disassembled during M phase (; ; ; ) and to reform thereafter (Fig. S4 B), an elongated cell shape (; and not depicted), and an abventricular position of the nucleus (; Fig. S4 A; ). Given this temporal relationship, we will refer to these as “aged” midbodies. Third, we observed pleiomorphic membrane structures that, as revealed by serial sectioning, were clearly detached from the apical surface of NE cells, but whose morphology was otherwise similar to that of the core of aged midbodies (). As these detached membrane structures were reminiscent in appearance of the P2 particles in the neural tube lumen (; Fig. S2, available at ; ), our observations collectively suggest that the latter particles may originate from the midbody. If P2 particles originate from the midbody, one would expect prom1 to be concentrated there. Immunogold labeling EM showed that this was indeed the case. Specifically, prom1 was clustered at the surface of the electron-dense central portion of the midbody (). Remarkably, prom1 labeling was particularly strong at buds emerging from the central midbody plasma membrane (, arrowheads). Comparison of various embryonic stages revealed that midbodies in which the electron-dense central region was connected to the daughter cells via relatively short stalks were observed both before (E8.5; ) and after (E12; ) the onset of neurogenesis in the telencephalon, whereas midbodies with long, thin stalks () were only detected at an early stage (E8.5). Prom1 immunolabeling of the various midbodies yielded three observations worth noting. First, in the case of short midbodies with an ordered array of microtubules, prom1 was mostly detected on the plasma membrane domain facing the lumen (, solid arrows), rather than on the membrane domain derived from the cleavage furrow (open arrows). This is consistent with the former midbody plasma membrane domain corresponding to the prom1-bearing, apical membrane of the mother cell. Second, in the case of aged midbodies, which had short or long stalks, prom1 appeared to surround the central core (), suggesting that midbody aging is accompanied by membrane microdomain redistribution. Third, in the case of the long, thin midbodies, prom1 was confined to the very central portion (), indicating that this represented a distinct membrane subdomain. In addition, the immunogold labeling of the apical surface of the neuroepithelium revealed that small (30–60-nm diameter) membrane buds with clustered prom1 were observed not only on the central portion of the midbody () but also on relatively large (1-μm diameter) structures detached from, but nonetheless in the vicinity of, the apical surface of NE cells (, gray arrowheads). This suggests that the small (50–80 nm) P4 prom1 particles in the neural tube fluid () may originate not only from apical membrane protrusions of NE cells such as microvilli (; for cilia, see ) but also from the larger P2 prom1 particles in the neural tube lumen that in turn have originated from the midbody of NE cells. A characteristic component of the midbody is anillin, an actin-binding protein associated with the contractile ring (; ). If the prom1 particles found in the neural tube fluid arise from the midbody, one might expect them to contain anillin. Indeed, double immunogold labeling of E10.5 telencephalic NE cells revealed that not only apical midbodies contained anillin (in addition to prom1; ), but also the prom1 particles that had detached from the apical surface () and could be isolated from the neural tube fluid (, arrowheads). By double immunofluorescence, a substantial portion (39 ± 8%) of the prom1 particles in the lumen of the E10.5 telencephalic ventricle contained anillin (). Furthermore, the vast majority of anillin-positive structures at the apical surface of the E11.5 telencephalic neuroepithelium also contained prom1 (, arrowheads), although within each structure, the distribution of the two markers relative to one another was distinct (). These findings corroborate our conclusion that prom1 particles in the neural tube fluid arise from the midbody of NE cells. Given these observations with fixed samples, we performed time-lapse confocal imaging of live dividing NE cells in slice culture after expression of GFP-anillin in the chick spinal cord (Fig. S1 and Video 3, available at ). This revealed that after its release from the nucleoplasm into the cytoplasm upon nuclear envelope breakdown, GFP-anillin clustered at the basal pole of the cell body, associated with the contractile ring, and became concentrated in particles that formed at the apical surface of the dividing cell at the end of cytokinesis. These apical GFP-anillin–containing particles eventually disappeared from the daughter cells, consistent with them being released into the neural tube lumen. It has previously been reported that, depending on whether an NE cell undergoes symmetric or asymmetric division, the cleavage furrow ingressing from the basal side will either bisect or bypass the apical membrane, resulting in its inheritance by both or only one of the daughter cells, respectively (). A corollary of this is that in a symmetric division, the midbody contains apical membrane, whereas in an asymmetric division, it does not. In light of the present findings, we extended previous observations related to this issue () and systematically determined the localization of midbodies relative to the apical, prom1-containing membrane of telophase NE cells. This analysis was performed in -GFP knockin mouse embryos, in which GFP is specifically expressed in NE cells undergoing neurogenic (rather than proliferative) divisions (); neurogenic (-GFP–positive) divisions at the ventricular surface are predominantly (but not exclusively) asymmetric, whereas proliferative (-GFP–negative) divisions are nearly always symmetric (). Our first approach was double-immunofluorescence analysis for anillin versus prom1 localization in E11.5 forebrain NE cells in telophase. shows an example of the frequently observed cases of an anillin-labeled midbody being colocalized with prom1, which is consistent with the results of and ; the corresponding NE cell lacked -GFP expression () and, hence, most probably underwent a proliferative division. , in contrast, shows an example of the fewer cases of an anillin-labeled midbody being located just underneath the ventricular surface and lacking prom1; here, the corresponding NE cell showed -GFP expression (), i.e., underwent a neurogenic division. These data on anillin versus prom1 localization were corroborated by 3D reconstruction (). Further support for the notion that the release of midbodies containing apical, prom1-enriched membrane is characteristic of NE cells undergoing symmetric, proliferative division was obtained when we identified the apical membrane of single mitotic NE cells by counterstaining for the lateral membrane protein cadherin (). Specifically, we determined by double-immunofluorescence analysis of E11.5 forebrain NE cells in telophase whether an anillin-labeled midbody was colocalized with the cadherin “hole” (), i.e., the apical membrane, or with the lateral membrane including the cadherin-stained adherens junction. and the 3D reconstruction () show an example of an anillin-labeled midbody being colocalized with the cadherin hole (apical midbody); together with the lack of -GFP expression (), this indicated that the corresponding NE cell underwent a symmetric, proliferative division. In contrast, and the 3D reconstruction () show an example of an anillin-labeled midbody being colocalized with the cadherin-stained adherens junction (subapical midbody); together with the presence of (albeit weak) -GFP expression (), this indicated that the corresponding NE cell underwent an asymmetric, neurogenic division. We quantified the localization of the midbody relative to the apical versus lateral membrane of telophase NE cells in the forebrain at three developmental stages (). After the onset of neurogenesis (E11.5) through midneurogenesis (E14.5), ∼90% of NE cell divisions at the ventricular surface were symmetric and ∼10% asymmetric. (Consistent with previous observations [], some of these symmetric divisions were -GFP positive [, green].) In contrast, before the onset of neurogenesis (E9.5), essentially all anillin-labeled midbodies were localized to the cadherin hole and none of the NE cells yet expressed -GFP, consistent with the notion that at this stage all NE cells undergo symmetric, proliferative divisions and the midbodies released from these cells contain apical membrane, accounting for the accumulation of the P2 prom1 particles in the neural tube fluid before neurogenesis (). Besides the midbody, cilia are microtubule-based protrusions of the plasma membrane (; ). Given that NE cells are known to bear a primary cilium on their apical surface (; ), and the presence of α-tubulin in the P2-type prom1 particles in the lumen of the neural tube (), we investigated whether prom1 is associated with the cilia of NE cells. Transmission EM analysis of mouse forebrain neuroepithelium subjected to preembedding immunogold labeling for prom1 showed that at E8.5, i.e., before the onset of neurogenesis, most (if not all) cilia lacked prom1 (). In contrast, at E10.5 () and E12.5–13.5 (), i.e., at the onset and during the early stages, respectively, of neurogenesis in the telencephalon, an increasing proportion (>50%) of cilia of NE cells contained prom1 on their surface. The amount and distribution of prom1 was variable, with some cilia being labeled over most of their surface () and others being labeled preferentially at their tips (). In some cases, the clustering of prom1 at the tip of the cilium was suggestive of a membrane budding process (, arrowheads). In addition, we occasionally observed electron-dense, strongly prom1-immunoreactive particles in the immediate vicinity of cilia that, remarkably, were rather short (). An overview by scanning EM of the immunogold-labeled ventricular surface of E10.5 forebrain neuroepithelium corroborated the presence of prom1 on cilia (, arrowheads) and other apical membrane structures (arrows). We report a novel role of the midbody and primary cilium, the removal of a stem cell–characteristic membrane microdomain from somatic stem and progenitor cells via the release of extracellular membrane particles. Our study shows that apical midbodies of NE cells are the source of the P2 particles in the neural tube lumen that show a ring-like prom1 immunostaining (). The evidence includes the resemblance in overall morphology between the electron-dense particles in the neural tube lumen () and the central portion of apical midbodies (); the enrichment of tubulin, which is known to be particularly concentrated in the central portion of the midbody (; ), in the P2 particles (); the presence of anillin, a marker of the midbody (; ), in the P2 particles; and the clustering of prom1 at the central portion of the midbody plasma membrane (). In fact, the ring-like appearance of the prom1-stained P2 particles in the neural tube lumen (; ; ; and ) is strikingly reminiscent of the midbody ring (). Considering the midbody ring structure and the present observations together, we suggest that the P2 particles in the neural tube lumen showing ring-like prom1 staining reflect the release of the central portion of apical midbodies from NE cells into the ventricular fluid (, pathway 1). The fate of the midbody after completion of cytokinesis, including its possible release, has been discussed in previous EM studies on NE and other cells (,; ; ; ; ; ; ; ). However, conclusive evidence showing that a midbody is actually detached from both daughter cells, which requires serial sectioning transmission EM, has so far been provided in only one of these studies, in which a human bone marrow–derived cell line was investigated in vitro (). Previous observations with NE cells have been inconclusive as to whether the midbody fully detaches or remains with one of the daughter cells, and their interpretation by the respective investigators has been contradictory (; ; ; ). Moreover, although some investigators have assumed that the midbody may be discarded after each cell division (), more recent studies with HeLa cells have supported the widely held view that the midbody is inherited by one of the daughter cells (; ). The evidence presented in this paper, obtained by serial sectioning EM of neuroepithelium ( and Fig. S4 C); immunoblotting, immunofluorescence, and immuno-EM analyses of isolated neural tube fluid ( and ); and live imaging of prominin-GFP and GFP-anillin release from NE cells ( and Videos 1–3), demonstrates that apical midbodies are released from NE cells into the extracellular space. Such a release implies two sets of membrane fusion events, one on either side of the midbody ring (, a, dashed lines). The first fusion constitutes abscission, which terminates cytokinesis () and, hence, the connection between the NE daughter cells. The second fusion then releases the central portion of the midbody containing the midbody ring as a P2 prom1 particle from one of the NE daughter cells into the neural tube fluid. At present, very little is known about the molecular machinery mediating midbody release, as compared with the abscission step in HeLa cells (; ). (Prom1, used in the present study as a marker to monitor the release of the midbody, does not appear to be essential for this release. Comparison of wild-type and prom1 knockout mice did not reveal an obvious difference in the number of P2 particles in the neural tube, using either immunofluorescence for the midbody protein CRIK [citron rho interacting kinase] or conventional EM [unpublished data].) However, time-lapse imaging of the release of prom1-GFP–labeled and GFP-anillin–labeled particles from NE cells provided clues as to when during the cell cycle this release takes place. We observed that the release of prom1- and anillin-bearing P2 particles occurred from NE cells whose nuclei had migrated from the lumenal surface toward the basal region of the neuroepithelium (, Fig. S1, and Videos 1–3). As the apical-to-basal migration of NE cell nuclei is known to occur in G1 (), this observation implies that the midbody is not necessarily released shortly after completion of mitosis but can be released when the daughter cell has progressed well into G1. Similarly, abscission may occur after the NE daughter cells have entered G1, as we observed aged apical midbodies (i.e., whose central region resembled P2 particles) that were still connected to daughter cells whose nuclei were already in an abventricular position (not depicted) and that had reformed a primary cilium ( and Fig. S4 B). The fate of the apical midbody of NE cells apparently is more complex than just to be released as a whole. We observed membrane buds of various sizes emerging from both cell-attached midbodies and midbody remnants detached from the cells ( and ). Prom1 was concentrated on these buds (). The larger of the membrane vesicles arising from these buds may well correspond to those P2 prom1 particles that lack tubulin and anillin ( and ) and thus do not appear to constitute complete midbodies (). The smaller of these buds may give rise to the P4 prom1 particles observed in the neural tube fluid and thus may constitute a second source of the latter particles in addition to microvilli (; ). The membrane vesicle budding from the detached midbodies probably corresponds to the previously observed midbody “deterioration” (). Importantly, given that plasma membrane protrusions showing prom1 clustering are known to contain specific membrane microdomains (), our observations imply that the apical midbody of NE cells is a specific donor structure for extracellular membrane traffic, both while being cell attached and after its detachment. The present study adds the primary cilium to the list of plasma membrane protrusions in which prom1 is concentrated. Interestingly, although the protrusions previously reported to bear prom1 (e.g., microvilli, lamellipodia, and filopodia; ; ; ) are actin based, the two types of structures found here to carry prom1, the midbody and the primary cilium, are microtubule based. Although the prom1-labeled apical midbodies account for the prom1-bearing pleiomorphic protuberances described previously (), the presence of prom1 on primary cilia has not been noticed so far. In particular, two of our observations in this regard deserve comment. First, the finding that most (if not all) of the primary cilia of mouse NE cells at E10.5–13.5, i.e., at the onset of neurogenesis and thereafter, contained prom1 (), whereas at an earlier stage (E8.5), prom1 was barely detectable on primary cilia (), raises the possibility that the presence of prom1 on cilia is linked to the state of differentiation of these neural progenitor cells. The appearance of prom1 on cilia with the onset of neurogenesis is also consistent with the time course of appearance of P4 particles in the neural tube fluid (). Second, the distribution of prom1 on primary cilia suggests that small membrane vesicles bud from their tips. In a preliminary report, Huang et al. (Huang, K., Diener, D.R., Karki, R., Pedersen, L.B., Geimer, S., and Rosenbaum, J.L. 2005. American Society for Cell Biology Annual Meeting. Abstr. 1513), independently reached a similar conclusion for the cilia/flagella of , consistent with earlier observations in this organism by others (; ; ; ). If the small, P4 prom1 particles known to exist in the neural tube fluid () indeed originate, at least in part, from the primary cilia of NE cells (), this would raise the possibility that these organelles not only exhibit a signal-receiving function for their cells (; ; ; ; ), but perhaps also a signal-sending function toward other cells. In addition, membrane budding from primary cilia may be part of the mechanism controlling their length, which varies between NE cells at different stages of development and at specific phases of the cell cycle (; ; ). It is interesting to note that primary cilia, which exist through interphase but not M phase (; ; ; ), and midbodies, whose existence is linked to M phase, share numerous components, including microtubule arrays and associated motor proteins, as well as specific protein kinases (; ; , ; ). The presence of the somatic stem cell membrane marker prom1 and the formation of prom1-bearing membrane buds on both primary cilia and apical midbodies of NE cells provide further support for a relationship between these two organelles. Processes in which these organelles may cooperate include via their ability to release membrane particles, the reduction in the apical surface of NE cells that occurs during development and with the switch to neurogenesis (; ) and extrapolating from the role of primary cilia in polycystic kidney disease (; ) the regulation of NE cell cycle progression. Our group previously showed that cleavages occurring approximately perpendicular to the ventricular surface of NE cells can be either symmetric or asymmetric, depending on whether the apical membrane is bisected or bypassed, respectively, by the cleavage furrow (). According to this concept, only cleavages bisecting the apical membrane will, at the end of cytokinesis, form apical midbodies that contain prom1 on their surface (i.e., midbodies emerging apical to the adherens junction belt and constituting a cytoplasmic bridge connecting the apical surfaces of the two daughter cells []). In contrast, cleavages bypassing the apical membrane would be expected to form midbodies at the apical-most end of the lateral membrane (including junctional complexes) that lack the apical membrane protein prom1. Our observations regarding apical () versus subapical (, and T′) anillin clusters are consistent with this concept. Importantly, these imply that the midbody-derived prom1 particles in the lumen of the neural tube predominantly originate from NE cells undergoing symmetric, proliferative rather than asymmetric, neurogenic divisions. The accumulation in the neural tube fluid of the P2 prom1 particles before the onset of neurogenesis and their decline thereafter () is fully consistent with this notion. There is an additional, major implication regarding the release of the lumenal prom1 particles from apical midbodies of NE cells. In HeLa cells, the midbody ring is inherited asymmetrically by one of the daughter cells and persists through interphase (; ). Our observations imply that neural progenitor cells, in contrast to HeLa cells, prevent such asymmetry by releasing their apical midbody. Why do NE cells, the primary progenitor cells of the mammalian central nervous system, cluster prom1, a somatic stem cell marker (; ; ; ; ) and defining constituent of a specific cholesterol-based membrane microdomain (), at the midbody and release these membrane domains from, or together with, the midbody into the extracellular space? We previously discussed two possible roles of the release of prom1 particles from NE cells (which are not mutually exclusive): disposal of a stem cell membrane microdomain and intercellular signaling (). In this regard, the present findings that the midbody of these neural progenitors is discharged into the neural tube lumen and is the source of the extracellular prom1 particles are particularly intriguing. In terms of disposal, the removal of the prom1-bearing membrane from the cell concomitant with the midbody would link the former precisely to the terminal step of the cell cycle, thereby ensuring the persistence of this membrane throughout, but not beyond, a given cell cycle. In terms of signaling, the release, by a given NE cell, of its apical midbody into the extracellular space could provide quantal information to the surrounding tissue about the history of division of that NE cell, i.e., that it underwent a symmetric, proliferative division. This may be important for controlling tissue growth. It will therefore be interesting to compare the fate of the midbody, including its specific membrane domains, between physiologically dividing and cancer cells. A fusion construct of GFP with the N terminus of mouse anillin (provided by Y. Kosodo, RIKEN, Kobe, Japan) was generated using standard cloning techniques. The GFP-anillin fusion construct was inserted into the pCAGGS expression vector. mRFP () was also inserted into the pCAGGS expression vector to monitor the transfection of the chick spinal cord. Chicken eggs (obtained from Lohmann Tierzucht GmbH) were incubated for 2 d in a humidified incubator at 38°C. At HH10–11, mouse prom1-GFP, GFP-anillin, and mRFP cDNAs (1–2 μg/μl), all driven by the cag promoter (CMV enhancer coupled to β-actin promoter; ), were injected (<0.1 μl) into the lumen of the chick spinal cord and electroporated unidirectionally using a BTX electroporator by applying five pulses of 25 V for 30 ms each (; ). After electroporation, eggs were incubated for 20–24 h, and spinal cord slice cultures were prepared and subjected to time-lapse fluorescence microscopy as described previously (), with the following modifications. Transfected spinal cord regions were dissected and sliced manually with a razor blade (∼400-μm-thick transverse slices). Slices were embedded in collagen (Cellmatrix type I-A; Nitta Gelatin, Inc.), which was diluted to 1–1.2 mg/ml in DME/F12 (Invitrogen), 26 mM NaHCO, 20 mM Hepes, and 5 mM NaOH; incubated for 20 min at 37°C to jellify; and covered with DME containing 100 U/ml penicillin/streptomycin, 10% chicken serum, and 5% fetal calf serum in a POC chamber at 37°C for imaging. Slices were imaged using an inverted confocal microscope (model IX81 [Olympus]; objective PlanApo 60× oil; NA 1.4) and imaging system (FluoView 1000; Olympus). Acquisition of fluorescent and differential interference contrast (DIC) images was performed simultaneously at a depth of 40–80 μm in the slice. Approximately 10 optical sections were collected in the z axis, with 1–1.2-μm steps. Imaging was performed either for 0.5–1.5 h, with the slice being scanned every 35–120 s, or for almost 7 h, with the slice being scanned every 5 min. Images were analyzed with Imaris software (Bitplane), level and gamma value were adjusted, and images were assembled with Photoshop and Image Ready (Adobe). knockin line (C57BL6 background; ), and the prom1 knockout line (generated in the laboratory of P. Carmeliet, University of Leuven, Leuven, Belgium), and electroporated chick embryos were fixed in 4% paraformaldehyde and cryosectioned. For immunostaining, sections (14–18 μm) were incubated with rat mAb 13A4 against mouse prom1 (1.2 μg/ml; ), affinity-purified rabbit anti-megalin antibody (1 μg/ml; obtained from S. Argraves, Medical University of South Carolina, Charleston, SC; ), affinity-purified rabbit anti-anillin antibody (1:400; obtained from C. Field, Harvard Medical School, Boston, MA; ), mouse mAb against α-tubulin (1:400, Sigma-Aldrich), and mouse mAb recognizing pan-cadherin (1:200; Sigma-Aldrich), followed by Cy3- and Cy5-conjugated secondary antibodies (1:1,000; Jackson ImmunoResearch Laboratories). Nuclei were stained with DAPI (125 ng/ml; Sigma-Aldrich). Images were acquired either on a conventional fluorescent microscope (model BX61 [Olympus]; camera and acquisition device obtained from Diagnostic Instruments, Inc., and ViSytron Systems with IPLab software) or with an inverted microscope (Axiovert 200; Carl Zeiss MicroImaging, Inc.) and laser-scanning confocal imaging system (LSM 510 [Carl Zeiss MicroImaging, Inc.]; PlanApo 63× oil; DIC objective, NA 1.4) used in conjunction with LSM 510 AIM acquisition software (Carl Zeiss MicroImaging, Inc.) using 1-μm z steps (pinhole at 1 area unit for prom1, cadherin, and anillin), and processed with Photoshop. For anillin versus prom1 localization by double immunofluorescence, only mitotic NE cells at the ventricular surface were analyzed, in which anillin was clustered at the apical surface, i.e., cells in late telophase. We first analyzed whether such apically clustered anillin was colocalized with (symmetric division) or distinct from (asymmetric division) prom1. This analysis was performed without knowledge by the investigator of -GFP expression, which was determined subsequently. For anillin versus cadherin localization by double immunofluorescence, only mitotic NE cells at the ventricular surface were analyzed, in which anillin was either clustered at the apical surface, i.e., cells in late telophase, or at least associated with the contractile ring of cells in anaphase or telophase. Following previously described methods (), the orientation of the cleavage plane was first deduced from the orientation of the sister chromatids and then corroborated by the orientation of the contractile ring as revealed by anillin immunostaining (in the case of incomplete [0–80%] ingression of the cleavage furrow) or the orientation of the cleavage furrow as revealed by cadherin immunostaining (in the case of complete [90–100%] ingression of the cleavage furrow). We then determined whether cleavage would be predicted to bisect (symmetric division) or bypass (asymmetric division) the apical membrane, revealed by cadherin immunostaining as cadherin “hole” (). In the case of complete ingression of the cleavage furrow, the localization of the midbody (revealed by anillin immunostaining) relative to the cadherin hole served as additional, decisive criterion. Symmetric versus asymmetric division was determined without knowledge by the investigator of -GFP expression, which was scored last as either weak or strong. Only mitotic NE cells in which symmetric versus asymmetric division could be determined unambiguously and that received the same scoring independently by two investigators were considered, and only cells showing 90–100% complete ingression of the cleavage furrow were eventually included in the quantification. 3D reconstruction of anillin versus prom1 or cadherin immunofluorescence and DAPI staining was performed from six to eight optical sections after applying a Gaussian filter, using Imaris 4.1.1 software, specifically the Iso Surface function with appropriate threshold settings. Neural tube fluid of wild-type E11.5 NMRI mouse embryo was collected and fractionated by differential centrifugation, and fractions were analyzed by immunoblotting using mouse mAb against α-tubulin (1:4,000; Sigma-Aldrich) as described previously (). Mouse embryos were fixed for 1–2 h in 4% paraformaldehyde and overnight in 2% glutaraldehyde, both in phosphate buffer. Defined regions of the embryonic brain were cut and processed for EM. The tissue was postfixed with 1% osmium tetroxide and dehydrated through a graded series of ethanol at room temperature (15–30 min for each step) before infiltration with EmBed resin (Science Services) and polymerization at 60°C for 2 d. Ultrathin sections (70 nm) were cut on a UCT microtome (Leica Microsystems) and viewed in an electron microscope (Morgagni; FEI Company). Micrographs were taken with a charge-coupled device camera (MegaviewII; Soft Imaging System) and AnalySis software (Soft Imaging System) or with plate negatives (SO163; Kodak), which were scanned on a flatbed scanner (PowerLook 1100; UMAX) with transmitted light and processed with Photoshop. Samples of mouse embryonic brain were processed for Tokuyashu cryosectioning and subsequent single or double immunogold labeling as previously described (). The following antibodies were used: 25–50 μg/ml mAb 13A4 followed by rabbit antiserum to rat IgG (Cappel) and protein A/5 nm gold (Utrecht University); rabbit antibodies against anillin and megalin (see Immunohistochemistry) followed in either case by protein A/10 nm gold; and mouse IgG1 against α-tubulin (1:600; Sigma-Aldrich) or mouse IgG2b against acetylated tubulin (1:100; clone 6-11B-1; Sigma-Aldrich) followed by a secondary goat antibody anti-mouse IgG/M coupled to 12 nm gold (Dianova). The P2 pellet, isolated by differential centrifugation of neural tube fluid isolated from 23 E10.5 mice (), was resuspended in 4% paraformaldehyde in phosphate buffer and adsorbed to 400 mesh formvar/carbon-coated grids. The samples were processed through immunogold labeling and negative contrasting as described previously (). E8.5–12.5 mice were fixed for at least 24 h in 4% paraformaldehyde in phosphate buffer and stored in fixative until use. E8.5 embryos were manually cut into small pieces, whereas older embryos were embedded in 3% low-melting agarose and 200-μm-thick vibratome sections (VT1000S; Leica Microsystems) were prepared. In the case of the vibratome sections, the surrounding agarose was removed before postfixation in 4% paraformaldehyde in phosphate buffer. Tissue pieces and vibratome sections were blocked with 0.5% BSA and 0.2% gelatine in PBS before incubation with 1 μg/ml rat mAb 13A4 for 2 h at room temperature. After washing in blocking buffer, samples were postfixed in 4% paraformaldehyde, blocked, and incubated with rabbit antiserum against rat IgG (1:1,000) followed by protein A/10 nm gold. Samples were washed with PBS and postfixed with 1% glutaraldehyde in phosphate buffer for 15 min at room temperature. Samples were then processed for plastic embedding as described above (see Conventional transmission EM). Contrast settings of scanned micrographs as whole (see Conventional transmission EM) were increased with Photoshop. Prom1 immunogold labeling for scanning EM was performed with E10.5 embryos as described (see Preembedding immunogold transmission EM), except that the brain ventricles were opened manually by a sagittal cut, and a goat anti-rabbit antibody coupled to 18 nm gold (Dianova) was used instead of protein A. Glutaraldehyde-postfixed samples were processed for detection of the backscattered electrons by scanning EM as described previously (). The formation of GFP-anillin–containing particles at the apical surface of NE cells at the end of cytokinesis is shown in Fig. S1 and Video 3. A midbody and electron-dense particles containing prom1 in the lumen of the chick spinal cord are displayed in Fig. S2. Fig. S3 and the corresponding text describe that the megalin-containing apical membrane subdomain does not overlap with the prom1-containing membrane subdomain. Fig. S4 provides the complete series of micrographs from which (A and E–H) was selected. Videos 1 and 2 show the release of prom1-GFP– bearing particles from the apical surface of NE cells. Videos 4–7 provide an animation of the 3D reconstructions shown in (E′, J′, O′, and T′). Online supplemental material is available at .

The rapid transition of a stationary axonal structure into a motile growth cone (GC) after axotomy involves massive restructuring of the cytoskeleton (; ), the recruitment and localization of membrane resources, and their insertion into the plasma membrane (; ; ). Although the sequence of microtubules (MTs) and actin network restructuring after axotomy was recently analyzed (; ), the mechanisms that regulate vesicle accumulation at the cut axonal end did not receive much attention. This is probably the result of the intuitive assumption that after axotomy, vesicles accumulate at the tips of the disrupted MTs simply because they cannot move efficiently beyond this point. Earlier studies revealed that axotomy is associated with massive membrane retrieval along the plasma membrane of the cut axonal end. The retrieved membrane appears to serve as part of the resealing mechanism in a variety of neurons (; ; ; ; ). Whole cell patch-clamp membrane capacitance measurements revealed that the axotomy of cultured neurons activates two processes in parallel: membrane retrieval and exocytosis. Surprisingly, it was demonstrated that axotomy-induced membrane retrieval quantitatively dominates exocytosis for >1 h after axotomy. Thus, although vigorous extension of a GC's lamellipodium was visualized, the total membrane surface area of the neuron decreased (). These observations implied that in order to permit effective extension of a lamellipodium after axotomy, the sites of membrane retrieval and exocytosis must be spatially separate. Attempts to determine the site of membrane insertion and retrieval in neurons yielded conflicting results. For example, suggested that membrane is added along the neurites of cultured neurons. On the other hand, and concluded that new membrane is added to the GC and that bulk membrane endocytosis occurs in the cell body. To the best of our knowledge, no information is available on the budgeting and spatial distribution of membrane resources or about retrieval and exocytosis in regenerating neurons after mechanical injury. Using cultured neurons, we report novel mechanisms that rapidly subdivide the cut axonal end into two structurally and functionally distinct compartments. In one compartment, Golgi-derived anterogradely transported vesicles accumulate and fuse with the plasma membrane in support of GC extension; in the other, retrieved plasma membrane is retained. We demonstrate that formation of the two compartments is generated by reorientation of the MT polarities at the cut axonal end. Our observations suggest that formation of the MT-based vesicle traps optimizes the rapid transformation of an axon into a motile GC after axotomy by sorting and concentrating different membrane resources to restricted sites on the cut axon. xref #text italic xref #text Leibovitz's L-15 medium (Invitrogen) was supplemented for marine species (ms L-15) according to by the addition of 12.5 g/L NaCl, 6.24 g/L D(+) dextrose, 3.15 g/L anhydrous MgSO, 344 mg/L KCl, 192 mg/L NaHCO, 5.7 g/L MgCl-6HO, and 1.49 g/L CaCl-2HO. Penicillin, streptomycin, and amphotericin B (Biological Industries) were added to make final concentrations of 100 U/ml, 0.1 mg/ml, and 0.25 μg/ml, respectively. Culture medium consisted of 10% filtered hemolymph obtained from (the specimens were collected along the Mediterranean coast) diluted in ms L-15. Artificial sea water consisted of 460 mM NaCl, 11 mM KCl, 10 mM CaCl, 55 mM MgCl, and 10 mM Hepes adjusted to pH 7.6. RH237 (-(4-sulfutyl)-4-(6-(p-dibutylamynophenyl) hexatrenyl)) pyridinum and inner salt (a gift from R. Hildeshiem, Weizmann Institute of Science, Rehovot, Israel; ) was diluted in ethanol to a concentration of 10 mM and further diluted before use in artificial sea water to a concentration of 10 μM. SR101 (Kodak) was prepared as a stock solution of 10 mM in double-distilled water and further diluted before use in artificial sea water to a final concentration of 40 μM. BFA (Sigma-Aldrich; ) was prepared as a stock solution of 5 mg/ml in methanol and was further diluted to a final concentration of 10 μg/ml in the experimental bathing solution. Retrograde transport of SR101-labeled vesicles was inhibited by bath application of EHNA and HCl (Calbiochem). For the experiments, EHNA was diluted in DMSO to a 1-M stock solution and was further diluted before use to final concentration of 2–3 mM in artificial sea water. Neurons B1 and B2 from buccal ganglia of were isolated and maintained in culture as previously described (; , ). In this study, we refer to these neurons collectively as B neurons. In brief, 1–10 g of juvenile supplied from the University of Miami's National Resource for was anesthetized by injecting 380 mM of isotonic MgCl solution into the animal's body cavity. Buccal ganglia were dissected and incubated in ms L-15 containing 1% protease (type IX; Sigma-Aldrich) at 34°C for 1.5–2.5 h. After the protease treatment, the ganglia were pinned and desheathed. The neurons were manually pulled out along with their original axon with the aid of a sharp glass microelectrode. The neurons were immediately plated in glass-bottom dishes coated with poly--lysine (Sigma-Aldrich) containing culture medium. All experiments were performed 24–48 h after plating at room temperature (21–25°C) after replacing the culture medium with artificial sea water. Axonal transection was performed by applying pressure on the axon with the thin shaft of a micropipette under visual control as previously described (, , 2003; ). mRNAs were in vitro transcribed using the recombinant transcription system as described previously (). In brief, human protein plus EB3 was prepared as EB3-GFP (), and SNAP-25 (provided by W.S. Sossin, Montreal University, Montreal, Canada) was prepared as EYFP–SNAP-25 or cherry–SNAP-25 (cherry was provided by R.Y. Tsien, University of California, San Diego, La Jolla, CA). These and superecliptic synaptopHluorin (provided by J.E. Rothman, Memorial Sloan-Kettering Cancer Center, New York, NY) were cloned in pCS2+ expression vector. 10 μg of those plasmids were linearized with NotI and purified using a DNA cleanup system (Promega). 1–3 μg of linearized DNA was transcribed using a RiboMax-sp6 kit (Promega). A typical reaction contains 8 μl of transcription buffer, 8 μl rNTP mix containing 25 mM CTP, ATP, UTP, and 12 mM GTP, 4 μl of 15 mM Cap analogue (Roche), 1 μl rRnasin (Promega), and 4 μl of enzyme mix. A final volume of 40 μl was incubated for 2–4 h at 37°C. RNA was purified by using an RNeasy Mini Kit (QIAGEN), and the clean RNA was eluted to a final volume of 25–40 μl and kept at −80°C until use. Two confocal imaging systems were used: the Radiance 2000/AGR-3 imaging system (Bio-Rad Laboratories) was mounted on an IX70 microscope (Olympus) with a plan-Apo 60X 1.4 NA oil objective (Olympus), and the D-Eclipse C1 imaging system (Nikon) was mounted on an Eclipse TE-2000 microscope (Nikon) with a plan-Apo 60X 1.4 NA oil objective (Nikon). SynaptopHluorin imaging was performed on the Nikon set. The protein was excited at 405 (blue diode laser) and 488 nm (argon laser). The emitted light was collected at 500–530 nm. Images were collected and processed using EZ-C1 software (Nikon). All other imaging was performed using the Bio-Rad Laboratories system. The images were collected and processed using LaserSharp and LaserPix software (Bio-Rad Laboratories), respectively. For simultaneous imaging of GFP fusion proteins and RH237, both chromophores were excited by 488 nm, and the emitted lights were collected at 500–530 nm for GFP and above 660 nm for RH237. For simultaneous imaging of GFP or EYFP fusion proteins and SR101 or cherry fusion proteins, the excitation wavelengths were 488 and 543 nm (green HeNe laser). The emission filter for GFP and EYFP was HQ 500–530 nm, and for SR101 and cherry, the filter was HQ 555–625 nm. Triple imaging of GFP- or EYFP-labeled proteins, SR101, and RH237 was collected by excitation wavelengths of 488 and 543 nm, and the emissions filters were HQ 500–530 nm for GFP or EYFP, HQ 560–580 nm for SR101, and HQ 660LP for RH237. The argon laser excitation intensity was usually lowered to 5–10%. The pinhole was set to 1.6–2.5 mm. Figures were prepared using Photoshop and FreeHand software (both from Adobe). Fig. S1 shows that the effective inhibition of SR101-labeled vesicle retrograde transport by EHNA does not inhibit formation of the plus end trap. Videos 1–3 show the formation of MT-based vesicle traps after axotomy. Videos 4 and 5 show the anterograde transport of EYFP–SNAP-25–labeled vesicles and their accumulation after axotomy. Video 6 shows the retrograde transport of SR101-labeled vesicles. Videos 7 and 8 show that the formation of a plus end trap depends on the anterograde transport of components from the cell body to the axon. Videos 9 and 10 show that the formation of MT-based plus and minus end traps cannot be correlated with retrogradely transported retrieved membrane. Video 11 shows that the formation of MT-based plus and minus end traps does not depend on the supply of Golgi-derived anterogradely transported vesicles. Online supplemental material is available at .

Neurons are highly polarized cells with morphologically and functionally distinct subcellular domains. The axon initial segment (AIS) and node of Ranvier are two axonal domains characterized by an “electron-dense” membrane undercoating consisting of voltage-gated Na and K channels (Nav and Kv, respectively), cell adhesion molecules (CAMs), and a specialized membrane cytoskeleton (; ). These domains act as the generator for action potential initiation and propagation (; ), and the diffusion barrier for maintaining axonal polarity (). Disruption of these membrane domains or their molecular composition contributes to the pathophysiology of many nervous system diseases, including epilepsy, multiple sclerosis, and spinal cord injury (; ; ). Consequently, any therapeutic strategy aimed at treating these diseases and reversing their devastating effects will require a detailed understanding of the mechanisms responsible for node and AIS formation and maintenance. From both molecular and functional standpoints, the AIS and nodes of Ranvier are very similar; they have nearly every protein component in common, and both provide the ionic currents necessary for membrane depolarization and action potential initiation/propagation. Despite these strong similarities, one major difference between these two membrane domains is that node formation requires myelination by Schwann cells or oligodendrocytes, but the AIS is intrinsically organized by the neuron. Thus, nodes form “outside-in,” whereas the AIS forms “inside-out” (for review see ). The ankyrin and spectrin protein families play important roles in regulating protein localization and membrane domain formation in many different cell types (). For example, in erythrocytes the spectrin-based membrane skeleton is essential for maintaining the cell's biconcave shape and restricting the lateral mobility of the anion exchanger through the scaffolding protein ankyrinR (ankR; ). The identification and localization of neuronal ankyrinG (ankG) and βIV spectrin provided important clues for the mechanism of AIS and node formation in axons (; ). During development, both ankG and βIV spectrin define putative nodes and initial segments before ion channels cluster (; ). In a mouse lacking ankG in Purkinje neurons, Nav and KCNQ2/3 Kv channels, neurofascin-186, and βIV spectrin all fail to cluster at the AIS (; ; ; ). Further, distinct protein domains in Nav channels, KCNQ2/3 Kv channels, and NF-186 have been identified that mediate their interactions with ankG (; ; ; ). Together, these results all point to ankG as a principal organizer of the membrane proteins located at the AIS. However, in mice lacking βIV spectrin, neither ankG nor Nav channels correctly localize to the AIS, indicating that like ankG, βIV spectrin is also indispensable for domain organization (). To determine whether βIV spectrin directs the formation of the AIS and nodes of Ranvier, we identified the molecular mechanisms regulating its recruitment to these domains. Throughout the central nervous system, AISs are characterized by high densities of Nav channels that colocalize with ankG (; ; ). Despite large dendrites and long axons, high densities of Nav channels and ankG are only found in short ∼20–40-μm-long domains at the proximal region of the axon adjacent to the cell body (). To determine how this specificity is achieved, we used the well-characterized embryonic hippocampal neuron culture system () to study the molecular mechanisms regulating AIS formation; this model has been used previously to elucidate the mechanisms regulating protein sorting and targeting in neurons (; Silverman et al., 2001; ; ; ). To determine if cultured hippocampal neurons form an AIS, we immunostained these neurons after 7–10 d in vitro (DIV) using antibodies against Nav channels, βIV spectrin, and ankG; antibodies against MAP2 were used to identify somatodendritic domains. These neurons typically had 1–2 axons that could be distinguished from dendrites by the enrichment for AIS proteins at the proximal region of the axon adjacent to the cell soma and the absence of MAP2 staining (). There was a marked increase in AIS proteins that corresponded to a complementary decrease in MAP2 (, fluorescence intensity profiles). Thus, hippocampal neurons in vitro form well-defined initial segments that are molecularly identical to those observed in vivo. Different ankyrin and spectrin isoforms are found in a variety of cell types, cytoplasmic organelles, and tissues (). Their unique spatial distributions are critical for the formation and maintenance of specific membrane domains. For example, ablation of the 190-kD isoform of ankG by RNAi disrupts lateral membrane domains in epithelial cells (), and mutations in ankB disrupt the localization of the Na/Ca exchanger, the Na/K-ATPase, and the inositol trisphosphate receptor in cardiomyocyte T-tubule/sarcoplasmic reticulum domains (). Elimination of presynaptic spectrin in results in the loss of synapse-associated CAMs and causes subsequent synapse disassembly (). Thus, ankyrins and spectrins function to link membrane proteins to the cytoskeleton and are indispensable for the formation and stabilization of membrane domains. What do our results reveal about the mechanisms underlying AIS and node of Ranvier membrane domain formation? Whereas in other cellular contexts β-spectrins have been proposed to dictate ankyrin localization (), our results clearly demonstrate that βIV spectrin recruitment to nodes and AIS depends on binding to ankG. Consistent with this idea, βIV spectrin failed to localize properly at the AIS when the dominant-negative AnkG-KK-GFP protein was expressed in neurons. Importantly, loss of βIV spectrin from the AIS did not disrupt Nav channel or endogenous ankG clustering, indicating that ankG directs βIV spectrin localization to the AIS and that βIV spectrin is not required for ankG or Nav channel clustering. This conclusion is consistent with previous studies demonstrating ankG binding is essential for the localization of many membrane proteins, including Nav channels, KCNQ2/3 Kv channels, and NF-186 at the AIS (; ; ). Our results extend the role of ankG to clustering and localization of both cytoplasmic and membrane proteins. Our results demonstrate that ankG directs βIV spectrin localization. However, this conclusion is in direct contrast to a recent study examining the mechanism of ankyrin and β-spectrin localization in (). To determine the function of distinct protein domains in β spectrin, mutant forms were introduced into a fly with a lethal mutation in β spectrin. Surprisingly, loss of the putative ankyrin-binding domain (equivalent to SR15 of βIV spectrin reported in this study) had relatively little effect on localization of β spectrin, and this mutant β spectrin rescued the lethal phenotype. However, deletion of only the PH domain failed to rescue the lethal phenotype. These results suggested that the PH domain may be critical for the membrane targeting and localization of β spectrins. However, our results argue against a role for the PH domain in membrane targeting and localization of βIV spectrin because the Myc- mutant βIV spectrin analyzed in this study was appropriately localized to the AIS. If βIV spectrin is not essential for the targeting and localization of ankG and AIS formation in general, why is there a loss of ankG and Nav channels from the AIS in mice (), and what is βIV spectrin's function? In this study, we examined early events in AIS formation. Previously, using mutant mice we showed that βIV spectrin is essential to maintain membrane structure and the proper molecular organization of nodes, the AIS, and the axonal cytoskeleton (; ). In the analysis of mice, early time points were not considered, and only initial segments of 3-mo-old mice were analyzed (). We speculate that during brain development the AIS forms properly in mice, but that with increasing age the lack of βIV spectrin destabilizes this membrane domain, resulting in the loss of other AIS components. Thus, βIV spectrin is important for node and AIS stability rather than formation. Consistent with this interpretation, compared with 1.5-mo-old mutant mice, there were decreased amounts of Nav channels and ankG at the AIS in 6-mo-old mutant mice. If ankG is the central mediator of AIS and node formation, what determines its localization to these sites? In the PNS, transinteractions between the CAMs NF-186 in axons and gliomedin on Schwann cells causes NF-186 to accumulate at the edges of myelinating Schwann cells (; ). These aggregates of NF-186 are the first axonal proteins detected at nascent nodes (; ). NF-186 binds to ankG, and mice lacking NF-186 fail to cluster ankG at putative nodes of Ranvier, indicating that NF-186 functions as a membrane attachment point for ankG recruitment and clustering (; ). AnkG is thought to act as a protein scaffold to retain Nav and KCNQ2/3 Kv channels at nodes (; ; ). Our results demonstrating that SR15 and ankG binding are required for localization of βIVΣ6 spectrin to nodes of Ranvier provides the first direct evidence that ankG plays a central role in organizing the nodal protein complex. Thus, at nodes of Ranvier, extrinsic glial-derived signals regulate the eventual clustering of ankG and the subsequent accumulation of βIV spectrin. In our experiments, we demonstrated that SR15 is necessary for βIVΣ6 localization at the AIS and nodes of Ranvier. However, we cannot rule out the possibility that additional N-terminal domains may contribute to the localization of βIVΣ1 in neurons. We consider this unlikely at the AIS because the AnkG-KK-GFP protein could block the proper localization of βIVΣ1 (both βIVΣ1 and βIVΣ6 are detected by the antibodies used in this study). However, showed that the addition of a soluble fusion protein (the ectodomain of the CAM IgSF4) to cultured DRG neurons caused clustering of βIV spectrin in the absence of any colocalized ankG. Although the mechanism responsible for this clustering is unknown, this observation suggests that in the PNS additional extrinsic factors may contribute to the localization of βIVΣ1. In contrast to nodes, much less is known about the mechanisms regulating recruitment of ankG to the AIS. Although one report indicated that multiple protein domains are required for ankG's proper AIS localization (), the specific protein–protein interactions involved are unknown. Intriguingly, when Nav channels are eliminated from motor neurons using RNAi, ankG could not be detected at the AIS, suggesting that Nav channels, or their β subunits, may participate in ankG targeting, retention, and/or stabilization (; ). Thus, although the AIS and nodes of Ranvier share a common ankG-based mechanism for membrane domain formation and recruitment of βIV spectrin, the intrinsic determinants regulating ankG localization and restriction to the AIS remain unknown. Identification of the molecular mechanisms regulating ankG localization at the AIS will require a more complete description of the ankG-interacting proteins located within this membrane domain. The full-length βIVΣ6 spectrin with N-terminal Myc tag was provided by M. Komada (Tokyo Institute of Technology, Tokyo, Japan). The C terminus deletion mutants were generated by introducing premature stop codons using the QuickChange mutagenesis kit (Stratagene) and were verified by sequencing. The following primers were used: (aa 1–943), forward, 5′-CGGACACCCCTTGGGACTCCGGC-3′, and reverse, 5′-GCCGGAGTCCCAAGGGGTGTCCG-3′; SR10-15 (aa 1–625), forward, 5′-GCTTCCACAGCTAGGCCCGCGACC-3′, and reverse, 5′-GGTCGCGGGCCTAGCTGTGGAAGC-3′; SR10-14 (aa 1–514), forward, 5′-TCTCGGGAGCTGCATTAGTTCTTCAGCGATGC-3′, and reverse, 5′-GCATCGCTGAAGAACTAATGCAGCTCCCGAGA-3′; SR10-13 (aa 1–401), forward, 5′-GTGAGCCTGGAACAGTAGTACTGGCTCTACCAG-3′, and reverse, 5′-CTGGTAGAGCCAGTACTACTGTTCCAGGCTCAC-3′; and SR10-11(aa 1–191), forward, 5′-CCGCTTGTTGCTGGCATAAAAGGAGCTGCATCAGG-3′, and reverse, 5′-CCTGATGCAGCTCCTTTTATGCCAGCAACAAGCGG-3′. SR14-15 (aa 399–610) was amplified by PCR from pCS3 + MTβIVΣ6 plasmid and inserted into the pCS3 + MT vector (a gift from D. Turner, University of Michigan, Ann Arbor, MI) using EcoRI and XbaI sites. The primers for amplifying SR14-15 are forward, 5′-GCCGAATTCACTGGAACAGCAGTACTGGCTC-3′, and reverse, 5′-CCGTCTAGATTACGAGCATCTTCACAGGCTGCT-3′. The number of amino acid residues is based on the βIVΣ6-A (; National Center for Biotechnology Information database accession no. ). The full-length rat 270-kD ankG with C-terminal GFP tag (AnkG-GFP) and mutant ankG construct (AnkG-KK-GFP) were gifts from V. Bennett (Duke University, Durham, NC). The βIII-spectrin-GFP construct was a gift from M. Stankewich (Yale University, Stamford, CT). For adenoviral constructs, the cDNA encoding Myc-βIVΣ6, Myc-SR10-15, and Myc-SR10-14 with pEF-1α promoter was inserted into pENTR11 vector. The pENTR11 plasmids were recombined with pAd vector using ViraPower Adenoviral Gateway Expression kit (Invitrogen). Adenovirus was produced using human embryonic kidney 293 cells. Hippocampal neurons were dissected and dissociated from E18 rat embryos, plated on 1 mg/ml poly-D-lysine (Sigma-Aldrich)/20 μg/ml laminin (Invitrogen)–coated glass coverslips at a density of 48,000 cells/cm. 3 h after plating, the medium was changed from normal medium (10% FBS in Neurobasal) to maintaining medium (2% B27 [Invitrogen], 0.5 mM -glutamine, 25 μM -glutamate, and 1× antibiotic antimycotic solution (Sigma-Aldrich) in Neurobasal). 2 d after plating, 1 μM cytosine arabinoside (Sigma-Aldrich) was added to inhibit nonneuronal growth. Half of the medium was replaced with an equal volume of maintaining medium without glutamate every 4 d. 7 d after plating, neurons were transfected with various cDNA constructs using Lipofectamine 2000 (Invitrogen) following the manufacturer's instructions. DNA/lipofectamine ratio was 1:3–4 (0.5–1 μg/1.5–4 μl for 35-mm dishes). 24 h after transfection, hippocampal cultures and myelinating DRG-Schwann cell cocultures were fixed with 1 or 4% PFA, respectively (it should be noted that the 1% PFA resulted in a less well-preserved cytoskeleton labeled by MAP2, but this was necessary because antigenicity for some AIS proteins decreased with stronger fixation). Cells were then permeabilized and blocked with 0.3% Triton X-100 in 5% nonfat milk before being stained with antibodies. Brain and optic nerve sections were fixed and immunostained as described in . Mutant mice were identified by PCR screening as previously described (). The mouse monoclonal pan–Na channel (NaCh) and Caspr antibodies have been previously described (; ). Rabbit, mouse, and chicken anti-MAP2 antibodies were purchased from CHEMICON International, Inc., Sigma-Aldrich, and Encor Biotechnology, Inc., respectively. The rabbit polyclonal Nav1.6 and βIV spectrin antibodies have been previously described (; ). The mouse monoclonal ankG antibodies were purchased from Invitrogen. Rabbit polyclonal ankG antibodies were provided by V. Bennett. The Myc and βII spectrin antibodies were purchased from Sigma-Aldrich. The anti-βIII spectrin antibodies were a gift from M. Stankewich. Anti–rabbit GFP antibodies and Alexa Fluor 488– and 594–conjugated secondary antibodies to rat and mouse primary antibodies were purchased from Invitrogen. AMCA-conjugated anti–chicken secondary antibodies were purchased from Jackson Immunoresearch Laboratories. Fluorescence images were collected on an Axioskop 2 (Carl Zeiss MicroImaging, Inc.) fluorescence microscope fitted with a camera (ORCA-ER; Hamamatsu). In some cases, images were taken using an Axiovert 200M (Carl Zeiss MicroImaging, Inc.) fitted with an apotome for optical sectioning, and an axiocam digital camera. Images were taken using 63×/1.4 NA Plan-Apochromat, 40×/1.3 NA Plan-Neofluar, or 10×/0.50 NA Fluar objectives (all Carl Zeiss MicroImaging, Inc.). Acquisition software packages used included both Axiovision (Carl Zeiss MicroImaging, Inc.) and Openlab (Improvision). In some cases, stacks of images were acquired and volume reconstructions were generated using Axiovision software. In some images, contrast and brightness were subsequently adjusted using Photoshop (Adobe). No other processing of the images was performed. Exposure times were controlled so that the pixel intensity in the measured AIS was below saturation; when averages were calculated, all exposure times were held constant. Fluorescence intensity values were measured along axons by tracing a line, beginning at the cell body and running through the AIS and axon. Fluorescence values along this line were measured using ImageJ software (National Institutes of Health). MAP2 staining was used to distinguish dendrites from axons. Only neurons with an axon isolated from other cell processes were chosen for analysis. For each construct, 3–5 independent transfection experiments were done. The expression of Myc-βIVΣ6 spectrin and AnkG-GFP constructs were detected by immunoblot of cell lysates from transfected CHO or COS cells (identical results were obtained in both CHO and COS cells). The lysates were incubated on ice for 30 min and centrifuged at 13,000 for 15 min at 4°C. The supernatants were denatured in SDS sample buffer, subjected to PAGE and electrophoretic transfer, and immunoblotted with Myc or GFP antibody. For coimmunoprecipitation experiments, CHO or COS cells cotransfected with spectrin constructs and ankyrin constructs were solubilized in 250 μl lysis buffer (containing 1% Triton X-100 and protease inhibitors). The soluble materials were incubated overnight with antibody and 25 μl of protein G or A agarose beads (GE Healthcare). The beads were washed six times with 1 ml lysis buffer, and then eluted with 50 μl 2× reducing sample buffer at 100°C for 5 min. The immunoprecipitates were resolved by SDS-PAGE, transferred to nitrocellulose, and subjected to immunoblotting with Myc or spectrin antibody. Fig. S1 shows (A) the molecular weights and expression of Myc-βIVΣ6 spectrin truncation mutants expressed in CHO cells, (B–D) cultured hippocampal neurons transfected with Myc-, Myc-SR10-13, and Myc-SR10-11, and (E) the distribution of βIII spectrin in cultured hippocampal neurons. Online supplemental material is available at .

The sympathetic nervous system regulates cardiovascular function through innervation of the heart, kidney, and blood vessels throughout the body. Sympathetic neurons deliver catecholamines to target tissues, and the adrenal gland releases catecholamines into circulating blood. Although sympathetic nerves are the principal source of catecholamines for cardiac adrenergic receptors (ARs), little is known about the cell–cell interactions between sympathetic nerves and cardiac myocytes. β and βARs, which are members of the G protein–coupled receptor (GPCR) family, form the interface between the sympathetic nervous system and cardiac muscle. However, the function and distribution of specific βAR subtypes at cardiac sympathetic synapses have not been addressed. These homologous receptors play distinct roles in regulating normal cardiovascular physiology (), and there is a growing body of evidence that they play opposing roles in the pathogenesis of heart failure (; ; ). A better understanding of the subtype-specific signaling of β and βARs in cardiac myocytes in response to sympathetic nervous system activation could have implications for the prevention and treatment of heart failure. β and βARs are highly homologous both structurally and functionally. They share 52% identity overall and 76% identity in the transmembrane domains. However, studies in both neonatal and adult cardiac myocytes provide compelling evidence that β and βARs signal through distinct pathways (; ). In neonatal myocytes, activated βAR couples only to Gs (guanine nucleotide–binding protein that stimulates adenylyl cyclase) and leads to a PKA-dependent increase in the contraction rate. In contrast, activated βAR undergoes sequential coupling to Gs and Gi (guanine nucleotide–binding protein that inhibits adenylyl cyclase), having a biphasic effect on the contraction rate that is independent of PKA activation (). Functional differences between β and βARs in cardiac myocytes can be attributed to subtype-specific targeting to different signaling compartments in the myocyte plasma membrane (). Activated βARs undergo robust endocytosis, whereas activated βARs remain at the plasma membrane (). In neonatal cardiac myocytes, endocytosis and recycling are both required for the switch in βAR coupling from Gs to Gi (; ; ). βARs are predominantly concentrated in caveolar structures, whereas βARs are mainly distributed in the noncaveolar membrane (). The cAMP phosphodiesterase PDE4D regulates signaling by the βAR but has no detectable effect on βAR signaling, suggesting that this phosphodiesterase isoform might be a component of the βAR signaling complex (). These observations suggest that distinct signaling domains exist in cardiac myocytes to conduct β and βAR signaling. The heart is richly innervated by sympathetic neurons, which are the principal source of catecholamines for cardiac ARs (Armour, 1994). As β and βARs are the primary sympathetic receptors in the heart, their distribution and function could be influenced by the sympathetic innervation of cardiac myocytes. Synapses in the central nervous system and neuromuscular junctions are formed by coordinated assembly and tight attachment of pre- and postsynaptic specializations (). At the site of contact, the postsynaptic plasma membrane develops into a specialized zone that contains accumulations of neurotransmitter receptors, channels, and anchoring and signaling molecules (). This colocalization is thought to provide a fast and efficient response to released neurotransmitter. Accumulation of receptors at the postsynaptic sites is regulated by synaptogenesis, whereas the dynamic behavior of receptors, such as endocytosis, exocytosis, and lateral movement, is regulated by activity-dependent cues (; ; ; ). In this study, we report the first detailed analysis of the organization of signaling molecules at the site of innervation of cardiac myocytes by sympathetic neurons. We demonstrate that sympathetic ganglion neurons (SGNs) regulate the contraction rate of cultured myocytes and provide evidence that sympathetic innervation influences the structure of the myocyte membrane and the organization and distribution of β and βAR signaling compartments. Cardiac myocytes induce presynaptic differentiation in contacting axons; synaptic vesicles accumulate at the sites of contact as delineated by synapsin I. On the postsynaptic side, the myocyte membrane develops into specialized zones that surround contacting axons and contain accumulations of the scaffold proteins SAP97 and AKAP79/150. In contrast, staining for caveolin-3, which is a marker of signaling caveolae microdomains in cardiac myocytes, is diminished at the sites of contact. The cardiac myocyte membranes are linked to contacting neurons by cadherin–catenin complexes. We have found striking differences in the trafficking of β and βAR-expressed myocytes that have been innervated by cultured sympathetic neurons. βARs accumulate at the synaptic zones, whereas βARs undergo local internalization from the synaptic sites in a neuronal activity-dependent manner. The subtype-specific differences in the localization of β and βAR at sympathetic synapses likely contribute to the regulation of cardiac performance by the sympathetic nervous system in vivo. To determine whether SGNs form functional connections with neonatal cardiac myocytes in vitro, we stimulated the cardiac myocyte contraction rate through the activation of neurons. The sympathetic cervical (autonomic) ganglia contain predominantly neuronal nicotinic acetylcholine receptors (nAChRs), which play a central role in neural transmission in the autonomic nervous system. In vivo, the transmitter released from the preganglionic axon terminals is primarily acetylcholine, which depolarizes postsynaptic neurons by the activation of nAChR (; ). The activation of nAChR with nicotine causes the depolarization of SGN, which, in turn, triggers neurotransmitter release. It has been shown that nicotine-induced changes in the heart rate are associated with an increase in plasma catecholamine concentrations, and they are attenuated by βAR blockade, indicating that the increase in heart rate is mediated by βARs (). Nicotine applied to cardiac myocytes in the absence of neurons did not produce any changes in the contraction rate (unpublished data). Myocytes were cocultured on the same coverslip with SGN or on separate coverslips (). SGNs were stimulated with 1 μM nicotine. This relatively low concentration of nicotine was used to minimize the release of catecholamines into the media because it has been shown that nicotine concentrations over 3 μM can lead to an accumulation of noradrenaline in the media of cultured SGNs (). We observed an increase (∼2.5-fold) in the contraction rate of cardiac myocytes cultured on the came coverslip with SGN (). The effect was much lower when myocytes and neurons were placed on different coverslips and, therefore, could not form connections with each other. Thus, stimulation of the myocyte contraction rate by activated SGNs requires that neurons form contacts with myocytes. #text Synapsin accumulation may not necessarily mark functional sites of transmitter release. To examine synaptic activity, we monitored activity-dependent internalization of an antibody to the luminal domain of synaptotagmin, a method that has been shown to mark the sites of synaptic vesicle endocytosis (). Synaptotagmin antibody was added to cocultures of myocytes and SGNs, and antibody internalization was stimulated by the addition of 500 μM nicotine. Cultures were then fixed and costained for synapsin I (). Synaptotagmin antibody internalization colocalizes with synapsin I throughout the axon. The uptake was completely abolished in Ca-free media. Therefore, synaptic vesicles undergo activity-dependent recycling at the sites of contact of SGNs and cardiac myocytes. We also performed analysis of the activity-dependent destaining of FM1-43 dye. The fluorescent styryl dye FM1-43 is taken up by synaptic vesicles in an activity-dependent manner (; ; ). Cocultures were loaded with FM1-43 in the presence of 500 μM nicotine for 5 min and were washed to remove background staining (Fig. S1 A, available at ). During a second round of nicotine stimulation, the fluorescence intensity of labeled puncta rapidly diminished by ∼80%, reflecting exocytosis of the dye from synaptic vesicles (Fig. S1, B and C). Therefore, synaptic vesicles undergo activity-dependent recycling at the sites of contact of SGNs and cardiac myocytes. As with the antibody to the luminal domain of synaptotagmin, FM1-43 uptake is punctate and is not uniformly distributed along the axon. Therefore, it is likely that release occurs at closely spaced but discrete sites along the entire axonal contact. xref #text Cadherins are calcium-dependent homophilic cell adhesion molecules that maintain the integrity of cell–cell junctions and promote the stability of synapses by linking pre- and postsynaptic membranes. Extracellular domains of cadherins interact directly to help hold the membranes of two cells together. The requirement of mechanical stability is most clear at the neuromuscular junction, which is maintained in the face of constant mechanical stress from muscle contraction (). We observed puncta of pancadherin immunostaining on the surface of cardiac myocytes localized along traces of axons, which were identified by staining with an antibody to tyrosine hydroxylase (). As shown on a two-photon image (), synapsin I puncta were often observed overlapping cadherin immunostaining (frequency of 74 ± 15% [SEM]). However, cadherin puncta are more abundant, and the majority of cadherin staining does not localize with synapsin I puncta. Nevertheless, cadherin puncta are not observed in myocytes cultured in the absence of SGN (unpublished data), suggesting that these puncta represent contacts between SGN fibers and myocytes. The fact that not all cadherin staining localizes with synapsin I puncta suggests that not all contacts between SGNs and myocytes are sites of synaptic activity. Cadherins are known to form complexes with β-catenin, which, in turn, associates with α-catenin and the actin cytoskeleton. Immunostaining of cocultures for synapsin I and β-catenin revealed that the pattern of staining for β-catenin () was very similar to the pattern of staining for pancadherin (), and the majority of synapsin I puncta colocalized with β-catenin (). Our culture results predict that catenin–cadherin complexes may be important for stabilizing interactions between myocytes and neurons in vivo. Therefore, we examined the distribution of β-catenin relative to synapsin I in intact ventricular muscle from the mouse heart. Tissue sections were stained with antibodies to synapsin I and β-catenin and were examined by two-photon microcopy. Consistent with our culture results, we observed the accumulation of β-catenin surrounding contacting axons (Fig. S2, available at ). These results suggest that cadherin–catenin complexes may be involved in stabilizing interactions between SGNs and myocytes in vivo; however, we have no evidence that they play a direct role in synaptic function. Collectively, the aforementioned data demonstrate that in coculture, cardiac myocytes and SGNs form specialized functional connections. Catecholamines (epinephrine and predominantly norepinephrine) are the major neurotransmitters released from SGN (), and βARs are the principal catecholamine receptors in cardiac myocytes. Therefore, we examined how interactions with SGN fibers and neurotransmission affect the localization of βARs in myocytes. Currently available antibodies are not adequate to detect endogenous βARs in cardiac myocytes. Thus, to visualize β and βARs in cardiac myocytes, we expressed FLAG-tagged βAR or HA-tagged βAR via recombinant adenovirus in cocultures. We have previously shown that the functions of epitope-tagged β and βARs expressed in βAR/βAR knockout (KO) myocytes are indistinguishable from endogenous receptors in βAR KO and βAR KO myocytes, respectively (; ). We took advantage of the fact that SGNs at 7–10 d in culture are highly resistant to adenovirus infection; therefore, we were able to express the receptors predominantly in cardiac myocytes and analyze the distribution of βARs relative to synaptic sites. To address activity-dependent events in cocultures, we stimulated neurotransmission through activation of the nAChRs of SGN with 500 μM nicotine. As shown in , there is an invagination of the myocyte membrane at sites of contact between cardiac myocytes and SGN cells. This bears a resemblance to the structure of neuromuscular junctions in skeletal muscle. Moreover, the subtype-specific localization of β and βARs and the exclusion of caveolin-3 suggest that sites of contact between myocytes and SGNs are specialized signaling domains. Therefore, we looked for other molecules that have been associated with synapses and might be involved in the trafficking, regulation, and signaling of βAR. Members of the SAP90/PSD-95 subfamily of membrane-associated guanylate kinase homologues have recently emerged as important players in the molecular organization of synapses in neurons (). In contrast to most other membrane-associated guanylate kinase homologues, the synapse-associated protein SAP97 is also expressed in nonneuronal tissues, including cardiac myocytes (). Indeed, we have found that SAP97 localized to the zones of contact between myocytes and SGNs (Fig. S4, available at ). We also examined the localization of SAP97 in intact ventricular muscle from the mouse heart. Tissue sections were stained for synapsin I and SAP97 and examined by two-photon microcopy. Consistent with our culture results, we observed the accumulation of SAP97 surrounding contacting axons (Fig. S5). The intensity of immunostaining for SAP97 associated with SGN was increased after 90 min of stimulation with nicotine in cocultures in which βAR was overexpressed in cardiac myocytes (Fig. S4). We conducted a blind analysis of the size and intensity of SAP97-positive spots (puncta). Cocultures that were stimulated by nicotine exhibited an increase in the area and intensity of SAP97-positive puncta when compared with the controls (Fig. S4, C and D). An increase in the size of SAP97-positive puncta was also observed in wild-type cardiac myocytes not expressing exogenous βAR, but the increase was not statistically significant. SAP97 was observed at the synaptic sites in βAR KO and β/βAR KO myocytes, but we did not detect an increase in the intensity of immunostaining for SAP97 after nicotine stimulation. The results suggest that βAR signaling can modulate SAP97 accumulation at synapses. Two-photon imaging revealed a partial overlap in immunostaining for βAR and SAP97 in cardiac myocytes (). Thus, the proteins may function within the same signaling compartment but may not directly interact. The postsynaptic density in neuronal synapses and postsynaptic membranes in neuromuscular junctions are known to contain accumulations of receptors, channels, and anchoring and signaling molecules. This colocalization is thought to provide a fast and efficient response to released neurotransmitter as well as facilitate retrograde signaling from targeting cells to the contacting neuron. PKA and phosphatase 2B anchored by AKAP79 are known to be associated with postsynaptic density in neurons (; ), and PKA and D-AKAP are also found in and around the neuromuscular junction (). By analogy with this, we have found the accumulation of AKAP79/150 (mouse orthologue of AKAP79) to the zones of contact between cardiac myocytes and SGNs (). When we expressed HA-tagged βAR in cardiac myocytes in a coculture and immunostained the cocultures for AKAP79/150 and HA, we observed both proteins at the zones of contact. Although there may be some overlap, the pattern of staining appears to be different (), suggesting that these proteins may not interact directly. Nevertheless, their proximity in the sympathetic synapse suggests that AKAP79/150 could play a role in βAR signaling. In contrast, two other PDZ scaffold proteins known to localize to cell–cell contacts, MAGI-3 and MUPP1, were not detected at sites of myocyte–SGN contact (unpublished data). sub #text SGNs were isolated from the cervical ganglia of newborn mouse pups by treating ganglia with collagenase type 1A-S (Sigma-Aldrich) and trypsin T XI (Sigma-Aldrich) followed by trituration. Neurons were plated on coverslips coated with laminin (Sigma-Aldrich) for immunocytometry as described previously (). Spontaneously beating neonatal cardiac myocytes were prepared from hearts of newborn mouse pups (from wild-type, βAR KO, βAR KO, and β/βAR KO mice) as described previously () and were added to already plated SGNs on the same day. After coculture for 24 h, cultures were treated with 1 μM cytosine arabinoside (Sigma-Aldrich) for 24 h to inhibit fibroblast growth. Cocultures were maintained in Leibovitz's L-15 medium supplemented with Nu serum (BD Biosciences), NGF (Invitrogen), and ITS liquid media supplement (Sigma-Aldrich). After cytosine arabinoside treatment, media were changed every 3 d as previously described (). Recombinant adenoviruses encoding HA-βAR and FLAG-βAR were prepared as previously described (). Cocultures of cardiac myocytes and SGNs were infected with viruses at a multiplicity of infection of 100 at a desired time point (between 1–10 d in culture) and were analyzed 24–72 h later. SGNs are resistant to adenovirus infection, so primarily cardiac myocytes were infected. Cocultures were fixed by adding PBS (Mediatech, Inc.) containing 8% PFA directly to the culturing media to achieve a final PFA concentration of 4%. Cells were permeabilized with 1% BSA solution in PBS containing 0.2% Triton X-100. Cells were then stained with the desired antibody. The antibodies used were as follows: anti-FLAG M1 antibody (mouse monoclonal IgG2b; 1:600; Sigma-Aldrich), anti-HA 16b12 antibody (mouse monoclonal IgG1; 1:600; Covance; and rabbit polyclonal; 1:1,000; Berkeley Antibody Company), antisynapsin I (rabbit polyclonal; 1:1,000; Chemicon International), anti-SAP97 (mouse monoclonal; 1:250; StressGen Biotechnologies), anti–β-catenin (mouse monoclonal; 1:300; Transduction Laboratories), antipancadherin (mouse monoclonal; 1:500; Sigma-Aldrich), and antityrosine hydroxylase (rabbit polyclonal; 1:1,000; Chemicon International; and mouse monoclonal; 1:800; Transduction Laboratories). The rabbit polyclonal antibody to the luminal domain of synaptotagmin I was a gift from P. Scheiffele (Columbia University, New York, NY). The primary antibodies were detected with AlexaFluor594-conjugated goat anti–mouse IgG (1:1,000; Invitrogen) and AlexaFluor488 goat anti–rabbit IgG (1:1,000; Invitrogen). The slices for imaging were mounted with Vectashield mounting media (Vector Laboratories). The images were acquired at room temperature on an imaging microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) using a plan-Apochromat 63X 1.40 NA oil lens (Carl Zeiss MicroImaging, Inc.), a camera (RTE/CCD-1300-Y/HS; Roper Scientific), and IPLab software (BD Biosciences). Confocal images were acquired using a confocal laser-scanning microscope (LSM510; Carl Zeiss MicroImaging, Inc.) equipped with a tunable Ti-Sapphire laser (Mira 900; Coherent) and a plan-Apo 63X 1.4 NA oil lens, and images were analyzed by Volocity software (Improvision). Measurement of the spontaneous contraction rate was performed as described previously () with some modifications. In brief, 2–3 × 10 cardiac cells were cultured on the coverslips coated with laminin (Sigma-Aldrich) and were placed in 35-mm petri dishes (Corning) to obtain a uniformly beating syncytium. SGNs were placed on the same coverslip or on a separate coverslip. On day 10–14, the culture dishes were placed in a temperature regulation apparatus positioned on the stage of an inverted microscope (TMS; Nikon) connected to a video camera (C2400-7; Hamamatsu). Cells were equilibrated at 37°C for 10 min before monitoring the contraction rate. Contraction rates of cells within the syncytium were determined at 60-s intervals for 10 min before and 20 min after the stimulation of SGNs by 1 μM (−)nicotine hydrogen tartrate salt (Sigma-Aldrich). The data were analyzed using Prizm software (Prizm Software). In experiments measuring the beat rate of myocytes cocultured with SGNs, 1 μM nicotine ([−]nicotine hydrogen tartrate salt; Sigma-Aldrich) was used to avoid activation of the myocytes by outflow of the released transmitter. In all other experiments, 500 μM nicotine was applied for 5 min followed by washout with prewarmed media. The Volocity program (Improvision) was used to quantify βAR removal from the sites of contact between myocytes with SGNs. Mean fluorescence intensity for the red channel representing immunostaining for FLAG-tagged βAR was measured in areas where the green channel fluorescence (immunostaining for synapsin I) was selected as a criterion. Axonal areas where residual fluorescence for green channel was present were included. Next, we excluded the area selected and measured the mean fluorescence for the red channel at the extrasynaptic regions. Measurements were performed in arbitrary units of a direct scale. Data were collected from 10 images obtained in three experiments. To compare the fluorescence intensity of βAR staining in the areas of synapsin I accumulation with areas along axons between synapsin I puncta, we used the Wizard tool of the Volocity program to select regions of interest. The data were obtained from three images using six different pairs of regions of interest on each image. Measurements were performed in arbitrary units of the direct scale. Statistic comparisons were performed with a test. Statistical significance was set as P < 0.05. We used the histograms of fluorescence intensity of the original LSM510 files for five different images to quantify the accumulation of βAR in the zones of contact with SGNs. Mean fluorescence intensity was quantified in the areas inside and outside the zones of contact along the selected straight line crossing a zone of contact (; two zones from each image). Measurements were performed in arbitrary units of the direct scale. Statistic comparisons were performed with a test. Statistical significance was set as P < 0.05. Colocalization of pancadherin immunostaining with synapsin I was quantified using the Volocity program (Improvision). Cadherin puncta were defined as discrete areas where fluorescence intensity was higher than threshold fluorescence in areas outside the sites of contacts. We first measured the total number of cadherin puncta and then measured the number of puncta that had colocalization with synapsin puncta. The percentage of colocalizing puncta was collected from seven images. Analogously, we measured the total number of synapsin puncta and the number of synapsin puncta that had overlap with cadherin puncta. All chemicals were purchased from Sigma-Aldrich unless mentioned otherwise. Cells were cultured on microscopy coverglass (Glaswarenfabrik Karl Hecht KG). After 7 d in culture, cocultures were transferred from medium to an imaging chamber containing 400 μl Tyrode (150 mM NaCl, 4 mM KCl, 2 mM CaCl, 2 mM MgCl, 10 mM Hepes, 10 mM glucose, pH 7.35; 310 osmol) supplied with 25 μg/ml NGF (Invitrogen). After mounting on a microscope (TE-200; Nikon), another 400 μl Tyrode containing 1 mM nicotine tartrate and 10 μM FM1-43 was added (load). After a 2-min incubation at room temperature, cells were perfused with the original Tyrode containing 0.1 mM ADVASEP-7 (CyDex, Inc.) at a speed of 1 ml/s for ∼7 min to wash out FM1-43 remaining on the plasma membrane. The destaining protocol (unload) during the imaging period was as follow: Tyrode for 30 s, 500 μM nicotine in Tyrode for 120 s, and Tyrode for 60 s. Mice were killed, and hearts were removed and placed in cold PBS buffer. The hearts were rinsed in buffer, cut into halves, and fixed in cold PBS containing 4% PFA for 24 h at 4°C. 50–70-μM slices were cut by a sectioning system (Series 1000; Vibratome) and incubated in blocking solution for 1 h followed by immunostaining. Imaging was performed by two-photon microscopy using 0.3-μM slices. 3D reconstruction was performed using Volocity software (Improvision). The antibody to the luminal domain of synaptotagmin (rabbit polyclonal) was a gift from P. Scheiffele. Cocultures of cardiac myocytes and SGNs were incubated with serum-free media containing 500 μM nicotine and 10 μg/ml synaptotagmin antibody for 15 min followed by washout with prewarmed complete media. Cells were then fixed and stained with mouse monoclonal to synapsin I. Fig. S1 shows uptake and release of the fluorescent dye FM1-43 by sympathetic neurons innervating cardiac myocytes. Fig. S2 shows immunostaining for β-catenin and synapsin I in the mouse heart. Fig. S3 shows the effect of a β blocker on the removal of βAR from sympathetic synapses. It demonstrates that the pretreatment of coculture with a β blocker abolishes the activity-dependent removal of βAR from synaptic sites. Fig. S4 shows the increase of SAP97 accumulation at the synaptic sites after 90 min of stimulation with nicotine in cocultures in which βAR was overexpressed in cardiac myocytes. Fig. S5 shows immunostaining for SAP97 and synapsin I in the mouse heart. Video 1 shows the rotation of the 3D-reconstructed image in E. Online supplemental material is available at .

Hyaluronan (HA) is a glycosaminoglycan that is broadly distributed in extracellular spaces (). HA is especially enriched in pericellular matrices surrounding migrating and proliferating cells during embryonic development, tissue repair, and inflammation (). The most widely distributed form of HA in normal tissues is a high molecular weight (HMW) extracellular and cell surface polysaccharide that usually consists of several million daltons (called HMW-HA). HMW-HA forms a highly viscous network that is important for molecular exclusion, flow resistance, tissue osmosis, lubrication, and hydration. Lower molecular weight forms of HA that are synthesized de novo or generated by either hyaluronidase-mediated degradation or oxidative hydrolysis of HMW-HA may accumulate at sites of inflammation (; ). The biological activities of HMW-HA and the lower molecular weight forms of HA are distinct (; ; ). HMW-HA and lower molecular weight forms of HA influence proliferation, migration, and adhesion of cells within matrices by binding to cell surface receptors such as CD44, RHAMM, LYVE-1, and layilin (; ; ). CD44, which is arguably the best studied HA receptor, is a type I transmembrane glycoprotein encoded by a single gene and expressed as multiple isoforms. The structural diversity of CD44 is generated by alternative RNA splicing as well as differential posttranslational modifications, including glycosylation and the attachment of glycosaminoglycans (; ; ). The HA-binding domain is present in all isoforms of CD44. We previously reported that HMW-HA and a lower molecular weight HA regulate cell cycle progression through G1 phase in vascular smooth muscle cells (SMCs; ). Interestingly, SMCs treated with HMW-HA were blocked in G1 phase, whereas the lower molecular form of HA ( = ∼2 × 10) stimulated progression through G1 phase (). CD44 was required for both of these effects (). Progression through G1 phase is regulated by cyclin-dependent kinases (cdks), cyclin D–cdk4 or cdk6 (hereafter called cdk4/6), and cyclin E–cdk2. Once activated, these cdks phosphorylate pocket proteins such as Rb (retinoblastoma) and p107, leading to the release of E2Fs and E2F-dependent gene transcription. E2F regulates several genes that are needed for the entry and completion of S phase as well as for further cell cycle progression (). Mitogenic stimulation activates cdk4/6 by allowing for the induction of at least one member of the cyclin D family (D1, D2, or D3), with D1 being a major D-type cyclin in mesenchymal cells such as fibroblasts and vascular SMCs. The mechanisms leading to the mitogen-dependent activation of cyclin E–cdk2 are more complex, involving decreases in the expression of p21 family cdk inhibitors (p21, p27, and p57). p21 and p27 (hereafter called p21 and p27) are the best characterized. These inhibitors bind to cyclin E–cdk2 complexes in G0 and early G1 phases to maintain them in a catalytically inactive state (). The down-regulation of p21 and p27 in late G1 phase contributes to the late G1-phase activation of cyclin E–cdk2. In addition, the p21 family cdk inhibitors promote the assembly () or stability () of cyclin D–cdk4/6 complexes without inhibiting their catalytic activity. In fact, the mitogen-dependent induction of cyclin D1 is thought to contribute to the activation of cyclin E–cdk2 indirectly by sequestering p21/p27 on cyclin D–cdk4/6 complexes and, thereby, preventing p21/p27 from inhibiting cyclin E–cdk2 (). This study was designed to identify the cell cycle regulatory mechanism responsible for the antimitogenic effect of HMW-HA in vascular SMCs, to determine the applicability of the mechanism to other mesenchymal cell types, and to assess the relevance of the mechanism during SMC proliferation in vivo. We conclude that HMW-HA binding to CD44 selectively inhibits cyclin D1 expression and p27 degradation in both vascular SMCs and fibroblasts. The effect on cyclin D1 is primary, whereas the effect on p27 is caused by an HA/cyclin D1–dependent inhibition of Skp2, the rate-limiting component of the SCF complex that degrades p27. These effects can account for the antimitogenic effect of HMW-HA on S-phase entry in vitro, are detected in vivo, and are associated with increased SMC proliferation after vascular injury in CD44-null mice. We previously reported that HMW-HA inhibits the S-phase entry of mouse vascular SMCs (). A similar antimitogenic effect was detected in early passage human vascular SMCs, as determined by BrdU labeling () and flow cytometry (). The effect was dose dependent and was largely reversed by anti-CD44 but not by an isotype-matched irrelevant antibody (). The small antimitogenic HMW-HA effect remaining in human vascular SMCs treated with neutralizing anti-CD44 may be caused by the actions of alternative HA receptors (see Introduction). Other ECM components with antimitogenic activity (e.g., the matricellular proteins thrombospondin, SPARC [secreted protein acidic rich in cysteine], and tenascin-C) disassemble focal adhesions, disrupt actin stress fibers, and can also inhibit cell spreading (; ). HMW-HA had none of these effects even under conditions in which it showed its typical inhibitory effect on S-phase entry (). We explored the mechanism underlying the CD44-dependent antimitogenic effect of HMW-HA by examining hyperphosphorylation of the Rb protein and induction of cyclin A, which are two cell cycle events that are closely linked to the completion of G1 phase and entry into S phase. HMW-HA blocked the mitogen-dependent hyperphosphorylation of Rb (shown by the gel shift; ). Rb hyperphosphorylation results in the release of activator E2Fs and E2F-dependent gene transcription, and cyclin A is one of the E2F-regulated genes. Indeed, cyclin A promoter activity was inhibited by HMW-HA (). Moreover, the expression of human papillomavirus E7, which inactivates pocket proteins and results in E2F release, rescued cyclin A promoter activity in HMW-HA–treated SMCs (), demonstrating that the inhibitory effect of HMW-HA on the cyclin A promoter is directly related to its effect on pocket protein phosphorylation. Consistent with its effect on the cyclin A promoter, HMW-HA inhibited the mitogen-dependent induction of cyclin A mRNA () and protein (). Cyclin D1–cdk4/6 and cyclin E–cdk2 are the G1-phase cdk complexes that are thought to control pocket protein phosphorylation. As discussed above (see Introduction), the rate-limiting step for the activation of these kinases is the mitogen-dependent induction of cyclin D1 (leading to active cyclin D–cdk4/6) and the depletion of cyclin/cdk2-bound p27 (leading to active cyclin E–cdk2). A time-course analysis showed that HMW-HA blocked both the mitogen-dependent induction of cyclin D1 and the mitogen-dependent down-regulation of p27 (). Cyclin D1 mRNA levels were inhibited by HMW-HA (), and the time course of this effect was similar to that seen for cyclin D1 protein (). In contrast, HMW-HA did not affect the levels of p27 mRNA (), indicating that its effect on p27 protein levels was likely posttranscriptional. These effects are selective because the mitogen-dependent changes in G1-phase p21 levels were unaffected by HMW-HA (). The levels of p27 are typically controlled by ubiquitin-mediated degradation, and a major ubiquitin ligase responsible for degrading p27 is the multimeric complex called SCF (; ; ; ). Mitogenic stimuli regulate the activity of SCF, at least in large part, by controlling the expression of Skp2 (). We studied the effect of HMW-HA on Skp2 gene expression and found that the mitogen-dependent induction of Skp2 mRNA and protein were strongly inhibited by HMW-HA (). The combined results in show that the antimitogenic effect of HMW-HA is associated with an inhibition of both cyclin D1 and Skp2 gene expression. Many studies have indicated that both of these effects would be expected to inhibit G1-phase progression and S-phase entry. Indeed, the ectopic expression of either cyclin D1 or Skp2 was sufficient to overcome the antimitogenic effect of HMW-HA on S-phase entry in vascular SMCs (). Recent studies indicate that Skp2 is an E2F target (; ; ). This finding raised the possibility that cyclin D1 is the primary target of HMW-HA and that the inhibition of Skp2 seen in SMCs treated with HMW-HA is a secondary consequence of decreased cyclin D1–cdk4/6 activity, decreased Rb phosphorylation, and decreased E2F release (see Introduction). To determine whether the effects of HMW-HA on cyclin D1 and Skp2 are related or independent, we asked whether the ectopic expression of cyclin D1 would override the effect of HMW-HA on Skp2 gene expression. We found that the ectopic expression of cyclin D1 () rescued the expression of Skp2 mRNA in HMW-HA–treated SMCs (). In contrast, the ectopic expression of Skp2 () was unable to rescue cyclin D1 gene expression in the presence of HMW-HA (). The results from these experiments indicate that the primary antimitogenic effect of HMW-HA is on the expression of cyclin D1 and that the effect of HMW-HA on Skp2 is likely a secondary consequence of inhibiting cyclin D1–cdk4/6 complex formation, Rb phosphorylation, and E2F release. HMW-HA is antimitogenic in early passage mouse vascular SMCs, as determined by both BrdU incorporation () and flow cytometry (). Moreover, HMW-HA inhibits the mitogen-dependent induction of cyclin D1, Skp2, and cyclin A mRNAs (). The down-regulation of p27 was partially inhibited by HMW-HA in wild-type mouse SMCs () without comparable change in the level of p27 mRNA (). Cyclin D1 and Skp2 proteins levels were also inhibited by HMW-HA (). All of these effects recapitulated the results we obtained in human SMCs. The use of mouse SMCs allowed us to compare the effect of HMW-HA in wild-type and CD44-null cells. We found that none of the aforementioned HMW-HA effects were recapitulated in vascular SMCs isolated from CD44-null mice (). Thus, CD44 is the major HA receptor mediating the antimitogenic effects of HMW-HA in mouse vascular SMCs. Similar CD44-dependent effects of HMW-HA were detected in early passage mouse lung fibroblasts (). Moreover, the ectopic expression of cyclin D1 overcame the inhibitory effects of HMW-HA on Skp2 mRNA, cyclin A mRNA, and S-phase entry in mouse embryonic fibroblasts (MEFs; ). Consistent with these results, HA binding to CD44 was detected in both primary mouse SMCs and fibroblasts (). Thus, the antimitogenic effects of HMW-HA and the critical role of CD44 in these effects are well conserved in mouse and human vascular SMCs as well as in mouse embryonic and lung fibroblasts. We then investigated whether CD44 regulates SMC mitogenesis in vivo. Aortae were isolated from 20-wk-old male wild-type and CD44-null mice, total RNA was isolated, and real-time quantitative PCR (QPCR) was performed to compare the levels of cyclin D1, Skp2, and cyclin A mRNAs. All three of these transcripts were increased in the aortae lacking CD44 (), with fold increases of 6 ± 2.7 for cyclin D1, 8 ± 2.1 for Skp2, and 7.5 ± 1.5 for cyclin A mRNAs (mean ± SD; represents data from two pools of four aortae for a total of eight aortae). Cyclin D1 and Skp2 mRNA levels were also increased (relative to wild-type controls) in CD44-null aortic arches and thoracic aortae from 10-wk-old female mice (Fig. S1, available at ), indicating that these effects of CD44 are independent of age and gender. Note that the mRNA levels of CD31 (platelet endothelial cell adhesion molecule; endothelial marker), CD68 (macrophage marker), and smooth muscle actin (smooth muscle marker) were similar in the wild-type and CD44-null aortae, with smooth muscle actin being the most abundant as expected (). Thus, these CD44-dependent changes in cyclin D1, Skp2, and cyclin A mRNAs are not caused by alterations in the cellular composition of the samples. We conclude that in the uninjured artery, the binding of HMW-HA to CD44 exerts a negative effect on the cell cycle that helps to maintain arterial SMCs in a quiescent state in vivo. Despite the increased expression of cyclin A mRNA, aortae isolated from CD44-null mice did not show gross evidence of SMC hyperplasia as assessed by morphometric analysis of hematoxylin and eosin– (, medial layer) or elastin (not depicted) -stained aortic sections. Thus, the increased expression of cyclin A mRNA in the CD44-null artery is apparently counterbalanced in vivo to maintain arterial homeostasis in the uninjured artery. It is well established that intimal endothelial cells release potent antimitogens for underlying medial SMCs (see Discussion), and we reasoned that the release of these intimal antimitogens might be responsible for the absence of cell proliferation in resting CD44-null arteries (). Therefore, we denuded the endothelial layer of the femoral artery in wild-type and CD44-null mice by performing fine wire injuries. Although the intact endothelium is eventually restored after injury, the transient denudation results in proliferation of the medial SMCs, which can be detected by an increased intimal/medial ratio and BrdU labeling. We asked whether the loss of CD44 increased SMC proliferation in this injury setting. Femoral arteries were isolated, fixed, and stained for elastin 14 d after arterial injury. The internal elastic lamina was restored in both wild-type and CD44-null mice (). However, neointimal thickening was clearly increased in the CD44-null mice relative to controls (). This proliferative effect was quantified by morphometric analysis, which showed an approximately twofold increase in the neointimal/medial ratio (). A similar increase was seen in the number of BrdU-positive neointimal nuclei (). In contrast, neither wild-type nor CD44-null mice showed evidence of neointimal formation () or SMC proliferation (not depicted) in sham-operated uninjured femoral arteries. Vascular SMCs synthesize HMW-HA, which is then released and deposited in the SMC matrix, where it can bind in an autocrine and paracrine fashion to the CD44 expressed on the SMC surface. We previously reported that the binding of HMW-HA to CD44 is antimitogenic for vascular SMCs (). Similarly, others have shown that HA inhibits PDGF-stimulated receptor activation and proliferation (; ) and that the ectopic expression of HA synthases (which leads to the production of HMW-HA) inhibits the proliferation of arterial SMCs (). The data presented here provide a mechanism for the antimitogenic effect of HMW-HA, show that this mechanism operates in vivo, and lead us to propose that the interaction between HA and CD44 on SMCs contributes to the maintenance of SMC quiescence in the healthy artery. First, our experiments with early passage human and mouse SMCs in culture show that HMW-HA can repress the mitogen-dependent induction of cyclin D1 and that this effect is associated with an inhibition of downstream cell cycle events such as Rb phosphorylation, Skp2 expression, p27 down-regulation, and cyclin A expression. Second, we show that the inhibitory effect of HMW-HA on cyclin D1 is causally related to its antimitogenic effect because ectopic cyclin D1 expression rescues S-phase entry in HMW-HA–treated cells. In agreement with our results using vascular SMCs in culture, our data in vivo show that the expression of cyclin D1, Skp2, and cyclin A mRNAs is increased in the aortae of CD44-null mice, a result that strongly supports the physiological relevance of the HMW-HA effects we and others () detect in vitro. The increased expression of cyclin A is typically associated with entry and progression through S phase, but resting aortae from CD44-null mice showed no gross evidence of SMC hyperplasia as compared with wild-type controls. There are several possible explanations for this result. First, increased proliferation of CD44-null SMCs could have been offset by increased turnover (). Second, the lack of proliferation in CD44-null arterial SMCs could have reflected a compensatory change that occurs during development in the absence of CD44. Third, the intimal endothelium, which is known to release potent antimitogens such as TGF-β, nitric oxide, and prostacyclin (; ; ), could have been obscuring the effect of CD44 deletion. Our ability to document the enhanced proliferation of CD44-null SMCs after arterial injury strongly supports the idea that the release of antimitogens from intact endothelium prevents the proliferation of medial SMCs in resting arteries of CD44-null mice. HMW-HA is widely distributed in ECM in vivo, and our data indicate that the presence of HMW-HA provides an active, repressive signal that cooperates with intimal antimitogens to maintain vascular SMCs in a resting state. Our results also suggest that the degradation of HMW-HA to lower molecular weight forms, which is thought to occur at sites of arterial injury and inflammation (), would eliminate the antimitogenic effects of HMW-HA and CD44. Additionally, the lower molecular forms of HA that may accumulate as a result of inflammation can exert a positive effect on G1-phase progression in and of themselves (; ; ). Together, the loss of HMW-HA and the appearance of its lower molecular weight isoforms could contribute to modulation of SMCs to the proliferative phenotype characteristic of vascular injury. Thus, we propose that the maintenance or degradation of HMW-HA can act as a rheostat, regulating the extent of growth factor– and ECM-dependent SMC proliferation. Much of our study focused on the actions of HMW-HA on vascular SMCs so that we could assess molecular effects both in cultured cells and in vivo. However, we also show that HMW-HA is a potent antimitogenic factor for both lung and embryonic fibroblasts. In both primary SMCs and fibroblasts, the antimitogenic effect of HMW-HA is mediated by CD44. Moreover, HMW-HA similarly affects cyclin D1, Skp2, and p27 in SMCs and fibroblasts, and cyclin D1 is the primary target in both cell types. The broad distribution of HMW-HA and the expression of CD44 indicates that this ligand–receptor system is likely to have an antimitogenic effect on multiple mesenchymal cell types. Primary human vascular SMCs isolated from human aortae were obtained from Cascade Biologics, Inc., maintained as described previously (), and used between passage 2 and 9. Near-confluent monolayers were G0 synchronized by incubation in 1 mg/ml of serum-free DME containing heat-inactivated fatty acid–free BSA (DME-BSA) for 48 h. 14 mg/ml of a concentrated HMW-HA solution (Healon GV; GE Healthcare) was diluted to 200 μg/ml with DME and 10% FBS. The medium containing HMW-HA was then directly added to the quiescent cells. In some experiments, HMW-HA was supplemented with 1 μg/ml Polymyxin B (Sigma-Aldrich) to control for potential lipopolysaccharide contamination. As expected from our use of patient- grade HA, the addition of Polymyxin B had no effect on any of the results. Primary mouse vascular SMCs were isolated from aortae of wild-type and CD44-null C57BL/6 mice, maintained as described previously (), and used at passages 2–5. The adhesion and morphology of the wild-type and CD44-null SMCs were similar under the conditions of our experiments. Primary lung fibroblasts were isolated from 8–12-wk-old wild-type and CD44-null C57BL/6 mice. Primary MEFs were isolated from embryonic day 12.5 C57BL/6 embryos. The explanted mouse fibroblasts were maintained in DME containing 10% FBS and used at passage 2–5. Each of the fibroblast cell types was serum starved by incubation in DME-BSA for 48 h. To control for specificity, we tested the effect of exogenous heparan sulfate and found that it did not inhibit BrdU incorporation in serum-stimulated mouse or human SMCs. For adenoviral infections, 3 × 10 human vascular SMCs or 6 × 10 MEFs in 10% FBS-DME were seeded overnight in 100-mm dishes and incubated in DME-BSA for 12 h. Adenoviruses encoding lacZ, cyclin D1 (a gift from J. Albrecht, Hennepin County Medical Center, Minneapolis, MN), or Skp2 (a gift from K. Nakayama, Kyushu University, Fukuoka, Japan) were directly added to the near-confluent cultures and incubated overnight. The medium was then replaced with fresh DME-BSA, and the cultures were incubated for an additional 24 h. The infected starved cells were washed once with serum-free DME before being directly stimulated with 10% FBS in the presence or absence of 200 μg/ml HMW-HA. The adenovirally expressed cDNAs were efficiently expressed in the serum-stimulated but not serum-starved cells. Cyclin A promoter-luciferase assays were performed after transient transfection of near-confluent SMCs (5 × 10 cells in 35-mm dishes) as described previously () using 1 μg p434/cyclin A promoter-luciferase expression vector (bases −270–164), 0.01 μg Renilla luciferase vector, and 1 μg of either empty vector or an E7 expression vector. Cyclin A promoter-driven luciferase activity and Renilla luciferase activities were measured twice for each sample, and the mean ± SD was plotted. Human or mouse vascular SMCs were seeded (3 × 10 cells per 100-mm dish), incubated overnight, serum starved, and stimulated with 10% FBS in the absence or presence of HMW-HA. The cells were then collected and lysed as described previously (). Total protein concentration was determined by Coomassie binding (Bio-Rad Laboratories). Equal amounts of protein (50–100 μg) were resolved on reducing SDS mini-gels and analyzed by immunoblotting using antibodies specific for the following proteins: cyclin D1, p21, cdk4 (all were obtained from Santa Cruz Biotechnology, Inc.), p27 (BD Transduction Laboratories), Skp2 (Zymed Laboratories), Rb (Zymed Laboratories), and COX-1 (Cayman Chemical). Rabbit polyclonal cyclin A antibody was prepared in our laboratory. The resolved proteins were detected using ECL (GE Healthcare). Autoradiograms were digitized by scanning, and figures were assembled using Photoshop (Adobe). The small numbers on the left side of each blot represent typical migration positions of molecular weight standards ( × 10). Human vascular SMCs, mouse vascular SMCs, mouse lung fibroblasts, and MEFs (70–80% confluent) were seeded on dishes containing autoclaved glass coverslips, serum starved, and stimulated with 10% FBS in the absence or presence of HMW-HA. In some experiments, 50 μg/ml of a species-specific blocking antibody to human CD44 (5F12; ) or an irrelevant isotype-matched antibody was added to the human SMCs at the time of FBS stimulation. Fixed cells were analyzed at room temperature by immunofluorescence for the incorporation of BrdU as described previously () using AlexaFluor594-conjugated donkey anti–sheep secondary antibody to visualize BrdU staining. The percentage of BrdU-positive cells was assessed by epifluorescence microscopy by counting the number of BrdU-positive nuclei relative to the total number of DAPI-stained nuclei and typically counting ∼150 nuclei in several separate fields of view per sample. Stress fibers were stained using fluorescein- phalloidin as described previously (). Focal adhesions were immunostained using antivinculin (Sigma Aldrich) and AlexaFluor594-conjugated goat anti–mouse secondary antibody. A representative result is shown for each figure. Images were acquired using a microscope (Eclipse 80i; Nikon), 20× 0.45 NA plan Fluor (ELWD; Nikon) and 40× 0.7 NA PL Fluotar (Leitz) objectives, a digital camera (C4742-95; Hamamatsu), and a camera controller. Images were converted to TIFF files using Openlab imaging software (Improvision) and were assembled using Photoshop (Adobe). Quiescent human or mouse vascular SMCs were stimulated with 10% FBS in 100-mm dishes in the absence or presence of HMW-HA. The cells were washed once with cold PBS, scraped, and collected in 0.5 ml of fresh PBS. Total RNA was then extracted from the collected cell pellet using 1 ml TRIzol reagent (Invitrogen) followed by reverse transcription using ∼100 ng of total RNA as described previously (). An aliquot (10%) of the cDNA was subjected to PCR using TaqMan Universal PCR Master Mix (Applied Biosystems; ). Real-time PCR for human cyclin D1 was performed using 900 nM of forward primer 5′-TGTTCGTGGCCTCTAAGATGAAG-3′, 900 nM of reverse primer 5′-AGGTTCCACTTGAGCTTGTTCAC-3′, and 250 nM of probe 5′-6FAM-AGCAGCTCCATTTGCAGCAGCTCCT-TAMRA-3′. Real-time PCR for mouse cyclin D1 was performed using 900 nM of forward primer 5′-TGCCATCCATGCGGAAA-3′, 900 nM of reverse primer 5′-AGCGGGAAGAACTCCTCTTC-3′, and 250 nM of probe 5′-6FAM-CTCACAGACCTCCAGCAT-MGBNFQ-3′. TaqMan gene expression assays (Applied Biosystems) were used to detect human Skp2 mRNA (assay ID Hs00261857_m1), human p27 (assay ID Hs00153277_m1), human cyclin A (assay ID Hs00171105_m1), mouse Skp2 (), mouse p27 (assay ID Mm00438167_g1), and mouse cyclin A (assay ID Mm00438064_m1). To detect both human and mouse 18S rRNA, we used TaqMan gene expression assay ID Hs99999901_s1 or 150 nM of forward primer 5′-CCTGGTTGATCCTGCCAGTAG-3′, 150 nM of reverse primer 5′-CCGTGCGTACTTAGACATGCA-3′, and 100 nM of probe VIC-TGCTTGTCTCAAAGATTA-MGB-NFQ. QPCR was performed as described previously using Taqman Universal PCR Master Mix (). Each sample was analyzed in duplicate PCR reactions, and mRNA expression was quantified using ABI PRISM 7000 SDS software (Applied Biosystems). Mean quantities and SD were calculated from duplicate PCR reactions. Near-confluent early passage primary MEFs or mouse vascular SMCs were lysed using 1.5% NP-40 in PBS containing a protease inhibitor cocktail (0.2 U/ml aprotinin and 0.2 mg/ml PMSF). Equal amounts of total protein were resolved on 7.5% SDS gels under nonreducing conditions and transferred to polyvinylidene difluoride membranes. Proteins were analyzed by immunoblotting with an antibody (KM81) that detects total mouse CD44. HA binding to CD44 was detected by incubating the polyvinylidene difluoride membranes with FITC-HA in the presence or absence of blocking anti-CD44 (KM81; ). Total CD44 levels and HA binding were visualized by ECL using HRP-conjugated secondary anti–rat IgG (Jackson ImmunoResearch Laboratories) and anti-FITC (Roche Diagnostics), respectively. The intimal layer of the left femoral artery was denuded in anesthetized wild-type and CD44-null C57BL/6 male mice (24–28 wk and 30–32 g) by insertion of a 0.36-mm angioplasty guide wire (Advanced Cardiovascular Systems) largely as described previously () except that the anesthesia was avertin, and the guide wire was kept in the femoral artery for 3 min. The right femoral artery was sham operated and used as control. The mice were maintained on a standard diet and water ad libitum. 14 d after injury, the mice were anesthetized, perfused with 0.9% NaCl (by placement of a 22-gauge needle in the left ventricle), and perfusion fixed in situ by infusion with Prefer fixative (Anatech Ltd.). In some experiments, the mice received four injections of 50 mg/kg BrdU intraperitoneally 12, 24, 48, and 72 h before perfusion. Both right and left femoral arteries were harvested and embedded in paraffin. All procedures were approved and animal husbandry was overseen by the Institutional Animal Care and Usage Committee of the University of Pennsylvania. 5-μm cross sections were made through the injured part of the isolated femoral artery, and identical cross sections were made in the sham-operated arteries. Three sections corresponding to the peak injury region of each artery were processed for elastin staining using the Accustain Elastic Stain (Sigma-Aldrich). Adjacent sections were stained with BrdU Labeling and Detection Kit II (Roche Diagnostics) according to manufacturer's instructions except that we first performed HCl-trypsin antigen retrieval generally as described previously (), incubated the sections in NHCl, and blocked them with 20 mg/ml BSA in PBS. Images of the elastin- and anti-BrdU–stained sections were captured using a microscope (Eclipse 80i; Nikon), 10× 0.3 NA plan Fluor or 20× 0.45 NA plan Fluor objectives (Nikon), and a color camera (MicroPublisher 5.0 RTV; QImaging). Each of the elastin-stained sections were subjected to a morphometric analysis with Openlab software (Improvision) to determine mean intimal and medial areas (). The total number of BrdU-positive nuclei in a peak injury section was determined by visual counting of the captured image. Samples showing the highest and lowest injury responses for each genotype were removed from the analysis, and the remaining data were plotted using Prism software (GraphPad). Statistical analysis used the one-tailed unpaired test. Fig. S1 shows the effect of CD44 on cyclin D1 and Skp2 mRNA in female mice. Online supplemental material is available at .

RNAi can unite the precision view from microscopy and the unbiased search from a genetic screen, said Ron Vale (University of California, San Francisco [UCSF], CA). The necessary tools for fly researchers are genome-wide RNAi, high throughput microscopy, and some fancy image processing. “It's a hard project for a single person, and it does take technology and a certain amount of money,” said Vale. “At the same time, whole genome screens are possible to do within individual labs. They do not need to be done within giant consortia.” Vale has previously used RNAi to knock down the activity of 26 microtubule motors in the fly S2 cell line. At the meeting, he presented the results from a genome-wide RNAi screen in S2 cells for proteins involved in spindle assembly. The numbers in the new screen were a lot higher: his group analyzed 14,400 genes and 4 million spindle images. To reduce subjectivity and the required labor, the UCSF team used automated microscopy. Automated does not mean no work, however. “It takes a lot of tweaking to get right,” said Vale. “A lot didn't work right out of the box.” Setting up the screen involved preparing RNAs, testing the screen itself, and writing image-processing programs. The double-stranded RNAs came from a new library, which was designed by the UCSF team to minimize overlap with genes other than the intended target. Assay optimization was comprehensive—“more like something industry does to really troubleshoot all of the assay design so that it works on a large scale,” said Vale. A custom MATLAB program written by collaborators Roy Wollman and Jon Scholey (University of California, Davis, CA) identified spindle images, sent them to a special database designed by the team, and pulled out key numbers on spindle dimensions and composition. A human had to come in at the end, however. For each gene, a researcher looked at the numbers and a panel of ∼200 spindle images. “Seeing all the spindles arrayed side by side made it much easier,” said Vale. Attrition of the initial hits was heavy: less than 5% of weak hits and approximately 50% of strong hits were reproducible with an RNAi made to another part of the same gene. Furthermore, “this screen, like a genetic screen, is just the beginning,” said Vale. Secondary screens were used to understand and prioritize the remaining hits. Those doing subsequent whole genome screens will have an easier time. The library created in this study is available from Open Biosystems both in the form of DNA template and dsRNA. But the software will not be as easily translated to a new task, especially for those used to the uniformity of other types of analyses such as microarray experiments. “With image-based RNAi screening there may be some overlaps, but the algorithms for mitotic spindles are not directly exportable to another screen for lysosomes or cell shape,” said Vale. “You learn tricks for one that are useful for the next, but it's still a big project.” Once the programming is done, however, a genomic screen can be executed in a matter of a few months. The phenotypes of the hits included monopoles, multipoles, long spindles, short spindles, dim microtubules, chromosome misalignment, and pole detachment. A group of RNAi hits backed up Vale's earlier claims that microtubule nucleation occurs not just at the spindle poles but also within the spindle itself. Vale sees the screen as a success, especially given that “the spindle is a very well-trodden area,” he said. “This idea that there is a vast sea of unknowns out there is probably unrealistic,” he continued. “However, one should not expect hundreds of hits or even dozens of hits. That is too much to deal with anyway. Even if you can identify just a few gems…that is a very successful outcome.” References: Vessels and ducts fill a tissue volume using tree-like structures. Building these structures requires not a complex prepattern of attractants and repellents, said Celeste Nelson and Mina Bissell (Lawrence Berkeley National Laboratory, Berkeley, CA), but something far simpler. They reported that the geometry of an existing branching structure directly determines the sites of new branching. “There is a big void in terms of understanding how these patterns form—why certain cells decide to branch out,” said Nelson. She started by micropatterning a collagen gel with an elastomeric stamp, and then embedding mammary epithelial cells in the resulting cavities. The first patterns were simple bar shapes. Prompted by EGF, which was added uniformly, the cells sprouted only from the ends of the bars. Cells also sprouted from the three termini of a Y-shaped structure, and from the convex face of a bent tubule. All these patterns mimic those seen in vivo, suggesting that cells surrounding the tubules in vivo are not needed to determine basic sprouting patterns. The various sprouting sites could be predicted by a simple model in which all tubule cells secrete a sprouting inhibitor. This inhibitor falls to local minima at the predicted (and actual) sprouting sites, where it is overcome by the uniform activity of the EGF. A gradient of the outgrowth inhibitor TGFβ1 was present in the predicted pattern. Branching was prevented by excess TGFβ1 and induced everywhere by inhibition of TGFβ1. Tubule structures started to inhibit the branching of a neighbor when they were brought to within ∼75μm—approximately the spacing seen between neighboring ducts in mouse mammary glands. This inhibition mechanism should maintain the open architecture of the mouse mammary gland. The new model is simple, but it may not be sufficient. Nelson saw that sprouting increased with increasing tubule length, whereas the simple model predicts either no increase or a decrease. Tension is one possible modifier of the inhibitor model, with tension between the cell mass and the surrounding substrate further encouraging outgrowth. Reference: The nuclear envelope (NE) retreats into the ER during mitosis and then reemerges during envelope reformation, according to findings presented by Martin Hetzer (Salk Institute, La Jolla, CA). The model isn't new but is fighting for acceptance. Those interested in the dynamics of the NE during open mitosis have held onto two models over the years: either the envelope vesiculates before division, or it is resorbed into the ER. The former theory was quickly and widely accepted when EM data revealed vesicles on the surface of chromatin before envelope reformation. The purification of vesicles with NE markers that distinguished them from ER vesicles supported the argument. The fact that vesicles would be easy to divvy up between daughter cells was a bonus. Many scientists thus believed that the NE reformed after mitosis via fusion of these vesicles. Still, some researchers worried that the vesicles in the EM images and biochemical experiments might have been created during sample preparation. And when the ER was shown to remain intact during mitosis, the NE field reconsidered its “everything vesiculates” hypothesis. Hetzer is now firmly entrenched in the resorption camp. Using a novel in vitro assembly system, he showed that NE assembly is more efficient when the membrane fraction is first preformed into an ER-like tubular network. Videos revealed tubules reaching down, tip first, to contact chromatin and then aligning lengthwise. Gaps between tubules were filled by a rapid expansion of the NE, fed by the tubular membrane. Inhibiting prototypical vesicle fusion did not interfere with NE formation. The vesicle fusion model posed one logistical problem that tubules bypass: how the volume inside two fused vesicles is flattened out into the shape of the envelope. Tubules make it easy—extra volume between the two bilayers can be squeezed back into the ER, or extra lipids can be pumped in from the ER. Reference: Intercellular communication usually relies on secreted messenger proteins or simple binding events between transmembrane proteins. But Jennifer Gillette and Jennifer Lippincott-Schwartz (NIH, Bethesda, MD) suggested that hematopoietic stem cells (HSCs) and their supporting osteoblasts might talk to each other via an intriguing intercellular transfer of membrane and associated proteins. HSCs are nurtured by osteoblasts. Gillette wanted to watch the two-cell types interacting, so she labeled some HSCs with quantum dots. “Her first reaction was that it was a bust,” said Lippincott-Schwartz. Many of the quantum dots had ended up on the cocultured osteoblasts. But there had been no mistake with the labeling. Instead, the HSCs were crawling on the outside of osteoblasts and depositing quantum dots on them. Transfer also occurred with liver cells, which are happy interacting with HSCs, but much less with other cell types. More physiological molecules were also transferred, including a lipid called lissamine that is found in signaling exosomes. The extracellular release of exosomes normally occurs when multivesicular bodies fuse with the plasma membrane. Alternatively, some viruses can induce the exosomal machinery to form exported vesicles directly from the plasma membrane. The NIH team does not yet know whether either or both of these mechanisms is operating in the HSCs. Intercellular transfer of molecules has been seen in other contexts. Notch and Delta are two transmembrane proteins that are expressed on different cells; after binding, cleavage, and endocytosis they end up together in one cell. But it is the other section of the cleaved Notch, operating by itself in the other cell, that is the active signaling participant. The other example occurs in T cells. The T cell receptor binds MHC and bound antigen and draws the whole complex into the T cell. Although this does down-regulate attachment, any downstream signaling is induced by binding rather than internalization. The NIH team is most intrigued by the possibility that, in their system, signaling may occur downstream of intercellular transfer. Consistent with this theory, the lissamine lipid ended up in signaling-competent endosomes within the osteoblasts. The downstream consequences of this are, however, unknown. The HSCs may be shedding molecules so that they can detach from the osteoblasts and differentiate; the osteoblasts may be gaining molecules that tell them to recruit and support new HSCs. Gillette has several interventions that should allow her to interrupt the intercellular transfer and thus test its function. Reference: Moving cells wobble forward, said Erin Barnhart and Julie Theriot (Stanford University, Stanford, CA). The oscillating pathway makes the cells look like a person stepping first on one leg and then the other. Cell movement is generally studied as either protrusion at the front of the cell or release of adhesions at the rear of the cell. Few if any researchers have concentrated on the links between front and back. Barnhart thinks she has seen signs of this link. She observed that fish keratocytes retracted their trailing edges on first one side and then the other. The result was an oscillation of the lengths of each side of the cell and a wavy path of the cell centroid. These cell strides are signs that the front and rear of the cell are connected by a flexible linker, said Barnhart. When the tension becomes too great for the adhesions at the rear to resist, the adhesions give way. Simultaneous release by both sides is probably unstable and subject to disruption by any noise in the system, suggests Barnhart. This leaves the cell to release first one side and then the other, in a more stable out-of-phase pattern. She is now concentrating on mathematical modeling and looking for the waddling in other cell types. Reference: Cells appear to measure and control the size of organelles, but the underlying mechanisms remain mysterious. Frank Neumann and Paul Nurse (Rockefeller University, New York, NY) reported that the size of the fission yeast nucleus can be manipulated by changing the amount of surrounding cytoplasm. The system should allow a genetic investigation of nuclear and cell size control. As early as 1901, researchers publishing in German noted a fixed ratio between the volume of the nucleoplasm and cytoplasm (the nucleocytoplasmic [N/C] ratio). Various developmental transitions also appear to take their cue from this ratio. In frog embryos, for example, a few nuclei in a large cytoplasm divide rapidly. Only once the N/C ratio increases to a critical level does the organism transition to a more measured cell division schedule. Neumann found that in fission yeast the N/C ratio was constant under a variety of conditions: throughout the cell cycle; in cells grown under very different conditions; in cell size mutants; and in cells with widely varying DNA content. When Neumann centrifuged multinucleate cells, he could pack several nuclei into a small amount of cytoplasm. These nuclei grew slowly, whereas nuclei in the same cell that were surrounded by more cytoplasmic space could grow much faster. It is not clear whether cell size controls nuclear size or vice versa. A clue to the molecular mechanism is also lacking, although nucleocytoplasmic transport might allow the two compartments to communicate information about their relative volumes. With fission yeast, Neumann should be able to address these problems using the power of genetics. Reference: Run, kill, and die—these commands from caspase-11 sound like those of an army sergeant ordering the troops to battle. This death- and inflammation-inducing caspase also encourages cell migration, said Juying Li and Junying Yuan (Harvard Medical School, Boston, MA). The caspases are apoptotic proteases that are conserved from worm to man. The worm situation is simple: caspases focus on death. But mammals are more complex; they express many more caspases and integrate them into other physiological pathways. Li's example of such integration is caspase-11. In mice, caspase-11 is induced by bacterial factors and stimulates the release of inflammatory cytokines and the activation of the death effector caspase-1. Direct substrates have been elusive, however, so Li went fishing for caspase-11 binding factors. She found Aip1, an actin-binding protein. Aip1 activates cofilin, which promotes actin destabilization. Li showed that Aip1 activity is increased by caspase-11. Both caspase-11 and Aip1 were necessary for the actin dynamics that drives cell migration. Immune cells lacking caspase-11 were slower to migrate in vitro and failed to home properly to organs in mice. How caspase-11 times its three commands—migration, inflammation, and apoptosis—is not clear. In the right order, however, this one caspase might drive immune cells to reach an infection site, synthesize antibacterial factors, and commit suicide before inflammation gets out of hand. Reference: Chromatin in meiotic cells uses Ran to tell the cell cortex where to establish polarity, said Rong Li (Stowers Institute, Kansas City, MO). Cell polarity in mitotic cells is cued mostly by external information, including contacts with the matrix or with neighboring cells. But for meiotic cells, the signal for polarization comes from their own chromosomes, said Li. The migration of chromosomes toward the cortex helps initiate the extrusion of the polar body, which contains the extra chromosome set. Li showed that beads coated with DNA added to oocytes induced ectopic spots of contractile cortex, including an actomyosin cap, which extrudes the polar body. The DNA didn't have to touch the cortex; it induced polarity from up to ∼20 μm away. Microtubules and actin were not bridging this distance, as their depolymerization did not prevent formation of the myosin cap, so Li and colleagues considered Ran. Ran is known to concentrate at meiotic chromatin, and Ran-GTP gradients help drive spindle formation and nuclear envelope assembly. Li showed that Ran also drives cortical polarity in oocytes. Interfering with the activation of Ran blocked the ability of added DNA to create cortical polarity. With its short diffusion distance (∼10–20 μm from the chromatin), Ran-GTP presumably creates only a localized contractile cap, thereby preventing too much cortex and cytoplasm from being lost during extrusion. The cap size depended on the number of DNA beads, which may tailor the size of the polar body according to the amount of DNA to be discarded. Oocyte cortical polarity is known to require MOS, an upstream kinase of the MAPK cascade in maturing oocytes. Ran seems to concentrate active MAPK around the chromatin, where it might locally activate myosin light chain kinase, which stimulates myosin cortical assembly. Meiotic chromatin was also recently noted by others to drive a gradient of another GTPase, Rac-GTP, which helps to stabilize the spindle and anchor it to the cortex. Inhibition of Rac activity caused spindle detachment and prevented polar body extrusion. Whether chromatin also induces a contractile cortex in somatic cells to help them round up before mitosis is unknown. References: Isolation makes dendritic cells (DCs) mature, according to a talk by Aimin Jiang (Yale University, New Haven, CT). The newly discovered maturation pathway may help prevent autoimmunity. Differentiated DCs cluster together in culture through homotypic interactions of E-cadherin, an adhesion protein that is more typically associated with epithelial cell junctions. Jiang, working with Yale's Ira Mellman, found that disruption of these cell clusters caused DCs to activate and develop into mature cells capable of migrating to lymph nodes and presenting antigen to T cells. Maturation was previously thought to occur only following the delivery of microbial stimuli. Loss of adhesion may lead to maturation by releasing β-catenin from junctions so that it can travel to the nucleus, where it activates differentiation-inducing transcriptional programs. Indeed, Jiang showed that overexpression or stabilization of β-catenin caused DC maturation. Unlike DCs that mature in response to pathogenic factors, those that matured via lost contact did not make inflammatory cytokines. They thus did not induce the strong immunogenic T cell responses in mice that bacteria-exposed DCs did. DCs constantly leave tissues such as the skin, lung, and intestine, even when no invaders are present. Loss of their previous contacts might activate this new maturation pathway and send the DCs to lymph nodes. Since they lack inflammatory cytokines and probably carry self-antigens, these cells are good candidates for disabling potentially autoreactive T cells. Reference: When Mike Blower started his project on mitotic spindle formation, he never would have guessed he'd end up studying RNA. Now, as head of his own lab at Harvard University (Boston, MA), Blower is elbow-deep in spindle-associated RNAs and is chasing down their functions. Mitosis was the focus of his research in Rebecca Heald's and Karsten Weis's laboratories at the University of California, Berkeley, CA. There, Blower discovered that a complex of protein and RNA was necessary for spindle assembly. One of the proteins in this complex is Rae1, which during interphase helps to export RNAs from the nucleus. Rae1 needed RNAs to stabilize microtubules and thereby build the spindle. Blower has now purified and identified many of the RNAs that are associated with microtubule asters. mRNA classes that were enriched on asters included those that encode mitotic regulators, DNA replication factors, and transcription factors needed for developmental patterning. Similar sets of RNAs were found in both frog and human cell extracts. Blower's previous work had shown that the RNAs themselves, not their translated protein products, were necessary for spindle assembly. But mRNAs also get translated locally at spindles, as they do at synapses. In his talk, Blower showed that polysomes associated with the spindle were translating the replication and mitosis regulators. Blocking translation during mitosis in cell extracts and in synchronized human cells caused subtle defects in chromosome condensation and alignment, although spindles themselves seemed to form normally. Local translation of mRNA at the spindle may be simply an efficient way to get mitotic regulators to where they are needed. It might be especially important in oocytes, in which the spindle occupies only a tiny portion of the cell volume. The function of untranslated mRNAs is a little more difficult to understand. Blower supposes that some of the RNAs, particularly those encoding developmental regulators, attach themselves to the spindle as a way to segregate evenly—or unevenly, in asymmetrically dividing cells—to the daughters. But just how the Rae1 partners help to stabilize microtubules is still unclear. Reference: Amethyltransferase silences worm X chromosomes from a distance, said Susan Strome (Indiana University, Bloomington, IN). By binding autosomes, this protein may prevent X chromosome silencing proteins from being distracted from their intended targets. In mammals, females silence one of their two X chromosomes so that their X gene expression levels match that of males. In the worm germline, however, every copy of the X chromosome—two in hermaphrodites and one in males—is silenced. Why worms work this way is unclear; compacting the male X chromosome, which has no pairing partner, into heterochromatin might be necessary for it to segregate intact into a daughter cell during division. Hermaphrodites may silence their two Xs to match the single silent X in males. Failure to silence the X in germlines makes worms sterile, as revealed by a genetic screen for maternal effect sterile mutants done previously in Strome's lab. Several of the proteins that were identified in the screen—MES-2, MES-3, and MES-6—form a complex that binds preferentially to the X chromosome, methylates lysine 27 of histone H3, and thereby silences this chromosome. In her talk, Strome discussed another sterility protein, MES-4, which her group recently found methylates lysine 36 of histone H3 on the autosomes of germline cells. MES-4 avoids nearly all of the X chromosome (due to the presence of MES-2/3/6), yet Strome found it is nonetheless needed to silence X chromosome loci. Microarray analysis showed that loss of MES-4 desilenced ∼60 X-linked genes in the germline. Very few autosomal genes were affected. Strome's next task will be to determine how MES-4 works from a distance. MES-4 might help the nucleus “gain specificity by preventing promiscuity,” as van Leeuwen and Gottschling proposed for other histone modifiers ( 2002. 14:756–762). The presence of MES-4 or its methyl mark on autosomes might repel a direct repressor. If the repressor is present in limiting amounts, it would be titrated away by the more abundant autosomes in the absence of MES-4 and thus fail to silence the X chromosomes. This limiting repressor is probably not the MES-2/3/6 complex, as its methylation patterns are unaffected by the loss of MES-4. Reference: Increased tissue stiffness can pull premalignant cells into invasion and tumorigenesis, said Kandice Johnson and Valerie Weaver (University of California, San Francisco). Amongst biologists seeking tumor promoters, mechanical forces have attracted less attention than individual gene products. Weaver has for some time wanted to correct this imbalance. At her former lab at the University of Pennsylvania, she was surrounded by bioengineers interested in mechanics and physical principles. “I was hearing this stuff day in and day out,” she said. “After a while, you start to think differently.” Weaver's first focus was tumors. Stiffer tumor lesions have been associated with poorer prognoses. In vitro, tension leads to various changes in two-dimensional cultures, but only recently has Weaver's group tested the effect of tension on three-dimensional cultures of mammary epithelial cells. They found that adding more collagen to increase matrix rigidity destroyed tissue organization: lumen formation was inhibited; and cell division increased. Matrix stiffness destabilized the cell–cell linkages of adherens junctions but promoted the cell–matrix links of focal adhesions and their associated pro-division signaling. In normal tissues, such forces may guide cell growth during development and direct cells into stiff wound tissue. To switch the focus to whole animals, Weaver lab members Johnson and Laura Kass used mice overexpressing the Her2/neu oncogene. They found that stiffness of the tumor and surrounding stroma increased during tumorigenesis and was higher than normal even in premalignant tissue. The collagen fibers became linearized and taut, suggesting that the organization and tension of fibers may be as important as their quantity. Tension can also be increased by cross-linking. Fibroblasts expressing the cross-linking protein lysyl oxidase promoted the growth and invasiveness of ras-expressing premalignant cells. Weaver's group now has preliminary results that lysyl oxidase production may be turned on in certain tumor cells and inhibiting it may restrict tumor formation in an animal model. Reference: For man and yeast alike, the stench of too much perfume can kill a dating opportunity. Matthieu Peil and Andrew Murray (Harvard University, Cambridge, MA) presented evidence that the Bar1 protease keeps the attractant α factor to a level that a partner yeast cell can interpret. Only in this narrow concentration range can the prospective partner polarize correctly and choose between equally attractive mates. This yeast strategy differs from that used by chemotactic cells, which can sense signals of widely varying intensity by adjusting the affinity and response of their receptors. These cells are seeking, and moving toward, targets that are far away. By contrast, mating yeast, neuronal growth cones, and pollen tubes are seeking to contact or fuse with cells that they can reach without moving their cell bodies. The proteolysis solution used by yeast may be common to several such systems, suggested Piel and Murray, as these systems share evidence of both protease involvement and narrow detection ranges. In budding yeast mating, Bar1 is the protease of interest. It is part of a simple attraction system: α cells make α factor to attract cells, their mating partners. The cells need Bar1 to degrade α factor because α factor concentrations rise with α cell number and concentration. Piel and Murray used flow chambers to show that yeast lacking Bar1 respond effectively to only a narrow range of concentrations of α factor. In mating mixes, adding more α cells reduced mating efficiency unless the cells could make Bar1. High levels of α factor induced production of Bar1, which reduced the concentration of surface α factor to a moderate level (as measured by a transcriptional readout). Cells lacking Bar1 performed better if protease was supplied exogenously, but this did not compensate entirely. Only the cells with endogenous Bar1 could efficiently pick between two identical partners. Modeling suggests that Bar1 targeted to the polarized cell surface helps distinguish between a single concentrated signal and two overlapping signals. References:

xref #text Until very recently, virtually nothing was known of the mechanisms by which proteins in the intermembrane space (IMS) of mitochondria are folded nor were any chaperones identified in this compartment. However, studies on the structure or biochemistry of several proteins in the IMS revealed the presence of disulfide bonds in IMS components, which were initially considered to be artifacts of aerobic oxidation during protein purification (). IMS proteins for which oxidized cysteine residues have been reported are listed in . In all of these proteins, the cysteine residues are highly conserved and, for those tested so far, are essential for functionality. Recently, machinery was identified in the IMS that catalyzes protein oxidation and presumably is responsible for all of the disulfide bonds present in the IMS. It consists of two known components: a flavin adenine dinucleotide (FAD)–containing sulfhydryl oxidase named Erv1 (essential for respiration and vegetative growth) and a redox-activated import receptor named Mia40 (). These proteins constitute a disulfide relay system that is designed to drive the translocation of cysteine-containing proteins from the cytosol into the IMS of mitochondria. Erv1 is a sulfhydryl oxidase of the IMS. The primary sequence of Erv1 does not show any recognizable similarity to that of DsbA–DsbB and Ero1, the sulfhydryl oxidases of the bacterial periplasm and ER, respectively (, ). Thus, these enzymes are either unrelated or are very distant relatives. However, sulfhydryl oxidases that share the FAD- binding domain of Erv1 are present in the ER of fungi (named Erv2 proteins) and of plants and animals (named quiescin/sulfhydryl oxidases; ). Moreover, some viruses contain Erv1-like sulfhydryl oxidases that catalyze the oxidation of capsid proteins in the cytosol of infected host cells (). Erv1 consists of two structural segments. The N-terminal segment, which in consists of 72 amino acid residues, contains an invariant CxxC motif but otherwise is hardly conserved. This segment is rich in glycine and proline residues and presumably represents a flexible, unstructured region that functions as a lever arm to bring the redox-active CxxC motif into the proximity of substrate proteins (). The C-terminal segment forms an FAD-binding domain that in consists of 117 amino acid residues. This domain is well conserved among Erv1-like sulfhydryl oxidases and also contains a redox-active CxxC motif (; ; ). Recent achievements in crystallization of the FAD-binding domains of Erv1 and Erv2 revealed a direct proximity of the isoalloxazine ring of FAD to this second CxxC motif (; ). This suggests that this CxxC is oxidized by transfer of its electrons to the FAD cofactor. In vitro, the electrons can be further passed on to molecular oxygen, resulting in the generation of peroxide. However, this reaction is slow but strongly enhanced in the presence of oxidized cytochrome , suggesting that Erv1 can transfer its electrons via cytochrome to the respiratory chain (; ). In baker's yeast, Erv1 is essential for viability, and mutations in the Erv1 protein lead to a wide variety of defects such as respiratory deficiency, an altered mitochondrial morphology, depletion of cytosolic iron-sulfur clusters, and the inability to import certain IMS proteins into mitochondria (; ; ; ; ; ). In addition, the mammalian Erv1 protein was proposed to function as a growth factor for hepatocytes because the addition of purified Erv1 can stimulate the regeneration of partially hepatectomized livers (for review see ). As a result of this observation, Erv1 is also named ALR (augmenter of liver regeneration) or hepatopoietin. The variety of defects observed in Erv1 mutants might point to a wide range of different substrate proteins of Erv1 or, alternatively, to a role for Erv1 in oxidation of a factor of general relevance. The only substrate of Erv1 identified so far is the IMS protein Mia40, which indeed is a factor of general importance, as Mia40 functions as a redox-activated import receptor for IMS proteins. Mia40 is ubiquitously present in the IMS of fungi, plants, and animals. All Mia40 homologues share a highly conserved domain of roughly 60 amino acid residues containing six invariant and essential cysteine residues (; ; ; ). In fungi but not in mammals or plants, this domain is tethered to the inner membrane by an N-terminal membrane anchor. This anchor is not critical for Mia40 activity and can be functionally replaced by unrelated sorting sequences that direct the conserved Mia40 domain to the IMS. The cysteine residues in Mia40 form a characteristic CPC-CxC-CxC pattern. In vivo, at least some of these cysteine residues are predominantly present in an oxidized state, forming intramolecular disulfide bonds (; ; ). The individual function of these cysteine residues is still not clear, but they have been suggested to constitute a redox-driven protein trap that is activated by Erv1-dependent oxidation and is used to import precursor proteins from the cytosol into the IMS (; ). Erv1 directly interacts with Mia40 via disulfide bonds, and this interaction is critical for the oxidation of Mia40. Depending on the Erv1 activity and the amount of imported protein, Mia40 cycles between oxidized and reduced states (). In vitro, reduced Mia40 can coordinate metal ions like zinc and copper, and it was suggested that the reduced state of Mia40 might be stabilized in vivo by metal binding (). Proteins of the IMS are involved in several fundamental reactions of the eukaryotic cell-like energy metabolism, the transport of metabolites, ions, and proteins, and apoptosis. All proteins of the IMS are encoded by nuclear genes and, after their synthesis on cytosolic ribosomes, need to be transported across the outer membrane of mitochondria. Some proteins of the IMS contain so-called bipartite presequences that allow import in an ATP- and membrane potential–dependent manner (for reviews see ; ). In contrast, many, if not most of the IMS proteins lack presequences or other classic mitochondrial sorting signals. Instead, these proteins contain characteristic patterns of cysteine residues that are essential for their stable accumulation in mitochondria (; ; ). All of these cysteine-containing proteins are of low molecular mass, mostly between 6 and 14 kD. This small size might allow them to diffuse rather freely across the protein-conducting channel of the protein translocase of the outer membrane (TOM) complex (). After their translocation into the IMS, they interact with Mia40, forming mixed disulfides (; ). Only the oxidized form of Mia40 is able to form these intermediates, and reduced Mia40 appears to be inactive. Upon reshuffling of the disulfide bonds from Mia40 to the imported precursor proteins, the substrate proteins are released into the IMS in an oxidized and folded state. Because folded proteins are unable to traverse the protein-conducting channel of the TOM complex, this leads to a permanent trapping of the precursors in the IMS (). The reaction is presumably completed by reoxidation of Mia40 by Erv1, which would explain why Erv1 is required for protein import. According to this model, Erv1 and Mia40 form a disulfide relay system that facilitates vectorial protein translocation across the outer membrane by use of an oxidative folding mechanism. In vivo, the process is presumably more complex and requires the role of additional factors. One of these factors might be Hot13, which influences the assembly and activity of small Tim proteins in the IMS (). Moreover, Erv1 also apparently plays a second role further downstream in the assembly of IMS proteins that is not understood (). It should be stressed that this model, which is depicted in , is still rather speculative, and many points remain to be clarified. The presented model matches the experimental observations, but alternative mechanisms by which Erv1 and Mia40 function are also possible. For example, some substrates might be directly oxidized by Erv1, and Mia40 might then function as an analogue of a protein disulfide isomerase. It will be necessary to establish in vitro assays with purified Mia40 and Erv1 to unravel the molecular function of both components in detail. The so far identified substrates of the Mia40–Erv1 relay system can be grouped into two classes that differ in their characteristic cysteine signatures. Members of the first group contain two pairs of cysteines that are spaced by three residues each; this pattern is called the twin CxC motif. Examples are the small Tim proteins, which serve as chaperones that usher hydrophobic inner membrane proteins through the hydrophilic IMS (for review see ). Mitochondria typically contain five different small Tim proteins that in fungi are called Tim8, 9, 10, 12, and 13. These proteins form hairpinlike structures in which two central antiparallel α helices are linked to each other by two parallel disulfide bonds (; ). The small Tim proteins form hexamers in which the central twin CxC motifs contact each other (; ). The intramolecular interactions between the cysteine residues play essential roles in complex formation, explaining why oxidation is vital for the assembly of these proteins (; ). Members of the second group of substrates are proteins containing twin CxC motifs. The best characterized representative is Cox17, a copper chaperone of cytochrome oxidase (). Cox17 contains six conserved cysteine residues that can undergo different intramolecular disulfide interactions, thereby influencing the affinity and capacity of Cox17 for copper ions. It was suggested that redox-regulated cycling through these different conformations drives the binding and release of copper ions (; ; ). Twin CxC motifs are present in several additional IMS proteins such as Cox19, Cox23, Mdm35, Mic14 (YDR031w), and Mic17 (YMR002w), which all require Erv1 and Mia40 for their biogenesis (). Interestingly, the twin CxC motif of these proteins mimics the cysteine motif of Mia40; the reason for this symmetry is not known. Recently, it was shown that the import of Erv1 requires the presence of Mia40 in the IMS (), suggesting that the Mia40–Erv1 relay can also be used for the import of proteins with cysteines that are not organized in twin CxC or CxC motifs. xref #text According to most of the cell biology textbooks, eukaryotic cells can be divided into two sections of different redox chemistry: the ER and, to some degree, also other secretory compartments are generally considered to favor the oxidation of thiol residues and thus to generate disulfide bonds between cysteine residues. In contrast, the cytosol, nucleus, and matrix of mitochondria are believed to counteract the formation of disulfide bonds by maintaining a high concentration of reduced glutathione and/or by the presence of thioredoxins (; for review see ). This is obviously also the case for the IMS because porin channels in the outer membrane presumably allow the free transfer of reduced glutathione. Recent studies challenged the view that disulfide bonds are limited to secretory compartments (; ). The results made it clear that the simple ratio of reduced to oxidized glutathione does not determine the fate of intracellular thiol groups. Instead, the specific nature of the respective proteins and their interactions with reducing or oxidizing enzymes decide their redox states. The proteins of the IMS might counteract their reduction by two means. First, the disulfide bonds in the IMS might be extremely stable and, thus, rather inert to glutathione reduction. The standard redox potential of Tim10 is very low, and the disulfide bridges in small Tim proteins resist even highly reducing conditions like incubation with 10 mM DTT (,; ). A second mechanism to counteract reductive unfolding might be provided by the specific arrangement of cysteine residues in IMS proteins; upon reduction, these patterns might be stabilized by the binding of metal ions like copper or zinc, which maintain the overall structure of the protein by keeping the cysteine residues in close proximity. The specific cysteine patterns providing four neighboring thiol groups are only found in IMS proteins and not in proteins of the periplasm or ER. It is conceivable that these patterns have developed specifically to promote oxidative protein folding in the presence of high concentrations of glutathione. Alternatively, it was proposed that the binding of zinc ions stabilizes nonimported precursors of small Tim proteins in the cytosol and that coordination of metal ions by the cysteine residues contributes to the import competence of IMS proteins (). s u m m a r y , t h e I M S o f m i t o c h o n d r i a c o n t a i n s a s y s t e m t h a t c a t a l y z e s t h e o x i d a t i v e f o l d i n g o f p r o t e i n s t o e f f i c i e n t l y t r a p i n c o m i n g p r e c u r s o r s . A l t h o u g h t h e p r i n c i p l e o f o x i d a t i v e p r o t e i n f o l d i n g i s c o n s e r v e d f r o m t h e p e r i p l a s m o f b a c t e r i a t o t h e E R a n d I M S o f e u k a r y o t i c c e l l s , t h e c o m p o n e n t s t h a t m e d i a t e t h e r e a c t i o n s d o n o t s h o w o b v i o u s s e q u e n c e h o m o l o g y . I n t h e f u t u r e , i t w i l l b e e x c i t i n g t o t r a c k t h e p h y l o g e n e t i c o r i g i n a n d r e l a t i o n s h i p o f t h e s e s y s t e m s t o u n d e r s t a n d h o w t h e y a r o s e d u r i n g e v o l u t i o n a n d f r o m w h e r e t h e y o r i g i n a t e d .

In mammals, DNA methylation is crucially involved in controlling gene expression, cell differentiation, silencing of transposable elements, X inactivation, imprinting, and neoplastic transformation (; ). DNA methylation patterns are established and maintained by three major DNA methyltransferases (DNMTs): DNMT1, DNMT3A, and DNMT3B. DNMT1 is the only mammalian DNMT that has a preference for hemimethylated CpG sites (; ) and localizes at both replication foci and repair sites because of its interaction with the proliferating cell nuclear antigen (PCNA; ; ; ; ; ). As a result of these observations and data from genetic manipulations in the mouse, DNMT1 is thought to be the major enzyme responsible for postreplicative maintenance of DNA methylation. Homozygous null deletions of mouse are lethal early in development and result in an 80% reduction of global genomic methylation in embryonic stem cells and embryos (). embryonic stem cells have reduced differentiation potential both in vivo and in vitro (; ). Later work using conditional deletion demonstrated that Dnmt1 is indispensable for the survival of differentiated cells (). An equivalent role of DNMT1 in human cells was questioned by the homozygous deletion of exons 3–5 of the gene in HCT116 colorectal carcinoma cells (). This deletion encompasses the sequence encoding the PCNA-binding domain (PBD) of DNMT1 and leaves the next exon out of frame. Thus, it was expected that this deletion would eliminate DNMT1 maintenance activity and cause a dramatic drop in genomic methylation levels. Surprisingly, HCT116 cells bearing this deletion (referred to here as MT1 knockout [KO] cells) showed only a 20% reduction of global genomic methylation levels and nearly no loss of methylation at CpG islands. The issue was further complicated by studies in which DNMT1 levels were knocked down by RNAi in human tumor cell lines (; ; ). In these studies, a drastic decrease of methylation at CpG islands was observed, including a study using HCT116 cells (; ; ). At the same time, both the transient and stable knockdown of DNMT1 in HCT116 cells seemed to have only a minor effect on methylation levels similar to those observed in MT1KO cells (). Simultaneous KO of and in HCT116 cells (double KO [DKO] cells) resulted in a dramatic reduction of genomic methylation levels, suggesting a cooperative effect of DNMT1 and DNMT3B on the maintenance of DNA methylation (). Interestingly, the combination of hypomorphic and null alleles in the mouse showed that animals expressing ∼20% of Dnmt1 wild-type (wt) levels are phenotypically inconspicuous and have normal levels of DNA methylation, whereas mice expressing ∼10% of Dnmt1 wt levels show severe hypomethylation, are runted, and develop aggressive T cell lymphomas (). Thus, there seems to be a threshold to the amount of Dnmt1 necessary for the maintenance of genomic methylation levels. This threshold amount of 10–20% roughly corresponds to the knockdown levels routinely achieved by RNAi and is hard to detect. Moreover, KO strategies may partially be frustrated by alternative splicing yielding biologically active proteins, albeit at low levels. Indeed, the first attempt to knock out Dnmt1 in mice eliminated only part of exon 4 and lead to a partial loss of function as a result of alternative splicing and weak expression of a truncated form of Dnmt1 (, ; ). We carefully revisited expression in MT1KO and DKO cell lines at the RNA and protein levels. Using RT-PCR and a newly developed antibody, we found that alternative splicing occurs in MT1KO and DKO cell lines that bypasses the KO cassette and allows the expression of a DNMT1 variant lacking the PBD. We show that this truncated variant is enzymatically active in vitro and in vivo and that its levels are crucial for the maintenance of global genomic methylation and cell survival. We first checked for the presence of DNMT1 transcripts in MT1KO and DKO cells. Northern blot analysis showed that MT1KO cells expressed low levels of an mRNA species with a slightly lower molecular weight than full-length DNMT1 mRNA (). Consistently, reverse transcription followed by PCR amplification revealed the presence of DNMT1 mRNA species in parental HCT116 cells and KO derivatives (). A primer pair spanning exons 32–35 (corresponding to sequence coding for the catalytic domain) yielded different amounts of a specific PCR fragment in parental, MT1KO, and DKO cells. A second primer pair spanning exons 1–15 and including the region targeted for deletion (exons 3–5; ) produced fragments with different sizes in parental and MT1KO cells, whereas no product was detectable in DKO cells. However, reamplification of these PCR reactions with a nested set of primers located in exons 2 and 10 produced the same two fragments for MT1KO and DKO cells. Direct sequencing revealed that these fragments represent alternative splicing events, specifically from the precise joining of exon 2 with either exon 7 (smaller fragment) or exon 6 (larger fragment; ). Splicing from exons 2–7 does not alter the reading frame and would result in a DNMT1 protein with an internal deletion spanning part of the DMAP1 interaction domain () and PBD. In contrast, exons 2 and 6 are not in frame, and their joining results in a reading frame terminating after 30 nt in exon 6 (). The same nested PCR approach with RNA from parental HCT116 cells produced three fragments. Direct sequencing showed that the smallest of these fragments corresponds to the transcript encoding the major DNMT1 somatic isoform, that the medium-sized fragment corresponds to the DNMT1b transcript isoform, which includes an additional 48-bp exon between exons 4 and 5 (exon 4a; ; ; ), and that the largest fragment represents heteroduplex molecules of these two isoforms generated during the PCR reaction. These results indicate that the alternative splicing is caused by genomic alterations of the KO allele. To verify whether the detected alternatively spliced mRNAs are actually translated in vivo, we subjected whole cell extracts from parental, MT1KO, and DKO cell lines to Western blotting with a new monoclonal antibody against DNMT1 (). A major band of ∼180 kD was detected in parental HCT116 cells, corresponding to the expected size of full-length DNMT1. In contrast, a single band with an approximate molecular mass of 160 kD was detected in MT1KO and DKO cells. This size fits the predicted molecular weight of the exons 2–7 splicing isoform. Quantification after normalization with an antibody to lamin B showed that the relative abundance of the variant DNMT1 protein expressed in MT1KO and DKO cells with respect to the wt DNMT1 in parental HCT116 cells is at most 17 and 11%, respectively (). We conclude that MT1KO and DKO cell lines express decreased amounts of a mutant DNMT1 protein that originates from an alternative splicing event bypassing the deletion of exons 3–5. This mutant, which we hereafter refer to as DNMT1, lacks part of the DMAP1 interaction domain and PBD. Next, we tested whether DNMT1 is a functional methyltransferase. To this aim, we first transfected HEK293T cells with expression constructs for either human wt DNMT1, human DNMT1, mouse Dnmt1 lacking the PBD, or mouse Dnmt1 all fused to GFP. In the latter construct, the cysteine responsible for the transient formation of a covalent bond with the C5 position of the cytosine ring during the methylation reaction is replaced by tryptophan, resulting in a catalytically inactive enzyme (). Extracts were made from the transfected cells, and the respective GFP fusion proteins were immunopurified and tested for their methyltransferase activity in vitro. Except for GFP-Dnmt1, which, as expected, had no substantial catalytic activity, all GFP-fused enzymes showed very similar levels of methyltransferase activity (). This result shows that DNMT1 is catalytically active in vitro. Importantly, GFP-DNMT1 that was transiently expressed in human cells localized to the nucleus (), which is consistent with the identification of multiple functional NLSs in mouse Dnmt1 (). This opened the possibility that the residual DNMT1 could contribute to maintenance of the relatively high genomic methylation levels in MT1KO cells. As the deletion in DNMT1 eliminates the PBD, we checked whether GFP-DNMT1 is still capable of interacting with PCNA. Endogenous PCNA coimmunoprecipitated with GFP-DNMT1 expressed in HEK293T cells but not with GFP-DNMT1 (). To test whether DNMT1 is catalytically functional in vivo, we used a recently developed trapping assay (). HeLa cells were cotransfected with expression constructs for RFP-PCNA and either GFP-DNMT1 or GFP-DNMT1. 24 h after transfection, cells were incubated in the presence of 5-aza-deoxycytidine (5-aza-dC). During replication, this cytosine analogue is incorporated into newly synthesized DNA. When DNMT1 engages in the methylation of 5-aza-dC, a covalent complex is formed that cannot be resolved. As a consequence, DNMT1 is immobilized (trapped) at these sites. The trapping rate can be measured by FRAP as a time-dependent decrease of the mobile fraction of GFP-DNMT1 fusions and reflects their enzymatic activity in vivo. In contrast to GFP-DNMT1, GFP-DNMT1 did not accumulate at replication foci during early to mid S phase (). This is consistent with our coimmunoprecipitation results and data obtained with a mouse Dnmt1 mutant lacking the PBD (). However, upon the addition of 5-aza-dC, a time-dependent focal accumulation of GFP- DNMT1 was observed. FRAP analysis revealed that GFP-DNMT1 is trapped with only an approximately twofold lower efficiency than GFP-DNMT1 (). These data strongly suggest that the DNMT1 enzyme expressed in MT1KO and DKO cells is fully catalytically active, localizes to the nucleus, and is capable of postreplicative methylation maintenance despite the loss of interaction with PCNA. We then sought to establish the contribution of DNMT1 to the maintenance of genomic methylation levels in MT1KO cells. To this aim, we knocked down DNMT1 by RNAi for a prolonged period of time and analyzed global genomic methylation levels (). At each collection point, a relatively modest but substantial decrease in DNMT1 protein levels was achieved (). Progressively fewer cells were found in cultures treated with DNMT1 siRNA with respect to cultures treated with control siRNA (). Microscopic inspection of cultures after 8 and 12 d of treatment with DNMT1 siRNA revealed large numbers of dead cells and cells with very long and thin cytoplasmic protrusions, whereas cells treated with control siRNA exhibited a normal morphology and death rate (). Interestingly, both of these phenotypes are reminiscent of the highly hypomethylated DKO cells. Global genomic methylation was assayed by restriction with the endonuclease McrBC, which selectively digests methylated sequences, and by HPLC analysis (). Progressively lower genomic methylation was detected from days 4 to 12 of treatment with DNMT1 siRNA, with cells treated for 8 and 12 d retaining only ∼10% of the methylated cytosines present in parental HCT116 cells, which is similar to the level found in DKO cells. Interestingly, the levels of DNMT3B remained unaffected until day 8 of RNAi treatment when global genomic methylation was already drastically decreased (). A slight reduction of DNMT3B levels was observed only after 12 d of RNAi treatment and is likely caused by secondary effects. Thus, a prolonged reduction of DNMT1 levels, although moderate, caused a drastic decrease of genomic methylation in the presence of normal levels of DNMT3B. These results indicate that the residual level of the DNMT1 mutant present in MT1KO cells provides most of the methyltransferase activity responsible for maintaining relatively high methylation levels in this cell line. Interestingly, DKO cells express a similarly reduced amount of DNMT1 (), which could explain their very low level of DNA methylation. Thus, the situation revealed here for the human MT1KO and DKO cell lines is reminiscent of transgenic mice bearing hypomorphic and null alleles. Although homozygotes for the hypomorphic allele were phenotypically normal and showed nearly no molecular alterations, the combination of a hypomorphic allele with a null allele resulted in severe hypomethylation, growth defects, and cancer (). Analogously, the contrasting results previously obtained with DNMT1 knockdown experiments in human cell lines are likely caused by the varying efficiency of RNAi that may or may not lower DNMT1 below the threshold level that is sufficient for the maintenance of normal methylation levels. Also, the reduced proliferation rate and viability of severely hypomethylated cells may lead to an enrichment of cells with less efficient knockdown of DNMT1, which may explain the variable results obtained in different RNAi studies. The drastically decreased viability of MT1KO cells that we observed upon the knockdown of DNMT1 is also consistent with the rapid cell death of mouse fibroblasts after the conditional KO of Dnmt1 (). While this manuscript was in preparation, a residual DNMT1 activity crucial for cell survival was independently identified in MT1KO cells (). These results indicate that DNMT1 plays a similar and prominent role in the maintenance of DNA methylation in mouse and human cells and that the dependence on DNMT1 for the survival of differentiated cells is similar in these species and likely in all mammals. We showed that the DNMT1 mutant expressed in MT1KO cells is enzymatically active and displays only a twofold reduced postreplicative methylation rate in vivo despite lacking the domain responsible for interaction with PCNA. We also showed by RNAi-mediated knockdown experiments that the expression level of this truncated DNMT1 is critical for the maintenance of genomic methylation. Collectively, these results demonstrate that the interaction of DNMT1 with the replication machinery is not strictly necessary for the maintenance of DNA methylation but improves its efficiency. As most cell types express an excess of DNMT1, this improved efficiency may not be critical for cell survival but may contribute to the faithful maintenance of epigenetic information and stable gene expression patterns in differentiated cells and developing organisms. HeLa, HEK293T, and HCT116 cells and their derivatives were maintained in DME supplemented with 10% FBS, 2 mM -glutamine, and 50 μg/ml gentamycine. All HCT116 cell lines were supplied by B. Vogelstein and K. Schuebel (Johns Hopkins University, Baltimore, MD). Cells were transfected with Transfectin (Bio-Rad Laboratories) according to the manufacturer's instructions. 5 μg poly (A) RNA from the indicated cell lines were subjected to Northern blotting and probed with a 1.1-kb BamHI fragment from DNMT1 cDNA according to standard procedures. For RT-PCR, total RNA was isolated form HCT116, MT1KO, and DKO cells with TRIzol reagent (Invitrogen), and reverse transcription was performed with the First Strand cDNA Synthesis kit (GE Healthcare) by priming with random hexamer oligonucleotides. PCR amplifications were performed in a 50-μl final volume containing 1.5–2.0 U SAWADY Taq DNA polymerase (PeqLab), 1× buffer (20 mM Tris, pH 8.55, 16 mM [NH]SO, 2 mM MgCl, and 0.1% Tween 20), 0.2 μM of each primer, and 0.2 mM of each deoxynucleotide triphosphate. For primary PCR reactions, cDNA template from ∼0.75 μg of total RNA was used with either primers HMT1catF (5′-TGCAACATCCTGCTGAAGCTGG-3′; forward, exon 32) and HMT1catR (5′-GACCCGAGCTCAACCTGGTTATG-3′; reverse, exon 35) or HMT1mF (5′-GTCTGCTGAAGCCTCCGAGATG-3′; forward, exon 1) and HMT1e15R (5′-TTTGAGGTCAGGGTCGTCCAGG-3′; reverse, exon 15) as well as the following touchdown cycling profile: 94°C for 2 min; 10 cycles at 94°C for 20 s; 68–60°C descending by 2°C every two cycles for 15 s; 72°C for 30 s plus a 3-s increment at each cycle; 30 cycles at 94°C for 20 s; 58°C for 15 s; and 72°C for 1 min plus a 5-s increment at each cycle. For nested PCR reactions, 1 μl of primary reaction product was used as a template with primers HMT1e2F (5′-AAAGATTTGGAAAGAGACAGCTTAACAG-3′; forward, exon 2) and HMT1e10R (5′-TCTCCATCTTCGTCCTCGTCAG-3′; reverse, exon 10) and a cycling profile consisting of 25 cycles of 94°C for 20 s, 58°C for 15 s, and 72°C for 30 s plus a 5-s increment at each cycle. Monoclonal antibodies were raised against purified recombinant DNMT1 and 6×His-tagged PCNA. Approximately 50 μg of antigen was injected both i.p. and subcutaneously into Lou/C rats using CPG2006 (TIB MOLBIOL) as adjuvant. After an 8-wk interval, a boost was given i.p and subcutaneously 3 d before fusion. Fusion of the myeloma cell line P3X63-Ag8.653 with the rat immune spleen cells was performed according to standard procedures. DNMT3B was detected with a mouse monoclonal antibody that recognizes the conserved catalytic domain within the DNMT3 family (clone 64B1446; Imgenex). The lamin B1 (H90), β-actin (clone AC-15), and GFP antibodies were obtained from Santa Cruz Biotechnology, Inc., Sigma-Aldrich, and Roche, respectively. Expression constructs for mouse GFP-Dnmt1, GFP-Dnmt1, human GFP-DNMT1, and RFP-PCNA were described previously (; ; ). The GFP-DNMT1 expression construct was derived from the GFP-DNMT1 construct by overlap extension PCR (). Extracts were prepared from transfected HEK293T cells in lysis buffer (20 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5 mM EDTA, 2 mM PMSF, and 0.5% NP-40), diluted with lysis buffer without NP-40, and incubated with an anti-GFP antibody for 30 min at 4°C with constant mixing. Immunocomplexes were pulled down with protein A–Sepharose beads (GE Healthcare), and the beads were washed extensively with dilution buffer containing 300 mM NaCl. For coimmunoprecipitation, beads were resuspended in SDS-PAGE loading buffer. For in vitro methyltransferase assay, beads were further washed with and resuspended in assay buffer (100 mM KCl, 10 mM Tris, pH 7.6, 1 mM EDTA, and 1 mM DTT), and 30 μl methylation mix (0.1 μCi [H]S-adenosyl-methionine, 1.67 pmol/μl hemimethylated double-stranded 35-mer oligonucleotide, and 160 ng/μl BSA in assay buffer) was added. After incubation at 37°C for 2.5 h, reactions were spotted onto DE81 cellulose paper filters (Whatman), and radioactivity was measured by liquid scintillation. The DNMT trapping assay was performed essentially as described previously (). In brief, transfected cells were incubated with 30 μM 5-aza-dC (Sigma-Aldrich) for the indicated periods of time before photobleaching experiments. FRAP analysis was performed with a confocal laser-scanning microscope (TCS SP2; Leica) equipped with a 63× 1.4 NA plan-Apochromat oil immersion objective (Leica). GFP and RFP were excited with a 488-nm Ar laser and a 561-nm diode laser, respectively. Image series were recorded with a frame size of 256 × 256 pixels, a pixel size of 100 nm, and laser power set to 1–5% of transmission with a detection pinhole size of 1 Airy U. For FRAP analysis, regions of interest were photobleached with an intense Ar laser beam (laser set to maximum power at 100% transmission of all laser lines) for 0.5 s. Image series were recorded before and after bleaching at 0.27-s intervals (typically 20 prebleach and 50–100 postbleach frames). Mean fluorescence intensities of the bleached region were corrected for background and total loss of nuclear fluorescence over the time course and were normalized to the mean of the last 10 prebleach values. Equal numbers of MT1KO cells were plated and transfected the next day (day 0) and every second day with 40 nM of either DNMT1 ShortCut siRNA Mix (New England Biolabs, Inc.) or a control siRNA (AllStars Negative Control siRNA; QIAGEN) using HiPerFect transfection reagent (QIAGEN). On days 2, 4, and 8, cells were harvested, aliquots were collected for protein extracts and DNA isolation, and equal numbers were replated and simultaneously transfected. On days 6 and 10, cells were transfected without splitting. The medium was never changed in between transfections. Genomic DNA was isolated by the phenol-chloroform extraction method. For the McrBC nuclease assay, 0.5-μg aliquots were incubated with or without 10 U of the enzyme (New England Biolabs, Inc.) in buffer supplied by the manufacturer for 1 h at 37°C, and digests were separated by agarose gel electrophoresis. ImageJ software () was used to quantify the McrBC-resistant fractions from digital images. The McrBC-sensitive fraction shown in was calculated as follows: (resistant fraction − mock-resistant fraction)/mock-resistant fraction. For HPLC analysis, DNA samples were further treated with RNase A and T1, phenol extracted, dialyzed extensively against 10 mM Tris, pH 7.2, and 0.1 mM EDTA, and hydrolyzed to nucleosides as described previously () except that 5 U Antarctic phosphatase (New England Biolabs, Inc.) was used for dephosphorylation. HPLC was performed on the Alliance system (Waters) using a Nucleosil C-18 column at a flow rate of 0.5 ml/min with a linear increase of buffer A (0.1 M HNEtOAc) from 0 to 20% in buffer B (0.1 M HNEtOAc in 80% acetonitrile) in 30 min.

Cell migration is a highly regulated and coordinated process. It is comprised of several coupled steps that include polarization, protrusion, adhesion formation and turnover at the cell front, and adhesion disassembly and tail retraction at the cell rear. Many of the major regulatory pathways that control these processes are known (; ; ; ). Most converge on Rho family GTPases, which in turn activate kinases like PAK or ROCK (; ). Recent studies point to other, analogous pathways that control protrusion, adhesion dynamics, and cell polarity. Cdc42 acting on MRCK, which is a kinase that phosphorylates MLC, regulates nuclear positioning in migrating cells (). In addition, PAK localizes to the centrosome, where it plays an essential role in MTOC positioning (). MII is a common effector for all of these pathways, and thus it is implicated as a key regulator of cell migration. MII is a bipolar, contractile protein composed of two myosin heavy chains (MHCs), two regulatory myosin light chains (MLCs), and two essential MLCs. Each MHC contains an N-terminal globular motor domain that moves actin as it hydrolyzes ATP and a C-terminal tail region that binds to the other MHC (). MLC phosphorylation regulates the ATPase activity of MHC (; ). In addition to its contractile properties, MII also cross-links, and thus stabilizes, actin through its bivalent binding to actin filaments (). In fibroblasts, two major isoforms of MHC-II have been described, MHC-IIA and -IIB. It is likely that they serve different roles in the regulation of the actin cytoskeleton because of their different ATPase activities, contraction rates, and subcellular localization (; ). Both MIIA and MIIB mediate stress fiber formation (; ). MIIB contributes to cell migration by controlling protrusion stability (), and MIIA is implicated in the regulation of actin retrograde flow (). Although these reports point to the participation of MII and its isoforms in migration, the mechanisms by which it controls and integrates its component processes are unclear. In this report, we reveal the integrative role of MII in migration and parse its isoform-dependent and contraction-independent activities. From these studies, MII emerges as a central, regulatory molecule that serves to integrate and coordinate diverse migration-related phenomena that comprise migration. Previous observations have shown the differential cellular localization of MII isoforms. In general, MIIA is present in regions distal to MIIB, and MII is largely absent from the lamellipodium of epithelial cells (; ). We have confirmed these observations using migrating CHO.K1 cells and reveal novel details (Fig. S1, available at ), as follows: (a) the two isoforms often decorate the same actin filaments in a stippled manner, suggesting that some functions might result from additive activities; (b) MIIA and MIIB likely mediate distinct functions because the two isoforms also occupy distinct areas, and therefore do not readily form cofilaments; and (c) MII resides well away from nascent adhesions; therefore, any effect on adhesion dynamics would result from an indirect rather than a local effect. To determine whether the spatial segregation of MIIA and MIIB results in different roles during cell migration, we generated knockdown vectors that inhibit MIIA and MIIB expression with high specificity (). For both isoforms, down-regulation was comparable and maximal 96 h after transfection, where it averaged 75–95% by immunoblotting, depending on transfection efficiency (). Immunofluorescence revealed >95% knockdown in individual cells (). When plated using migration-promoting conditions (see Materials and methods), MIIA-deficient cells exhibited broader lamellipodia than control cells and did not retract their trailing edge ( and Video 1, available at ). This resulted in cells with extended tails (, arrowheads). This phenotype is reminiscent of the effect of Rho-kinase inhibitors in macrophages () and overexpression of a paxillin mutant with the LD4 domain deleted (). In contrast, MIIB-deficient cells were round and occupied a large area without distinguishable leading and trailing edges (; and Videos 2 and 4, available at ), e.g., front–back polarity. Depletion of MIIA or MIIB in Rat2 fibroblasts yielded similar results (Fig. S2). This phenotype differs somewhat from that reported for MEFs from MIIB knockout mice (). Although both showed inhibited migration, the MIIB MEFs also showed long, unstable protrusions. This could arise either from incomplete ablation of MIIB by the RNAi knockdown or an uncharacterized adaptation. Off-target effects of the RNAi seem unlikely because our phenotypes were rescued by ectopic expression of RNAi-insensitive MIIA or MIIB, respectively ( and not depicted). In addition to the round morphology, the MIIB-deficient cells also showed a defect in nuclear, centrosomal, and Golgi anchoring. More than 95% of the nuclei in the knockdown cells rotated clockwise (∼2 h/cycle; and Video 4). The centrosome accompanied this rotation, and the Golgi apparatus was distributed around the nucleus rather than polarized, as observed in nontransfected, migrating cells ( and not depicted), suggesting a more general role of MIIB in cell polarization. Although the origin of this nuclear rotation is not known, it suggests that MIIB is part of a balanced mechanism of nuclear anchoring. The increased area of MIIB-deficient cells and the broader lamellipodia in MIIA-deficient cells pointed to alterations in protrusion. By kymography, MIIA- and MIIB-deficient cells exhibited 2–3-fold increased rates of protrusion (). In addition, the protrusion was continuous, resulting in kymograms that showed a near linear, uninterrupted slope (, bottom). In contrast, wild-type cells often showed an interrupted, stepwise pattern, as previously reported (). Interestingly, during these periodic interruptions in protrusion, the adhesions stabilized and grew as MII began to localize in the previously protrusive region (Video 5, available at ), suggesting a causal link. Finally, when the protrusions in MIIA- and MIIB-deficient cells stopped advancing, they did not retract efficiently (Videos 6 and 8), suggesting that both MII isoforms regulate retraction of the lamellipodium. Thus, both MII isoforms control the speed, stepwise pattern of extension, and retraction of protrusions. To determine whether the abnormal protrusion is accompanied by alterations in adhesion dynamics, we knocked down MIIA or MIIB in paxillin-GFP– ( and Videos 6–8, available at ) or vinculin-GFP–expressing cells (not depicted). Control cells showed numerous well-defined adhesions in protrusions, as well as some small adhesions near the leading edge (, top; and Video 7) that assembled and turned over as described previously (). In contrast, MIIA- and MIIB-deficient cells showed few discrete adhesions in the protrusions; instead, there was a nearly continuous band of adhesions very close to the leading edge (, and Videos 7 and 8). The small individual adhesions that comprise this band were readily apparent at higher magnification (, , and Video 9). These adhesions disassembled and reformed rapidly ( < 15 s) as the leading edge progressed (, C and D; and Videos 7–9). More importantly, they did not evolve into larger adhesions when lamellipodial growth halted. The defects were rescued when RNAi-insensitive mCherry (mChe)-MIIA or -MIIB were expressed in MIIA- or MIIB-deficient cells, respectively (). Because both isoforms of MII regulate adhesion dynamics at the leading edge, but only MIIA inhibits rear retraction, we investigated the effect of MIIA and MIIB knockdowns on adhesions in other cellular regions. MIIB-deficient cells exhibit central adhesions comparable to those in control cells (unpublished data). In contrast, MIIA-deficient cells showed abnormally small, but static, adhesions in the central region of the cell ( and Video 10, available at ). At the cell rear, where MIIA inhibits retraction, adhesion disassembly is greatly inhibited, e.g., the adhesions slide slowly and do not disassemble ( and Video 10), thereby showing that adhesion sliding and disassembly are coordinately regulated by MIIA. Because both MIIA and MIIB mediate contraction and actin bundling (), we used cross-rescue experiments to determine whether their functions were overlapping in the regulation of adhesion assembly and disassembly. First, mChe-MIIA was expressed in MIIB-depleted cells coexpressing paxillin-GFP. mChe-MIIA localized in regions very similar to those in unperturbed cells (not depicted), and it restored the maturation of nascent adhesions (, left). However, the polarity defects and the appearance of multiple protrusions around the cell periphery remained (not depicted). We then expressed mChe-IIB in MIIA-deficient cells. mChe-MIIB localized in the central areas of the cell, as it does in unperturbed cells. However, it did not rescue the inhibited maturation of nascent adhesions induced by the MIIA knockdown. Instead, a band of small, dynamic adhesions remained near the leading edge, as in MIIB knockdowns (, right). In contrast, mChe-MIIB rescued the effect on the central adhesions, i.e., they were larger (, top right). Thus, increased MIIB in central areas (where endogenous MIIB resides) of MIIA knockdown cells rescues the maturation of adhesions in the central regions of the cell, but not the nascent adhesions at the cells periphery. In contrast, increased MIIA in MIIB knockdown cells rescues the maturation of adhesions at the leading edge. This points to a mechanism in which the central MIIB activity requires MIIA for its translation to the periphery, perhaps by organizing the actin so that tension produced in the middle of the cell propagates into protrusions. This suggestion is supported by our observation that overexpressed MIIA in wild- type cells localizes primarily in actin bundles and produces more and larger adhesions (Fig. S3, available at ), indicating that MII activity can dial up or down adhesion assembly, depending on its expression level. To separate the bundling from the contractile functions of MII on cell migration, we produced ATPase-inhibited mutants of MIIA and MIIB fused to GFP and mChe. The ATPase activity of N93K-MIIA and R709C-MIIB are inhibited 80 and 75%, respectively, in vitro (; ). However, both mutants bind and cross-link, but do not move, actin filaments in vitro (). We used FRAP to show that the mutants exhibit increased time in the actin-bound state, as expected from their inhibited ATPase cycling. Both MIIA N93K and MIIB R709C showed decreased rates and fractional recoveries (). The fractional recovery observed for both mutants was the same, suggesting that the two isoforms bind actin similarly. Interestingly, wild-type MIIA exhibited faster and higher fractional recovery than MIIB (). This suggests that MIIA is more active than MIIB, and therefore binds actin strongly in a smaller fraction of time, as shown previously in vitro (). It also points to the use of FRAP as a method to determine MII activity in living cells. When expressed in MIIB-deficient cells, MIIB R709C did not effectively restore adhesion maturation (). However, MIIB R709C partially restored the front–back polarity and localized at the back of the cell (). Thus, maturation of adhesions at the leading edge requires MIIB activity; but its role in determining front– back polarity suggests a cross-linking contribution. In contrast, MIIA N93K localized like its wild-type counterpart (unpublished data), rather than the rearward localization of MIIB R709C, and did not rescue the increased protrusiveness observed in MIIA-deficient cells. However, it did restore leading edge retraction and the concomitant growth of adhesions in protrusions pointing to its actin-binding function in these activities (). Adhesive signaling through integrin receptors both stimulates and responds to tension through Rho GTPases, thus constituting a feedback loop connecting adhesion and contraction through MII regulation (). The phosphorylation of paxillin on Y31, Y118, and S273 and the phosphorylation of FAK on Y397 are part of this signaling mechanism and serve as markers for the activation of this pathway (; ; ). The small dynamic adhesions near the leading edge of MIIA- and MIIB-depleted cells were prominently phosphorylated on tyrosine (). They also stained positively for Y397-FAK and Y31-paxillin, defining an almost continuous band of adhesions (). The staining of these phosphomarkers decreased in the stable adhesions that reside in regions removed from the leading edge (). Thus, depletion of myosin function at the lamellipodium enhances an adhesive signaling pathway that regulates adhesion turnover and MII activity, providing a mechanistic link between myosin-generated tension in the control of adhesion maturation at the leading edge. The complex interplay between myosin-mediated contraction, protrusion, adhesion, and polarization underscores the central role of MII and its integrative properties in cell migration. Although the protrusion rate is determined by factors that regulate actin polymerization (), it is also influenced by the rate of retrograde flow, which, in turn, is regulated by MII activity and serves to counterbalance actin polymerization (; ). The retrograde flow rate is also influenced by adhesion, through a clutch-like mechanism, which links actin filaments to the substratum and can inhibit retrograde flow (; ; ). The net protrusion rate is also influenced by cycles of retraction and adhesion maturation at the leading edge. Highly motile cells protrude and move nearly continuously (; ), whereas other cells can show cycles of protrusion and retraction (). MII is also involved in a feedback loop that links adhesion, protrusion, and tension. Adhesion initiates signaling through Rho family GTPases that leads to the formation of adhesions and protrusions and generates tension. Tension also acts on adhesions to promote their maturation and the formation of actin filament bundles (). Highly motile cells tend to have small, highly dynamic adhesions that turnover rapidly, whereas the adhesions in slower moving cells stabilize and grow in response to increased tension before turning over (). Signaling components, such as phosphorylated paxillin and PAK, localize in the small, dynamic adhesions near the leading edge, where they function in a signaling pathway that inhibits adhesion maturation and promote protrusion (). Interestingly, in retracting regions, MII mediates the disassembly, rather than the assembly, of adhesions. Finally, MII polarizes and connects spatially segregated activities. Myosin acts “at a distance” in regulating protrusion and adhesion. It also contributes to the overall polarity of the migrating cell and establishes front and rear. The former is through the role of MII in orienting microtubules, Golgi, and the nucleus, and the latter is through actin bundling at the rear and sides (). GFP-MIIA and GFP-MIIB were gifts from R.S. Adelstein (National Institutes of Health, Bethesda, MD; ). Where indicated, GFP was replaced by mChe, which was obtained from R. Tsien (University of California, San Diego, La Jolla, CA; ). siRNA-insensitive MIIB was generated by site-directed mutagenesis (QuickChange kit; Stratagene) introducing a silent mutation (TCA → AGC = Ser → Ser) in the RNAi target region of human MIIB. The MIIB R709C and MIIA N93K mutants were generated by site-directed mutagenesis using the appropriate primers. The following antibodies were used: MIIA, MIIB, and GIT1 (rabbit, pAb) were purchased from Covance; paxillin (mouse, IgG1) was obtained from BD Biosciences; α-actinin (mouse, IgM) was purchased from Sigma- Aldrich; phosphotyrosine 4G10 (mouse, Ig2b) was obtained from Millipore; phosphoTyr31-paxillin (rabbit, pAb) was purchased from BioSource; and phosphoTyr397-FAK (rabbit, pAb) was obtained from CHEMICON International, Inc. Bodipy FL C5-ceramide (for Golgi detection) was obtained from Invitrogen and used as described by the manufacturer. CHO-K1 cells and Rat2 cells were cultured in standard conditions and transfected using Lipofectamine (Invitrogen). For cotransfection experiments, plasmids containing the siRNA sequences were used in 10:1 excess to GFP or mChe-containing plasmids to ensure knockdown in fluorescence-positive cells. Cells were allowed to adhere to 2 μg/ml fibronectin-coated coverslips for 60 min, fixed using 4% paraformaldehyde, and permeabilized with either 0.5% Triton X-100 for 5 min or ice-cold methanol for 10 min. Coverslips were incubated with primary antibodies and a species-appropriate secondary antibody coupled to either Alexa Fluor 488 or 568 (Invitrogen). Protrusion parameters were quantified using kymography (). Images were captured every 5 s for 3 min. Kymographs were generated using Metamorph software along 1-pixel-wide regions oriented in the protrusion direction and perpendicular to the lamellipodial edge. Confocal images for FRAP analysis were acquired using the FluoView system. Initially, a cellular area (34.72 μm) that contained GFP-MII–decorated stress fibers was scanned 3 times and bleached using 15 scans at 100% laser power. To image the FRAP, we did 15 scans every 0.1 s, 15 scans every 3 s, 14 scans every 5 s, and 2 scans every 10 s. Background subtraction and normalization were calculated, and normalized intensity versus times were fitted by a single exponential equation (R > 0.98). Fig. S1 shows the spatial localization of MIIA and MIIB in a migrating cell. Fig. S2 shows the migratory phenotypes of MIIA-depleted, MIIB-depleted, and control Rat2 fibroblasts. Fig. S3 shows that MIIA expression levels affect the number of adhesions. Video 1 (corresponding to ) is a time-lapse video of MIIA-deficient CHO.K1 cells. Video 2 (corresponding to ) is a time-lapse video of MIIB-deficient CHO.K1 cells. Video 3 (corresponding to ) is a time-lapse video of pSUPER-transfected (control) CHO.K1 cells. Video 4 (corresponding to ) is a time-lapse video of a MIIB-deficient CHO.K1 cell, highlighting nuclear spinning. Video 5 is a dual-color TIRF time-lapse video of a protrusion of a MIIB-deficient CHO.K1 cell cotransfected with mChe-MIIB (magenta) and paxillin-GFP (green). Video 6 (corresponding to ) is a TIRF time-lapse video of a protrusion of a MIIA-deficient CHO.K1 cell cotransfected with paxillin-GFP. Video 7 (corresponding to ) is a TIRF time-lapse video of a protrusion of a pSUPER-transfected control CHO.K1 cell cotransfected with paxillin-GFP. Video 8 (corresponding to ) is a TIRF time-lapse video of a protrusion of a MIIB-deficient CHO.K1 cell cotransfected with paxillin-GFP. Video 9 (corresponding to ) is a TIRF time-lapse video of a protrusion of a MIIA-deficient CHO.K1 cell cotransfected with paxillin-GFP. Video 10 (corresponding to ) is a TIRF time-lapse video of a MIIA-deficient CHO.K1 cell cotransfected with paxillin-GFP (shown).

Bacterial RecA and its yeast orthologue RAD51 are the founding members of the RecA/RAD51 protein family, which plays a crucial role in DNA repair by homologous recombination (for review see ). After DNA ends are resected at the site of a DNA break, these proteins form a nucleoprotein filament on the single-stranded DNA and catalyze a strand invasion and strand exchange reaction with a homologous region on another DNA molecule to ensure faithful DNA repair. In addition to RAD51, a few other RAD51-like proteins were found in eukaryotes, with their number increasing from three in budding yeast (Rad55, Rad57, and Dmc1) to six in most of the higher eukaryotes (RAD51B, RAD51C, RAD51D, XRCC2, XRCC3, and DMC1; ). These proteins have 20–30% amino acid sequence similarity and share common functional domains (). The core C-terminal domain contains two functionally important ATP-binding Walker A and B motifs and is linked to a small globular N-terminal domain via a linker region that is important for protein–protein interactions. Except for DMC1, a meiosis-specific protein with functions overlapping those of RAD51 (), the other five paralogues are auxiliary to RAD51. In yeast, the purified Rad55/Rad57 heterodimer functions as a mediator of Rad51, enabling it to nucleate on single-stranded DNA in the presence of replication protein A (RPA; ). Chicken DT40 cells and hamster cell lines deficient in any one of the five RAD51-like proteins are extremely sensitive to DNA damage caused by DNA cross-linking agents and the topoisomerase-I inhibitor camptothecin (). In such cells, RAD51 foci formation after DNA damage is attenuated, and the frequency of homologous recombination is reduced (). RAD51C and XRCC3 directly interact with RAD51, and RAD51C has been shown to stabilize RAD51 after DNA damage by protecting it from ubiquitin-mediated degradation (). These proteins interact with one another and form multiple protein complexes (; ). Two stable complexes have been biochemically purified. One includes RAD51B, RAD51C, RAD51D, and XRCC2 (BCDX2 complex), and the other contains RAD51C and XRCC3 (CX3 complex). The functional relevance of these complexes is supported by studies of double mutants in DT40 cells, in which two mutations in different complexes as opposed to mutations in the same complex had an additive effect on sensitivity to camptothecin (). Despite all of the aforementioned similarities and a general understanding that all RAD51 paralogues are involved in the homologous recombination process, little is known about the specific roles played by any of them. However, each of these proteins is believed to have a nonredundant function, as the loss of any of these genes cannot be compensated for by the overexpression of other family members (). Mutations in these genes have an unequal impact on drug sensitivity in DT40 cells (). Although mice that are null for any of the paralogues are embryonic lethal, they vary in the severity of the phenotype (; ; ). In , and mutants are sterile, whereas other mutants are fertile (). One family member, RAD51D, has been specifically linked to telomere maintenance (). Although all RAD51 paralogues are implicated in RAD51 foci formation, it is unclear how this function is shared between different paralogues and their complexes (). RAD51C is part of both BCDX2 and CX3 complexes and, therefore, is believed to play a central role. Being part of two different protein complexes, RAD51C may also have several distinct functions. On one hand, it is involved in the early steps of recombination associated with RAD51. On the other hand, it may also be involved in late steps of homologous recombination, as RAD51C, together with XRCC3, was purified from HeLa cells as part of a small protein complex possessing a Holliday junction (HJ) resolvase activity (). RAD51C is an essential part of this complex because the resolvase activity was lost upon depletion of RAD51C from the fraction or when protein extracts from hamster cells lacking functional were tested. However, the role of RAD51C as a resolvase remains controversial. RAD51C lacks an endonuclease domain, and efforts to demonstrate resolvase activity for the recombinant RAD51C protein have not been successful so far. Also, there is no in vivo evidence supporting this late role of RAD51C in recombination. In this study, we report the generation of a viable mouse model carrying a hypomorphic allele of , which enabled us to overcome the early embryonic lethality of a null allele and demonstrate the role of RAD51C in meiotic recombination. We demonstrate that RAD51C is involved in the recruitment of RAD51 at early stages of homologous recombination in spermatocytes. In oocytes, we describe a defect in sister chromatid cohesion at metaphase II. We speculate that RAD51C may also be required at late steps of homologous recombination. The mouse gene consists of eight exons and encodes a 366–amino acid protein (). Exons 2 and 3 code for a 142–amino acid region, including a linker and the Walker A ATPase domain. The deletion of these exons resulted in a functionally null allele, , which caused an early postimplantation lethality of mouse embryos (unpublished data). In addition to a null allele, we also generated a hypomorphic allele, , by inserting a neomycin (neo) resistance gene under the control of a phosphoglycerate kinase (PGK) promoter into intron 1 (). Disruption of normal splicing by the presence of the cassette has been reported previously for other genes (). Aberrant splicing was demonstrated for the allele by RT-PCR analysis using primers to amplify a region between the first and fourth exons (Fig. S1, available at ). The sequence analysis predicted that the misspliced transcript was likely to encode for a 76–amino acid polypeptide that included 39 amino acids from the first exon of and an additional 37 amino acids from the 3′ region of the cassette. Aberrant splicing resulted in an ∼60% reduction in RAD51C protein levels (). The combination of a hypomorphic and a null allele () produced mice expressing only 5–30% of the normal level of RAD51C protein (). The residual amount of RAD51C protein in mice is sufficient to allow normal growth and development. However, 36.6% ( = 82) of such males and 11.6% (= 173) of the females were infertile. , , and (from here on, these genotypes are referred to as controls for simplicity), only one male and one female out of 105 mating pairs were infertile. To determine the fertility status, 6–8-wk-old mice were mated with wild-type mice for up to 6 wk. Infertile animals were sexually active as indicated by the repeated detection of vaginal plugs, which did not result in pregnancy. In contrast, fertile males and females produced normal sized litters compared with the control group (7.1 ± 2.7 pups [ = 68] and 7.3 ± 2.6 pups [ = 21], respectively). Testes of infertile males were significantly reduced in size (P = 0.042) and weighed 18–48 mg at 8–10 wk of age in contrast to 43–90 mg for fertile males. Furthermore, testes of infertile males also had reduced levels of RAD51C protein compared with that of fertile males (). Unlike testes of the control males, histological examination of testes from 4-wk-old males revealed seminiferous tubules that were deformed and were often almost devoid of germ cells (). There was a marked increase in the number of apoptotic spermatocytes in the testes of infertile mice as determined by TUNEL assay (). Testis histology of the 12-wk-old infertile males revealed a few tubules containing some mature spermatozoa, but the majority still showed abnormal structures compared with control testes (). Seminiferous tubules from control animals were comprised mostly of multiple layers of round spermatids and usually a single layer of spermatocytes undergoing meiotic divisions (). In contrast, most tubules from mutant males contained no mature sperm, few, if any, spermatids, and multiple layers of spermatocytes in the zygotene and pachytene stages of meiotic prophase (). Although testes morphology and cell composition improved with age, adult males remained functionally infertile for up to 8 mo of age. To determine the precise nature of spermatogenesis failure in mice, we analyzed surface spreads of spermatocytes using molecular markers specific for different stages of meiosis. SCP3 and SCP1, which are components of the lateral and central elements of the synaptonemal complex, respectively, were used to identify cells at different stages of meiotic prophase based on the degree of chromosome condensation and synaptonemal complex formation (). In control samples, >60% ( = 184) of the spermatocyte population is comprised of cells at late zygotene, when chromosomes are not fully synapsed at their ends (), and at pachytene when synapsis is complete (). In infertile mutants, we found that some spermatocytes progressed to the pachytene stage (), but 63% displayed reduced numbers of MLH1 foci ( and Fig. S2, B–E; available at ), which are markers of the crossover sites (). In contrast, only 11% of the control spermatocytes lacked MLH1 foci on one or more chromosomes, whereas most spermatocytes displayed one or two foci on each chromosome ( and Fig. S2, A and E). Mutant spermatocytes lacking MLH1 are likely to be at the pre-MLH1 early pachytene stage (Fig. S2 D). Overall in mutant males, a shift in spermatocyte distribution toward early stages of prophase I was observed (). The number of late zygotene and pachytene spermatocytes was reduced to 40% ( = 208), whereas cells at leptotene (22% in the mutant vs. 7.6% in the control) and at early to midzygotene (30% in the mutant vs. 21% in the control) became the major fractions (). The fact that cells at all stages (leptotene-pachytene) of meiosis could be found in these preparations suggests that the phenotype observed in infertile males reflects an impairment of spermatogenesis during meiosis I rather than an absolute developmental arrest. Formation and repair of DNA double strand breaks (DSBs) by homologous recombination ensures a correct pairing and subsequent segregation of homologous chromosomes during the first meiotic division. A successful generation of DSBs and initiation of their repair is indicated by the presence of phosphorylated histone H2AX (γ-H2AX; ) and RPA () protein staining in mutant and control spermatocytes at leptotene and zygotene (; ). At pachytene, normally only the sex chromosomes stain positively for γ-H2AX (; ). However, in mutants, more than half of all spermatocytes displayed multiple γ-H2AX foci at pachytene, indicating the persistence of DSBs at this stage (). We found no marked difference in the number of RPA foci in mutant and control spermatocytes at various stages of prophase I (; and Fig. S3, available at ). As RAD51C has been implicated in RAD51 foci formation in mitotic cells, we tested whether RAD51 foci formation was defective in spermatocytes of infertile males. RAD51 plays an important role in the initial steps of homologous recombination by mediating homologous pairing and strand exchange and is normally observed in multiple foci first appearing at leptotene and sharply decreasing at pachytene (Fig. S4, A–C; ). littermate control; ; and Fig. S4, D–G), indicating an early defect in the homologous recombination process. γ-H2AX staining marks not only unrepaired DNA but also unsynapsed regions along homologous chromosomes. Although some mutant spermatocytes showed only a few γ-H2AX foci on autosomal chromosomes, we found others that had entire chromosomes staining positively for γ-H2AX at pachytene (, arrowheads). Such chromosomes also appeared thinner than others when stained for SCP3 (, arrowheads) and lacked SCP1 staining (, arrow), suggesting that these were completely unpaired autosomal univalents. Abnormal synapsis between homologous chromosomes was further confirmed by karyotyping spermatocytes at metaphase I. littermate control. We found that although a few metaphase I spermatocytes (16%; = 114) from mutant males appeared normal, similar to the control spermatocytes (), the majority (84%) were morphologically abnormal, possibly undergoing apoptotic fragmentation of the chromatin (). Unpaired autosomal univalents were visible in each of these spreads (; arrowheads). Such univalents are known to induce apoptosis at metaphase I (). Although metaphase I and II spermatocytes were almost equally represented in control animals (57% and 43%, respectively; = 90), metaphase II spermatocytes were rarely found in mutant males (9%; = 125). This suggests that mutant spermatocytes rarely progress beyond metaphase I, as they undergo apoptosis at this stage. Ovaries from infertile females were similar in size compared with fertile littermates, and histological examination showed the presence of morphologically normal follicles at all stages of development (). However, ovaries from 6- and 12-wk-old animals revealed the absence of corpora lutea in infertile females, suggesting that failure to ovulate may be the cause of infertility (). This ovulation block could be overcome by hormonal (intraperitoneal injection of 5 IU [0.1 cc] of pregnant mare serum [PMS] followed by a second injection of 5 IU [0.1 cc] of human chorionic gonadotropin [hCG] 48 h later) treatment (see ovary histology after superovulation; ), after which infertile females could become pregnant when mated with wild-type males. When embryos from the uterine horns of these mice were dissected at embryonic day 8.5, we found that the number of embryos obtained from infertile females was greatly reduced (5 ± 0; = 2) compared with the number of embryos from heterozygous superovulated littermates (15.5 ± 4.94; = 2; P = 0.05). In addition, 7/10 embryos (70%) from the mutant females displayed a wide range of developmental abnormalities compared with only two abnormal embryos out of 31 (6.5%) from superovulated control mice (). All of the embryos from the control females appeared to develop normally (). We hypothesized that the developmental defects observed in these embryos resulted from gross chromosomal defects in the oocytes caused by abnormal meiosis (). To determine the cause of such abnormalities in oocytes, we examined the meiotic progression of germinal vesicle (GV)–intact oocytes by in vitro maturation. oocytes were aligned at the metaphase plate after 8 h of maturation (). Mutant oocytes progressed to metaphase I, and some aligned normally along the metaphase plate (). littermates. The reason for the relatively high number of abnormal oocytes observed in control samples is unclear. One possibility is that this might be an effect of haploinsufficiency, as these animals were heterozygous for the hypomorphic allele. females ( = 6; ). Evaluation of the centromeric cohesion between sister chromatids was facilitated by using a centromere-specific probe (Fig. S5, A and B; available at ). We then karyotyped the oocytes that were allowed to progress to metaphase II in vivo after hormonal treatment. At metaphase II, oocytes from fertile females show the presence of 20 pairs of chromatids, each consisting of two sister chromatids that are attached together at their centromeres (). In oocytes from infertile mice, we found a variety of chromosomal abnormalities. A majority of the mutant oocytes (85%; = 34) showed precocious separation of sister chromatids (PSSC), indicating a problem with chromatid cohesion (). The degree of the cohesion defect varied from cases in which all chromosomes were affected (approximately half of all cases) to those in which only a few chromosomes were affected (for more examples, see Fig. S5, D–G). Only one oocyte from the control littermate was found to have a PSSC phenotype (5%; = 20). Mutant oocytes with a PSSC phenotype often had a few acentric chromatids (40%; = 10; , open arrowhead; and Fig. S5 F, arrowheads). On the other hand, a few other chromatids in the same spreads had two centromeres (, double arrows; and Fig. S5, E–G; double arrows). No such phenotype was observed in control oocytes. In addition, 20% of mutant oocytes ( = 10) in which chromosomes could be counted revealed >20 chromosomes or chromatid pairs, whereas others had 20 or fewer pairs of centromeres, indicating the abnormal segregation of homologous chromosomes during anaphase I of meiosis (Fig. S5 G). The chromosomal defects observed in oocytes after metaphase I suggest a late role for RAD51C in recombinational repair. We did not find any evidence to support a direct role of RAD51C in sister chromatid cohesion (see Discussion; unpublished data). However, based on a biochemical study (), the late function of RAD51C is likely to be associated with the resolution of HJs. To test whether HJ resolvase activity is indeed affected in -deficient mouse cells, we examined protein extracts from mouse embryonic fibroblasts (MEFs) in an in vitro HJ resolution assay. Two MEF lines were established on a -null background, which partially rescued the early embryonic lethality of -null embryos (unpublished data). We used MEFs derived from null and wild-type embryos as controls. HJ resolution activity of protein extracts from the two mutant MEFs was at the background levels between 1.9 and 3.8% and did not correlate with the increasing amounts of protein extract (). The HJ resolution activity of the -null control cells increased from 5.4 to 11.3% as the amount of protein extract increased from 0.9 to 3.5 μg. This was slightly higher than the activity of the wild-type primary MEFs (between 1.0 and 7.5%; ). These activity differences correlated with the amount of RAD51C protein detected in these extracts by Western blotting (). Thus, these data further support a RAD51C role in the resolution of HJs. To test whether the premature separation of chromatids occurred even in males, we examined the spermatocytes that progressed to metaphase II ( = 39 in control and = 11 in mutant samples). In control spermatocyte samples, we found 20 pairs of chromatids attached at the centromeres as observed in oocytes ( and , respectively). Although the chromosomes in some mutant spermatocytes appeared to be normal (), others revealed fragmented chromosomes with broken centromeres or had aberrant chromosome numbers (). However, none of the mutant spermatocytes displayed the sister chromatid cohesion defect. Thus, the PSSC defect is a sexually dimorphic feature restricted only to females. It is unclear whether this phenotype in males is caused by a defect in DSB repair or is caused by the defect in HJ resolution. We present a model to explain how a defect in HJ resolution may result in PSSC in oocytes (see Discussion and ). To date, the only well established function of RAD51C is its role in the process of homologous recombination by facilitating RAD51 foci formation after DNA damage. The precise role of RAD51C in the recruitment of RAD51 and how it differs from other RAD51 paralogues is still unclear. More importantly, recent studies of - and -deficient plants and fruit flies deficient in a RAD51C-like gene, , point to their unique requirement in meiosis unlike other RAD51 paralogues (; ). Characterization of the mutant plants showed that this defect is associated with the repair of Spo11-induced DSBs, leading to abnormal synapsis and severe chromosomal fragmentation at the pachytene stage of prophase I (; ). In this study, we describe a meiotic defect in mutant male mice that is associated with abnormal synapsis between homologous chromosomes at pachytene. At this stage, abnormal synapsis is indicated by the presence of γ-H2AX foci on some of the autosomal chromosomes. Occasionally, entire chromosomes appeared positive for γ-H2AX, indicating completely unsynapsed chromosomes (). Unlike , mouse spermatocytes with unsynapsed chromosomes undergo apoptosis either at pachytene or metaphase I (; ). Consistent with this, we observed a massive increase in abnormal metaphase I spermatocyte spreads. Chromosome synapsis is dependent on RAD51-mediated recombination machinery, which helps bring homologous chromosomes together to repair DNA breaks introduced at leptotene using homologous regions as a template. We found that in spermatocytes, the number of RAD51 foci was reduced about threefold as early as at leptotene. This is consistent with observations of attenuated RAD51 foci formation in RAD51C-deficient cell lines and with a recent finding that RAD51C may interact with RAD51 directly or as a complex with XRCC3 (; ). A similar role for Rad55/Rad57, the budding yeast paralogues of RAD51, has been described previously (). The lack of RAD51 foci revealed in spermatocytes of -deficient mice results in developmental arrest at late zygotene (). However, spermatocytes do not completely arrest at this stage but clearly accumulate at leptotene and at early to midzygotene. This suggests that the RAD51C defect in these mice causes impairment of the RAD51 function; however, because of the hypomorphic nature of the mutation, some spermatocytes escape an early arrest at zygotene, progress further with unrepaired DSBs, and exhibit partially or completely unsynapsed chromosomes so that most of them are eventually blocked at metaphase I. Like the males, a subset of the females failed to produce any litters. Female infertility was associated with an ovulation failure, as no corpora lutea were found in ovaries of such mice. In ovaries of adult mice, oocytes are normally arrested in meiotic prophase I (). As ovaries of the infertile mice were of the same size as the wild-type animals and contained follicles at all stages of maturation, we conclude that -deficient oocytes are able to progress to pachytene normally without an early arrest like in males. However, oocytes from infertile mice do suffer a maturation defect preventing ovulation, which could be overcome only by external hormonal stimulation. Although the number of GV-intact oocytes isolated from ovaries of such females was not considerably reduced compared with littermate controls, they often appeared more fragile to handle during the in vitro maturation experiments (unpublished data). Mutant oocytes display other early meiotic defects like an increased incidence of dysregulated chromosome alignment at the metaphase plate during metaphase I. Such defects may result from the impaired DSB repair as observed in mutant spermatocytes, but, unlike spermatocytes, oocytes could progress to metaphase I even with unrepaired DSBs as a result of sexually dimorphic checkpoint mechanisms (; ). Sexually dimorphic phenotypes in mice have been reported for several other meiotic mutations. Mutations in genes like , , and result in male meiosis arrested during a zygotene to pachytene transition, whereas mutant oocytes can progress through pachytene all the way to metaphase I (; ; ). Similarly, sexual dimorphism in -deficient mice is reflected by the fact that 37% of males were infertile, whereas only 12% of females were infertile. Despite a slightly dysregulated chromosome alignment at the metaphase plate, mutant oocytes were karyotypically normal and did not display unsynapsed chromosomes, which is characteristic of metaphase I oocytes with early meiotic dysfunctions (such as MLH1 and Mei1-deficient mice; ; ). Thus, abnormal embryos produced by infertile females after superovulation must be caused by defects occurring after the metaphase I stage. Indeed, major chromosomal abnormalities were found in mutant oocytes later at metaphase II such as PSSC, aneuploidy, and broken chromosomes. PSSC is the most prominent defect at metaphase II, affecting almost all oocytes of the mutant infertile females. The PSSC phenotype has been previously demonstrated for several genes regulating meiotic cohesion such as β in mice and Sgo1, Bub1, and PP2A in yeast (; ; ; ). Sister chromatid cohesion is mediated by the multiprotein complex cohesin, comprising REC8, STAG3, SMC1β, and SMC3 proteins (). During prophase I of meiosis, cohesin is required for synaptonemal complex formation and recombination to occur between homologous chromosomes rather than between sister chromatids (). Although cohesin is cleaved and released from chromosome arms by separase at metaphase I, Shugoshin protects it at centromeres until anaphase II (; ). This ensures a correct segregation of homologues at anaphase I and sister chromatids at anaphase II into separate daughter cells. mice () to the occasional breakup of bivalents at metaphase I and premature sister chromatid separation that is detectable at metaphase I and is fully exposed at metaphase II for mice (). This sister chromatid cohesion defect at metaphase II seen in infertile females is remarkably similar to defects found in β knockout mice (). Interestingly, a hamster cell line lacking functional RAD51C has also been reported to have defects in sister chromatid cohesion (). Altogether, these findings suggested a possible role for RAD51C in sister chromatid cohesion. We tested whether RAD51C protein can directly interact with proteins involved in sister chromatid cohesion like RAD21, REC8, SMC1β, and Shugoshin (Sgo1) by coimmunoprecipitation but found no evidence to support this hypothesis (unpublished data). This does not eliminate the possibility that RAD51C affects cohesins indirectly via other binding partners or signaling pathways. However, other evidence suggests that RAD51C may not be directly involved in sister chromatid cohesion. First, immunostaining of spermatocyte spreads for SMC1β, RAD21, and REC8 did not reveal any obvious abnormalities during prophase I in male meiosis (; and not depicted). mice (). Third, cohesin failure alone does not explain the presence of acentric chromatids and chromatids with two centromeres seen in the -deficient oocytes. Fourth, the few mutant spermatocytes that reach metaphase II do not show any defect in sister chromatid cohesion (). embryos (unpublished data). mice (). Also, we have shown that protein extracts from -null MEFs lack HJ resolution activity, which further supports this finding (). If RAD51C indeed plays a role in the resolution of HJ, why do RAD51C mutant oocytes exhibit a sister chromatid cohesion defect? A precise phenotype of an HJ resolvase deficiency in the mouse is difficult to predict because so far no other mouse protein has been implicated in this process. Mus81-Eme1 was identified as an HJ resolvase in fission yeast (). Yeast cells deficient for any of these genes are sterile as a result of a defect in chromosome segregation. However, in higher eukaryotes, HJ resolution is likely to be performed by other molecules because Mus81-deficient mice are fertile and do not show any meiotic defect (). We propose that the PSSC phenotype reflects the response of unresolved chromosomes to the increased tension exerted by the spindle at the kinetochores when the chromosomes are pulled to the opposite poles. This model is supported by the behavior of dicentric chromatids in mouse oocytes, which revealed a range of meiotic defects that were surprisingly similar to those of -deficient oocytes, including PSSC as well as broken or acentric chromosomes and aneuploidy at metaphase II (). Such dicentric chromosomes were generated in strains of mice that were heterozygous for an inversion in a large region of chromosomes X or 19. Meiotic recombination in the inverted region led to the generation of dicentric chromatids. When pulled in opposite directions during anaphase I, such chromatids resulted in PSSC in 90–97% of cases and rarely in broken chromosomes (3–10%). Interestingly, the meiotic behavior of such chromosomes in female mice was different from that in male mice, flies, and maize, which were more prone to breakage or selective loss (). Consistent with this observation, metaphase II spermatocytes from mutant mice do not show any sister chromatid cohesion defect but do show chromosomes with broken centromeres (). We speculate that a similar situation would be created if homologous chromosomes failed to dissolve chiasmata. Chiasmata are the cytological manifestation of crossovers, which are believed to arise from double HJs (). Therefore, we conclude that the late meiotic defects found in oocytes of infertile mice can be associated with the HJ resolution failure and propose a model to explain its mechanism (). In normal meiocytes, HJs are established between homologous chromosomes by the pachytene stage of prophase I using the homologous recombination machinery. At metaphase I, bivalents align in the middle of the spindle so that homologous chromosomes can be pulled in opposite directions by microtubules attached to kinetochores of sister chromatids that are oriented toward the same pole. Correct chromosomal alignment triggers anaphase-promoting complex activation, and cohesion is released along the chromosome arms, whereas Sgo1 protects it at the centromere to ensure that sister chromatids stay together during the reductional division. In fission yeast, a spindle checkpoint regulator, Bub1, is essential for the recruitment of Sgo1 to centromeres and, together with Sgo2, promotes sister kinetochore coorientation during metaphase I (). After chiasmata are dissolved, homologous chromosomes can segregate in separate cells. When chiasmata are not formed, homologous chromosomes cannot align properly at the metaphase plate. This activates a spindle checkpoint leading to metaphase I arrest (). We propose that in -deficient oocytes, meiosis proceeds normally until anaphase I. However, there is an accumulation of recombination intermediates such as double HJs that hold the homologous chromosomes together even though the chiasmata fail to fully mature. At the onset of anaphase, there is an increase in tension at the centromere as a result of the persistence of unresolved double HJs. The increased tension may disrupt sister chromatid cohesion at the centromere, as has been demonstrated for dicentric chromosomes and the unpaired X chromosome of XO mice (; ). The exact mechanism of this process is unclear. However, centromeric sister chromatid cohesion is sensitive to chemicals interfering with the metaphase I to anaphase I transition (; ). In addition, Sgo1 has been shown to be a sensor of kinetochore tension in mitotic cells (). Although it remains to be shown in meiosis, it is likely that Shugoshin degradation/disruption caused by increased tension plays a critical role in the PSSC phenotype that is so prominent in -deficient oocytes. In addition to the disruption of centromeric cohesion, chromosome breakage is another consequence of increased tension. Indeed, half of all -deficient oocytes displayed broken chromosomes ( and Fig. S5, E–G). We predicted that in such cases, the centromeric cohesion would remain unaffected. As shown in (inset), in several oocytes, we did observe single chromatids with an extra centromere that is likely to have originated from a sister chromatid. It is also possible that both homologous chromosomes may segregate into one daughter cell, which would lead to aneuploidy detectable at metaphase II. We did see aneuploidy in 2/10 oocytes in which individual chromosomes could be counted (Fig. S5 G). Based on the meiotic defects observed in infertile females, we speculate that RAD51c functions in late stages of meiotic recombination, possibly participating in the resolution of HJs. We cannot rule out the possibility that RAD51C may play a role in sister chromatid cohesion, which may explain the PSSC defect in oocytes. Similarly, it is possible that the primary defect may be the same (i.e., a defect in RAD51-mediated DSB repair) in males and females, but the phenotypes are different because of sexually dimorphic checkpoints. However, based on the evidence presented here, we have proposed a model connecting the impairment of HJ resolution function in infertile females to the observed phenotype in oocytes. The allele was generated in embryonic stem cells in which a resistance gene flanked by two sites was inserted into intron 1 and a single was inserted into intron 3. Heterozygous offspring in the C57BL/6J × 129/Sv mixed genetic background were crossed with β-actin– transgenic mice () to obtain the allele. or fertile mice to mice. To determine the fertility status of mice, 6-wk-old males and females were mated with fertile animals for 4–10 wk. To monitor the mating behavior, mice were checked for vaginal plugs daily in a course of at least 6 wk. Mice that plugged several times but failed to produce any offspring were identified as infertile. Infertile mice are referred to as mutant mice in the text for simplicity. Protein lysates from control and mutant testes were prepared in cold radioimmunoprecipitation assay buffer. Samples containing 80 μg of protein were separated in NuPAGE 4–12% Bis-Tris polyacrylamide gels (Invitrogen) and transferred onto a nylon membrane. The membrane was probed with mouse or rabbit α-RAD51C antibody (Novus Biologicals or Chemicon International, respectively) at a 1:500 dilution or were probed with α-panactin antibody (NeoMarkers) diluted 1:400 according to standard procedures. Secondary α-mouse IgG-HRP antibody (1:2,000; Santa Cruz Biotechnology, Inc.) and an ECL chemiluminescence system (GE Healthcare) were used for signal visualization. Testes and ovaries were fixed in Bouin's solution. Samples were dehydrated through an ethanol series, embedded in paraffin, serially sectioned, and stained with hematoxylin and eosin. Slides were examined using brightfield microscopy. For TUNEL staining, testes were fixed in 10% neutral buffered formalin and were stained using the ApopTag kit (Chemicon International) according to the manufacturer's instructions. Surface spreads of spermatocytes from the testes of mutant and control animals were prepared and stained as described previously (). Another method was used to prepare spermatocytes for the RAD51 and RPA staining and was described previously (). The difference between the two protocols is that no enzymatic treatment is involved in the second method, and cells are separated by pipetting and spread in a hypotonic solution directly on a slide. The following primary antibodies were used for immunofluorescence: rabbit anti–γ-H2AX (1:1,000; obtained from W. Bonner, National Cancer Institute, Bethesda, MD), mouse anti-MLH1 (1:10; BD Biosciences), rabbit polyclonal α-SCP3 (1:500), α-SCP1 (1:1,000), α-RPA (1:100; all provided by P. Moens, York University, Toronto, Canada), mouse monoclonal α-SMC1β (1:10; provided by E. Revenkova, Mount Sinai School of Medicine, New York, NY), and rabbit α-Rad51 (1:500; obtained from S. West, Cancer Research UK, South Mimms, UK). Secondary antibodies used were goat anti–rabbit AlexaFluor488, goat anti–rabbit AlexaFluor568, goat anti–mouse AlexaFluor488, and goat anti–mouse AlexaFluor568 (Invitrogen). Secondary antibodies were used at a 1:250 dilution. Images were acquired with a microscope (Axioplan 2; Carl Zeiss MicroImaging, Inc.) using an oil plan Neofluar 100× 1.3 NA objective (Carl Zeiss MicroImaging, Inc.). Images were taken with a CCD camera (Quantix; Photometrix) and processed using SmartCapture software (Desksoft). Images were further processed with Photoshop software (Adobe) to adjust for size and contrast. Females were superovulated, and oocytes and embryos were collected as described previously (). In brief, 0.1 ml PMS (5 IU; Sigma-Aldrich) was injected intraperitoneally into female mice. GV-intact stage oocytes were collected 44–48 h later in flushing and holding medium with 3 mg/ml BSA (Specialty Media) by puncturing ovaries with a 27-gauge needle. Oocytes were incubated briefly in 0.1 mg/ml hyaluronidase type IV-S (Sigma-Aldrich), and cumulus cells were removed by pipetting before transfer into Chatot, Ziomek, and Bavister media containing -glutamate and 3 mg/ml BSA (Specialty Media) for 8 h at 37°C and 5% CO to obtain metaphase I stage oocytes for immunocytochemistry and karyotyping. Alternatively, 48 h after PMS treatment, 6-wk-old female mice were given an intraperitoneal injection of 0.1 ml hCG (5 IU; Sigma-Aldrich), and, 14 h later, metaphase II–stage oocytes were collected from ampullae. To obtain embryos, 6-wk-old female mice given both PMS and hCG injections were placed with stud males. Metaphase I and II stage oocytes were washed in Dulbecco's PBS without CaCl or MgCl (Invitrogen) and were fixed for 10 min in cold methanol. Oocytes were incubated in blocking buffer (4% BSA; Sigma-Aldrich), 10% normal goat serum (Vector Laboratories) overnight at 4°C, monoclonal anti–β-tubulin clone TUB2.1 (Sigma-Aldrich) for 1 h at RT, AlexaFluor488 goat anti–mouse IgG (Invitrogen) for 1 h at RT, and in a 1:5,000 dilution of 1 mg/ml DAPI (Roche) for 20 min at RT before mounting in Vectashield Mounting Medium (Vector Laboratories) on Shandon Multi-Spot microscope slides (Thermo Savant). Metaphase spreads of spermatocytes were prepared by the Evans method from the testes of 3-wk-old mutant and control males (). Oocytes were karyotyped by an air-drying method as described previously (). In brief, mutant and control females were injected with PMS followed by hCG as described above (see Superovulation and collection of oocytes and embryos), and, 14 h later, oocytes arrested at metaphase II were collected from the ampullae in Weimouth media containing penicillin-streptomycin, 10% FBS, and 2.5 mg/ml sodium pyruvate. Chromosome spreads were prepared from the oocytes after a 2-min incubation in 0.9% sodium citrate solution. Protein extraction was performed as follows: exponentially growing MEFs were collected from ∼20 15-cm tissue culture dishes after trypsinization. Approximately 1 g of a cell pellet was resuspended in prechilled lysis buffer (10 mM Tris, pH 8.0, 1 M KCl, 1 mM EDTA, and 1 mM DTT) in the presence of complete EDTA-free protease inhibitor cocktail (Roche). Cells were homogenized by sonication followed by incubation on ice for 1 h. The insoluble pellet was removed by centrifugation at 13,000 rpm for 1 h. Solid (NH)SO (25%; 134 g/liter) was added to the supernatant and dissolved by gentle stirring on ice for 30 min. Insoluble materials were removed by low speed centrifugation (30 min at 9,000 rpm). The (NH)SO concentration was then raised to 55% (an additional 179 g/liter), and the proteins were precipitated during 30 min of stirring on ice. The precipitate was recovered by centrifugation and resuspended in buffer A1 (50 mM KHPO/KHPO, pH 6.8, 10% glycerol, 1 mM EDTA, 1 mM DTT, and 0.01% NP-40) containing 250 mM KCl and was dialyzed against the same buffer for 2 h before storage at −80C. The resolvase assay was performed as follows: for each 10-μl reaction, 1 μl of extract containing 0.9, 1.8, and 3.5 μg of protein was assayed on 1 ng 5′-[P]end-labeled synthetic HJ DNA (X0) in phosphate buffer (60 mM NaHPO/NaHPO, pH 7.4, 5 mM MgCl, 1 mM DTT, and 100 μg/ml BSA) in the presence of 200 ng of competitor poly[dI:dC] DNA. Incubation was performed for 30 min at 37°C. Products were deproteinized and analyzed by native 10% PAGE. The percentage of resolution was quantified by a phosphoimager (Typhoon 9400 Scanner; Molecular Dynamics). Fig. S1 shows aberrant splicing of the transcript as a result of the presence of the cassette. Fig. S2 shows quantitative analysis of MLH1 foci formation in spermatocytes from control and infertile mice. Figs. S3 and S4 show quantitative analysis of RPA foci (Fig. S3) and RAD51 foci (Fig. S4) in spermatocytes from control and infertile mice. Fig. S5 shows chromosomes from control and oocytes at metaphase I hybridized with a pancentromeric probe as well as sister chromatid cohesion and other defects at metaphase II. Online supplemental material is available at .

Chromosomes occupy discrete, mutually exclusive volumes in interphase nuclei (). These chromosome territories do not intermingle, except at their boundaries (; ; ), and their size is approximately related to DNA content (). Although most proteins diffuse rapidly throughout nuclei (), the diffusion of individual chromosome territories is constrained (; ). Territories are not randomly distributed, and the rules governing positioning are incompletely understood and vary between cell types () and growth state (). The most constant finding is of a conserved radial organization; gene-poor chromosomes (e.g., human chromosome 18) are found at the periphery, and gene-rich chromosomes (e.g., human chromosome 19) are found more centrally (; ; ). A similar although weaker radial organization with gene-poor chromosomes more peripheral has recently been reported for the mouse, based on position analysis of chromosomes 1, 2, 9, 11, 14, and X in a range of cell types (). No specific nuclear envelope (NE) component has been implicated in these large-scale localizations, and the inner nuclear membrane protein emerin and lamina protein lamin A have been excluded (; ). The significance of the nonrandom distribution of chromosome territories within the interphase nucleus remains poorly understood. The nonrandom organization of territories may reflect functional associations between specific chromatin regions and other nuclear structures, the nucleolus being one obvious example. The inner face of the nuclear periphery, consisting of an inner nuclear membrane decorated with a nuclear lamina pierced by nuclear pore complexes (NPCs) represents another large domain for potential chromatin interaction. Experimental evidence from several groups has shown that associations between certain gene loci and the nuclear periphery may play important roles in the transcriptional regulation of those genes. The association has been shown to act at the level of chromosomal subregions and generally has a repressive or silencing role, with activation involving movement away from the periphery into the nuclear interior (; ; ). However, peripheral association does not always result in repression; in yeast, transcription-dependent association with the NPC couples transcriptional activity with message export (; ). Significantly, engineered enhancement or weakening of the association of the hexokinase I test locus with the NPC modulated the gene expression changes seen in response to physiological triggers, suggesting a positional effect operating in tandem with regulation because of transcription factor binding (). Using live-cell imaging of the yeast gene marked by adjacent operator sequences detected with GFP-tagged repressor protein, demonstrated that transcription is necessary but not sufficient for perinuclear confinement of active loci, which requires, in addition, direct interactions between components of the histone acetyltransferase complex and NPC components. Intriguingly, in this study, the subnuclear localization of the activated test locus demonstrated a peripheral confinement that was not static but, rather, involved a 2D sliding motion along the nuclear periphery, suggestive of molecular interactions continually formed and released. In parallel with the emerging picture of active gene association with NPCs in yeast, association of chromatin with the nuclear lamina is increasingly recognized in mammalian cells. Known chromatin–lamina protein interactions (; ), together with the observed genetic defects in the human laminopathies (; ; ), also implicate this structure in additional levels of nuclear organization and regulation (). Recently, a lamina-dependent chromatin position effect has been observed during analysis of the localization of the human 4q35.2 region implicated in fascioscapulohumeral muscular dystrophy (FSHD; ). In this case, the tightly peripheral position of the subtelomeric FSHD locus requires a lamina containing functional lamin A. The overall nuclear position of the chromosome 4 territory, however, is unchanged between wild-type (WT) and lamin A–null fibroblasts, suggesting that the 4q35.2 region migrates within the chromosome territory when lamin A–dependent peripheral association is lost. The nuclear lamina is a protein network that underlies the inner nuclear membrane, where it maintains nuclear shape and plays roles in attaching heterochromatin (). The mammalian nuclear lamina contains lamins A and C (alternatively spliced products of a single gene, ), together with lamins B1 and B2, products of two additional genes, and . B-type lamins are expressed in all cells, whereas A-type lamins are developmentally regulated. Lamin A contains a C-terminal CAAX motif (cysteine, aliphatic, aliphatic, any of several residues) that undergoes a -like processing comprising farnesylation, endoproteolysis, and carboxymethylation, but this modified C terminus is then removed by a Zmpste24-dependent maturation cleavage, which removes the C-terminal 15 amino acids. Mature lamin A therefore lacks the farnesylated and carboxymethylated C-terminal anchor. A point mutation that alters splicing to generate lamin A lacking the maturation cleavage site produces “progerin,” a lamin A with the CAAX attachment site intact. The resulting severe multisystem premature aging phenotype (Hutchinson-Gilford progeria syndrome) suggests that modified C-terminal anchoring of the nuclear lamina is important for its function (). The dominant toxic effect of progerin may be the result of competition for limited binding sites on the inner nuclear membrane, and it has recently been shown that reducing the amount of progerin within the nucleus using farnesyl transferase inhibitors (FTIs) restores nuclear shape (; ; ; ) and may be of potential use in the treatment of progeria (). Lamin B1 also undergoes CAAX processing; the mature protein retains the hydrophobic processed CAAX anchor and remains stably associated with the lamina and with the inner nuclear membrane. CAAX processing is important for functional lamin B1 expression at the nuclear periphery (). Processing has three stages; farnesylation of the cysteine at −4 by a unique farnesyl transferase, endoproteolysis to remove the last three residues by the Ras converting enzyme 1 (Rce1), and carboxymethylation of the newly terminal cysteine by isoprenylcysteine carboxyl methyltransferase (Icmt). Endoproteolysis only occurs after farnesylation. The first two steps are essential for stable association of lamin B1 with the nuclear periphery, whereas the final carboxymethylation step is only important in the context of the isolated C terminus lacking the coiled-coil domains (). In the current study, we hypothesize that the stability of associations of the nuclear lamina with chromatin are important for gene expression. We first use live-cell imaging to show that the stability of the lamin B1 network of the nuclear lamina is dependent on lamin B1 processing by the endoprotease Rce1. Using three MIAME (minimum information for annotation of microarray experiments)–compliant microarray datasets, we show that either absence of full-length lamin B1 or lack of C-terminal processing affects gene expression and that some of the dysregulated genes form clusters on certain chromosomes. We identify a significant cluster of three dysregulated genes within an ∼4–mega bp (Mbp) region on chromosome 18 and use this chromosome as a model for how loss of interaction with the nuclear lamina affects chromosome position and, hence, gene expression. This is, to our knowledge, the first report linking a defect in the NE, an altered chromosome position, and changes in gene expression and supports the view that peripheral nuclear architecture is important for aspects of genome organization that play a role in the regulation of gene expression. We previously demonstrated the importance of lamin B1 processing for the integrity of the nuclear lamina and its association with the NE using sequential extraction of nuclear proteins (). Here, we first looked at the effects of defects in lamin B1 processing on the stability of the NE using fluorescence loss in photobleaching (FLIP) of GFP-tagged lamin B1 expressed in cells lacking either of the CAAX processing enzymes, Rce1 or Icmt (). FLIP shows that lamin B1 endoproteolysis by Rce1 is important for its stability within the nuclear lamina (), whereas loss of carboxymethylation by Icmt has little effect on the GFP–lamin B1 stability (). Restoring tagged full-length lamin B1 expression to cells gives a stability indistinguishable from WT, confirming that secondary changes due to selection on the transgenic cells are not responsible for altered stability. Lamin C dynamics are not altered in the absence of lamin B1 or lack of its processing, as FLIP of YFP–lamin C in all the knockout cells used in the current study shows no significant difference from that in WT cells (Fig. S1, available at ). These results suggest that farnesylation and endoproteolysis are essential for integrity of lamin B1 in the NE and, hence, for those aspects of nuclear architecture that depend on it. Given the association between chromosome position and gene expression, we set out to investigate the effect of the lack of lamin B1 or defects in its processing on gene expression. To study the effect of lamin B1 and its processing on gene expression, we performed 18 genome-wide microarray experiments to compare the gene expression profiles of , , and cells with their WT backgrounds. We reasoned that genes with altered expression in both and cells are dependent on endoproteolyzed lamin B1, whereas those dysregulated in , , and cells are dependent on carboxymethylated lamin B1. In contrast, genes dysregulated in either or cells but not in cells would be dependent on Rce1 and/or Icmt processing of proteins other than lamin B1. The microarray datasets were MIAME complaint and included six biological replicates of RNA preparations of each cell type, with duplicate comparisons on each slide. Dye swaps were also included to eliminate any dye-specific effects on the microarray hybridizations. All of the raw data may be downloaded from Array Express (). The microarray datasets were also validated by quantitative real-time PCR (qRT-PCR) for samples of genes up-regulated, down-regulated, or unchanged for each cell type (; qRT-PCR values are shown in parentheses). All of the data discussed in this paper were statistically significant at P < 0.05 for a 1.5-fold change in expression between the test and WT samples. We used a 1.5-fold change as a cutoff for biological significance; by this criterion, there were significantly up- and down-regulated genes for each knockout type. For example, in the cells, 834 genes were down-regulated, in contrast to only 129 genes that were up-regulated. For cells, the corresponding values were 42 genes down-regulated compared with 422 up-regulated. Three-way comparisons of expression changes were then made to find genes affected in more than one knockout cell type. First, genes with altered expression in only one of the knockout cell types were identified. We found 614 genes with altered expression (predominantly down-regulation) only in cells. Because the normal expression of these genes depends on lamin B1 expression but is not altered when the CAAX processing machinery is defective, we conclude that their expression is influenced by an unmodified lamin B1 pool. Similarly, we found 249 genes with altered expression (predominantly up-regulation) only in cells. Because the normal expression of these genes requires cellular CAAX processing but is not altered when lamin B1 is absent, we conclude that their expression is modulated by CAAX processed proteins other than lamin B1; ras is an obvious candidate. shows part of a tree view from a clustering analysis that takes account of all three knockout cell types. The section shown includes genes with altered expression (in either direction) in any two of the cell types. Using the 1.5-fold change as a cutoff, we identified a group of 16 genes that were coordinately dysregulated in the and cells, 11 up-regulated and 5 down-regulated (). To confirm the similarity in the pattern of expression change in the two cell types, we performed correlation analysis on the fold changes for these 16 genes in and cells. The resulting Pearson product–moment correlation coefficient was 0.897, with a value of 1.0 representing perfect correlation; this confirms that this group of genes is dysregulated in a similar way in the two transgenic cell types. Because these genes were coordinately either up- or down-regulated in each of the knockout cells, we conclude that their normal expression requires both lamin B1 expression and an intact CAAX farnesylation and endoproteolysis machinery, indicating that their expression depends on processed lamin B1. To confirm that this dysregulation is indeed a consequence of interfering with lamin B1 and its processing, we performed additional experiments on WT fibroblasts in which CAAX processing was abolished by treatment of the cells with an inhibitor of the farnesyl transferase enzyme. qRT-PCR on several of these genes in FTI-treated WT cells confirmed that they showed similar changes as a result of interfering with the first step of lamin B1 processing as they do in the cells (). FTI treatment is expected to alter the gene expression of several genes, but the subset of genes selected here are also dysregulated in the cells, which means that we are examining genes that are dysregulated as a result of interfering with the farnesylation of lamin B1. After identifying this group of 16 coordinately dysregulated genes (), we sought common features or characteristics between them. First, we tried to cluster these dysregulated genes according to function using the Gene Ontology database but were unable to find a common pattern. To exclude any functional clustering more completely, we took all genes showing significant dysregulation in either direction in one or more of the knockout cell types (a total of 4,144 genes) and repeated the Gene Ontology clustering. The entire dataset is given in Table S1 (available at ), but a small representative area is shown in . This analysis should find clusters of dysregulated genes associated with specific functions or pathways even if expression levels are changing in opposite directions in different knockout cells. We could find no evidence for clustering of dysregulated genes by function even by this more relaxed criterion. To continue the search for shared characteristics in this small group of genes with expression dependent on processed lamin B1, the position of each gene was mapped onto the mouse karyotype. Inspection of the result suggested that the distribution, especially of the up-regulated genes, was not random but bunched (). In particular, a group of three up-regulated genes were found within a short 4-Mbp region on chromosome 18. To test whether this apparent clustering was statistically significant, we used a bootstrap sampling method to repeatedly draw 16 genes at random from the set of genes common to both lamin B1 and Rce1 knockout experiments. This method takes account of the genes present in both datasets from the microarray experiment and their nonuniform distribution within the karyotype. Analysis of clustering of the results of 10,000 trials confirmed that the experimental cluster found on chromosome 18 is significant (P < 0.02). We then considered this significant cluster of up-regulated genes on chromosome 18 and reasoned that its presence might be a consequence of movement away from the periphery as a result of the loss of fully processed lamin B1. There have been no previous reports of the radial position of the gene-poor mouse chromosome 18. We therefore studied its position using FISH. The method used to determine its radial position involved dividing the nucleus into five concentric shells and measuring the distribution of the chromosome 18 signal within these shells in 50 nuclei. We found that in three independent WT mouse embryonic fibroblast populations, chromosome 18 occupies a strongly peripheral location (, WT), possibly in association with the nuclear lamina. Indeed, when Con A was used as a marker for high-mannose glycoproteins in the intermembrane space of the NE in further FISH experiments, we found no detectable separation between the NE and the outermost border of the chromosome 18 territory (). As the same result was obtained in three independent WT fibroblast populations, the peripheral location of mouse chromosome 18 is unlikely to be an artifact because of inadvertent clonal selection during culture. We also performed FISH for chromosome 19, a very gene-rich chromosome that would not be expected to be associated with the nuclear periphery. We found that chromosome 19 has a more central location in all the studied cell types (), consistent with the observation that no chromosome 19 genes are dysregulated as a result of the lack of processed lamin B1 (). Once the peripheral position of chromosome 18 in normal primary fibroblasts was established, we repeated the chromosome 18 FISH in and cells, to test whether detectable chromosome territory movement had occurred. reported that 38–39% of the Lmnb1 cells they examined had misshapen nuclei, compared with 2–8% of the WT cells. We also observed many misshapen nuclei in the Lmnb cells. To exclude any secondary effects of gross morphological abnormalities on measured chromosome distribution, we only considered nuclei that exhibited normal morphology in all of our FISH analyses. Strikingly, in both and knockout cell types, chromosome 18 was no longer found at the periphery (, and ). In contrast, however, cells showed a peripheral distribution for chromosome 18 that was very similar to WT (, ), suggesting that just as carboxymethylation is not essential for the stability of lamin B1 in the lamina (), it is also not important in the maintenance of chromosome 18 at the nuclear periphery. To confirm this striking result, we sought an additional experimental test. In particular, we were concerned with excluding the possibility that the observed new position of chromosome 18 was the result of selection of an unusual subclone during culture, although this would have had to occur independently for both and cell types. We reasoned that a demonstration of chromosome 18 relocation as a result of an acute perturbation of WT cells would offer strong evidence that the altered position of chromosome 18 in both transgenic knockout cells was not the result of inadvertent simultaneous selection of both and subclones with altered chromosomal position. Accordingly, we used a selective inhibitor of farnesylation, the first and mandatory step of CAAX processing, in WT cells and repeated the FISH experiment for chromosomes 18 and 19. As shown in , the relocation of chromosome 18 was also observed in FTI-treated WT cells, confirming that it is indeed processed lamin B1 that is essential for the localization of chromosome 18 at the nuclear periphery. Previous sequential extraction studies of lamin B1 association with the nuclear lamina after inhibition of CAAX processing demonstrated the importance of these posttranslational modifications for stable lamina formation (). Quantitative live-cell microscopy using GFP-tagged lamin B1 in FLIP experiments in WT, , , and cells confirms and extends this biochemical data. Using FLIP analysis, we show that the stability of full-length lamin B1 lacking only carboxymethylation is almost identical to that of fully processed lamin B1 (). In contrast, the stability of nonendoproteolyzed lamin B1 in is lower, reflected by the faster loss of fluorescence in FLIP experiments (). These results suggest that lamin B1 is much less able to contribute to a stable lamina when CAAX endoproteolysis does not occur, with consequences for any aspect of nuclear organization or function that depends on a stable lamin B1 scaffold. This effect is specific for the lamin B1 component, as defects in the lamin B1 processing machinery do not affect the stability of lamin C (Fig. S1), which does not undergo the same series of posttranslational modifications. We hypothesized that cells with defects in lamin B1 expression or in the CAAX processing machinery would exhibit altered gene expression levels, partly as a result of the absence of processed lamin B1 in the lamina. We examined this proposal using microarray analyses to compare gene expression profiles in WT murine fibroblasts with those in cells deficient in full-length lamin B1 () or components of the processing pathway. Although cells deficient in the converting enzyme endoprotease have defective CAAX processing and exhibit the gene expression consequences of loss of processed Ras, cells deficient in lamin B1 exhibit normal CAAX processing and normal Ras function. The cells, which are deficient in the carboxymethyl transferase, occupy an intermediate position because the final carboxymethylation step is important for Ras targeting and function (), whereas the FLIP analysis () suggests that this step is less critical for proper deployment of lamin B1 to a stable NE. Thus, we reasoned that genes showing coordinate dysregulation in both the lamin B1 and Rce1 knockouts would likely represent genes whose expression is dependent on an NE containing correctly processed lamin B1. On this basis, we identified 16 genes that showed a 1.5-fold or greater change in expression in the same direction in both and cells; of these, five also showed altered expression in cells. In most cases, the change represented up-regulation in the knockout cells (). Taking a more stringent twofold cutoff, no genes were either up- or down-regulated in all three cell types, suggesting that few, if any, of the genes examined were affected by a failure of carboxymethylation of lamin B1. In contrast, at this stringent twofold cut-off, seven genes were up-regulated and two were down-regulated in both and cells (), suggesting that their expression depends on the presence of farnesylated, proteolyzed lamin B1, irrespective of carboxymethylation. When the cutoff for the p-values associated with the fold changes is raised from 0.05 to 0.1, we find that 51 genes are coordinately dysregulated in and cells, 32 of which are up-regulated and 19 of which are down-regulated. Conversely, genes with altered expression only in the lamin B1 hypomorph cells are presumably responding to a deficit other than the lack of processed lamin B1 in the NE; otherwise, they would also be dysregulated in cells that expressed normal lamin B1, but which were unable to process it to the mature form. The discovery of 498 genes with at least 1.5-fold change in expression that are unique to the cells () is consistent with a model in which unprocessed full-length lamin B1 in the nuclear interior plays an important role in regulating gene expression. The mechanism by which nucleoplasmic lamin B1 might play a role in regulated gene expression remains unknown, although observations in other systems provide some clues (). Intranuclear lamin B1 might associate with the machinery of transcription and RNA processing in a way similar to that observed for lamin A. A possible direct interaction with a polymerase complex is suggested by the discovery that the germline-specific lamin of oocytes (Liii) associates with RNA polymerase II, and that Pol II activity is inhibited by dominant-negative lamin mutants (). A more indirect effect via specific transcription factors is suggested by reports of lamin B1 binding to the repressor protein, Oct1, although in this case the presence of elevated levels of Oct1 at the nuclear periphery may implicate processed lamin B1 in the lamina, rather than an unprocessed nucleoplasmic pool (). We concentrated next on the small group of 16 genes showing a consistent expression dependence on processed lamin B1 (listed in ). First, we attempted to cluster them by function using the Gene Ontology database. No clustering into any functional grouping could be detected ( and Table S1). Second, the genes were mapped to identify their chromosomal positions; clustering at specific chromosomal locations was observed (). In particular, a cluster of three genes within ∼4 Mbp coordinately up-regulated in both and cells was found on chromosome 18, a chromosome that we found to have a strongly peripheral localization in WT cells (). This result is consistent with an envelope-dependent suppression of gene expression that is lost when processed lamin B1 is unavailable. In contrast to this up-regulated cluster on chromosome 18, none of the genes on chromosome 19 showed such up-regulation. We have shown that chromosomes 18 and 19 have peripheral and central locations, respectively. The presence of the up-regulated cluster on chromosome 18 might therefore be due to disruption of peripheral localization because of the absence of proteolyzed lamin B1. We therefore analyzed the localization patterns of chromosomes 18 and 19 in the three knockout cell types and their WT background cells by two-color FISH (). Because mouse chromosome 19 has one of the highest gene densities in the genome (14.1 genes/Mbp) and gene-dense chromosomes are usually more central, and no coordinate dysregulation of its genes was detected, its localization would not be expected to be altered in the knockout cells; therefore, it serves as an internal control. Mouse chromosome 18 has the lowest gene density of any mouse chromosome (7.5 genes/Mbp), suggesting that a peripheral localization was likely. We found that chromosome 18 is indeed located at the nuclear periphery in three independent WT mouse embryo fibroblast populations. This peripheral localization was preserved in cells, but in both and cells (), it became centrally located, like chromosome 19. This suggests that farnesylated and proteolyzed lamin B1 anchors chromosome 18 to the periphery and that this positioning in turn plays a role in the expression of a group of genes on this chromosome. Chromosome 19, on the other hand, showed a central location in the WT cells and in all the knockouts, and none of its genes were coordinately up-regulated in both and cells. This is the behavior expected of a chromosome that is indifferent to the presence or absence of peripheral processed lamin B1. recently reported that the distribution of chromosome territories is cell type specific. Although they observed that the gene-rich mouse chromosome 11 is generally more centrally located, they reported that it also had contact points with the nuclear periphery. Our current observation of two dysregulated genes on chromosome 11 in the absence of processed lamin B1 from the periphery supports this observation because chromosome 11 (121.7 Mbp), which is larger than chromosome 18 (90.7 Mbp), may be anchored at the nuclear periphery at specific points but still exhibit an internal location. The published gene expression changes in the absence of lamin A or lack of its processing by Zmpste24 () are distinct from genes that are coordinately dysregulated in and cells. Only 12 genes are coordinately dysregulated in , , and cells, with 8 up-regulated and 4 down-regulated. Therefore, the genes that we report to be dysregulated in the absence of processed lamin B1 are specific to that defect and are not due to a general abnormality in the nuclear lamina. Furthermore, the change in chromosome position as a result of the absence of lamin B1 or its processing is specific, as cells lacking the inner nuclear membrane protein emerin do not show altered chromosome organization (). It has more recently also been reported that cells from patients with emerin and lamin A mutations do not show a significant change in chromosome locations (). Although lamin B1 deficits have not to date been associated with any human disease, such defects are certainly not inconsequential. mutant mice, from which the embryonic fibroblasts used in the current study were obtained, die shortly after birth, with lung and bone abnormalities (). Although the groups of up-regulated genes on chromosome 18 do not seem to have a common function, heparin-binding EGF-like growth factor (), plays important roles in development (; ), which might explain some of the phenotypic aspects of the lamin B1 mutant mice. Furthermore, although no functional clustering is observed when the genes that are coordinately dysregulated in two or more of the transgenic cells used in the current study are considered, genes that are dysregulated only in the absence of full-length lamin B1 do show some functional clustering. A group of down-regulated genes—, , , , , , , , , , , and —are involved in lung development (Gene Ontology accession no. 0030324), which is consistent with the mice failing to survive after birth as a result of respiratory failure and having reduced numbers of alveoli and thickened mesenchymal tissue. It was recently reported that overexpression of lamin B1 because of a genomic duplication covering the 5q31 region containing the lamin B1 gene causes autosomal dominant leukodystrophy (ADLD; ). The increased gene dosage results in increased lamin B1 message and protein in brain tissue of affected individuals. The clinical result is a late (adult) onset progressive, symmetrical demyelinating disease that resembles multiple sclerosis except that oligodendroglia are preserved in lesions and there is no astrogliosis. This result suggests that altered lamin B1 expression can be associated with severe human disease and that ADLD should be added to the list of laminopathies. Alongside lamin B1, lamin A and B2 are major components of the nuclear lamina of most cells. We therefore speculated that a compensatory mechanism involving the up-regulation of their corresponding genes might be taking place to maintain the structure of the nuclear lamina. qRT-PCR revealed about a threefold up-regulation in the transcript in cells, suggesting that this may indeed be an attempt to compensate for the loss of functional lamin B1. This, however, does not reverse the effect of the lamin B1 defect, as observed by abnormalities in gene expression and chromosome localization. This indicates that there are functions that are quite unique to lamin B1 that cannot be compensated for by an excess of lamin A. In contrast, transcription of remains almost unchanged, suggesting, perhaps surprisingly, that lamin B2 is subject to entirely independent regulation. Some investigators have observed an activation of some genes while still at the nuclear periphery but by changing location within that vicinity (; ). Therefore, the nuclear periphery probably contains regions of effective suppression and other regions where additional factors may be contributing to the regulation of gene expression. This might explain why not all the genes that we analyzed on chromosome 18 are up-regulated although the whole chromosome moves away from the periphery in the knockout cells. Such a conclusion would be supported by recent high-resolution studies of a short, 4-Mbp chromosome segment showing that zigzagging of the chromatin can bring discontiguous genes together, while intervening genes are looped out (). A similar arrangement of the genes within the 4-Mbp cluster on chromosome 18 would permit lamina-dependent regulation of three genes that are not contiguous, although they are clustered in a small region. In summary, our results represent the first report of a role for a nuclear lamina component, specifically, farnesylated and endoproteolyzed lamin B1, in the positional organization of chromosomes in the interphase nucleus. We demonstrate that processed lamin B1 is required to anchor chromosome 18 at the nuclear periphery and that disruption of this interaction (directly or indirectly) results in dispersion of this chromosome from the nuclear periphery together with an up-regulation of certain genes on the chromosome, consistent with a context-dependent gene-silencing role for the NE on these genes. Alterations to the global organization of chromosomes in the nucleus may lead to the severe consequences observed in laminopathies and may also provide insights into the normal process of aging. Mouse embryonic fibroblasts , (), , (an insertional mutation lacking 6 exons of the lamin B1 gene encompassing the C-terminal 273 amino acid residues, including the chromatin interaction and CAAX domains; ), , and () were cultured in DME supplemented with 10% FCS, -glutamine, and nonessential amino acids at 37°C in a humidified atmosphere. The GFP-tagged full-length lamin B1 construct (GFP–lamin B1) has been described before (). Cells were grown in 25-cm flasks for RNA isolation, on glass-bottomed 35-mm dishes (MatTek) for photobleaching experiments, and on glass coverslips for FISH. For photobleaching experiments, cells were transfected with GFP–lamin B1 using Lipofectamine 2000 (Invitrogen), and experiments were performed 48 h after transfection. For microarray analyses, early passage cells were seeded at 50% confluency, and RNA was extracted when cells reached ∼90% confluency. FTI-treated cells were incubated with 100 μM FPT inhibitor III (Calbiochem) for 48 h. FLIP was performed using a confocal laser-scanning system (Radiance 2000 MP; Bio-Rad Laboratories) on an inverted microscope (Eclipse TE300; Nikon) at 37°C using the 488-nm line of a Kr/Ar laser with a 60× 1.4 NA objective. Some cells with distorted nuclear morphology were observed in transgenic knockout populations (); only cells with normal nuclear morphology were selected for FLIP analysis. A region of interest (ROI) was photobleached at full laser power while scanning at 5% laser power elsewhere with 1-s intervals between scans over a period of 250 s. Image acquisition was controlled by Lasersharp (Bio-Rad Laboratories), and images were analyzed using MetaMorph (Universal Imaging Corp.). For quantitative analysis, background intensity was subtracted and intensities of a specific ROI outside the photobleached area were measured over time and normalized using intensities of an ROI in a transfected but nonbleached cell. RNA was extracted from six biological replicates of early passage cultures for each cell type, using TRIZOL reagent (Invitrogen), further purified with RNeasy mini columns (QIAGEN), and quantitated on a Nanodrop spectrophotometer. RNA integrity was confirmed before labeling using Nanochips on an 2100 Bioanalyzer (Agilent Technologies), according to the manufacturers' instructions. Two protocols for microarray labeling and processing were used during these experiments, an indirect amino-allyl dUTP Alexa Fluor labeling system and, subsequently, a 3DNA dendrimer-based system. For the indirect amino-allyl dUTP-based method, RNA (5 μg of total RNA) was labeled and hybridized using the HiSpot RT kit (Genetix) according to the manufacturer's instructions, with the exception that Superscript III was used in place of the supplied reverse transcriptase, followed by labeling with the ARES kit (Invitrogen) with Alexa Fluor 555 and 647 dyes, according to the manufacturer's instructions. For the 3DNA dendrimer-based system, RNA (1 μg of total RNA) was labeled using the 3DNA Array 900 kit (Genisphere), using Superscript III reverse transcriptase (Invitrogen) in the first strand cDNA synthesis. The hybridization and detection steps were performed using a two-step hybridization procedure on a SlideBooster (Advalytix), each with a power setting of 25 and a pulse ratio of 3:7 at 55°C. The first hybridization was for 16 h using hybridization buffer EB, and the second hybridization was for 4 h using SDS buffer. Microarrays containing probes for 6,482 mouse genes were fabricated using the Mouse Known Gene SGC Oligo set, printed in duplicate, designed by Compugen, synthesized by Sigma-Genesys, and printed and supplied by the MRC Human Genome Mapping Project Resource Centre. After the hybridization and washing steps, microarray slides were scanned using the ScanArray ExpressHT system (PerkinElmer), and images for analysis were obtained using autocalibration with 100% laser power, a variable PMT, and a target saturation of 90%. Spot features were identified, poor quality spots were manually flagged, and intensity values were extracted using BlueFuse for microarrays version 2 (BlueGnome). Full details of the slide layout, culture conditions, detailed protocols, and primary extracted data files have been submitted to, and are publicly available in a MIAME-compliant form from, ArrayExpress (; experiment references E-MEXP-538 DJVaux_MEF_Lmnb1, E-MEXP-539 DJVaux_MEF_Rce1, and E-MEXP-540 DJVaux_MEF_Icmt). Intensity values, extracted using BlueFuse, were analyzed using BASE (). Only median fold ratio values with P < 0.05 using test were used for subsequent analysis. Cluster and TreeView (rana.lbl.gov/ EisenSoftware.htm) were used to generate the tree diagrams. For the functional cluster analysis, Gene Ontology IDs associated with genes that are differentially expressed in the three cell types were obtained using Ensembl MartView (). Ensembl KaryoView was used for mapping gene positions. The ordering of fold change ratios in selected groups of dysregulated genes was compared for different cell types and tested for significance using the Pearson product–moment correlation coefficient. To assess the significance of groups of genes identified as the intersection of independent experimental datasets, we drew genes at random from each dataset and counted the number of individual genes that appeared in both lists. This sampling was repeated 100,000 times and enabled us to assign a probability for the observed intersection subsets occurring by chance. The significance of the distribution of selected groups of genes was tested using a bootstrap method in which 16 genes were drawn at random from the list of genes meeting specific criteria and mapped onto the genome. The sampling was repeated 10,000 times, and the distribution of each of these 16-mer gene sets was assessed using predetermined test criteria. To assess apparent clustering of genes, this approach was used with the test criteria of three genes (out of the 16 selected in each trial) mapping within a 5-Mbp region as the definition of a cluster. Quantitative PCR was performed using a Rotor-Gene 3000 (Corbett Research) using the Platinum two-step qRT-PCR kit with SYBR green according to the manufacturer's instructions (Invitrogen). Primers were designed using the OligoPerfect designer (Invitrogen) and tested for single product generation in control end-stage PCR before qRT-PCR. The housekeeping gene β () was used as an internal standard for the qPCR verification of , , , , , and . Relative gene expression values were determined using the method (). Cells were fixed in 3:1 methanol acetic acid as described previously (; ; ) with the modification that the cells were grown and fixed as adherent monolayers rather than in suspension. Denaturation was performed for 4 min at 65°C in 70% deionized formamide/2× SSC. Mouse chromosome 18 and 19 paints labeled with Cy3 and FITC, respectively, were denatured according to the manufacturer's instructions (Cambio) followed by hybridization to coverslips for 16 h at 37°C in a humidified box. Coverslips were washed three times for 5 min at 45°C in 50% deionized formamide/2× SSC, washed two times for 5 min at 50°C in 1× SSC, and mounted in Mowiol supplemented with DAPI. Cells stained with Con A–Alexa Fluor 633 (Con A 633; Invitrogen) were incubated with 100 μg/ml of the Con A conjugate for 1 h at room temperature after the 1× SSC washing step and washed three times for 5 min in PBS before mounting. Cells were examined using a fluorescence microscope (Axioplan 2e; Carl Zeiss MicroImaging, Inc.) or a Radiance 2000 MP confocal laser-scanning microscope. The Con A and DAPI stains were used to identify cells with distorted nuclear morphology, and these cells were excluded from the analysis. Images were viewed in MetaMorph or ImageJ 1.33u. For chromosome position analysis, images were successively partitioned into five shells. For each partition, a length, , was determined such that all pixels with a distance less than away from the immediate outer partition (or the boundary if they constitute the outermost partition) are grouped as one partition, and such that the resulting partition has ∼20% of the overall area of the nucleus. The intensity associated with each shell is the sum of the “bright” pixels within that shell. Because of the discrete nature of the image, shells may not have exactly 20% of the total area; hence, the intensity corresponding to each shell is renormalized by the actual area of the shell. Unlike some conventional analyses, this method does not depend on prior definition of a centroid for the cell. Fig. S1 shows FLIP of YFP–lamin C expressed in WT, , , and cells. The figure shows that there is no significant difference between lamin C dynamics in the three different cell types. This indicates that the loss of lamin B1 or any of its processing steps does not affect the stability of lamin C interactions or the lamin C lamina. Table S1 provides a summary of the gene expression data (expressed as fold ratios) for , , and cells. Online supplemental material is available at .

As nascent polypeptides enter the ER lumenal space, N-linked glycans modify asparaginyl residues in the context Asn-Xaa-Ser/Thr (). This is catalyzed by oligosaccharyltransferase (OT), which transfers a preformed oligosaccharide unit, glucosemannoseGlcNAc (GMGn), from the lipid-linked oligosaccharide (LLO) glucosemannose N-acetylglucosamine-P-P-dolichol (GMGn-P-P-Dol). Synthesis of GMGn-P-P-Dol starts with dolichol-P (Dol-P), and sequentially requires 2 residues of GlcNAc from UDP-GlcNAc (the first transfer also forming the pyrophosphate linkage), 5 residues of mannose from GDP-mannose, 4 residues of mannose from mannose-P-Dol (synthesized from Dol-P and GDP-mannose), and 3 residues of glucose from glucose-P-Dol (synthesized from Dol-P and UDP-glucose). During biosynthesis, Dol-P and GMGn-P-P-Dol are oriented at the cytoplasmic and lumenal faces, respectively, of the ER membrane, with flipping of the key intermediate MGn-P-P-Dol (; ). Upon transfer of GMGn by OT, Dol-P-P is released, and is recycled to Dol-P for additional rounds of glycosylation (). N-linked GMGn is sequentially digested by ER glucosidases and mannosidases to generate high-mannose processing intermediates with functions in protein folding, quality control, and degradation (; ). Inhibition of LLO synthesis by tunicamycin, accumulation of LLO intermediates such as MGn-P-P-Dol caused by glucose deprivation, mutations affecting mannosyl precursor synthesis, and interference with ER-processing glycosidases all disturb ER homeostasis (). To minimize damage, mitigate the source of stress, and restore ER homeostasis to normal, the resulting ER stress activates a set of coordinated signals known collectively as the unfolded protein response (UPR). UPR signaling uses resident ER membrane proteins with lumenal stress-sensing domains that control activation of their respective cytoplasmic effector domains (). Of particular significance here, the cytoplasmic domain of the stress-sensor PKR-like ER kinase (PERK; ), also termed PEK (), is a kinase activated by transautophosphorylation that phosphorylates eukaryotic initiation factor (eIF)-2α. The resultant eIF2α-P interferes with translation initiation, which is sufficient to inhibit protein synthesis by 70–90% after robust ER stress. Importantly, translation attenuation by PERK reduces stress by diminishing the load of ER client protein (). Several lines of evidence suggest that metabolic deficiencies affecting GMGn-P-P-Dol synthesis or N-linked glycosylation might be compensated for by ER stress responses, implying homeostatic adaptation (). The goal of this study was to investigate the potential role of PERK in such adaptation. To do so, we took advantage of the fact that, for many cell types normally maintained in physiological (≥4 mM) glucose, brief incubations with 0.3–0.5 mM glucose hinder conversion of undermannosylated LLO intermediates to GMGn-P-P-Dol. This is distinguished from the glucose-starvation effect, which requires glucose-free medium and causes a rather discrete shift from GMGn-P-P-Dol to MGn-P-P-Dol (). We focused on our prior finding that dermal fibroblasts incubated 20 min in medium with 0.5 mM glucose accumulated MGn-P-P-Dol. Although the improperly glycosylated proteins that resulted were expected to compromise ER function, the treatment by itself was too brief to activate an ER stress response. Significantly, ER stress induced by dithiothreitol (DTT), thapsigargin (TG), castanospermine, azetidine-2-carboxylic acid, or geldanamycin all restored GMGn-P-P-Dol levels in the fibroblasts to normal (; ). The underlying mechanism was not determined, although regulated glycogenolysis was later proposed (). The brief treatments used argued against considerable contributions of UPR transcriptional programs (). In this study, we identify a surprisingly simple protective mechanism by which the ER stress response modulates GMGn-P-P-Dol synthesis and N-linked glycosylation. It is known that ER stress from aberrant GMGn-P-P-Dol production activates PERK. PERK is shown to reduce LLO consumption by attenuating synthesis of glycoprotein precursor polypeptides. This facilitates extension of undermannosylated intermediates to GMGn-P-P-Dol, restoring correct N-linked glycosylation. In this way, PERK balances glycoprotein synthesis with LLO flux. Most mammalian cells in conventional media containing physiological glucose concentrations synthesize GMGn-P-P-Dol efficiently, a state we operationally term unrestricted LLO synthesis. However, as we already mentioned, brief incubations with low glucose concentrations can hinder extension of undermannosylated intermediates. We categorize conditions such as these, which impede extension of LLO intermediates, as restricted LLO synthesis (). Consequently, the flux of LLO intermediates through the LLO pathway is reduced. We distinguish LLO flux from two other factors, LLO capacity (total LLO that can be synthesized, dependent on availability of Dol-P) and LLO consumption (dependent on transfer of glycan to polypeptide). Deficits in neither capacity nor consumption would be expected to cause LLO intermediates to accumulate. As presented below, restricted conditions were used to determine whether ER stress-activated PERK could compensate for reduced LLO flux. Unrestricted conditions were used to determine whether any such compensation was related to attenuated consumption of LLO, altered LLO capacity, or enhancement of LLO flux. LLOs in cultured cells are conveniently labeled with [2-H]mannose. Established solvent extraction procedures can then recover the entire [H] LLO pool (i.e., LLO intermediates plus GMGn-P-P-Dol), or HPLC can be used to examine glycans of individual [H] LLOs. Efficient incorporation of [H]mannose requires low-glucose medium, resulting in restricted or unrestricted LLO synthesis, depending on the cell type and the glucose concentration. As discussed in the Introduction, LLO synthesis in dermal fibroblasts is restricted with 0.5 mM glucose, and ER stress promotes extension of the accumulated LLO intermediates to GMGn-P-P-Dol. To simulate the contribution of translation attenuation by PERK during the ER stress response we used CHX. We found that mild treatment (4 μM; inhibiting protein synthesis by 25–30%; unpublished data) reduced accumulation of [H] LLO intermediates and greatly shifted the LLO profile in favor of [H]GMGn-P-P-Dol (). Treatment with 100 μM CHX (inhibiting translation by ∼50%) caused even greater enhancement, comparable to the effect of 2 mM DTT. Because CHX slows consumption of LLOs for protein N-glycosylation (), we surmise that 4–100 μM CHX may have allowed more time for extension of LLO intermediates, thereby compensating for reduced flux. This also implicated translation attenuation by PERK as a key factor in the ER stress response–mediating extension of undermannosylated intermediates. To assess PERK's potential contribution to LLO synthesis, the transfected CHO-K1 line “Fv2E-PERK” was used. This line expresses a cytoplasmic fusion protein with the PERK kinase domain joined to dual FKBP12-derived domains that bind AP20187, a cell-permeant bifunctional “dimerizer.” Such fusion proteins are normally monomers, and are inactive. Addition of dimerizer oligomerizes the Fv2E-PERK fusion proteins, resulting in transautophosphorylation and activation of the kinase domains. The kinase domains then phosphorylate Ser of eIF2α, inhibiting translation (). AP20187 caused graded, regulated inhibition of protein synthesis in Fv2E-PERK transfectants, but not untransfected CHO-K1 cells (Fig. S1, available at ). We used 0.5 nM AP20187 (inhibiting protein synthesis by ∼35%) to replicate translation attenuation expected to occur with moderate ER stress. Fv2E-PERK cells underwent restricted LLO synthesis when incubated in 0.3 mM glucose medium for 20 min (), with AP20187 () causing a robust effect on extension of [H] LLO intermediates to [H]GMGn-P-P-Dol. Moreover, N-linked glycans derived from [H]GMGn-P-P-Dol (labeled and in ) were sparse under restricted conditions (C), but were increased greatly by AP20187 (D). Thus, moderate translation attenuation by PERK was sufficient to drive extension of LLO intermediates to GMGn-P-P-Dol, and reestablish correct N-linked glycosylation. Compared with mouse embryonic fibroblasts (MEFs) expressing normal eIF2α (eIF2α), MEFs with alanine substitutions at Ser (eIF2α) have greatly reduced translation attenuation in response to ER stress because Ser is phosphorylated by PERK (). Incubation of eIF2α or eIF2α MEFs with 0.3 mM glucose for 20 min resulted in restricted LLO synthesis (). This provided an opportunity to formally demonstrate the role of eIF2α phosphorylation, and therefore translation attenuation, in stimulation of LLO intermediate extension by ER stress. We considered analogous experiments with PERK MEFs, but reasonable conditions causing restricted LLO synthesis were not identified (unpublished data). Rather, in these cells, glucose deprivation diminished GMGn-P-P-Dol without accumulation of intermediates, with reversal by CHX (Fig. S2, available at ). Consistent with prior results with dermal fibroblasts, treatments of eIF2α MEFs with DTT (25 min) or TG (30 min) promoted extension of [H] LLO intermediates (). However, ER stress did not enhance extension in eIF2α MEFs (), although treatment with 20 μM CHX () demonstrated that the LLO pathway in these cells could respond to translation attenuation. Splicing of XBP1 mRNA is mediated by IRE1, and is a quantitative measure of ER stress (). Assays for this reaction verified that TG and DTT caused robust ER stress with both MEF lines (). Ostensibly, PERK could aid LLO synthesis if a labile protein inhibited the pathway, or if a stimulatory protein was made upon attenuation of translation initiation (). However, another explanation comes from work by , which was extended by us (), showing that under unrestricted conditions of LLO synthesis, the translation inhibitors CHX and puromycin prevent synthesis of [H]mannose-labeled GMGn-P-P-Dol. This is because GMGn-P-P-Dol consumption is inhibited in the absence of nascent glycoprotein precursor polypeptides, and the GMGn-P-P-Dol pool does not turn over. Consequently, Dol-P is not regenerated, so new LLO cannot be made with [H]mannose. Thus, the results shown in – (under restricted conditions) may be explained by slowed LLO consumption, with more time for extension of undermannosylated intermediates to GMGn-P-P-Dol. Two types of measurements were required to determine whether ER stress slows GMGn-P-P-Dol consumption. For inhibition of GMGn-P-P-Dol synthesis, metabolic labeling with [H]mannose was used. In parallel, to detect unlabeled GMGn-P-P-Dol, we used fluorophore-assisted carbohydrate electrophoresis (FACE), by which glycans cleaved from LLOs are tagged at their reducing termini with the anionic fluorophore 7-amino-1,3-naphthalenedisulfonic acid (ANDS; ). The negatively charged ANDS-glycan conjugates can be separated by electrophoresis with a high-percentage polyacrylamide gel, and detected with ultraviolet light. FACE allows glycans from individual LLO species to be measured quantitatively, and can be used regardless of the medium's glucose concentration. CHO-K1 cells (and CHO-K1–derived Fv2E-PERK cells; see below) underwent unrestricted GMGn-P-P-Dol synthesis in 0.5 mM glucose medium (Fig. S2 C and not depicted). Synthesis of total [H] LLO in CHO-K1 cells with 0.5 mM glucose was inhibited by DTT and TG (), but no breakdown products were observed (Fig. S2 C). Importantly, no losses of GMGn-P-P-Dol were detected by FACE (; increases were apparent, and are explored in Fig. S2 B). PERK's activity was sufficient to block GMGn-P-P-Dol consumption because 10 nM AP20187 strongly inhibited both protein () and total [H] LLO () synthesis in Fv2E-PERK transfectants, but not in CHO-K1 cells, whereas GMGn-P-P-Dol detected by FACE was unaffected in all cases (). These results show that PERK's kinase activity is sufficient to inhibit GMGn-P-P-Dol consumption under unrestricted conditions, just as it is sufficient to drive extension of undermannosylated intermediates under restricted conditions (). The necessity of PERK was addressed because IRE1 has also been reported to attenuate translation in response to ER stress (). HeLa S3 cells were transfected with siRNA duplexes directed against distinct regions of PERK mRNA. The PERK-A duplex was ineffective for RNA interference, and was therefore used along with sham transfection as a negative control. The PERK-B duplex efficiently knocked down PERK mRNA (losses of 58 ± 3% with 5 h of transfection and 77 ± 1% with 16 h of transfection). Because some effects on cell viability were noticed with the 16-h transfection, subsequent experiments were done with 5-h transfections. DTT and TG each inhibited synthesis of protein and total [H] LLO under unrestricted conditions by about half in sham and PERK-A–treated cells (), whereas the PERK-B duplex fully prevented DTT- and TG-induced translation arrest and [H] LLO synthesis inhibition. For reasons that are unclear, DTT treatment tended to elevate total [H] LLO labeling in PERK-B–transfected cells above that in nonstressed cells, but the key point is that LLO synthesis was not inhibited. Splicing of XBP1 mRNA (inset) verified that the PERK-B duplex did not prevent DTT or TG from inducing ER stress. Because eIF2α−Ser can be phosphorylated by the kinase PKR in response to double-stranded RNA (), the potential concern over the use of RNA interference was addressed with MEFs bearing two normal PERK alleles (PERK) or two disrupted alleles (PERK). The results () confirmed those obtained by RNA interference. Synthesis of protein and total [H] LLO under unrestricted conditions were both strongly inhibited by ER stress inducers in the presence, but not the absence, of PERK. XBP1 mRNA splicing assays (unpublished data) verified that DTT and TG induced similarly robust ER stress in both MEF types. PERK's effects on LLO flux might be explained entirely by a compensatory reduction of LLO consumption because PERK activity was replicated by attenuating translation with either CHX () or activators of cytoplasmic eIF2α kinases (see the following section). As a more direct test of PERK's effect on LLO flux, we performed pulse-chase experiments with Fv2E-PERK cells. After incubation in the absence or presence of AP20187 or DTT, the cells were labeled for 2 min with [H]mannose in medium containing 1 mM glucose (which allowed GMGn-P-P-Dol to be made efficiently, yet permitted sufficient uptake of [H]mannose), and then chased in medium lacking [H]mannose for up to 10 min. Although several [H] species were detected, the only ones confirmed as LLOs (by sensitivity to tunicamycin and comparison with standards) were [H]MGn-P-P-Dol and [H]GMGn-P-P-Dol. Neither LLO was detected during the 2-min pulse (unpublished data). In untreated cells, both [H] LLOs were detected after 4 min of chase, and [H]GMGn-P-P-Dol was the only [H] LLO significantly detected after 10 min of chase (). By comparison, after activation of PERK with AP20187 or introduction of ER stress with DTT, [H]GMGn-P-P-Dol was the predominant LLO detected after 4 min of chase. These results suggest that PERK may have a direct effect on LLO flux, in addition to the aforementioned compensatory effect on consumption. In contrast, there is no evidence that ER stress affects LLO capacity because in CHO-K1 cells in 10 mM glucose undergoing unrestricted GMGn-P-P-Dol synthesis (FACE analysis; Fig. S2 B), GMGn-P-P-Dol quantity was not altered by ER stress (we did notice, however, that 0.5 mM glucose incubation caused an unexpected reduction of GMGn-P-P-Dol that was reversed by ER stress, perhaps because of reduced consumption). Specific activation of PERK's kinase activity in Fv2E-PERK cells with AP20187 also failed to alter GMGn-P-P-Dol content (). Eukaryotes contain multiple cytoplasmic eIF2α Ser kinases distinct from PERK (), suggesting an alternative way to modulate LLO biosynthesis. Arsenite (ARS) and diamide (DIA) induce transcription of the cytoplasmic stress marker HSP70 mRNA, but not the ER stress marker GRP78 mRNA (Table S1, available at ), which is a result opposite to that obtained with ER stress inducers (). By incubating dermal fibroblasts as described for , ARS and DIA greatly enhanced the extension of [H] LLO intermediates to GMGn-P-P-Dol (), and diminished N-linked glycoproteins with undermannosylated glycans (). Their effects rivaled those of DTT (). Because the effects of ARS and DIA on LLO synthesis correlated with their abilities to inhibit protein synthesis (Figs. S3 and S4, available at ), but not with changes in mannose uptake or hexose-phosphate metabolism (not depicted), the requirement for eIF2α-Ser phosphorylation was tested with eIF2α and eIF2α MEFs (as described for ). Disulfiram (DIS), which is another cytoplasmic stress inducer (Table S1), was included. All three agents inhibited protein synthesis in eIF2α MEFs () by at least half, and robustly promoted extension of [H] LLO intermediates (). However, their responses were quite disparate in eIF2α MEFs. ARS failed to appreciably affect protein () or [H] LLO () synthesis in eIF2α MEFs, showing that ARS acted mainly through an eIF2α-Ser kinase, with a specificity comparable to that of DTT (). DIA and DIS both inhibited protein synthesis () and promoted LLO extension () in eIF2α MEFs, but not as well as with eIF2α MEFs, indicating that they acted partly through eIF2α-Ser phosphorylation, and partly through a second means of translation attenuation. Collectively, translation attenuation by eIF2α-Ser kinase activity explains the effects of cytoplasmic stress inducers on LLO synthesis, and represents a merge point with the mechanisms of ER stress inducers. How is the synthesis of GMGn-P-P-Dol controlled? Evidence exists for regulation of specific reactions in the LLO pathway (; ) and for developmental induction of key enzymes (; ; ). However, mechanisms that might acutely regulate the pathway based on the availability of GMGn-P-P-Dol have been elusive. Because GMGn-P-P-Dol is ultimately needed for N-linked glycosylation, and hence ER homeostasis, processes that sense ER stress are candidate regulatory inputs for adjustment of GMGn-P-P-Dol synthesis. This study shows that decreased synthesis of polypeptide acceptors by activation of PERK reduces LLO consumption and, consequently, enhances extension of LLO intermediates, replenishing GMGn-P-P-Dol (). Because PERK is activated by extended periods of ER stress resulting from hindered GMGn-P-P-Dol synthesis, and hence, aberrant N-glycosylation (), PERK can balance glycoprotein precursor polypeptide synthesis with LLO pathway flux. Thus, in addition to reducing ER stress by lessening the load of client protein, this mechanism allows PERK to ensure proper N-glycosylation of the polypeptides that continue to be made. The importance of this synergy is emphasized by a recent study showing that maintenance of favorable diffusional properties in the ER lumen is much more dependent on efficient functioning of the lectin-chaperone system (which requires proper N-glycosylation) than the total load of polypeptide (). In addition to decreasing LLO consumption, pulse-chase experiments suggested that PERK stimulates LLO flux itself. The mechanism responsible for this is unclear at this time. Though used here to accumulate undermannosylated intermediates, glucose deprivation may also be a physiological cause of ER stress (). A 30-min reduction of glucose concentration to 2.5 mM (just below the typical fasting level of 4 mM) can cause significant accumulation of LLO intermediates in fibroblasts (). Although the brief (20 min) incubations in low-glucose media used here were insufficient to appreciably inhibit protein synthesis (unpublished data) or cause ER stress (; ), seminal studies showed that extended glucose deprivation can trigger an ER stress response, as well as interfere with protein glycosylation (; ). In our hands, incubation of dermal fibroblasts for 12 h in medium with 0.5 mM glucose (a restricted condition when used for only 20–30 min) triggered an ER stress response, and both GMGn-P-P-Dol synthesis and proper protein N-glycosylation were restored (). However, at such extended time-points, UPR-dependent transcription of LLO biosynthetic enzymes () and increased glucose transport () would be expected, as well as PERK, to participate in enhancement of GMGn-P-P-Dol synthesis. Other conditions that interfere with LLO synthesis (; ) include exposures to glucosamine and 2-deoxyglucose, which trigger ER stress responses (; ; ), and to tunicamycin (), which induces ER stress and causes PERK-dependent translation arrest (). In all cases, a compensatory role of PERK would be anticipated. The vigorous effects of only 20–35% translation attenuation on LLO intermediates were surprising, suggesting influence by translational variations within the physiological range (). The failure to detect acute ER stress effects on LLO synthesis in eIF2α MEFs suggests that, absent of translational control, there may be no other strongly stimulatory mechanisms in these cells during the first ∼30 min of the response. This also argues against GMGn-P-P-Dol metabolism being influenced by potential secondary effects of ER stress (such as misfolding of polypeptide acceptors or disruption of OT). LLO extension in ER stressed-dermal fibroblasts correlated temporally with loss of glycogen and elevation of glucosyl phosphates, suggesting that regulated glycogenolysis might elevate sugar precursor pools and drive LLO extension (). However, a direct link was not established. Given our current results, changes in glycogen metabolism do not appear to have major importance for regulation of LLO synthesis by robust ER stress in MEFs. The mechanism for PERK reported in this study, termed “translational balancing” in , does not involve complex ER stress-signaling pathways and may be analogous to regulation by other eIF2α kinases, notably HRI (). Iron deficiency hinders conversion of protoporphyrin IX to heme and releases HRI from its heme-inhibited state to phosphorylate eIF2α-Ser. This, in turn, reduces α- and β-globin chain synthesis to balance hemoglobin synthesis with heme. In addition to reducing undesired malfolded globin chains (), translational balancing should increase heme relative to protoporphyrin. Accordingly, translational balancing may be a simple, general mechanism by which eIF2α-Ser kinases adjust metabolic pathways whose end products interact with newly synthesized proteins. Translational control also has implications for the type I congenital disorders of glycosylation (CDG-I), which involve mutations in genes required for GMGn-P-P-Dol synthesis (; ), resulting in aberrant glycosylation of serum proteins. Fibroblasts from CDG-I patients exhibit several criteria of chronic ER stress, suggesting that their LLO defects may be partially offset by beneficial effects of the ER stress response (). CDG-I may be amenable to correction, as GMGn-P-P-Dol production is not completely impaired because of the presence of at least one partially active allele. Because Ib is the only treatable subtype of the 12 CDG-Is (a–l; ), the compensatory effects of DIS () are particularly interesting. DIS is a clinically approved drug used to discourage alcoholism (), and is innocuous unless alcohol is consumed. Though toxic, ARS also has a history of therapeutic use (). In preliminary experiments with CDG-Ia fibroblasts, we noted that DIS, ARS, and DIA all had some ability to restore synthesis of GMGn-P-P-Dol, although the effects were highly variable (unpublished data), advocating further development of such agents. In conclusion, we find that PERK can balance ER glycoprotein synthesis with flux through the GMGn-P-P-Dol pathway. Upon accumulation of LLO intermediates, aberrant N-linked glycosylation would create ER stress and activate PERK. PERK's kinase activity would then reduce the load of glycoprotein precursor polypeptides, slow LLO consumption, facilitate extension of LLO intermediates to GMGn-P-P-Dol, and reestablish correct N-linked glycosylation. All stress inducers were obtained from Sigma-Aldrich. An ARGENT Regulated Homodimerization kit containing AP20187 was a gift from Ariad Pharmaceuticals (). [2-H]mannose (15 Ci/mmol) and [H]leucine (164 Ci/mmol) were purchased from GE Healthcare. Cell culture media were obtained from Invitrogen, and sera were obtained from Atlanta Biologicals. CHO-K1 (), PERK-Fv2E–expressing CHO-K1 transfectants (), PERK and PERK MEFs (), eIF2α and eIF2α MEFs (), HeLa S3 (), and dermal fibroblasts (; ) were obtained and grown in the culture media described. However, to aid adhesion, MEFs were grown on standard 100-mm tissue culture dishes pretreated with 10 ml autoclaved 0.1% type B bovine gelatin (Sigma-Aldrich) for at least 1 h. After removal of the solution, the dishes were dried for at least 30 min. Cell cultures were incubated (see previous section) for 20–30 min (except for and ; 5 min) in media with 0.3–0.5 mM glucose containing 10% dialyzed FBS and [2-H]mannose. [H] LLOs were extracted with chloroform/methanol/water (10:10:3). Either the total LLO-associated tritium was measured by liquid scintillation spectroscopy or the [H] LLOs were treated with weak acid to release water-soluble glycans, which were then fractionated and detected by HPLC with in-line liquid scintillation spectroscopy (; ). The HPLC system resolves glycans on the basis of single-sugar differences, with the largest glycans eluting the latest. The proteinaceous pellets remaining after organic extraction were digested with pronase and N-glycanase (Calbiochem), and the released N-glycans were analyzed by HPLC (). For LLO pulse-chase studies with Fv2E-PERK cells, both [H]mannose and [H]glucosamine were evaluated, but only [H]mannose was deemed suitable. Conditions were optimized by varying the times and [H]mannose concentrations for pulse labeling. In most [H]mannose pulses, we detected multiple species eluting from the HPLC column earlier than 20 min, but these were disregarded because they were refractory to inhibition by tunicamycin (an inhibitor of LLO synthesis). For the experiment presented in , Fv2E-PERK cells were incubated in F-12 medium with 10% dialyzed FBS, 1.0 mM glucose, and 250 μCi/ml [H]mannose for 2 min. Labeling was then terminated by removal of [H] medium and addition of methanol (no chase), or cells were washed twice with prewarmed phosphate-buffered saline, and the incubation continued in the same medium, but without [H]mannose before terminating the reactions (chase). LLOs were recovered from methanolic suspensions in sequential chloroform/methanol (2:1) and chloroform/methanol/water (10:10:3) extracts, which were combined for recovery of all LLO species, and processed for HPLC as described in the previous paragraphs. LLO glycans from unlabeled cells were recovered by techniques described in the preceding section, coupled to ANDS, and analyzed with FACE oligosaccharide profiling gels. Gel images were acquired with a Fluor-S MultiImager (Bio-Rad Laboratories) using a 530DF60 filter. When necessary, individual ANDS conjugates were quantified with Quantity One software supplied with the scanner (). For clarity, some images were adjusted with brightness and contrast tools in PowerPoint 2003 (Microsoft), treating all data from a single gel identically. Cropping and joining of lanes in a single gel is indicated by vertical lines. For critical direct comparisons, only samples loaded on the same FACE gel were considered. This limited most experiments to duplicate determinations. Thus, we present original FACE data for the reader's inspection. Per 10 cells, 50% of the sample was usually loaded per gel lane. Incorporation of [H]leucine into total protein involved incubation in media with 10% dialyzed FBS and 5 μCi/ml [H]leucine for 5 min, collection of the material insoluble in 5% trichloroacetic acid (), and determination of tritium by liquid scintillation spectroscopy. Incorporation of 125 μCi/ml [S]methionine for 20 min was done exactly as described () by phosphorimager analysis of polyacrylamide gels. We have not noticed any differences in the validity or reliability of these two assays. siRNA duplexes targeting human PERK (accession no. ) were PERK-A (sense 5′-CAAGAGGAAGACAUCCUGCtt-3′, antisense 5′-GCAGGAUGUCUUCCUCUUGtt-3′) and PERK-B (sense 5′-UGGACCAUGAGGACAUCAGtt-3′, antisense 5′-CUGAUGUCCUCAUGGUCCAtt-3′), corre- sponding to coding region nucleotides 691–709 and 2,237–2,255, respectively (synthesized by the RNA Oligonucleotide Synthesis Core of University of Texas Southwestern Medical Center). Individual oligonucleotides were resuspended in 500 μl DEPC-treated HO (concentrations determined by OD), mixed, and diluted to 20 μM each in annealing buffer (100 mM potassium acetate, 30 mM Hepes-KOH, pH 7.4, and 2 mM magnesium acetate), heated for 3 min at 90°C, and incubated for 1 h at 37°C to form duplexes. HeLa S3 cells were plated in DME with 10% FBS without antibiotics in 60-mm dishes, and used at ∼30–50% confluence. Transfection of siRNAs was done with Oligofectamine (Invitrogen) according to the manufacturer's instructions. Cells were incubated with buffer only (“sham”) or with duplexes for 5 or 16 h. Cells were passaged once into the desired numbers of 100-mm dishes. 3 d after the initiation of transfection, cells were treated with the stress inducers indicated or used to harvest RNA. PERK mRNA was quantified by northern analysis and normalized to actin mRNA (). Total mRNA was isolated with a RNeasy Mini kit (QIAGEN). Splicing of XBP1 mRNA was assessed by RT-PCR (). PCR products representing spliced XBP1 (XBP1), unspliced XBP1 (XBP1), and a hybrid (formed during the chain reaction, composed of one strand each of XBP1 and XBP1, [XBP1]) were resolved by agarose gel electrophoresis. Fig. S1 shows the inhibition of protein synthesis in Fv2E-PERK cells by AP20187. Fig. S2 shows ancillary LLO analyses. Fig. S3 shows the inhibition of protein synthesis by cytoplasmic stress inducers. Fig. S4 shows the recovery from DIA treatment. Table S1 shows mRNA responses in dermal fibroblasts for cytoplasmic stress inducers. Online supplemental material is available at .

Early in the secretory pathway of eukaryotic cells, integral plasma membrane (PM) proteins are inserted into the lipid bilayer of the ER. The mechanisms enabling polytopic membrane proteins, composed of multiple transmembrane segments (TMSs), to integrate and fold in the ER membrane are not well documented. It is known that hydrophobic amino acids within N-terminally localized TMSs are sufficient to target membrane proteins to the ER, where they interact with and enter a protein-conducting channel, or translocon, with a hydrophilic core. The translocon is a highly conserved heterotrimeric integral membrane complex, i.e., the Sec61 complex in eukaryotes and SecY complex in bacteria (). The successful ability to crystallize membrane proteins has greatly enhanced the understanding of two basic processes associated with membrane protein biogenesis. First, the structure of the SecY translocon has been elucidated (), which revealed that the protein-conducting channel is contained within SecY, the largest translocon subunit and the Sec61 homologue. Despite unresolved questions regarding the structure of an active Sec61/ SecY translocon in vivo (), the protein-conducting channel, although perhaps quite flexible, is too small to accommodate multiple TMSs (; ). Second, the structures of bacterial PM transporters LacY and GlpT, each composed of 12 TMSs (; ), have clearly shown that interactions between TMSs within the lipid phase of the membrane are not limited to immediately flanking TMSs. Consequently, to fold properly, N-terminally localized TMSs that partition into the membrane of partially translated proteins must await the insertion of C-terminal TMSs. Together, these advances are consistent with the idea that polytopic membrane protein biogenesis occurs essentially in two discrete stages, i.e., membrane insertion and folding (; ; ). To bridge these stages, and to account for the fact that polytopic membrane proteins cannot fold in a strict cotranslational manner, the existence of chaperone-like proteins has been postulated (; ; ). The identification of membrane-localized chaperones in bacteria () and yeast () provided strong support for this notion. Consistent with the two-stage model of biogenesis, these chaperones do not appear to be necessary for the insertion of TMSs (; ) but, nonetheless, they appear to interact early during the translocation process to enable their substrate proteins to obtain native conformations (; ). Although it remains to be determined, this novel class of chaperones may prevent TMSs of polytopic membrane proteins that do not normally interact in the mature protein from engaging in nonproductive interactions with flanking TMSs, or with other ER components, as they sequentially partition into the membrane. In the yeast , the recently characterized membrane-localized chaperones exhibit a striking degree of substrate specificity. The best studied of these chaperones, Shr3, is specifically required for proper folding of amino acid permeases (AAPs; ). AAPs comprise a conserved family of 18 proteins with 12 TMSs (). AAPs localize to the PM in an Shr3-dependent manner (), where they function to transport amino acids into cells. In the absence of Shr3 chaperone activity, AAPs aggregate, forming large molecular weight complexes (), which are excluded from coatomer protein II transport vesicles (, ). Consequently, AAPs accumulate in the ER of null mutant strains. Shr3 has two well-defined domains, a membrane domain composed of four TMSs and a hydrophilic C-terminal domain. The chaperone activity of Shr3 is associated with the membrane domain; the hydrophilic C-terminal domain is dispensable for function (). The steady-state levels of AAPs are similar in both wild-type and -null mutant cells (; ; ). In wild-type cells, AAPs are degraded in the vacuole (; ); the fate of aggregated AAPs in mutants has not been investigated. PM and secretory proteins are subject to ER quality-control systems that operate to ensure that only properly folded and assembled proteins exit the ER and enter into subsequent stages of the secretory pathway (). The ER lumen contains numerous factors that assist folding reactions (; ). Terminally misfolded, or incompletely assembled soluble or integral membrane proteins are degraded in a process referred to as ER-associated degradation (ERAD; ). ERAD substrates are passed through or extracted from the membrane, ubiquitylated, and ultimately presented to cytoplasmic proteasomes for degradation. Several ERAD pathways have been defined (; ; ). These pathways are composed of multimeric protein complexes organized around two membrane-localized E3 ubiquitin ligases, i.e., Doa10 and Hrd1 (; ). These pathways appear to differentially recognize ERAD substrates, and thus far, all characterized ERAD substrates exhibit a strong dependence on either Doa10 or Hrd1 pathways. With respect to polytopic membrane proteins, little is known regarding how ERAD surveillance mechanisms differentiate between proteins in the process of membrane insertion and folding from proteins that are terminally misfolded. The potential involvement of membrane-localized chaperones in ERAD has not been examined. We have investigated the chaperone activity of Shr3 from two perspectives. Using split Gap1 constructs, we examined the temporal requirement of Shr3 during membrane insertion and folding. Our results indicate that Shr3 interacts with an N-terminal fragment composed of TMSs I–V and maintains it in a conformation that enables it to functionally assemble with a coexpressed C-terminal fragment with TMSs VI–XII. Second, we assessed the potential role of Shr3 in quality-control mechanisms that monitor AAP folding. We report that in the absence of Shr3, Gap1 aggregates are redundantly targeted to Doa10- and Hrd1-dependent ERAD pathways. Cells lacking these ERAD pathways exhibit restored amino acid uptake capacities. Thus, given sufficient time, AAPs are able to attain functional conformations independent of Shr3. These findings highlight the intimate link between folding and degradation during the biogenesis of polytopic membrane proteins. Previous studies examining the ER retention of AAP in mutants have relied on microscopic evaluation (; ). The intracellular distribution of Gap1 in wild-type and mutant strains was examined by subcellular fractionation of whole cell lysates. In lysates from wild-type cells (, left) the majority of Gap1 cofractionated with the PM marker protein Pma1 (fractions 2 and 3) and the late Golgi marker Kex2 (peak fraction 6). The distribution of Gap1 was distinct from the ER marker Wbp1, which exhibited the highest concentration in fraction 5. Gap1 migrated as two bands: a top band present in PM-containing fractions and a bottom band that was observed in fractions containing internal membranes. Gap1 is known to be posttranslationally modified by phosphorylation, which correlates with increased amino acid uptake (). Also, phosphorylation is important for proper targeting of Gap1 from the Golgi to the PM and for preventing Gap1 down-regulation (). The pattern of mobility we observed likely correlates with Gap1 becoming phosphorylated as it progresses through the secretory pathway on its way to the PM. In lysates from cells (, left) the bulk of Gap1 (bottom band form) cofractionated with the ER marker Wbp1 (fractions 5 and 6), and there was minimal overlap with the Golgi marker Kex2 (peak fraction 7). Interestingly, a small amount of Gap1 (top band form) cofractionated with Pma1, the PM marker protein (fractions 2 and 3), suggesting that a portion of Gap1 is able to fold independently of Shr3. However, based on our extensive phenotypic analysis of mutants, including quantitative amino acid uptake assays, microscopic analysis (indirect immunofluorescence and Gap1-GFP), and biochemical approaches, the fraction of folded and correctly localized Gap1 in mutant cells is too low to confer detectable Gap1- dependent phenotypes (; , ; ; ). These results demonstrate that Shr3 is required to enable Gap1 to efficiently exit the ER and correctly localize to the PM. In early studies regarding polytopic membrane protein biogenesis, it was found that truncated N- and C-terminal fragments of bacteriorhodopsin could incorporate into membranes and assemble to attain catalytic active conformations (). Subsequent to this pioneering work, there have been numerous examples of functional “split” polytopic membrane proteins reported in the literature, including bacterial, yeast, and mammalian proteins. To more fully understand the chaperone-dependent folding of AAPs, and to investigate the regions of Gap1 that require Shr3, we sought to take advantage of the possibility that Gap1 could be functionally expressed as a split protein. The two-dimensional structure of Gap1 is depicted in . The N and C termini and even-numbered hydrophilic loops (L2–L10) are oriented toward the cytoplasm. The odd-numbered loops (L1–L11) are luminally oriented during biogenesis and extracellular when Gap1 is correctly localized to the PM. With the exception of L5 (35 amino acids) and L7 (22 amino acids), the odd-numbered loops are short and are composed of <10 amino acids. Consequently, the bulk of the non–membrane-associated amino acids are on the cytoplasmic side of the membrane. We constructed alleles encoding two matched pairs of N- and C-terminal Gap1 fragments (). The first pair, encoded by the and alleles, expresses nonoverlapping Gap1 fragments truncated between TMS V and VI. The allele encodes an N-terminal Gap1 fragment composed of the first 263 amino acid residues, and the allele encodes a C-terminal fragment of Gap1 with a methionine residue placed immediately before residues 264–602. The second pair, encoded by the and alleles, expresses partially overlapping Gap1 fragments truncated between TMS VI and VII. The encoded fragment is composed of the N-terminal 320 amino acid residues, and the allele has a methionine residue placed before residues 311–602. Each of these alleles is expressed under control of the endogenous promoter. The truncated Gap1 alleles were separately introduced into the -null mutant strain FGY15, and the growth of transformants was assessed to determine whether the individually expressed N- and C-terminal fragments exhibited Gap1 transport activity (). Gap1 mediates the uptake of toxic -amino acids (); consequently, cells lacking Gap1 activity are able to grow in media containing -histidine (dilution series 1), whereas cells expressing functional full-length Gap1 cannot (dilution series 2). As expected, cells expressing the truncated N- and C-terminal fragments of Gap1 grew (dilution series 3–6), indicating that the individually expressed truncated proteins are nonfunctional. Next, plasmids encoding gap1 TM1-5 and gap1 TM6-12 or gap1 TM1-6 and gap1 TM7-12 were introduced together as pairs into FGY15 (). The growth of both strains carrying the matched pairs of N- and C-terminal fragments was inhibited by -histidine (; dilution series 3 and 4). All strains grew well in the absence of -histidine (, top); thus, the trivial explanation for the lack of growth, e.g., inhibitory secondary effects resulting from the expression of truncated Gap1 constructs, can be ruled out. These results demonstrate that in both instances, the truncated N- and C-terminal fragments assemble into an active split Gap1 that correctly localizes to the PM and facilitates -histidine uptake. A slight difference in growth of the split Gap1 expressing strains was noted; the growth of the split Gap1TM1-5/TM6-12 expressing strain (dilution series 3) was nearly as poor as the strain expressing the intact full-length allele (dilution series 2), whereas the growth of the split Gap1TM1-6/TM7-12 expressing strain was better (dilution series 4). The noticeable difference suggests that split Gap1TM1-5/TM6-12 is more active. To facilitate subsequent biochemical analysis of the C-terminal fragment, we inserted a thrice-reiterated myc epitope immediately preceding the stop codon of the allele. We examined the functionality of the C-terminal–tagged split gap1 TM6-12myc construct by coexpressing it with gap1 TM1-5 in strain FGY15 (). Growth of this strain (dilution 12) on media containing -histidine was poor and similar to the untagged Gap1TM1-5/TM6-12 expressing strain (dilution 11). Because growth was inhibited by -histidine, the myc-tagged split Gap1 was able to assemble together with the N-terminal fragment into an active Gap1. Finally, we tested whether the untagged and tagged split Gap1 proteins require Shr3 for functional expression (; dilution series 5–8 and 13–16, respectively). The matched plasmid pairs were introduced into an strain (FGY135). Strain FGY135 is unable to take up -histidine and grows well because of the lack of Shr3 (dilution series 5 and 13), even when full-length is reintroduced on a plasmid (dilution series 6 and 14). The transformants expressing either of the functional untagged and tagged split Gap1 proteins exhibited robust growth (dilution series 7–8 and 15–16, respectively). The strict requirement for Shr3 indicates that the split proteins, similar to full-length Gap1, require the chaperone activity of Shr3 to attain functional conformations. To further investigate the role of Shr3 in Gap1 biogenesis, we examined whether split Gap1TM1-5/TM6-12 and Gap1TM1-5/TM6-12myc aggregate in the absence of Shr3. Cell lysates from strains FGY15 () and FGY135 () expressing full-length Gap1, untagged split Gap1 TM1-5/TM6-12, or tagged Gap1 TM1-5/TM6-12myc were solubilized in the presence of dodecyl-β--maltopyranoside (DM), and soluble proteins were separated by blue native PAGE (BN-PAGE; ). Consistent with our previous results (), monomeric full-length Gap1 was readily extracted from membranes prepared from wild-type cells (, lanes 1 and 3), but not from membranes prepared from cells (, lanes 2 and 4). In the absence of Shr3, Gap1 aggregates form high molecular weight complexes that migrate as a diffuse smear. Similar results were obtained when we examined untagged split Gap1 (); monomers of the N-terminal fragment were readily extracted from membranes containing Shr3 (, lanes 1 and 3), whereas fewer monomers were extracted from membranes lacking Shr3 (, lanes 2 and 4). We were unable to detect intact assembled split Gap1 monomers, indicating that DM extraction disrupts N- and C-terminal fragment association. The decreased efficiency to solubilize monomeric gap1 TM1-5 in mutant cells was not a consequence of reduced levels of gap1 TM1-5; both and strains expressed similar amounts of the N-terminal fragment (, bottom). Antibodies to native Gap1 recognize the N-terminal hydrophilic sequence preceding TMS I (anti-NT-Gap1; ). To investigate the aggregation state of the C-terminal fragment, we coexpressed Gap1 TM1-5/TM6-12myc in and cells. In a manner entirely consistent with the untagged split Gap1, the N-terminal fragment was readily extracted from membranes containing Shr3 (, lanes 1 and 3), and fewer monomers were extracted from membranes lacking Shr3 (, lanes 2 and 4). Next, we assessed whether monomers of the C-terminal fragment could be extracted (). Similar to the N-terminal fragment, C-terminal monomers were more readily extracted from membranes with Shr3 (, lanes 1 and 3) compared with membranes lacking Shr3 (, lanes 2 and 4). However, the difference between gap1 TM6-12myc monomer levels in extracts from and cells was not as striking as for the N-terminal fragment. The and strains expressed similar amounts of the N- and C-terminal fragments (, bottom). These results indicate that in the absence of Shr3, the N- and C-terminal fragments of split Gap1 are prone to aggregation. To investigate the temporal requirement of Shr3 during the assembly of split Gap, we examined the aggregation state of individually expressed N- and C-terminal fragments. In cells expressing only the N-terminal fragment, gap1 TM1-5 monomers were readily extracted from membranes with Shr3 (, lanes 1 and 3), but fewer monomers were extracted from membranes lacking Shr3 (, lanes 2 and 4). These results, analogous to what we observe when membranes with coexpressed N- and C-terminal fragments are solubilized (), indicate that Shr3 interacts directly with the N-terminal fragment and prevents its aggregation. In contrast to the N-terminal fragment, we were unable to find conditions to efficiently extract gap1 TM6-12myc monomers from membranes in cells expressing only the C-terminal fragment (). Only small amounts of monomeric gap1 TM6-12myc were extracted, and importantly, the presence (lanes 1 and 3) or absence (lanes 2 and 4) of Shr3 did not affect the levels of soluble C-terminal monomers. These results indicate that in the absence of the first five N-terminal TMSs of Gap1, Shr3 is unable to associate with and prevent aggregation of C-terminal domain of Gap1 containing TMSs VI–XII. The clear Shr3 dependence of the N-terminal fragment is consistent with Shr3 interacting early during the biogenesis of Gap1, an interaction that is likely to occur before the partitioning of all 12 TMSs into the lipid phase of the ER membrane. The observed aggregation of full-length and truncated fragments of Gap1 is the likely consequence of misfolding, which accounts for Gap1 retention and accumulation in the ER of mutants. Despite the accumulation of misfolded proteins, mutants do not exhibit an activated ER unfolded protein response (). This may be explained by the fact that each of the TMSs of Gap1 are correctly inserted in the membrane independently of Shr3 (), and that the luminally exposed hydrophilic loops are rather short (). Thus, it is likely that Gap1 aggregates do not expose sequences that are recognized by Kar2, the luminal Hsp70 BiP homologue and key regulator of the unfolded protein response (). Prototrophic mutants grow as well as wild-type cells, even under conditions when AAP expression is induced (), suggesting that mutants are not negatively affected by the accumulation of AAP aggregates. To examine how mutants cope with AAP aggregates, we examined the turnover of Gap1 using pulse-chase analysis (). Similar to previous reports (; ), the half-life of Gap1 in wild-type cells was ∼40 min (, solid circles). The half-life of Gap1 in mutants was similar to wild-type cells (, solid circles). Thus, despite the fact that Gap1 is differentially localized and present in different forms, i.e., PM and monomeric in versus ER and aggregated in , Gap1 is efficiently degraded. The results suggest that AAP aggregates are degraded by ERAD. To test this possibility, we initially examined the degradation of Gap1 in mutants with an impaired capacity to degrade proteins in the vacuole (). In wild-type cells, Gap1 degradation was markedly impaired (, open squares), clearly demonstrating that in the presence of Shr3, Gap1 exits the ER and is degraded in the vacuole. This is consistent with the well-characterized trafficking pathways underlying the turnover of PM proteins (). In contrast, Gap1 was efficiently degraded in cells (, open squares), indicating that Gap1 aggregates are degraded independently of vacuolar hydrolases. This finding is consistent with AAP aggregates being degraded by ERAD. We examined this directly by constructing and strains carrying single- and double-null alleles of the genes encoding ER-localized E2 ubiquitin-conjugating enzymes Ubc6 and Ubc7. Ubc6 and Ubc7 are required for ubiquitylating most characterized luminal and transmembrane ERAD substrates (). As expected, in cells, the rate of Gap1 degradation was unaffected in , , or strains (). However, in strains lacking Shr3, the rate of Gap1 degradation was diminished in the presence of either or mutations (). In comparison to , the mutation exhibited a more pronounced stabilizing affect. Gap1 exhibited the greatest stability in the strain carrying both mutations. The clear dependency on Ubc6 and Ubc7 demonstrates that aggregated Gap1 is an ERAD substrate. Doa10- and Hrd1-dependent ERAD pathways are able to recognize and ubiquitylate membrane protein substrates. However, to date, all characterized ERAD substrates appear to be preferentially ubiquitylated by either Doa10- or Hrd1-dependent pathways. Thus, these E3 ligases appear to have distinct and nonredundant roles (). To determine which ligase is primarily responsible for ubiquitylating misfolded and aggregated Gap1, we constructed strains carrying - or -null alleles. Unexpectedly, the stability of Gap1 increased only slightly and to the same extent in both of these strains (). These results suggested that both Doa10- and Hrd1-dependent pathways might function in a redundant manner to ubiquitylate Gap1 aggregates. This possibility was examined by following the degradation of Gap1 in an strain carrying both - and null alleles. In this strain, Gap1 was significantly stabilized, indicating that with respect to ubiquitylating Gap1 aggregates, Doa10 and Hrd1 are equally effective and, indeed, functionally redundant. Thus, multiple ERAD pathways are involved in the degradation of Gap1 aggregates. We sought to test whether our findings regarding the turnover of Gap1 can be applied to other AAPs and chose to investigate the turnover of Agp1 (). We previously showed that Agp1 aggregates and is retained in the ER of cells lacking Shr3 (). Using a simplified experimental approach relying on cycloheximide, we found that, in wild-type cells, Agp1 is primarily degraded in the vacuole (); Agp1 was stabilized in a mutant, whereas degradation was unaffected by the combination of mutations. In cells lacking Shr3, the stability of Agp1 was dependent on ERAD (). In these cells, the turnover of Agp1 was largely Pep4 independent and, rather, was Ubc6 and Ubc7 dependent. As was found for Gap1, mutations alone stabilized Agp1 aggregates almost as efficiently as double mutations (not depicted). Also, both Doa10- and Hrd1-dependent pathways redundantly targeted Agp1 for degradation; aggregates were degraded in and single mutants with equal efficiency but were stable in double mutants. The data demonstrate that Agp1 aggregates are degraded in a manner identical to Gap1 aggregates. Little is known regarding surveillance mechanisms that monitor the folding of nonglycosylated polytopic membrane proteins, such as AAPs (). Although assumed to exist, mechanisms that distinguish between partially folded and misfolded states have not been identified, and the potential role of membrane-localized chaperones in such quality-control mechanisms has not been examined. We suspected that the chaperone activity of Shr3 would have a critical role in discriminating between these two states and perhaps function to shield partially folded AAPs from being prematurely targeted for degradation by ERAD pathways. To evaluate this possibility, we monitored the growth of strains with impaired ERAD using two growth-based assays that provide an extremely sensitive measure of the activity and, hence, the folding state of three AAPs, i.e., Gap1 () and Agp1 and Gnp1 (). We reasoned, based on our finding that low levels of Gap1 can apparently fold and exit the ER and reach the PM independently of Shr3 (), that if Shr3 normally prevents precocious degradation of AAPs, then mutations that impair ERAD would enable more AAPs to fold properly and lead to increased amino acid uptake. Consistent with this notion, we observed that in comparison to mutants with intact ERAD pathways (; dilution series 2), strains carrying mutations that inactivate both Doa10- and Hrd1-dependent ERAD pathways display increased sensitivity to -histidine and the toxic proline analogue azetidine-2-carboxylate (AzC; ; combinations of or ; dilution series 5 and 8, respectively). As observed in our studies analyzing the degradation of Gap1 and Agp1 ( and ), we noted that strains carrying only the mutation were almost as sensitive to amino acid analogues as the double mutant strains (; dilution series 4). The data indicate that under conditions of impaired ERAD, cells have increased levels of properly folded and functional AAPs in their PMs. We tested this possibility directly by examining the aggregation state of Gap1 in the ERAD-defective strains. Membranes from , , , and strains were solubilized in the presence of DM, and soluble proteins were separated by BN-PAGE (). In comparison to membranes from the Shr3 wild-type strain, we observed diminished levels of Gap1 monomers in membranes lacking Shr3 (, compare lanes 1 and 2). Substantially more (two- to threefold) Gap1 monomers were extracted from membranes prepared from strains carrying or mutations (, compare lanes 3 and 4 with lane 2). The increased ability to solubilize Gap1 monomers accounts for the increased amino acid uptake in strains with impaired ERAD (). In summary, our data support the idea that in addition to facilitating folding of AAPs, Shr3 functions to shield partially folded AAPs from being prematurely targeted for degradation. Yeast possesses highly specialized ER membrane-localized chaperones (). In this paper, we have experimentally addressed two questions regarding the chaperone activity of Shr3. First, is there a temporal requirement for the chaperone activity of Shr3 during the membrane insertion and folding of AAPs? Second, does the chaperone activity of Shr3 affect surveillance mechanisms that monitor the status of membrane protein folding? In answering these questions, we document that Shr3 has a central role in the biogenesis of AAPs, and our results illuminate the tight coupling between folding and quality-control mechanisms operating during biogenesis of polytopic membrane proteins. We probed the temporal requirement of Shr3 during AAP folding by exploiting the fact that independently coexpressed truncated N- and C-terminal fragments of Gap1 can fold and assemble forming functional split Gap1 proteins (). Importantly, the functional assembly of N- and C-terminal fragments of Gap1 exhibited a strict dependence on Shr3. This finding enabled us to directly analyze whether Shr3 maintains the N- and/or C-terminal fragments in productive assembly competent conformations. Consistent with our previous data regarding full-length Gap1 (), we found that the N-terminal fragment exhibited an increased propensity to aggregate in membranes isolated from cells lacking Shr3, and importantly, its aggregation was not affected by the presence or absence of the C-terminal fragment (; and ). In marked contrast, the ability to solubilize monomers of the C-terminal fragment exhibited not only a dependence on Shr3 () but, more significantly, on the presence of the N-terminal fragment (). These data are consistent with a model in which Shr3 interacts early during the membrane insertion of AAPs. Furthermore, as the N- and C-terminal fragments are likely to individually insert into the membrane, Shr3 is able to maintain the N-terminal fragment in a conformation that enables the C-terminal fragment to assemble with it. These observations raise many interesting questions regarding how these individual fragments, perhaps in analogy to subunits of multimeric membrane protein complexes, find each other before their functional assembly. During translocation, exclusively hydrophobic TMSs rapidly partition into the lipid phase of the membrane (). In contrast, less hydrophobic TMSs containing charged or polar residues partition into the membrane less readily and are retained in proximity to the translocon or to translocon-associated proteins, e.g., TRAMs (). Recent results regarding the membrane insertion of aquaporin-4 are consistent with the view that individual TMSs exhibit distinct requirements during translocation (). After individually passing through a single entry site in Sec61, several TMSs of aquaporin-4 were found to interact with secondary peripheral sites on Sec61. These results suggest that the translocon may transiently retain certain TMSs to facilitate early folding events and to control their release into the membrane (). The intrinsic chaperone-like activity of Sec61 translocon may thus suffice to facilitate correct folding of many polytopic membrane proteins. However, our results regarding membrane-localized chaperones indicate that complex, or larger, polytopic proteins require the assistance of additional chaperones to fold. An appealing hypothesis, consistent with AAP aggregation observed in the absence of Shr3, is that Shr3 facilitates the partitioning of TMSs of AAPs containing charged or polar amino acid residues as they emerge from the translocon. According to this hypothesis, Shr3 may physically shield charged or polar residues within TMSs, enabling them to more rapidly partition into the membrane and fold, thereby preventing them from engaging in nonproductive interactions. Gap1 possesses five TMSs that have a single charged residue (). Although we have not systematically investigated the role of these charged residues, in several instances they are conserved, and mutations that exchange noncharged residues into their positions abolish catalytic activity (unpublished data). Consistently, a mutant Gap1 protein with a lysine replacing the conserved glutamate residue (E300) in TMS VI is inactive and retained in the ER (). In cells lacking Shr3, both Gap1 and Agp1 aggregates are efficiently recognized by ERAD surveillance systems that redundantly activate Doa10- and Hrd1-dependent pathways (, , and ). In general, our results are consistent with the recently proposed unifying concept for ERAD (). However, to our knowledge, AAP aggregates are the first substrates that target to both Doa10- (ERAD-C) and Hrd1- (ERAD-M/L) dependent pathways with equal efficacy. Our data indicate that these pathways function in parallel to degrade AAP aggregates. However, it is important to note that AAP aggregates may in fact be composed of a nonhomogeneous mix of kinetically linked folding intermediates, which raises the possibility that ERAD pathways preferentially degrade discrete folding intermediates. Previous reports regarding the organization of ERAD pathways suggested that membrane proteins with misfolded cytoplasmic domains are substrates of the Doa10 pathway, whereas soluble secretory proteins and membrane proteins with misfolded luminal or membrane domains are substrates of the Hrd1-dependent pathway (; ; ; ). Although it remains to be determined, the involvement of the Doa10 pathway can be explained by the fact that the bulk of the non–membrane-associated amino acids of AAPs are on the cytoplasmic side of the membrane; presumably the folding of cytoplasmic domains are dependent on the proper folding of TMSs. With respect to the Hrd1 pathway, we suspect that in the absence of Shr3, misfolded TMSs of AAPs are recognized. Given the high level of sequence homology and dependence on Shr3, the other members of the AAP protein family are likely to be similarly degraded. Consequently, our findings substantially augment the number of known substrates of the Doa10- and Hrd1-dependent pathways. We have found that AAPs can attain native structures, albeit inefficiently, independent of the chaperone activity of Shr3 ( and ). Our finding that mutations that impair ERAD enhance functional expression of AAPs highlights the intimate link between folding and degradation. There are two possible explanations for the ability of AAPs to fold independently of Shr3. First, a small portion of AAPs may fortuitously fold and attain native conformation as they insert in the membrane. In this case, other chaperones present in the ER membrane may substitute for the lack of Shr3, or the inherent chaperone-like functions associated with the Sec61 translocon may suffice. Alternatively, AAP aggregates may not be solely composed of terminally misfolded proteins. This raises the possibility that low levels of aggregated AAPs are able to reenter a productive folding pathway. Because the topological orientation of each TMS of Gap1 is correctly fixed as they insert into the membrane (), AAPs initiate folding at a point that is already quite constrained (). Thus, if provided sufficient time (e.g., in cells with impaired ERAD), aggregated AAPs may spontaneously rearrange from a kinetically trapped conformation into their native lower energy conformations. The idea that Shr3 interacts with TMSs containing charged residues is consistent with it having a role in surveillance mechanisms that monitor folding. Previous work has demonstrated that charged residues within membrane-spanning segments provide key signals for targeting misfolded proteins for ERAD. In the case of unassembled α-subunits of the T cell receptor, which are retained and degraded in the ER, the degradation determinant was mapped to two basic amino acid residues in the TMS (), and the placement of a single charged residue in a TMS of a cell surface protein caused it to be retained and degraded in the ER (). Thus, by shielding charged amino acid residues, Shr3 not only facilitates folding but also prevents incompletely folded AAPs from being targeted for degradation. Furthermore, the fact that Shr3 does not itself exit the ER (), and newly synthesized AAPs copurify with Shr3 (), suggests that Shr3 retains actively folding AAPs in the ER. Thus, Shr3 may function analogously to the well-characterized N-linked glycosylation-based quality-control system, which provides temporal cues regarding folding of glycoproteins, preventing their premature exit out of the ER and targeting of folding intermediates to ERAD (). Our findings underscore the importance of ER membrane-localized chaperones in governing the stability of membrane proteins. In mammalian cells, insulin-induced gene (INSIG) proteins appear to have chaperone-like activity that influences the sterol-dependent degradation of 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase and ER retention of Scap (for review see ). When sterol levels are high, INSIG proteins bind sterol-sensing membrane domains of HMG CoA reductase and Scap. As a consequence of binding, HMG CoA reductase is targeted to ERAD and Scap is prevented from exiting the ER, two events that lead to decreased cholesterol biosynthesis and uptake. In yeast, the INSIG homologues Nsg1 and Nsg2 associate with HMG CoA reductase, facilitating the folding of the sterol-sensing membrane domain and inhibiting Hrd1-dependent degradation (). Finally, we note striking similarities between the inefficient folding of AAPs in cells lacking Shr3 and the folding of the mammalian Cl channel cystic fibrosis transmembrane conductance regulator (CFTR). The folding of native CFTR is inefficient (): only 25% of synthesized CFTR folds properly, and the remaining 75% is degraded by multiple ERAD mechanisms (). The similarities to the situation in yeast suggest that CFTR folding occurs in the absence of dedicated Shr3-like chaperones; consequently, CFTR folding intermediates are more or less constitutively and prematurely targeted for degradation. The ability of small molecules to act as chaperones, and the temperature-dependent characteristics of CFTR folding supports this notion (; ; ). Also, the folding intermediate that accumulates as a result of the disease-causing ΔF508 mutation is kinetically trapped. Interactions with Hsp90, which are positively and negatively modulated by cochaperones, determine the fate of this misfolded intermediate (). The down-regulation of cochaperone Aha1 enables ΔF508 CFTR to fold, exit the ER, and become functionally expressed in the PM. Presumably, reduced levels of Aha1 circumvent surveillance mechanisms that normally target ΔF508 CFTR for ERAD. It will be interesting to continue to exploit the yeast system described here to dissect out further mechanistic details regarding the quality-control surveillance systems that monitor the folding of polytopic membrane proteins. Yeast strains and plasmids are listed in . FGY206 was created by transforming PLY123 with a PstI–BamHI DNA fragment containing from pGR172 (). FGY205 was created by transforming PLY123 with a HindIII–HindIII DNA fragment containing from pTX33 (). PLY127 was transformed with a EcoRI–XhoI DNA fragment containing from pAS173 (provided by A. Sachs, University of California, Berkeley, Berkeley, CA), and a Ura transformant was propagated on medium containing 5-fluoroorotic acid (FOA), resulting in the unmarked strain FGY217. PLY127, FGY205, FGY206, and FGY217 were transformed with a linear EcoRI–SalI DNA fragment containing from pPL288 (); Ura transformants were propagated on FOA-containing medium; and strains FGY212, FGY209, FGY210, and FGY219, each carrying the unmarked deletion, were obtained. FGY257 was constructed by transforming FGY212 with a SphI–SalI DNA fragment containing (). Strains JKY29 and JKY39 were constructed by deleting the entire sequence of in FGY212 and FGY257 with a PCR-amplified cassette (primers F-doa10D and R-doa10D). JKY36 and JKY37 were constructed from FGY205 and FGY209 by deleting the coding sequence of with a PCR-amplified cassette (primers F-ubc7D and R-ubc7D). Plasmid pJK92 was created by inserting a 3.4-kb SalI–NotI fragment encoding Gap1 from pPL247 into SalI–NotI–restricted pRS317 (). Plasmids expressing split Gap1 constructs under the control of the endogenous promoter were created by homologous recombination in yeast. Plasmids pJK97 () and pJK96 () expressing N-terminal fragments of Gap1 were created by introducing Bsu36I–BglII–restricted pPL247 together with primer pairs F:1-5TM/R:1-5TM and F:1-6TM/R:1-6TM, respectively. Plasmids pJK98 () and pJK99 () expressing C-terminal fragments of Gap1 were created by cotransforming BsiWI–BsaBI restricted pJK92 with primer pairs F:7-12TM/R:7-12TM and F:6-12TM/R:6-12 TM, respectively. A thrice-reiterated myc epitope was inserted at the C terminus, immediately before the stop codon, creating plasmid pJK100; 3×myc was PCR amplified from plasmid pPL329 (primers F-Gap1TM6-12–3×MYC/R-Gap1TM6-12–3×MYC) and cotransformed together with SphI–SalI–restricted pJK99. Standard media, YPD (yeast extract, peptone, dextrose) and SD (synthetic complete), were prepared as described previously (). Ammonium-based synthetic complex () and minimal media containing urea (SUD), proline (SPD), and allantoin (SAD) as the sole nitrogen source were prepared as described previously (; ). Minimal media were supplemented as required. Media were made solid with 2% (wt/vol) bacto Agar (Difco). Cells were grown overnight in SD to an OD of 2–3, harvested, washed once in water, and resuspended in SUD to a starting OD of 0.1. Cells were grown at 30°C and harvested when cultures reached an OD of 0.8, and protein extracts were prepared and fractionated on 12–60% sucrose gradients essentially as described previously (). 1-ml fractions were collected from the bottom of the gradients using a Fraction Recovery System (Beckman Coulter). Fractions 1 and 2 and 10 and 11 were separately pooled, and proteins from equal aliquots of the nine resulting fractions were concentrated by TCA precipitation. Proteins were separated by SDS-PAGE and blotted onto nitrocellulose membranes. Radiolabeling and immunoprecipitations were conducted essentially as described previously (; ). In brief, cells grown in SAD (plus uracil) to an OD of 0.5–0.8 were used to inoculate fresh SAD (plus uracil) at an OD of 3. Cultures were incubated for 30 min before the addition of S-methionine (25 μCi/OD cells; GE Healthcare). Cells were labeled for 30 min, after which an aliquot of 100× chase solution (25 mM -methionine and 25 mM -cysteine) was added. At the times indicated, 1-ml aliquots of culture were placed in Eppendorf tubes containing 60 μl ice-cold lysis solution (1.85 M NaOH and 7% β-mercaptoethanol), rapidly mixed by rigorous vortexing, and after a 10-min incubation on ice, 60 μl of 50% TCA was added. Precipitated proteins, pelleted by centrifugation for 15 min at 12,000 , were washed once in 60 μl of 1 M Tris, pH 7.5, and solubilized in 0.5% SDS for 10 min at 37°C. Proteins were diluted with TNET buffer (1% Triton X-100, 50 mM Tris, pH 7.4, 150 mM NaCl, and 5 mM EDTA) to a final volume of 0.6 ml. Insoluble material was removed by centrifugation for 30 min at 12,000 . Rabbit anti-Gap1 antibodies were added (1:1,500), and samples, continuously mixed by inversion, were incubated overnight at 4°C. 40 μl of a 12.5% (vol/vol) suspension of protein A–Sepharose CL-4B beads (GE Healthcare) was added to each sample, and incubations were continued for 3 h at 4°C. Immunoprecipitated material was collected by centrifugation and washed three times with TNET buffer and once with TNET without Triton X-100. Precipitated proteins were eluted by incubation for 10 min at 39°C in 2× SDS-PAGE sample buffer. Eluted proteins were separated by SDS-PAGE in a 10% gel. Gels were fixed in glacial acetic acid/methanol/HO (10:20:70), rinsed briefly in water, and dried. Radiolabeled proteins were detected and quantified by phosphorimaging (Fujix Bio-Image Analyzer BAS1500; Fuji). Cultures grown at 23°C in SAD (plus uracil) to an OD of 0.8–1 were treated with 400 μg/ml cycloheximide. At the times indicated, 1-ml aliquots were removed to Eppendorf tubes containing 250 μl ice-cold lysis solution (1.85 M NaOH), rapidly mixing by rigorous vortexing, and after a 10-min incubation on ice, 250 μl of 50% TCA was added. Precipitated proteins, pelleted by centrifugation for 15 min at 12,000 , were washed once in 100 μl of 1 M Tris, pH 7.5, and solubilized in 2× SDS-PAGE sample buffer. Proteins were resolved by SDS-PAGE and analyzed by immunoblotting. BN gels and whole cell protein extracts were prepared as previously described (). Protein extracts from cells grown in SAD at 25°C were solubilized at 4°C for 35 min in the presence of DM at the concentrations indicated. Extracts containing full-length Gap1 and split Gap1 fragments were separated using 4–15% and 4–20% gradient gels, respectively. High molecular weight marker proteins (GE Healthcare) were used as standards.

γ-Secretase catalyzes the final cleavage of the amyloid precursor protein (APP), liberating the amyloid β (Aβ) peptide () probably causing Alzheimer's disease (; ). Besides APP, γ-secretase cleaves other biologically important substrates like Notch and many more type I membrane proteins and is involved in that way in development, neurogenesis, and cancer (; ; ). γ-Secretase is a multisubunit complex consisting of presenilin (PS), nicastrin (NCT), PS enhancer-2 (PEN-2), and anterior pharynx defective-1 (APH-1; ). These proteins are minimally required to generate a functional active complex. It is far from fully understood where in the cell the complex is assembled and activated and what the sequence of events are leading to full assembly. Nevertheless, given its complexity and indispensable role in intramembrane proteolysis, mechanisms must be present to tightly control and regulate this assembly. The prevailing hypothesis suggests that assembly is initiated by the NCT–APH-1 subcomplex and is followed by the sequential incorporation of PS and PEN-2 or, alternatively, the PS–PEN-2 subcomplex (; ; ; ). Because all four hydrophobic components are synthesized and translocated in the ER, this is the earliest membrane compartment where assembly can theoretically be initiated, but whether the complex is already activated in this organelle remains a question. This has been formulated originally as the spatial paradox (): although most of the endogenous PS1 resides in the ER and in the intermediate compartment (IC; ) in coat protein complex I (COPI)–coated areas, (), activity seems mainly to occur at the cell surface or close to the cell surface in endosomal compartments (; ). The slow maturation of NCT in the Golgi (; ) and the relative long turnover of this and other γ-secretase complex partners once they are incorporated in the complex indicate that these components are actively retrieved from Golgi compartments or retained in the ER. In support of this, suggested that a hydrophobic stretch in the C terminus of PS1 constitutes a retention signal for unassembled PS1 in the ER, the basis of which remains to be investigated. Moreover, absence of one component results in destabilization or reduced expression of other components and impaired maturation (NCT) or PS endoproteolysis (; ; ), predicting a substantial role of early biosynthetic compartments in complex assembly. The intriguing question, therefore, is what the molecular mechanism is that governs active recycling of γ-secretase components. This knowledge is of utmost importance, as it may converge with the regulation of the stepwise assembly of the complex in early compartments (). Some proteinous components have been identified recently that modulate γ-secretase activities, including CD147 (), TMP21 (), and phospholipase D1 (). However, they emerged as binding partners of mature complexes, acting likely in later trans-Golgi network and endosomal compartments. On the contrary, not a single binding partner or factor has been identified that mediates through its interaction early steps of complex assembly. In this paper, we show the interaction of NCT with Rer1p (retrieval to ER 1 protein), a membrane receptor operating in Golgi retrieval. Binding to Rer1p requires critical residues in the transmembrane domain (TMD) of NCT also used by APH-1, thereby lifting a first important corner of the veil in the molecular regulation of complex assembly in early compartments. This interaction highlights the requirement of secondary ER quality control in complex assembly. Several signal motifs in membrane proteins mediate retrieval from IC or Golgi compartments, including dilysine and -arginine motifs in intracellular or the KDEL motif in luminal domains (; ). However, none of these “classical” motifs are identified in individual γ-secretase components. We therefore hypothesized that individual components are retrieved through interaction with membrane proteins bearing such motifs. Candidate retrieval proteins were tested for their interaction with NCT in HeLa cells by coimmunoprecipitation using anti-NCT mAb 9C3 or affinity-purified pAb B59.4. Because CHAPS extraction preserves the integrity of the complex, mAb 9C3 pulled down NCT with all γ-secretase complex members (). Although only little hRer1p coimmunoprecipitated from CHAPS extracts, these levels increased dramatically in Triton X-100 extracts, i.e., under conditions that disrupt the interaction between γ-secretase components. Other retrieval proteins like BAP31 () and the KDEL receptor did not interact with NCT, underscoring the specificity of the pull down. In a reciprocal experiment, anti-hRer1p pAb pulled down small amounts of endogenous immature NCT from Triton X-100 HeLa and mouse embryonic fibroblast (MEF) extracts (). Adding the chemical cross-linker dimethyl 3,3′-dithiobispropionimidate (DTBP) before solubilization and coimmunoprecipitation even slightly increased NCT levels in the precipitate. The specific interaction with immature NCT is even more apparent from CHO extracts that express higher levels of endogenous immature NCT (). Other γ-secretase complex components like PS1 C-terminal fragment (CTF; unpublished data) and PEN-2 did not bind, underscoring the selectivity of the interaction. Similar coimmunoprecipitations were performed upon overexpression of either NCT or PEN-2 in HeLa cells (). Although overexpression levels were similar for both proteins, anti-hRer1p pAb pulled down NCT but not PEN-2. The amount of immature NCT bound to hRer1p increased dose dependently with increasing concentrations of DTBP (0.5–3 mM; ). Collectively, we can conclude that hRer1p specifically interacts with immature NCT, most likely in its uncomplexed status. Rer1p is an integral membrane protein of ∼23 kD, including four TMDs with both termini facing the cytosol. The gene emerged originally from a yeast screen for mutants defective in retention of Sec12, a protein involved in COPII-coated vesicle formation (; ). Since then, several yeast proteins have been identified next to Sec12p () that use the Rer1p-dependent retrieval system, including subunits of the translocation machinery, like Sec63p and Sec71p (), and the iron transporter subunit Fet3p (). The latter is of particular interest for the current study, as yRer1p only binds Fet3p in its unassembled state. Like in yeast, epitope-tagged hRer1p localized to the Golgi in HeLa cells (). However, at the endogenous level, hRer1p only partially colocalized with the cis-Golgi marker GM130, as demonstrated by high-resolution confocal microscopy (). Instead, a far better colocalization was observed with ERGIC-53, indicating that, at steady state, hRer1p essentially localizes in the vesicular tubular elements of the IC. This was confirmed by subcellular fractionation (unpublished data) and by incubating cells at 15°C, a condition that blocks transport from the IC. Under this condition, hRer1p remained entirely colocalized with ERGIC-53–positive structures (). Interestingly, highly overexpressed hRer1p accumulates in the ER (; unpublished data), indicating that hRer1p requires other (limiting) sorting determinants enabling it to cycle rapidly between ER and Golgi. In conclusion, the localization of hRer1p in the IC predicts a functional role for the hRer1p–NCT interaction in the transport of NCT between early biosynthetic compartments. To better understand the physiological relevance of this interaction, we first decided to analyze in detail the binding characteristics. As the Rer1p interactions in yeast are mediated through the TMD of the reported binding proteins, we generated a series of NCT deletion constructs to delineate possible interaction sites, focusing on the TMD of NCT (). We transiently coexpressed these constructs with hRer1p in HeLa cells followed by coimmunoprecipitation using anti-hRer1 pAb (). Deleting the intracellular domain (NCTΔIC) considerably reduced binding compared with full-length (FL) NCT. Removing the TMD in addition (NCTΔ[TMD + IC]) further interfered with binding (), whereas interaction with hRer1p was completely abolished when an additional short hydrophobic stretch before the putative TMD region was deleted (NCTΔ[long TMD + IC]; ). On the other hand, NCTΔEC, a construct lacking the entire ectodomain, displayed a relatively increased interaction compared with wild-type (WT) NCT, indicating that the ectodomain is dispensable for Rer1p interaction (). We next swapped the TMD of NCT and from another unrelated type I transmembrane protein, namely, telencephalin (TLN; ), which does not bind hRer1p () and coexpressed the two chimeric proteins each with hRer1p. hRer1p interacted with TLN harboring the TMD of NCT (TLN/TMD) but not with NCT containing the TMD of TLN (). However, in the latter, some binding was still observed, further confirming the contribution of regions adjacent to the NCT TMD to hRer1p interaction. In conclusion, this experiment demonstrates that the NCT TMD is required and sufficient for efficient binding to hRer1p. In yeast it has been shown that interaction with yRer1p (e.g., Sec71p) critically depends on precisely spaced polar residues within the TMD. We identified in the NCT TMD several polar residues that generate a charged patch on one side in an α-helical wheel model (, left, orange residues). When aligned, it becomes clear that the spacing of polar residues in the NCT TMD is similar to that seen in other Rer1p binding partners, including Sec12p and Sec71p (, right, red boxed residues). Polar residues can also be found in the TMD of the other γ-secretase components, e.g., in the second TMD of PEN-2; however, these lack the critical spacing. This striking structural resemblance of the NCT TMD with the TMD of other Rer1p binding partners urged us to investigate the polar residues of NCT in more detail. Therefore, polar residues were mutated individually or together to leucine residues using site-directed mutagenesis (). All mutants were coexpressed with hRer1p in HeLa cells and tested for their ability to coimmunoprecipitate with anti-Rer1p pAb. To determine the extent to which each mutation affects binding to Rer1p, FL-NCT and NCT/TMD were used as positive and negative controls for binding, respectively (). Single (S/L) or double mutations (TY/LL and ST/LL) already substantially interfered with hRer1p binding, indicating that hRer1p may use mechanisms similar to yeast Rer1p in the recognition of binding partners. Excitingly, the same polar residues, S, T, and Y, but also T and G, have been shown to be critical for the interaction of NCT with APH-1 (). We reasoned that a competition at the level of this binding motif between APH-1 and Rer1p could provide a molecular mechanism that couples ER-Golgi trafficking of NCT (through its interaction with Rer1p) to subcomplex assembly with its cognate APH-1 partner. As such, this would constitute a secondary ER quality control for the multisubunit assembly of the γ-secretase complex. To test the requirement of these residues, we generated NCT TMD mutants with three to all five residues mutated to leucine (TGS/3L, TGST/4L, and TGSTY/5L). In our coimmunoprecipitation assay, these mutants displayed a gradual decrease of interaction with hRer1p (), indicating that these five amino acids are required for hRer1p binding. Our data demonstrate that Rer1p and APH-1 require the same structural features (or the same residues) in the TMD that mediate binding to NCT and that both proteins would be in competition for binding to this motif. If this holds true, binding of Rer1p to NCT should increase in the absence of APH-1. We therefore used MEFs deficient for PS, NCT, APH-1a, or APH-1a,b,c () to check this intriguing possibility. As shown in , a substantial increase of mRer1p coimmunoprecipitating with NCT is observed in single APH-1a and triple APH-1 KO MEFs, respectively, strongly arguing in favor of our interpretation that APH-1 and Rer1p compete for binding to the same site in NCT. The absence of any effect in PS1 and -2 KO MEFs moreover underscores the specific interaction of Rer1p with NCT and not with any other γ-secretase component. These findings extend our previous conclusion that Rer1p not only mediates the retrieval of NCT from Golgi compartments but also may couple this retrieval of NCT with the assembly of the γ-secretase complex in early biosynthetic compartments. To test this, we investigated the effect of changing Rer1p expression levels on the trafficking of NCT and γ-secretase complex formation. The functional relevance of the interaction between NCT and Rer1p becomes apparent after knock down of hRer1p in HeLa cells. Specific RNAi duplexes efficiently suppressed hRer1p levels up to 80–90% at 48 h after transfection (). Interestingly, we noticed a slight shift in the mobility of mature glycosylated (endoglycosidase H [endoH] resistant) NCT. This hypoglycosylation suggests that Rer1p knockdown affects the residence time of NCT in early compartments. To test this, we performed metabolic pulse-chase experiments that confirmed the higher mobility of newly synthesized mature NCT after Rer1p down-regulation (). Moreover, quantification of the ratio of mature over total NCT at different time points revealed that Rer1p knockdown considerably delayed mature glycosylation. This is opposite from what is expected if Rer1p mediates retrieval of NCT from cis-Golgi or IC. After all, ablation of such a retrieval mechanism would result in higher kinetics of NCT passaging to post-Golgi compartments. Our results, therefore, suggest that NCT is controlled by additional sorting mechanisms that become apparent after scaling down the Rer1p-dependent retrieval. This is confirmed in NCT MEFs stably expressing NCT/TMD, i.e., an NCT variant bearing the TMD of TLN (see also ). In contrast to NCT, NCT/TMD failed to become mature glycosylated, indicating that it is fully retained in the ER. Moreover, NCT lacking its TMD fails to restore any aspect of γ-secretase, including PS1 endoproteolysis, APH-1 and PEN-2 stabilization, complex formation, and activity (). On the other hand, introducing the NCT TMD into TLN (TLN/TMD) sharply decreased mature glycosylation, suggesting that Golgi passage of TLN/TMD was substantially delayed as compared with TLN (). Excitingly, Rer1p knockdown (72 h) partially restored mature glycosylation, demonstrating that the decrease in TLN/TMD glycosylation is directly mediated through interaction of Rer1p with the NCT TMD. Hence, although the NCT TMD mediates an Rer1p-dependent retrieval of NCT in early compartments, other yet-to-be-determined domains in NCT confer more ER retention independently of Rer1p. Finally, we explored the effect of Rer1p knockdown on the post-Golgi localization of NCT applying cell surface biotinylation. Surprisingly, Rer1p down-regulation resulted in substantially more mature (though hypoglycosylated) NCT, but not the transferrin receptor, at the cell surface (). This raises the question of how a longer residence time and slower kinetics in pre-Golgi compartments still results in increased surface expression. Interestingly, the co-increase of other γ-secretase components, like PS1 N-terminal fragment (NTF) and APH-1a, fosters the idea that Rer1p-dependent and -independent trafficking of NCT in early compartments may be directly linked to the degree of γ-secretase complex assembly. To explore this, we applied blue native PAGE (BN-PAGE) to study the effect of Rer1p expression on complex formation. Rer1p was down-regulated using RNAi (48 h) in HeLa, WT, and PS1 and -2 KO MEFs or, alternatively, up-regulated by transient overexpression (36 h) in HeLa cells (). In mock-transfected cells, NCT immunostaining typically reveals four bands on BN-PAGE. The ∼140-kD band is stained with only anti-NCT antibody, whereas the other two bands between 150 and 440 are subcomplexes of NCT–APH-1 and NCT–APH-1–PS1-CTF, respectively, and the ∼440-kD band is the mature complex. When Rer1p is down-regulated, the relative amount of mature γ-secretase complex is clearly increased both in HeLa cells and in WT MEFs, in line with the cell surface biotinylation data discussed in the previous paragraph. In addition, we observe a decrease of the NCT–APH-1 subcomplex and an increase of the NCT–APH-1–PS1-CTF subcomplex. This suggests that in the absence of Rer1p, the NCT–APH-1 subcomplex is more rapidly converted to more complete complexes. This was confirmed in an overexpression experiment showing that high expression of Rer1p in HeLa cells resulted in relatively lower levels of mature complexes and an increase of subcomplexes (). Because altering the levels of Rer1p affects the NCT–APH-1 subcomplexes, we hypothesized that Rer1p interacts with NCT at a very early stage in γ-secretase complex assembly. The simplest interpretation of our data is that the Rer1p–NCT interaction titrates the amount of NCT available for binding to APH-1. Further support for this idea comes from an experiment in PS1 and -2 KO MEFs, in which no mature complex is generated. Because Rer1p down-regulation results in a sharp decrease of remaining NCT–APH-1 subcomplexes, and no PS and almost no PEN-2 is present in these cells, Rer1p must interfere at the level of the NCT–APH-1 subcomplex. In conclusion, the longer residence time in early compartments as a result of Rer1p knockdown () appears to also promote complex assembly (). Hence, we provide strong evidence that the regulation of full complex assembly is tightly linked with ER-Golgi recycling. Consequently, complex assembly may ultimately mask Rer1p-dependent and -independent retrieval/retention motifs, enabling complexed NCT to escape quality control in early compartments. Hence, increased complex formation due to Rer1p knockdown may therefore also explain the increased surface expression of γ-secretase components, as these are stable and long-lived components (). If hRer1p is rate limiting for the assembly of the complex, it becomes important to know whether it is also rate limiting for γ-secretase activity. To test this, APP-C99, a direct γ-secretase substrate, was overexpressed in HeLa cells together with hRer1p or in conjunction with down-regulation of hRer1p, followed by metabolic labeling and immunoprecipitation of newly produced Aβ (). Significantly more Aβ was secreted from hRer1p down-regulated cells, whereas the reverse effect could be observed when hRer1p was overexpressed. However, the possibility that these changes in Aβ production are indirectly caused by an altered trafficking of APP-C99 due to a change in hRer1p levels cannot be excluded. Therefore, direct γ-secretase activity was tested in a cell-free assay (). Here, cell extracts were mixed with recombinant flag-tagged APP-C99 (). More APP intracellular domain (AICD; ) was produced from membranes generated from cells that were down-regulated for hRer1p, arguing for a higher total γ-secretase activity in these cells. In conclusion, the change in mature γ-secretase complex levels caused by altering the expression of hRer1p as observed by BN-PAGE correlated well with the changes in γ-secretase activity, indicating that hRer1p regulates γ-secretase activity. In this study, we identify Rer1p as a novel limiting factor in the stepwise assembly of the γ-secretase complex. Rer1p binds NCT in a region that also involves the binding with other γ-secretase components, notably, PS1 and APH-1. We also show that Rer1p is a resident protein of the IC/cis-Golgi, indicating that Rer1p acts as a retrieval receptor for ER-Golgi recycling of NCT. We propose that it negatively regulates the availability of NCT for assembly in the NCT–APH-1 subcomplex. By interfering with the initial steps of complex assembly, Rer1p expression controls total levels of γ-secretase complexes and, hence, activity in the cell. The evidence that Rer1p plays a key role in complex assembly is, in the first place, based on the identification of important binding requirements in the TMD of NCT. Although short regions adjacent to the TMD contribute to binding, especially the polar residues within the NCT TMD are crucial, and mutating one or more of them affects binding to Rer1p to a large extent. These residues are also important for the interaction of NCT with APH-1 () and indicate that both proteins compete for binding to NCT. This is strongly supported by our findings that Rer1p binding to NCT is substantially increased in APH-1 KO MEFs. The presence of polar residues alone, however, is not sufficient for interaction, as the second TMD of PEN-2 bears three polar residues but fails to interact with Rer1p (). The additional structural feature becomes apparent when we project the TMD of NCT in a helical wheel. Now the critical residues form a hydrophilic patch on one side of the α-helix, implying that their spacing in the TMD is critical for binding. When aligned, the polar residues within the TMD of NCT flank a highly hydrophobic core domain, and this spaced distribution is also encountered in the TMD of other Rer1p-binding proteins, like Sec12p, Sec71p, and Fet3p. Moreover, in the case of Sec71p-Rer1p, the length of this hydrophobic core is indeed essential for binding (). Collectively, it can be easily envisaged, also in our study, that mutating any of the flanking polar residues expands the length of the hydrophobic core as well as decreases the size of the hydrophilic patch, both accounting for a decreased interaction with Rer1p (). Our study not only extends the findings in yeast but also proves that the mechanism Rer1p uses to target TMDs is conserved from yeast to mammals. Mechanistically, the binding of Rer1p to NCT may decrease its availability to bind APH-1 and form the initial NCT–APH-1 subcomplex. Conversely, the binding of APH-1 can sterically mask Rer1p interaction, allowing the NCT–APH-1 subcomplex to escape the Rer1p-dependent retrieval mechanism. However, to get a full complex assembled, mechanisms must exist that secure the NCT–APH-1 binding, preventing it from shifting back to Rer1p interaction. Because the NCT–APH-1 subcomplex likely constitutes the earliest step in assembly, it may act as a scaffold for the sequential association of PS1 and PEN-2 (; ). PEN-2, eventually in coordination with APH-1 (), acts to stabilize and promote PS1 endoproteolysis. Here, the binding of the C terminus of PS1 to the TMD of NCT () could provide a molecular mechanism to “lock” the NCT–APH-1 interaction into a maturing complex. The incorporation of PS1 could definitively prevent Rer1p from binding to NCT and reversing the interaction with APH-1. This idea is supported by the effects of modulating Rer1p expression on complex formation. Overexpression of Rer1p predicts a more active deprivation of NCT for APH-1 binding and a slowing down of complex formation, as indicated by a decrease in mature complexes. On the other hand, Rer1p knockdown may facilitate the binding of APH-1 to NCT, thereby shifting the equilibrium and promoting a transition to more complex formation (). The fact that we do not see the NCT–APH-1 subcomplex accumulating in these cells is in agreement with the very rapid association of PS1 (and PEN-2) with NCT–APH-1 and fast transition to mature complexes. Furthermore, we demonstrate that Rer1p knockdown in PS1 and -2 KO MEFs results in substantially lower levels of the NCT–APH-1 subcomplexes. Apparently, the combined absence of active recycling by hRer1p and the stabilizing PS1–PEN-2 interactions cause rapid degradation of both NCT and APH-1 under those conditions. Although we have no proof, we find it probable that NCT and APH-1 in these cells are degraded, not unlike what is observed in yeast deficient for Rer1p () and Ftr1p (; ). Here, unassembled Fet3p, a bona fide target of Rer1p, is more rapidly transported to the vacuole for degradation. We show that endogenous Rer1p resides at steady state in the IC/early cis-Golgi, indicating that Rer1p contributes in complex assembly during ER-Golgi recycling. This localization is, however, dynamic, as Rer1p readily exits the ER in COPII vesicles in a cell-free ER-budding assay (unpublished data). This is in agreement with its rapid recycling and function as a retrieval receptor in yeast and mammals (; ). Interestingly, in γ-secretase–deficient MEFs (PS1 and -2 or APH-1 KO), NCT still exits the ER (unpublished data), and the fact that it remains fully immature can only be explained by an active retrieval mechanism preventing it from passing through the Golgi. That Rer1p is involved herein is indicated by its effects on complex formation outlined in the Results section. Moreover, the observed slow maturation of NCT () indicates that NCT retrieval is a repetitive process, increasing or even titrating its chance to bind the cognate APH-1. Also, PS1 has a long half-life and is concentrated over COPI-coated areas of the IC, indicating that it is actively retrieved (). Except for an ER-retention sequence in the C terminus (), no retrieval motifs are recognized in PS1, suggesting that PS1, like NCT, also depends on retrieval receptors. If so, Rer1p is most likely not involved here because we did not coimmunoprecipitate PS1 (and also PEN-2) with Rer1p antibodies. A potential candidate is TMP21, a member of the p24 family of retrieval receptors that was found to interact with PS1 (). Surprisingly, the functional relevance of this interaction was apparently not related to the established role of TMP21 in protein transport in the early secretory pathway. Nevertheless, at this stage, our study unequivocally demonstrates that γ-secretase complex assembly extends from ER to IC and cis-Golgi compartments and is at least under the active control of a Rer1p-dependent recycling mechanism. Our findings may, therefore, explain the existing controversy on the subcellular location of γ-secretase assembly ranging from the ER () to Golgi/trans-Golgi network compartments (). Although the yeast homologue Rer1p has been known for quite some time (; ; , , ), NCT is the first mammalian membrane protein identified that utilizes this Rer1p-dependent recycling. Irrespective of the topology, Rer1p binds to similar structural motifs within the TMD of identified targets (, , ), except for Mns1p (). A surprising observation, however, is that depending on the ligand the Rer1p retrieval mechanism serves different purposes. Rer1p functions in the retrieval of escaped ER-resident proteins, such as Sec12p, Sec71p, and Sec63p or can recycle components of the vesicle fusion machinery, like Sed4p. Strikingly, Rer1p also interacts with proteins that are not yet assembled in their corresponding multisubunit complexes, for example, Fet3p () and NCT (this study). Fet3p, a subunit of the yeast iron transporter, is only retrieved by Rer1p as long as it is not assembled with its cognate subunit Ftr1p (). These authors proposed that interaction with Ftr1p conceals Rer1p from binding to Fet3p, thereby triggering escape from the retrieval mechanism. This mechanism constitutes a secondary ER quality-control system () that couples the assembly stage of multimeric protein complexes to forward transport. Control at the stage of complex assembly has also been demonstrated for other multisubunit complexes, including major histocompatibility complex class II, K channels, and cystic fibrosis transmembrane conductance regulator (). In addition to detecting the TMDs of unassembled subunits, a common alternative mechanism is based on masking of Arg-based ER-sorting motifs. However, these motifs are localized in cytosolically exposed regions of individual subunits (). In any case, the quality-control systems ensure that only correctly assembled complexes could leave their place of synthesis and reach the compartments where their action is required. Although the Rer1p-dependent retrieval of NCT is most reminiscent of that of unassembled Fet3p, it appears to be more complicated. Indeed, and aside from the Rer1p-dependent retrieval, we show that NCT trafficking is counterbalanced by an Rer1p-independent ER retention–based mechanism mediated through its ecto- or intracellular domain (). Therefore, NCT recycling in early compartments is subject to at least a dual ER quality-control system that serves to control the residence time for NCT in early compartments. This time window defines the chance to interact with properly folded APH-1 and, thus, the amount of initial NCT–APH-1 subcomplex formation. Excitingly, our findings clearly demonstrate that these secondary ER quality controls provide the cell with a mechanism to tightly control the quantitative levels of the γ-secretase complex and, hence, activity in distal compartments, such as the cell surface and endosomes (; ). In conclusion, our data establish Rer1p as a limiting factor and transport regulator in initial complex assembly through binding to NCT. Moreover, we demonstrate the feasibility to interfere with complex assembly and activity by altering Rer1p expression levels. Controlling Aβ production via modulation of γ-secretase is an important therapeutic strategy in Alzheimer's disease. Herein, PS1 has caught most of the focus partly because it harbors the catalytic site of the complex. Additionally, the fact that proteolysis and binding/docking occurs in spatially distinct domains has increased at least the chance to develop specific inhibitors (). Our findings, together with the role of NCT as a substrate receptor (), definitely augment the critical role of NCT in the complex. The identification of the binding motif for Rer1p in NCT and the competition with APH-1 may open opportunities for drug development yet to be explored. Rabbit pAb against human PS1-NTF (B14.5), mouse PS1-NTF and -CTF (B19.2 and B32.1, respectively; ), NCT (pAb B59.4; ), APH-1a (B80.2; ), PEN-2 (B95.1), APP (B63.3; ), Aβ (B104.1; ), TLN (B36.1 and biotinylated B36.1; ), and mAb 9C3 against NCT () have been characterized before. pAbs SB129 (anti-PS1), anti-KDELr (erd2), and -BAP31 were provided by C. Van Broeckhoven (University of Antwerp, Antwerp, Belgium), H.-D. Schmitt (Max Planck Institute for Biophysical Chemistry, Göttingen, Germany), and M. Reth (Max Planck Institute for Immunology, Freiburg, Germany). mAbs 5.2 against PS1 and Ergic-53 were obtained from B. Cordell (Scios, Fremont, California) and H.-P. Hauri (University of Basel, Basel, Switzerland), respectively. Polyclonal anti-murine Rer1p was generated in New Zealand white rabbits using the C-terminal sequence CKRRYKGKEDVGKTFAS coupled to KLH (Pierce Chemical Co.) as the antigen (PickCell Laboratories). Additional antibodies used follow: mAb against NCT ectodomain (amino acids 168–289; BD Biosciences), the transferrin receptor (Tfr; cl H68.4; Zymed Laboratories), GM130 (BD Biosciences), and actin (cl AC-15; Sigma-Aldrich). Studies were performed in HeLa cells, WT MEFs, and MEFs deficient for PS (PS1 and -2 KO; ), APH-1a, APH-1a,b,c (), or NCT (a gift from P. Wong, Johns Hopkins University, Baltimore, MD; ). All cell lines were routinely grown and maintained in DME/F12 supplemented with 10% FCS. After HeLa cells were incubated at 37 or 15°C for 3 h, they were fixed in 4% paraformaldehyde and processed for double immunofluorescent labeling as described before using Alexa 488 and 568 conjugated secondary antibodies (Invitrogen) and TOPRO-3 to label nuclei (). Images were captured on a confocal microscope (Radiance 2100; Carl Zeiss MicroImaging, Inc.) connected to an upright microscope (Eclipse E800; Nikon) and using an oil-immersion plan Apo 60× A/1.40 NA objective lens. Final processing was done using Lasersharp 2000 (Carl Zeiss MicroImaging, Inc.) and Photoshop (Adobe) and restricted to limited linear color balance adjustments to interpret merged pictures. To obtain NCT without intracellular domain (NCTΔIC) and NCT without both intracellular and transmembrane domains (NCTΔ[TMD + IC]), FL mouse NCT, subcloned in pGEM-T (Promega), was used as a template in PCR reactions with the same sense primer (including a SalI restriction site) and two different antisense primers, including a stop codon, a flag-tag sequence, and NcoI restriction site. In NCTΔIC, 16 C-terminal amino acids are missing, and in NCTΔ(TMD + IC), an additional 24 amino acids are omitted. PCR products were cloned first in pGEM-T to generate SalI ends and then further into the XhoI restriction site of pSG5** (pSG5 with extended multiple cloning site). The NCT construct lacking the ectodomain, except the signal peptide (NCTΔEC), was generated using gene splicing by overlap extension reaction PCR (SOE-PCR; ). Restriction enzymes SalI and NcoI were used for subcloning the final PCR product from SOE reaction into pMSCV (CLONTECH Laboratories, Inc.); using BglII and BamHI restriction enzymes, the NCTΔEC construct was further subcloned from pMSCV into the BamHI restriction site of pSG5**. NCTΔEC contains the last 43 amino acids of NCT. The NCTΔ(long TMD + IC) construct was made by introducing a stop codon at position 656 of mouse NCT (subcloned in pSG5**) using site-directed mutagenesis (Stratagene). cDNAs encoding mouse NCT and TLN, subcloned in pSG5**, were used for introducing restriction sites MluI and AgeI by site-directed mutagenesis at codon positions 660 and 690 of NCT and at codon positions 828 and 858 of TLN, respectively. These codon positions correspond to the flanking residues of the NCT and TLN TMDs. The introduced restriction sites were used to swap the coding sequence of the TMDs of the two proteins (NCT/TMD and TLN/TMD). Amino acids T670, G674, S681, T685, and Y686 were mutated to L by site-directed mutagenesis using NCT subcloned in pSG5** as a template. Human is subcloned in pCB6 using HindIII and XbaI. All constructs were transiently overexpressed in HeLa cells for interaction studies. The cDNAs encoding TLN, NCT, NCT/TMD, and TLN/TMD were cloned in the retroviral vector pMSCV and packaged in retroviruses, which were used to transduce MEF NCT cells, all according to established procedures of virus production and puromycin selection (). For coimmunoprecipitation experiments, cells were harvested in PBS, centrifuged (800 for 10 min), and lysed in 125 mM NaCl, 50 mM Hepes, pH 7.4 (supplemented with 1% Triton X-100 or CHAPS and Complete protease inhibitors [Roche]) for 30 min at 4°C. After centrifugation (16,000 for 15 min), cleared cell extracts were incubated overnight at 4°C with protein A beads and specific antibodies (anti-NCT mAb 9C3 or affinity-purified pAb B59.4) or with anti-rabbit IgG beads (Rabbit IgG TrueBlot set) and anti-hRer1 pAb. In the case of biotinylated B36.1 pAb, streptavidin-Sepharose beads (GE Healthcare) were used (). Immunoprecipitates were solubilized in NuPAGE sample buffer (Invitrogen) under reducing (for anti-hRer1 and biotinylated B36.1) or nonreducing conditions (for anti-NCT), electrophoresed on 4–12% NuPAGE Bis-Tris gels in MES running buffer (Invitrogen) and processed for Western blotting and immunodetection using ECL (PerkinElmer). When mature glycosylation was studied, fractions were treated with endoH (10 mU for 18 h at 37°C) as described previously () before SDS-PAGE. Cells were harvested in PBS, centrifuged (800 for 10 min), and homogenized in 250 mM sucrose, 10 mM Hepes, and 1 mM EDTA, pH 7.4, supplemented with protease inhibitors using a ball-bearing cell cracker (10 passages; clearance 10 μm; Isobiotec). After low-speed centrifugation (400 for 10 min), the postnuclear supernatant was ultracentrifuged (100.000 for 1 h). Microsomal membranes resuspended in 125 mM NaCl and 50 mM Hepes, pH 7.4, were used for reaction with DTBP (Pierce Chemical Co.) at room temperature for 30 min. DTBP is thiol-cleavable, water-soluble cross-linker. The reaction was quenched with 50 mM Tris, pH 7.5, at room temperature for 15 min. Proteins were extracted by adding Triton X-100 to a final concentration of 1%, followed by a 30-min incubation at 4°C. After centrifugation (16,000 for 15 min), cleared cell extracts were used for coimmunoprecipitation as indicated in the previous paragraph. To cleave DTBP, immunoprecipitates were solubilized in NuPAGE sample buffer with 4% β-mercaptoethanol and incubated at 100°C for 10 min before electrophoresis. For RNAi, the target sequence 5′-AATATCAGTCCTGGCTAGACA-3′ was used in human cDNA and 5′-CCTGGTGATGTACTTCATCATGCTT-3′ in mouse cDNA. The GL2 luciferase RNAi duplex was used as a nonspecific control. Oligonucleotides (human and Stealth [Dharmacon Research, Inc.] and mouse [Invitrogen]) were annealed and transfected using Lipofectamine (MEF) or Oligofectamine (Invitrogen; HeLa) as described previously. The cells were analyzed 48 or 72 h after transfection or further processed for metabolic labeling. In this case, control and RNAi down-regulated HeLa cells were pulse labeled with [S]methionine/cystein (Translabel; MP Biomedicals) for 30 min, washed, and chased in DME supplemented with 1% FCS for 0, 3, and 6 h. Cleared 1% Triton X-100 extracts were subjected to immunoprecipitation with 9C3 mAb (anti-NCT). Bound fractions were denaturated (10 min at 70°C) and incubated with endoH (1 U/50 μl for 18 h at 37°C). Labeled proteins were separated by SDS-PAGE on 7% Tris-Acetate gels (Invitrogen) and quantified using phosphorimaging (Typhoon; Molecular Dynamics) and ImageQuant software. Microsomal membranes were resuspended in buffer containing 0.5% dodecylmaltoside, 20% glycerol, 25 mM Bis-Tris/HCl, pH 7.0, and protease inhibitors and solubilized for 1 h on ice. After ultracentrifugation (100,000 at 4°C for 30 and 15 min), the cleared supernatant was collected. For each sample, the same amount was mixed with 5× BN sample buffer (2.5% Coomassie brillant blue G-250, 50 mM BisTris-HCl, 250 mM 6-amino-caproic acid, pH 7.0, and 15% sucrose) and stored overnight at 4°C. BN-PAGE was performed as described previously () with some modifications. Samples were loaded on a 5–16% polyacrylamide gel and electrophoresed (100 V for 30 min followed by 1 h at 200 V). For the final run (1 h 50 min), the Coomassie brilliant blue G-250 was omitted from the cathode buffer. For cell surface biotinylation, 0.5 mg/ml NHS-SS-biotin (Pierce Chemical Co.) was used as described previously (), and biotinylated proteins were isolated overnight (4°C) using streptavidin-Sepharose beads (GE Healthcare). Bound material was processed for Western blotting as described. Subconfluent (70–80%) HeLa cells were cotransfected with pcDNA3.1-APP-C99 and either a control empty pSG5 vector, pCB6-hRer1 using GeneJuice (VWR), or RNAi duplex oligonucleotides for hRer1p (see RNAi section) using Lipofectamine 2000 according to the manufacturer's instructions. After 24 h of hRer1p overexpression and 48 h of hRer1p down-regulation, cells were metabolically labeled (100 μCi 35S-Translabel/dish in Met/Cys-free DME) for 3–4 h in a CO incubator. Next, conditioned media were collected while cell layers were scraped and extracted in TBS containing 1% Triton X-100. Both cleared media and cell extracts were subjected to immunoprecipitation using 20 μl protein G–Sepharose beads, including anti-Aβ pAb B104.1 () or anti-APP B63.3 (1:200), respectively, and bound material was separated on 10% NuPAGE gels in MES buffer (Invitrogen), dried, and analyzed by phosphorimaging (Typhoon) and ImageQuant software (Molecular Dynamics). For the cell-free γ-secretase assay, CHAPS-extracted microsomal fractions derived from HeLa cells either overexpressing hRer1p or with down-regulated hRer1 levels (see RNAi section) were mixed with recombinant APP-C99-FLAG affinity purified from transiently transfected APH-1a,b,c MEFs exactly as described previously (). After incubation, newly produced AICD was separated on 10% NuPAGE gels in MES buffer and processed for Western blotting and ECL.

All cells maintain nonrandom distributions of cytoplasmic components, including membranous organelles and macromolecular RNA or protein complexes. This cytoplasmic organization is accomplished by a concerted effort of molecular motors, which are proteins that transport cargo along microtubules or actin filaments using the energy of ATP hydrolysis (; ; ). The docking of motors to their specific cargoes is essential for proper cargo distribution. Several docking proteins that are essential for binding cytoplasmic dynein, kinesin, or myosin motors to cargo have been described recently (; ; ; ). The most versatile and ubiquitous adaptor is the dynactin complex (; ; ; ). The main function of dynactin is to facilitate the attachment of cytoplasmic dynein to its cargo (; ). In addition, dynactin can function as an adaptor for at least two motors of the kinesin superfamily, heterotrimeric kinesin-2 () and mitotic kinesin Eg-5 (). Dynactin can also act independently of cytoplasmic dynein to anchor microtubules at the centrosome (; ) and organize radial microtubule arrays (). The dynactin complex consists of two morphologically distinct structural domains: a rod-shaped domain that binds to the cargo and an extended projection that mediates an interaction with cytoplasmic dynein and microtubules. The rod-shaped part consists of an Arp-1 filament and actin-capping proteins, whereas the projection is formed by a homodimer of a p150 protein subunit. These two parts of the dynactin complex are bridged by the p50 subunit dynamitin. p150 interacts with other subunits of the dynactin complex through its C terminus and with cytoplasmic dynein and other motors through its coiled-coil domains (). Remarkably, in addition to providing a platform for motor binding, p150 has the ability to interact with microtubules independently of cytoplasmic dynein. The microtubule-binding region of p150 is localized at the extreme N terminus and consists of a CAP-Gly (cytoskeleton-associated protein glycine rich) domain and a basic region, both of which are positioned within the first 200 amino acid residues (; ; ). Analysis of p150 isoforms in the mammalian brain showed that in addition to the full-length p150, neurons express an alternatively spliced 135-kD isoform lacking the microtubule-binding domain (). This suggests that this domain could be dispensable for at least some of the dynactin functions in nondividing cells. In vitro motility analysis using beads coated with a mixture of cytoplasmic dynein and dynactin demonstrated that dynactin could function as a processivity factor for dynein (), presumably by providing an extra site for microtubule binding and, thus, preventing cargo dissociation from microtubules (; ; ). Another function of the microtubule-binding domain of p150 is to localize the dynactin complex to the plus ends of growing microtubules (). p150 is a member of a family of microtubule plus end–binding proteins () and colocalizes with other proteins of this class such as CLIP-170 and EB1 to the plus ends of growing microtubules (; ; ). Its binding affinity to microtubules is regulated by phosphorylation (). It has been postulated that the accumulation of p150 at the plus ends of microtubules facilitates the loading of retrograde cargo on microtubules () and linking microtubule plus ends to specific sites, such as mitotic kinetochores and the cell cortex (). Both the tip binding and enhancement of motor processivity by dynactin require the N terminus of p150. However, the existence of the shorter p135 isoform of p150, which lacks the microtubule-binding motif, suggests that the dynactin complex can perform at least some of its functions even without this microtubule-binding activity. In this study, we examine the role of the microtubule-binding domain of p150 in the cargo transport and organization of microtubules. In cultured S2 cells, we replaced the full-length p150 protein with a truncated form lacking the microtubule-binding domain. We then examined effects of the deletion of the microtubule-binding domain on cargo transport (membranous organelles and mRNA–protein complexes) and the organization of microtubules. To eliminate the effect of the actin-based component on transport, we treated cells with cytochalasin D. Our results demonstrated that truncation of the first 200 amino acid residues from p150 eliminated its binding to microtubules but had no effect on the rate, processivity, or step size of cargo transport by either kinesin-1 or cytoplasmic dynein. However, truncation of the microtubule-binding domain resulted in defects in organization of the mitotic spindles, including the formation of multipolar spindles and free microtubule-organizing centers (MTOCs). Thus, we conclude that the microtubule-binding domain of p150 is not required for microtubule-dependent transport but is essential for the proper organization of radial microtubule arrays. To replace the wild-type p150 with a truncated form, we generated S2 cell lines expressing a fusion protein of monomeric red fluorescent protein (mRFP; ) or EGFP with the N terminus of either full-length p150 or p150 with a deletion of residues 1–200 (ΔN-p150). Stable cell lines were selected by hygromycin. We then treated cells with double-stranded RNA corresponding to the 3′ untranslated region (UTR) of p150 mRNA to deplete the endogenous protein. Western blotting with an antibody that recognizes the C-terminal fragment of p150 showed that in addition to the endogenous protein, stable cell lines expressed new proteins with the molecular weights expected for fusions of either full-length p150 or ΔN-p150 tagged with mRFP (, lanes 1 and 3) or EGFP (not depicted). The antibody generated against residues 1–200 of p150 does not recognize ΔN-p150 (, lane 7), demonstrating that our construct is indeed lacking its N terminus. As mentioned above, to deplete endogenous p150, we treated cells with a double-stranded RNA corresponding to the 3′ UTR of p150 mRNA. As shown in (lanes 2, 4, 6, and 8), such treatment dramatically reduced the level of endogenous p150 but did not affect the expression of mRFP-tagged p150 fusion proteins, as mRNAs encoding these proteins do not have the 3′ UTR. Serial dilutions of samples demonstrated that the level of endogenous p150 in RNAi-treated cells dropped below 10% of the control untreated cells (unpublished data). Thus, by using a combination of the stable expression of tagged p150 constructs and RNAi-mediated knockdown of the endogenous p150, we can replace the endogenous protein with mRFP or with EGFP-tagged p150 or ΔN-p150. To examine whether truncation of the microtubule-binding domain affected the ability of p150 to form a dynactin complex and interact with cytoplasmic dynein, we analyzed the sedimentation behavior of the dynein–dynactin complex in sucrose density gradients (). We performed these experiments using wild-type S2 cells and cells expressing EGFP-tagged p150 or ΔN-p150. Endogenous p150 was depleted from cells expressing EGFP-tagged proteins by RNAi. To monitor the sedimentation behavior of p150, we used either an N-terminal p150 antibody (for untransfected cells) or an antibody against EGFP (for transfected cells). The distribution of cytoplasmic dynein in gradient fractions was probed using an anti–dynein heavy chain (DHC) antibody. Western blot analysis of the fractions demonstrated that endogenous p150, EGFP- p150, and EGFP–ΔN-p150 have identical sedimentation profiles in sucrose gradients, with a peak in fractions 18–20 that coincides with the peak of cytoplasmic dynein (). Kinesin heavy chain (KHC) was probed as a control protein that does not sediment with DHC. These results indicate that neither truncation of the microtubule-binding domain nor fusion with EGFP affects the ability of p150 to incorporate into the dynactin complex. To further confirm formation of the dynactin complex by truncated p150, we performed an immunoprecipitation assay using anti-EGFP antibody to pull down EGFP-p150 or EGFP–ΔN-p150 and probed the precipitates for other dynactin subunits. shows that both p50-dynamitin and Arp1 were detected in the precipitates from cells expressing EGFP-p150 or EGFP–ΔN-p150, but not from untransfected S2 cells. Analysis of these precipitates with the DHC antibody demonstrates the presence of cytoplasmic dynein in the samples precipitated with EGFP antibody from cells expressing EGFP-p150 or EGFP–ΔN-p150. DHC was not detected in the anti-EGFP immunoprecipitates from untransfected S2 cells or in precipitates with a preimmune serum. We conclude that ΔN-p150 incorporates into the dynactin complex and that this complex interacts with cytoplasmic dynein. To examine microtubule binding by EGFP-p150 or EGFP–ΔN-p150, we performed a pelleting assay with microtubules in vitro and colocalization studies with microtubules in S2 cells. For microtubule pelleting assays, we expressed recombinant proteins that contain either amino acid residues 1–600 or 200–600 of p150 fused to EGFP and His tags. Both proteins were purified by using a Talon affinity column, incubated with taxol-stabilized microtubules, and pelleted by centrifugation through a glycerol cushion. As shown in , p150 (amino acids 1–600) bound to microtubules, whereas the truncation of 200 residues from the N terminus abolished microtubule binding. These results confirm previous studies demonstrating that the microtubule-binding domains of p150 are localized within residues 1–160 of the protein (; ). For in vivo analysis of microtubule binding, we depleted endogenous p150 by RNAi and transiently transfected cells with either EGFP-p150 or EGFP–ΔN-p150 constructs. Cells were extracted with Triton X-100 to remove soluble proteins, fixed in methanol, and double stained for microtubules and EGFP. As shown in , EGFP-p150 decorated cytoplasmic microtubules. On the other hand, in agreement with in vitro results, EGFP–ΔN-p150 did not show any microtubule binding. In addition, both EGFP-p150 and EGFP–ΔN-p150 were found in small clusters in the cytoplasm. We do not know the nature of these clusters, but they are probably identical to the cortical p150/APC clusters previously observed by in association with microtubules in MDCK cells. The interaction of p150 and ΔN-p150 with microtubules was also examined in stable cell lines using a spinning disc confocal microscope. Similar to fixed cells, EGFP- p150 in live S2 cells was localized along microtubules and formed clusters that move bidirectionally along microtubules (Video 1, available at ). In some cells, EGFP-p150 also accumulated at microtubule tips, although this phenotype was not as prominent as in mammalian cells (Video 2; ). On the other hand, in the cell line expressing EGFP–ΔN-p150, microtubule decoration and microtubule tip accumulation were not observed in any focal plane, and p150 clusters seemed to move randomly in the cytoplasm (Video 3). Thus, we confirm that the first 200 amino acid residues of p150 contain microtubule-binding activity, and truncation of this domain completely eliminates the ability of p150 to interact with microtubules. To investigate the role of dynactin in the process of microtubule-dependent transport, we used two types of cargo: membranous organelles (peroxisomes, endosomes, and lysosomes) and nonmembranous mRNA–protein (messenger RNP [mRNP]) complexes ( homologue of the fragile X mental retardation protein [dFMRP]). Both EGFP-tagged peroxisomes and EGFP-tagged dFMRP particles have a well-defined morphology and are transported along microtubules by cytoplasmic dynein and conventional kinesin (kinesin-1) as we demonstrated previously (; ). To study the movement of cargo along microtubules without the interference of any myosin-dependent components, S2 cells were plated on a concanavalin-A–coated substrate, and actin filaments were depolymerized with cytochalasin D. Under these conditions, cells formed long and thin processes that had a length of 5–20 μm with a diameter of 0.5–1 μm. In addition, as expected, cytochalasin D treatment eliminated ruffling of the lamella and retrograde flow of actin in the lamelloplasm that could contribute to the movement of organelles and could, therefore, interfere with analysis of microtubule-dependent movement. Immunofluorescent staining with an anti–α-tubulin antibody showed that these processes contain microtubule bundles. Microtubules in these bundles have uniform polarity with their plus ends directed toward the tips of the processes (). Both EGFP-tagged peroxisomes ( and Video 4, available at ) and dFMRP particles () moved bidirectionally in these processes. In agreement with our previous results (; ), knockdown of either kinesin or cytoplasmic dynein by RNAi abolished the bidirectional motility of peroxisomes along microtubules (Fig. S1, available at ). In addition, knockdown of either kinesin-1 or dynein by RNAi completely inhibited the motility of lysosomes and endosomes along microtubules (unpublished data). Inhibition of bidirectional movements after the RNAi-induced depletion of one motor is not caused by the depletion of the motor of the opposite polarity because the other motor is still present both in the soluble pool and in the organelle fraction (Fig. S2). These results agreed with a previous study showing the coordination of plus and minus end–directed movements of lipid droplets in embryos () as well as a study showing an interdependence of cytoplasmic dynein, the dynactin complex, and kinesin in fast axonal transport in neurons (). To study the role of dynactin in cargo transport, we knocked down either p150 or the p50-dynamitin subunit of the dynactin complex. Such treatment effectively reduced the bidirectional transport of both types of cargo, demonstrating that dynactin is required not only for the transport of membranous organelles but also for the transport of mRNP particles along microtubules ( and Table S1, available at ). RNAi against a mitotic kinesin, Klp61F, was used as a control. To further confirm the role of dynactin in cargo transport, we overexpressed mRFP fusion proteins with either p50-dynamitin or the first coiled-coil region (amino acid residues 232–583) of p150. Both constructs act as dominant-negative inhibitors of dynactin-dependent cellular processes (; ; ). As shown in , the overexpression of either protein dramatically inhibited the transport of peroxisomes and dFMRP particles. It is worth noting that as in the case of dynactin knockdown, both the plus and minus end movements were blocked by the overexpression of dynactin subunits. The binding of both kinesin and dynein to dynactin could be one potential explanation of the inhibition of bidirectional movement in cells after the knockdown of dynactin subunits or overexpression of dominant-negative constructs. We performed immunoprecipitation assays to test this possibility. An antibody against p150 pulled down DHC but not KHC from S2 extracts. Similarly, a kinesin antibody did not pull down p150, although, in agreement with a previous study (), it did pull down DHC (Fig. S3, available at ). Thus, we conclude that dynactin is absolutely required for the bidirectional transport of both membranous (peroxisomes) and nonmembranous (mRNP) cargoes along microtubules, and this result cannot be explained by simultaneous binding of dynactin to the motor proteins of opposite polarity. To determine the role of the microtubule-binding domain of p150, we analyzed the movement of EGFP-tagged peroxisomes and dFMRP particles in S2 cells expressing either mRFP-p150 or mRFP–ΔN-p150. As mentioned above (see Generation of cell lines and RNAi procedure), we treated cells with RNAi against the 3′ UTR of p150 mRNA to deplete endogenous p150. The observation of cargo movement in these cells indicated that the replacement of wild-type p150 with either mRFP-p150 or mRFP–ΔN-p150 had no effect on peroxisome or dFMRP particle transport. Similar to the wild-type cells, cells expressing ΔN-p150 showed multiple EGFP-labeled cargo moving bidirectionally within cellular processes (Videos 5 and 6, available at ), whereas cells depleted of p150 demonstrated no movement (not depicted). Quantitative analysis demonstrated that removal of the microtubule-binding domain of p150 did not affect the long-range movement of peroxisomes or dFMRP particles along microtubules. Unlike the total depletion of p150, replacement with mRFP–ΔN-p150 had no effect on the mean run length of either peroxisomes or dFMRP particles (). We also compared two other parameters of movement: the relative number of runs and the mean velocity of runs (, Fig. S4 A, and Table S1, available at ). The relative number of runs was determined as the number of runs longer than a threshold value normalized to the number of analyzed organelles (see Materials and methods). This parameter should be most dramatically affected if the processivity of organelle movement was changed. In agreement with the mean run length data, the number of runs and velocity of peroxisome and dFMRP movements in both directions were not affected by removal of the microtubule-binding domain of p150. To exclude the possibility that the aforementioned results are specific to these two particular kinds of cargo, we measured the effects of removal of the microtubule-binding domain of p150 on the movement of endosomes and lysosomes. Endosomes were labeled by the incubation of cells with Texas red–conjugated dextran, and lysosomes were labeled with Lysotracker. We found that similar to peroxisomes and dFMR, the movement of lysosomes and endosomes in cytochalasin-treated cells was not affected by truncation of the microtubule-binding domain (Fig. S4 B and Table S1). We also examined the distribution of organelles in cells not treated with cytochalasin D. We could not see any effects of truncation of the microtubule-binding domain on the steady state distribution of any organelles studied here (Fig. S4 C), suggesting that even in the presence of actin, organelle movement is not affected by truncation of the microtubule-binding domain. Collectively, these results demonstrate that the presence of the microtubule-binding domain of p150 is not important for the movement of at least four different types of cargo along microtubules in S2 cells. Previously, we showed that both kinesin and cytoplasmic dynein move peroxisomes along microtubules in discrete 8-nm steps in vivo that correspond to the step size of both motors in vitro (). Similar discrete steps were clearly seen in traces of peroxisomes in cells expressing mRFP-p150 or mRFP–ΔN-p150 and depleted of endogenous p150. As shown in , the mean step sizes for kinesin and cytoplasmic dynein in cells expressing mRFP–ΔN-p150 are 7.7 ± 1.8 nm and 8.9 ± 1.4 nm, respectively. These numbers are not substantially different from the mean step sizes of kinesin and dynein in cells expressing mRFP-p150 (9.6 ± 3.1 nm and 8.8 ± 2.4 nm, respectively) or in wild-type cells as previously shown (). We conclude that deletion of the microtubule-binding domain of p150 does not affect the characteristics of the mechanochemical cycle of molecular motors in vivo. cytoplasmic dynein and dynactin play critical roles during cell division that include spindle assembly and elongation, anaphase chromosome movements, and removal of spindle checkpoint components from attached kinetochores (; ,; ; ; ). To examine the mitotic contribution of the microtubule-binding domain of p150, S2 cells expressing EGFP-p150 or EGFP–ΔN-p150 were treated with either control or double-stranded RNA from the 3′ UTR of p150 and immunostained for both microtubules and the mitosis-specific phosphorylated histone H3. Depletion of endogenous p150 in cells expressing EGFP–ΔN-p150 elevated the mitotic index threefold and increased the frequency of prometaphase stage figures as compared with control RNAi–treated cells. A similar prometaphase-like arrest was also described for cultured mammalian cells overexpressing the dynactin inhibitor p50/dynamitin (). Unexpectedly, we also observed a substantial increase in multipolar spindles in cells expressing EGFP– ΔN-p150 (), a phenotype not previously described for dynactin inhibition. We attribute multipolar spindle formation to the failure to properly coalesce MTOCs during prometaphase, a mechanism that may depend on a p150–microtubule interaction. A previous study found that 39% of untreated prophase S2 cells contain four or more γ-tubulin–staining MTOCs, and live imaging of S2 cells expressing GFP-tubulin revealed a clustering and fusion mechanism to eliminate extra MTOCs after nuclear envelope breakdown (). Consequently, failure to cluster and fuse extra MTOCs resulted in multipolar spindle formation. Furthermore, identified cytoplasmic dynein as a critical component of a centrosome-clustering mechanism present in human tissue culture cells that, when mislocalized in certain tumor cells, results in multipolar spindle formation. Consistent with this hypothesis, we found that cells expressing EGFP– ΔN-p150 exhibited a dramatic increase in the number of free MTOCs surrounding mitotic spindles (). Thus, the microtubule-binding domain of p150 is required to suppress multipolar spindle formation, probably by coalescing extra MTOCs that are frequent in many early mitotic S2 cells. Adaptors between motors and cargo are a diverse group of cellular proteins, but the most universal of them is the dynactin complex. An unusual property of this complex is its ability to bind microtubules independently of motor proteins. In this study, we directly addressed the functional significance of the microtubule binding of dynactin by replacing endogenous p150 with a p150 mutant lacking residues 1–200, which was thus unable to bind microtubules (ΔN-p150). We selected stable cell lines expressing mRFP-tagged p150 or ΔN-p150 and depleted endogenous p150 by using RNAi. This approach allowed us to study the contribution of the microtubule-binding domain of p150 to cargo transport and the organization of mitotic microtubule arrays. We demonstrated that the movement of two types of cargo, membranous organelles (peroxisomes, endosomes, and lysosomes) and mRNP complexes (dFMRP particles), is absolutely dependent on the dynactin complex, but, to our surprise, ΔN-p150, which lacks the ability to bind microtubules, was fully functional in supporting microtubule-dependent transport in vivo. All of the measurements were performed in the cells treated with cytochalasin D to eliminate actin-based motility. It should be stressed that this study showed for the first time that dynactin is directly involved in RNA transport in somatic cells. Although a previous study suggested that dynactin is required for the proper localization of morphogen RNA in oocytes (), direct evidence has been lacking. On the other hand, we observed a profound effect of truncation of the microtubule-binding domain on mitotic spindle structure, including the generation of multipolar spindles and free MTOCs. Although dynactin plays an important role in organizing the interphase radial microtubule arrays at the centrosome in mammalian cultured cells (; ; ), we observed no change in the pattern of interphase microtubules in S2 cells expressing ΔN-p150. This was expected given that S2 cells lack functional centrosomes capable of nucleating and organizing microtubules during interphase. Instead, microtubules nucleate at random in the cytoplasm and do not form a focused radial array (unpublished data). It is worth noting that one of the p150 isoforms in the mammalian brain is an alternatively spliced version called p135, which naturally lacks the microtubule-binding motif (). As neurons are terminally differentiated cells that do not divide or contain radial arrays of interphase microtubules, they do not need dynactin activity to facilitate microtubule organization. On the other hand, the results presented here suggest that this alternatively spliced isoform is as effective as full-length p150 in supporting transport. Examination of the database shows that there is a second gene in (gi|23093121|gb|AAF49148.2|; ). We performed additional experiments addressing whether the existence of this second gene rescues and compensates for the loss of the normal p150 protein. RNAi against this gene did not change organelle movement either in the wild-type or mutant background. On the other hand, RNAi against conventional p150 alone completely blocked movement. We assume that this second gene is not important for organelle transport or is not even expressed in S2 cells. It has been reported recently that a G59S substitution in the microtubule-binding domain of p150 results in decreased microtubule binding and enhanced dynein and dynactin aggregation (). It is known that this substitution results in a slowly progressing motor neuron disease in humans (), suggesting the impairment of dynactin functions. Although this mutation could directly affect organelle transport by altering the microtubule-binding affinity of the dynactin complex, it is also possible that it could inactivate dynactin by forming aggregates that recruit both wild-type and mutant dynactin, resulting in a decrease in the level of functional dynein–dynactin complex available for cargo transport. Three possible functions have been proposed for the microtubule-binding domain of p150. First, demonstrated in vitro that the dynactin complex increases the processive movement of beads coated with cytoplasmic dynein. This idea is further supported by two recent studies (; ) showing that p150 is capable of one-dimensional diffusion along microtubules and, thus, maintains contact with microtubules. Therefore, enhanced processivity in vitro can be achieved by the p150–microtubule interaction. In contrast, the results shown here indicate that the effect of dynactin on motor processivity is sufficiently minor, and it is not detected by tracking cargoes in vivo even with the 1-nm, 1-ms accuracy of the FIONA technique (). There are several potential explanations of an apparent discrepancy between our results and the previous in vitro studies (; ). First, it is not known whether the four types of cargo studied here are transported by single or multiple motors. If cargo were transported by more than one motor, dynactin would not contribute to the processivity. Furthermore, several redundant mechanisms may be involved in the regulation of dynein processivity in cells, and, thus, the effects of truncation of the microtubule-binding domain might not be immediately obvious if other putative mechanisms are in action. Second, previous studies (; ) were performed in vitro not with the native dynein–dynactin complex but with two proteins bound separately to the surface of carboxylated beads. It is unclear whether all of the properties of the dynein–dynactin complex on the surface of organelles can be faithfully recapitulated by the binding of purified proteins with highly charged beads. Finally, in vitro assays were performed under no load condition, and it is possible that the load applied to organelles in the cytoplasm affects the processivity of transport. The second potential role of the microtubule-binding domain of p150 was proposed by ; ). These authors suggested that p150 tethering to the plus ends of growing microtubules facilitates the loading of retrograde cargo on the plus ends of microtubules. This function would require the interaction of dynactin with microtubules. Obviously, the truncated form of p150, ΔN-p150, could not load cargo onto microtubules. However, at least in the case of the two types of cargo we examined here, such a search and capture mechanism is not a major contributor to the retrograde transport by dynein because the truncation of microtubule binding from dynactin had no effect on cargo transport. In agreement with these results, showed that depletion of either EB1 or CLIP-170 resulted in a loss of p150 from microtubule plus ends but had no effect on the trafficking of membrane organelles. It is likely that a combination of microtubule dynamics and a Brownian motion of cargo is sufficient for making initial contact and loading of cargo onto microtubules. Thus, microtubule plus end targeting of p150 is not required for cargo loading. Still, it would be interesting to examine the effect of ΔN-p150 on cargo transport in the cellular regions containing single microtubules instead of cytochalasin D–induced microtubule bundles to detect the kinetic advantages of search and capture if they existed. Finally, the microtubule-binding domain of p150 interacts directly with the C terminus of CLIP-170 () or EB1 (; ). Dynactin has potential roles in regulating microtubule dynamics at the plus end complex and is required for microtubule anchoring and focusing in a cell cycle–dependent manner. It is likely that p150 links proteins at the cell cortex with cytoplasmic dynein or tip-binding proteins to pull and focus microtubules, thus preventing the formation of multipolar spindles or free MTOCs during cell division. We observed a dramatic effect of truncation of the microtubule-binding domain on mitotic spindle structure, stressing the essential role of the N-terminal microtubule-binding domain of dynactin in microtubule organization. This supports the role of centrosomal dynactin for the microtubule anchoring function proposed by . Collectively, our data demonstrate that the microtubule-binding domain of p150 has a substantial role in spindle microtubule orientation and focusing but is not required for long-range movement of cargoes along microtubules by motor proteins in vivo. The cDNA for p150 (clone AY118377; Open Biosystems) and the ΔN-p150 construct encoding amino acid residues 201–1,280 were amplified and subcloned into the pAc5.1/V5-HisA vector (Invitrogen). The EGFP or mRFP sequence was introduced at the N terminus of p150. To make His-tagged p150, the sequence of p150 encoding residues 1–600 or 200–600 was subcloned into pET28 (a+; Novagen/EMD Biosciences). The first coiled-coil domain (CC1; amino acids 232–583) of p150 or full-length p50/dynamitin was fused to mRFP to create the dominant-negative constructs mRFP-CC1 and mRFP-p50 in the pMT5.1/V5-HisA vector. To select stable cell lines expressing EGFP-SKL (peroxisome targeting signal) and mRFP-p150 or mRFP–ΔN-p150, S2 cells were cotransfected with three plasmids: pGG101 encoding EGFP-SKL (a gift from G. Goshima, University of California, San Francisco, San Francisco, CA), pCoHygro (Invitrogen) as a selection plasmid, and mRFP-p150 or mRFP–ΔN-p150 at a 20:1:20 molar ratio. Transfection was performed using Cellfectin reagent (Invitrogen). 40 h after transfection, 300 μg/ml hygromycin was added for selection. Selection was performed for 4–5 wk, and protein expression was confirmed by fluorescence microscopy and immunoblotting. Lysosomes in S2 cells were labeled by staining with 100 nM Lysotracker red DND-99 (Invitrogen) for 10 min. For endosome labeling, S2 cells in suspension were incubated with 1 mg/ml Texas red dextran (Invitrogen) for 6 h. RNAi treatment was performed as described previously (). Templates for in vitro transcription were generated by using the primers 5′-TAATACGACTCACTATAGGGGTATCGTGGCAATGGAATCG-3′ and 5′-TAATACGACTCACTATAGGGGAGTTATACAACATCAGCAA-3′ to amplify the 500 bp from the 3′ UTR of p150, and the primers 5′-TAATACGACTC-ACTATAGGGACCACCTGCAAAGCGATATCG-3′ and 5′-TAATACGACTCACTA-TAGGGTTGCAGATACTCCGTCAGGAT-3′ were used to amplify the 550-bp segment from the C terminus of the gene. The recombinant His-tagged p150 fusion proteins containing residues 1–190 or 1,073–1,280 were expressed in and purified by Talon affinity chromatography. Rabbit immunization was performed by Proteintech Group, Inc. For Western analysis, a 1:5,000 dilution of antiserum was used. An antibody against DHC was provided by J. Scholey (University of California, Davis, Davis, CA), and HD antibody against KHC was provided by A. Minin (Institute of Protein Research, Russian Academy of Sciences, Moscow, Russia). An antibody against p50 was a gift from R. Warrior (University of California, Irvine, Irvine, CA; ), and Arp1 antibody was provided by L. Goldstein (University of California, San Diego, La Jolla, CA). SUK4 and 9E10.2 (anti-KHC and anti-myc antibody, respectively) were obtained from the Developmental Studies Hybridoma Bank. Cells were incubated in extraction buffer containing 1% Triton X-100 in PBS for 2 min and fixed with 1% glutaraldehyde or cold methanol for 10 min. For microtubule or EGFP staining, we used monoclonal antibody DM1-α against α-tubulin (1:2,000) and polyclonal affinity-purified EGFP antibody (1:200), respectively. For analysis of the mitotic phenotype, cells were immunostained for both microtubules and mitosis-specific phosphorylated histone H3. Approximately 10 cells were used for the immunoprecipitation assay. Cell pellets were resuspended at a ratio of 1:2 (wt/vol) in a homogenization buffer (50 mM Tris, pH 7.5, 50 mM KCl, 0.1 mM EDTA, 2 mM MgCl, 1% NP-40, 5% glycerol, 1 mM DTT, 1 mM PMSF, and 10 μg/ml each of chymostatin, leupeptin, and pepstatin) and homogenized by using a 25-gauge syringe needle. Cell extracts were centrifuged at 15,000 for 10 min and then at 200,000 for 15 min. The resulting supernatants were incubated with antibodies (EGFP or preimmune rabbit IgG) prebound to protein A–Sepharose beads (GE Healthcare) for 2 h at 4°C. The beads were washed, and proteins were eluted in SDS sample buffer and analyzed by Western blotting. Images of live cells were acquired as described previously () using a microscope system (U2000 Perfect Focus; Nikon). A 100-W halogen bulb was used for fluorescence excitation to minimize photobleaching and phototoxicity. Images were captured every 1 s for 2 min for EGFP-tagged peroxisomes, endosomes, and lysosomes and every 2 s for 2 min for EGFP-tagged dFMRP. The movement of particles was analyzed by using the automatic tracking software Diatrack (version 3.01; Semasopht). The threshold speed was 0.2 μm/s for peroxisomes, endosomes, and lysosomes or 0.15 μm/s for dFMRP particles, and movements slower than this threshold were excluded from the calculations. We measured all runs longer than 2 μm for peroxisomes, longer than 1.6 μm for endosomes and lysosomes, and longer than 1 μm for dFMRP particles. The number of runs above the threshold was divided by the mean number of particles in the analyzed areas of the image. This value was defined as the relative number of runs. At least three independent experiments were analyzed for each condition, and five to six cells in each experiment were randomly chosen for recording and analysis. Approximately 3 × 10 cells were pelleted, and cell extract was prepared as described in the Immunoprecipitation section. ∼500 μl of the clarified supernatant was layered on top of 12 ml of 5–20% linear sucrose density gradient prepared in the homogenization buffer without NP-40. After centrifugation at 150,000 for 18 h in a SW40 rotor (Beckman Coulter), 0.5-ml fractions were collected and analyzed by Western blotting using antibodies to EGFP, p150, DHC, or KHC. Microtubules were prepared by polymerization of bovine brain tubulin in the presence of 1 mM GTP and 20 μM taxol at 37°C for 1 h. The polymerized microtubules were mixed with recombinant proteins and layered on top of a 30% glycerol cushion in BRB80 buffer (80 mM Pipes, pH 6.9, 1 mM EGTA, and 1 mM MgCl) with 10 μM taxol. Microtubules were pelleted by centrifugation at 150,000 for 40 min in a SW55 rotor (Beckman Coulter). The pellets were washed and resuspended in 30 μl SDS sample buffer. Fig. S1 shows that the depletion of cytoplasmic dynein or kinesin stops peroxisome movement. Fig. S2 shows that the depletion of one motor (kinesin or dynein) does not deplete the other motor of opposite polarity. Fig. S3 shows that kinesin can interact with dynein but not with dynactin. Fig. S4 A shows that truncation of the microtubule-binding domain of p150 has no effect on the mean velocity of peroxisome or dFMR particles. Fig. S4 B shows that truncation of the microtubule-binding domain of p150 has no effect on mean run length, relative number of runs, and mean velocity of endosomes or lysosomes. Fig. S4 C demonstrates that the steady-state distribution of organelles in S2 cells not treated with cytochalasin D is not affected by the expression of mRFP-p150 or mRFP–ΔN-p150 . Table S1 shows absolute and relative numbers of long runs in cells treated with RNAi against motors or dynactin components, cells overexpressing dynactin subunits, and cells expressing EGFP-p150 or EGFP–ΔN-p150. Videos 1 and 2 demonstrate that EGFP-p150 decorates tips and the length of microtubules. Video 3 demonstrates that EGFP–ΔN-p150 is distributed randomly in the cytoplasm. Videos 4–6 show bidirectional movements of peroxisomes in the processes of S2 cells expressing endogenous p150, mRFP-p150, and mRFP–ΔN-p150, respectively. Online supplemental material is available at .

Intraflagellar transport (IFT) is the bidirectional movement of granule-like particles, termed IFT particles, along the length of eukaryotic cilia and flagella (; ). IFT was first reported in the green alga (), and has subsequently proven to be a conserved process among ciliated organisms (). IFT moves axonemal components, such as cargo, to the tip of the flagellum (; ), where axonemal assembly occurs (; ). IFT is also involved in the turnover of flagellar components (), in the movement of flagellar membrane components in the plane of the membrane (), and in cilium- generated signaling (). Consequently, mutations in IFT affect ciliary and flagellar assembly, maintenance, and function (; ; ). Considerable progress has been made in understanding the structure and composition of the IFT particles and the motors that move them. IFT from the base to the tip of the flagellum is powered by kinesin-2 motors (; ); IFT in the opposite direction is generated by cytoplasmic dynein 1b (, ; ; ). The IFT particles themselves are composed of at least 16 proteins organized into two complexes, complexes A and B (; ), which sediment at 16S in sucrose density gradients. Biochemical analysis has revealed that complex B contains an ∼500-kD core composed of IFT88, IFT81, IFT74/72, IFT52, IFT46, and IFT27 (). IFT172 appears to be a peripheral component, as it often dissociates from complex B during the latter's purification (). Much less is known about the functions of the individual IFT particle proteins, and how they interact with the IFT cargoes. Many of the IFT particle proteins have been sequenced, but the sequences provide few hints as to the proteins' functions. Recognizable domains consist mainly of WD repeats, TPR domains, and coiled-coil domains, all of which are thought to be involved in protein–protein interactions (). Mutations have been identified for IFT52, IFT88, and IFT172, but these generally block flagellar assembly (; ; ; ), and thus have not been informative in regard to the proteins' specific functions. Mutations in genes encoding IFT proteins in and mammals have been similarly uninformative (; ). In , IFT172 interacts with EB1, which is located at the tip of the flagellum (, ), and an IFT172 temperature-sensitive mutant is defective in IFT particle turnaround at the tip of the flagellum (); based on these results, it was proposed that IFT172 is involved in regulating the transition between anterograde and retrograde IFT (). In mammalian cells, IFT20 was found to be unique among IFT particle proteins examined in that it is localized to the Golgi apparatus, as well as the cilium, and moderate knockdown of IFT20 reduced the amount of the membrane protein polycystin-2 in the cilium (), suggesting that IFT20 is involved in trafficking of ciliary membrane proteins. Other than this, nothing is known about the specific roles of the IFT particle proteins and how, or even whether, they interact with specific cargoes. To obtain more information on the function of a specific IFT particle protein, we have focused on IFT46, which was previously briefly reported to be a complex B protein in (). We have cloned and characterized IFT46 from and mouse, and find that it is a homologue of DYF-6, a protein very recently reported to undergo IFT in and to result in truncated dendritic cilia when mutated in the worm (). We also describe the phenotype of a -null mutant for IFT46, and of a suppressor of that mutant. The observations provide new insights into the role of IFT46 in IFT particle stability and transport of a specific IFT cargo. Portions of this work were previously reported in abstract form (). The gene and cDNA encoding IFT46 were cloned as described in Materials and methods. The cDNA (accession no. ) contains a 1,035-nt ORF predicted to encode a 37.9-kD protein () with a pI of 4.61. The cDNA has a stop codon 18 nt upstream of the predicted start codon, and a polyA consensus sequence at nt 1,703–1,707. ESTs have a polyA tail 12–14 nt downstream of the polyA consensus sequence. Therefore, the ORF is complete. No structural domains or motifs were identified within the sequence. IFT46 was initially identified as an IFT complex B protein, based on its cosedimentation with other complex B proteins in sucrose gradients (). Characterization of our cloned protein indicates that it behaves exactly as expected of a 46-kD complex B protein, as follows: a) six unique peptides corresponding to the cloned protein were identified in the membrane plus matrix fraction of the flagellar proteome, but no peptides from it were found in any other fraction, which is a distribution typical for IFT particle proteins but unusual for non-IFT proteins (); b) using real-time PCR, we found that expression of the protein is up-regulated 10.4 ± 2.03-fold (SEM) upon deflagellation, which is characteristic of flagellar proteins, including other complex B proteins (); c) an antibody to a peptide contained in the cloned protein was generated and shown to react specifically with a single protein of M ∼46,000 in Western blots of whole cell lysates (); d) immunofluorescence microscopy with the antibody as probe showed that the majority of the cloned protein is located in the basal body region, with a lesser amount in puncta along the length of the flagella (), is a distribution identical to that of other IFT particle proteins (; ); moreover, the protein colocalized with complex B protein IFT172, but not complex A protein IFT139 (see the section Complex A and B proteins are located in distinct compartments in the basal body region; ); and e) when the flagellar membrane plus matrix fraction was analyzed by sucrose density gradient centrifugation, the cloned protein co-migrated precisely with IFT81, a complex B protein, but not with IFT139, a complex A protein (). These results indicate that the cloned protein is indeed IFT46, and confirm that IFT46 is a complex B protein. Database searches revealed that IFT46 is conserved across organisms that have cilia, including (XP_694278; BLAST E, 1e-47), (XP_396519; BLAST E, 1e-47), (NP_609890; BLAST E, 1e-17), (NP_076320; BLAST E, 2e-61), and (CAB66868; BLAST E, 8e-62). The protein is also homologous to DYF-6 (NP_741887; BLAST E, 9e-34), which was recently reported to undergo IFT in dendritic cilia when fused to GFP (). No similar sequence was found in nonciliated organisms, including and . The middle portions of the and mammalian proteins are highly similar (51% identity, 72% similarity; ), making it likely that they are orthologous proteins. To test this, the putative mouse orthologue was Flag tagged and expressed in IMCD3 cells, a mouse kidney cell line. Immunofluorescence microscopy revealed that the Flag-tagged protein localized specifically to primary cilia (). To determine if the mammalian protein was part of IFT complex B, lysates were prepared from IMCD3 cells expressing either IFT46-Flag, IFT20-Flag (positive control), or GFP-Flag (negative control) and immunoprecipitated with an anti-Flag antibody. In each case, the Flag-tagged proteins were highly enriched in the immunoprecipitates (, top). Western blots showed that complex B proteins IFT88 and IFT57, but not complex A protein IFT140 or a control protein, were coimmunoprecipitated from lysates of the IFT46-Flag–expressing cells (, bottom). Similarly, in the positive control, IFT88 and IFT57, but not IFT140, were coimmunoprecipitated. No IFT proteins were coimmunoprecipitated from the lysate of cells expressing GFP-Flag. Therefore, mammalian IFT46 localizes to cilia and is tightly associated with complex B proteins, but not with complex A proteins. To investigate the role of IFT46 in intraflagellar transport, we screened a collection of insertional mutants by Southern blotting and identified one strain, T8a44-11, with a defect in the gene. Analysis by PCR showed that the mutant allele, which we term , has a deletion or disruption somewhere between the fourth and the seventh exon of the gene (). This mutant has short, stumpy flagella. Strain T8a44-11 was backcrossed to wild-type cells, and a progeny, YH6, which lacked the mutation carried by the parental strain, was selected for detailed characterization. As in T8a44-11, the gene in YH6 is disrupted, as shown by Southern blotting (, lane 2). No IFT46 can be detected in lysates of YH6 cells (, lane 2), indicating that IFT46 is not expressed in YH6 cells. Thus, the mutant allele is a null allele (see also the section The 3′ end of the gene is transcribed in Sup1 cells). Like T8a44-11, YH6 cells have short, stumpy flagella that barely extend beyond the flagellar collar (). The flagella are nonmotile. To confirm that the mutant phenotype of YH6 was caused by the disruption of the gene, YH6 cells were transformed with a 4.8-kb fragment containing only the wild-type gene (). Numerous wild-type swimmers with full-length flagella were recovered. Southern blotting revealed that the exogenous gene had integrated into the genome at different sites in several of these rescued strains (, lanes 3–9), confirming that the rescued strains were independently derived. Therefore, restoration of motility was caused by incorporation of the transgene, and not caused by disruption of some other gene. Western blotting confirmed that expression of IFT46 was restored in the transformants (, lanes 3–9). These results demonstrate that the short flagella phenotype of YH6 is caused by the absence of IFT46. Hereafter, this strain will be referred to as the mutant. Although very short, flagella are longer than those of mutants with defects in the complex B proteins IFT88 () and IFT52 (), which do not form flagella beyond the transition region. Because flagella are formed in the mutant, we were able to compare them with flagella from wild-type and rescued cells by EM to identify flagellar defects associated with loss of IFT46. Serial sections of wild-type flagella have shown that the outer doublet microtubules are connected by rodlike “peripheral links” in the most proximal part of the flagella (); the rows of dynein arms begin at a level slightly more distal, but still within the flagellar collar (; ). The flagella have nine outer doublet microtubules and frequently extend to the limits of the flagellar collar or beyond (), but we never observed dynein arms in longitudinal or cross sections of these axonemes. In addition, the mutant flagella lack the projections into the lumens of the B-tubule () and frequently have defects in the central pair of microtubules (). In contrast to mutants with defects in the retrograde IFT motor (; ; ), few, if any, IFT particles accumulate in the mutant flagella. It is important to note that the flagella from the rescued cells () have typical wild-type morphology with normal inner and outer dynein arms and central pair microtubules; this confirms that the ultrastructural defects seen in the mutant are caused by loss of IFT46. To determine whether the dynein arm deficiency in the mutant is caused by degradation of dyneins within the cell body, by an inability to transport dyneins into the flagella, or by an inability to assemble them onto the axonemes, we analyzed whole cell lysates and flagella from and wild-type cells by Western blotting (). The cell lysates of the mutant contained normal levels of the outer arm dynein intermediate chain IC2 and the inner arm dynein intermediate chain IC138. Therefore, the dyneins are present in the mutant cells. However, both IC2 and IC138 were completely absent from the mutant flagella, indicating that the dyneins are not transported into the flagella. In contrast, DC2, which is a component of the outer dynein arm docking complex that is transported into the wild-type flagellum and assembled onto the doublets independently of the outer dynein arm (), is transported into the flagella. The presence of DC2 in the flagella provides additional evidence that the lack of dynein arms is not simply because of the short length of the mutant flagella or to a general failure to transport proteins into the flagellum. These data confirm that the ultrastructural findings that the mutant has a defect in transporting dynein arms into the flagella. To investigate the role of IFT46 in IFT complex assembly, we examined the cellular levels of IFT complex B and A proteins in cell lysates of wild-type and cells (, lanes WT and ). When normalized with tubulin, the levels of complex B proteins IFT20, IFT57, IFT72, and IFT81 were greatly decreased in the mutant cells relative to the wild-type cells. The only complex B protein not reduced in the absence of IFT46 was IFT172, the level of which was the same as or greater than in wild-type cells, depending on the preparation, suggesting that the level of IFT172 is controlled independently from that of the other complex B proteins. In contrast to the decrease in most complex B proteins seen in the mutant, the levels of complex A proteins IFT139 and IFT140 were greatly increased in cells relative to wild-type cells. To examine if these differences in protein levels were caused by changes in synthesis or stability, we used real-time PCR to measure transcript levels for several complex A and B proteins in wild-type and cells. Transcript levels in the mutant were increased by ∼2.8- and ∼2.0-fold for complex A proteins IFT140 and IFT139, and by ∼1.8- and ∼1.6-fold for complex B proteins IFT81 and IFT72, respectively. Therefore, the mutant responds to its defect by increasing the mRNA levels of at least these complex A and B proteins. The increase in complex A proteins in the mutant presumably reflects this increase in mRNA abundance. However, because the levels of most complex B proteins are drastically decreased in the mutant, even though complex B mRNA levels in general appear to be increased, it is likely that these proteins are specifically degraded in the absence of IFT46. To determine if the absence of IFT46 and the accompanying large decrease in most other complex B proteins affected the transport of IFT172 or complex A into or out of the flagellum, we used immunofluorescence microscopy to examine cells that were double labeled with antibodies to tubulin and IFT172 or IFT139 (). In both cases, the IFT particle proteins were concentrated around the basal bodies and in the short flagella. Thus, both proteins are transported into the flagella in the absence of IFT46. Surprisingly, however, the distributions of IFT172 and IFT139 differed from each other in the cell body, with that of IFT172 () appearing to have almost no overlap with that of IFT139 (), which was more anterior and often concentrated into two distinct lobes. To clarify whether this difference in distribution was normal or caused by loss of IFT46, wild-type cells were double labeled with antibodies to IFT46 and IFT172 or IFT139. In most cases, IFT46 and IFT172 colocalized precisely with each other in the peribasal body region (), which is consistent with the other evidence that IFT46 is a complex B protein. In contrast, although IFT139 colocalized with IFT46 at the extreme apical end of the cell, the labeling of IFT46 almost always extended more posteriorly than that of IFT139 (). This is the first observation that complex A and B proteins differ in their distribution, and indicates that the complexes, or at least a subset of them, are not physically associated in the cell body. This, together with the results for the mutant, also shows that in the absence of IFT46, the colocalization of IFT172 with complex A proteins at the extreme apical end of the cell is lost. Our observation that IFT172 was transported into the short flagella of the mutant () raised the question of whether IFT172 was being transported into the flagella independently of other complex B proteins or in association with incomplete complex B particles, possibly assembled from the small amount of complex B proteins still present in the mutant. To address this question, we used immunofluorescence microscopy to examine the distribution of another complex B protein, IFT57, the levels of which are greatly reduced in the mutant. Like IFT172, the residual IFT57 was transported into the short flagella of the mutant (). These results support the hypothesis that a small number of incomplete complex B particles assemble from the residual complex B proteins and are capable of being transported into the flagellum in the absence of IFT46. The resulting low level of IFT may account for the ability of the cells to assemble their short flagella. cells are completely nonmotile. However, on one occasion, swimming cells were observed in an unaerated, stationary phase culture of cells. Cells from this culture were cloned, and a partially suppressed strain, Sup1, was isolated. Sup1 cells are usually nonmotile when grown in M media with aeration, but are stimulated to form flagella of variable length () and swim with a slow jerky movement in the absence of aeration. The suppressed phenotype is the result of a rare spontaneous mutation that allows transcription of the 3′ end of the IFT46 gene (see next section). To elucidate the effect of the partial suppressor mutation on IFT46 and other IFT particle proteins, the levels of the proteins in stimulated Sup1, wild-type, and cells were compared by Western blotting. In the Sup1 cells, complex B proteins IFT20, IFT57, IFT72, and IFT81 were increased to a level between those of and wild-type cells (), whereas the levels of the complex A proteins IFT139 and IFT140 were decreased to a level between those of and wild type. Importantly, IFT46 is still undetected in the suppressed strain. This result indicates that the partial suppression of involves an increased stability of IFT complex B in the absence of full-length IFT46. It is possible that a C-terminal fragment of IFT46 is expressed in Sup1 cells and incorporated into complex B, thereby stabilizing it. Such a fragment would not be detected by our antibody to the N terminus of IFT46. The slow, jerky swimming of Sup1 is typical of outer dynein arm mutants. Therefore, we used immunofluorescence microscopy to check for the presence of outer and inner dynein arms in Sup1 flagella. No outer arm dynein was detected using an antibody to the α heavy chain of outer arm dynein (). In contrast, labeling of Sup1 flagella by an antibody to inner arm dynein I1 intermediate chain IC138 was normal (). These results show that transport into the flagellum of inner arm dynein I1, but not outer arm dynein, is restored in the suppressed strain. To determine the extent to which the ultrastructural defects of were restored in the partially suppressed strain, Sup1 cells and flagella were examined by electron microscopy (). The inner arms, radial spokes, and central microtubules were present and appeared normal. However, no outer dynein arms were observed. Therefore, the suppressor strain assembles flagella that lack the outer dynein arm but appear normal in every other way. Western blotting showed that the levels of both outer arm dynein and outer dynein arm docking complex proteins, as represented by IC2 and DC2, respectively, were normal in Sup1 cells (). Thus, the inability to transport and assemble outer arms in the flagella is not simply caused by an absence of these components from the cell cytoplasm. These results indicate that IFT46 is specifically needed to transport outer dynein arm components into the flagellum. The inner dynein arm and central pair defects observed in the mutant are likely attributable to a more general deficiency in IFT caused by the reduced number of complex B particles. Analysis by PCR revealed that the suppressor mutation involved a rearrangement or deletion somewhere in the region between the 3′ end of the inserted gene and the seventh intron of the gene (unpublished data). To determine if this mutation caused a change in the transcription of the gene, wild-type, , and Sup1 cells were examined by real-time PCR using primer pairs designed to assay for the presence of the 5′ end, middle, and 3′ end of the mRNA (). In wild-type cells, all three regions were detected. In the mutant, only the 5′ end was detected, indicating that the 5′ end of the gene is transcribed, but a full-length mRNA is not made. Because our antibody to the N terminus of IFT46 did not detect a product, it appears that the truncated mRNA is not translated into a stable protein. This provides further evidence that is a null allele. In Sup1 cells, both the 5′ and the 3′ end, but not the middle, were reproducibly detected. Therefore, the suppressor mutation results in transcription, and possibly translation, of the 3′ end of the IFT46 gene. Our antibody would not detect a product containing the C-terminal end of IFT46, but lacking its N-terminal end. However, we can rule out the possibility that the suppressor mutation results in translation of the N-terminal part of IFT46 fused with the C-terminal part of IFT46, because our antibody did not detect any product in Sup1 cells. Transcripts encoding the 3′ end of the IFT46 gene were detected in Sup1 cells in both the presence and absence of aeration, so the suppressor mutation, not stress, causes transcription of the 3′ end of the gene. The IFT46 sequence, which is reported for the first time in this study, demonstrates that this protein, like other IFT particle proteins, is highly conserved among ciliated organisms. IFT46 was previously reported to cosediment with other IFT complex B proteins (). We confirm that IFT46, in both and mammals, is a complex B protein based on cosedimentation, coimmunolocalization, and coimmunoprecipitation with other complex B, but not complex A, proteins. Our analysis of a mutant and a suppressed strain of the mutant indicate that IFT46 is necessary for complex B stability and is specifically required to transport outer dynein arm complexes into the flagella. Our sequencing of IFT46 also revealed that it is a homologue of DYF-6 (), the sequence of which was recently reported by . DYF-6-GFP moves at IFT rates in sensory cilia and is required for ciliary assembly (), providing independent and complementary evidence that IFT46/DYF-6 is an essential IFT protein. Although both cilia and mammalian primary cilia lack outer arms, it may be that IFT46 is required for transport of other cargos in these immotile cilia, or simply is needed for complex B assembly, as in . strains 137c (, , ), CC124 (, , ), and S1D2 were obtained from the Genetics Center (Duke University, Durham, NC). T8a4-11 (, , , ) was generated by K. Kozminski and J. Rosenbaum (Yale University, New Haven, CT) by transforming KK30A3 (, , , ) with the plasmid pMN24 linearized with EcoRI. YH6 (, , ) is an offspring of a cross between T8a4-11 and CC124. Sup1 is a spontaneous partial suppressor for YH6. Cells were grown in M ( medium I altered to have 0.0022 M KHPO and 0.00171 M KHPO), M-N (M medium without nitrogen), or TAP () media. Murine IMCD3 cells (CLONTECH Laboratories, Inc.) were grown as described in . The antibodies used are listed in Table S1 (available at ). The rabbit antibody to IFT46 was generated against a synthetic peptide corresponding to the protein's N-terminal 19 amino acids (Pocono Rabbit Farm) and affinity purified using the same 19 amino-acid peptide. 16S IFT particles were purified from flagella () and the particle proteins separated by PAGE. A band corresponding to IFT46 was excised and microsequenced. Two peptides, VPRPDTKPDYLGLK and IKPFIPDYIPAVGGIDEFIK, were obtained; these identified the protein C_2130007 predicted by the genome (v. 2; ). Several EST clones in Genbank that contain these two peptides in their ORF were used to clone the IFT46 cDNA. A 4.8-kb fragment that contains only the full-length gene was cloned from genomic DNA after its amplification by PCR with primers IFT46-5 and IFT46-6 (the sequences of all primers are given in Table S2, available at ) using eLONGase Enzyme Mix (Invitrogen). The sequences of the 5′- and 3′-UTRs of the IFT46 cDNA were verified by sequencing the cloned gene; the sequence of the coding region was verified by sequencing a PCR product from a cDNA library. IFT46 homologues were identified by searching the translated nr database at . Sequences were analyzed as described in . DNA was isolated from as described in . DNA gel electrophoresis was carried out by standard procedures (). Southern blotting was performed using the DIG High Prime DNA Labeling and Detection Starter Kit II (Roche); instead of using the kit's hybridization buffer, we used Church buffer (7% SDS, 1 mM EDTA, and 0.25 M NaHPO, pH 7.2; ) and hybridized it at 65°C. IFT46 gene induction upon deflagellation was analyzed by real-time PCR, as described in , using primers IFT46-3 and -4. The ratio of the amount of IFT46 message after deflagellation to that before deflagellation was calculated for each trial. Three independent sets of mRNA were isolated and analyzed three times each. To measure the mRNA levels of IFT140, IFT81, and IFT72, cDNAs were prepared from cells at the mid-log phase of growth, and quantitated by real-time PCR as described in using the primer pairs IFT140F/IFT140R, IFT81F/IFT81R, or IFT72F2/IFT72R2. Two independent sets of mRNA were isolated and analyzed three times each. To assay transcription of the gene, cDNAs were prepared from wild-type cells and from and Sup1 cells with or without aeration. Real-time PCR was performed using primer pairs to the 5′ end of the gene (IFT46-11/IFT46-2), middle part of the gene (IFT46-9/IFT46-28), and the 3′ end of the gene (IFT46-3/IFT46-4). Samples were normalized using G protein β subunit (). The end products were examined on a 1.5% agarose gel. Three independent sets of mRNA were isolated and analyzed three times each. A collection of insertional mutants having motility defects was screened for a defect in by Southern blotting, using an 864-bp partial cDNA fragment amplified by PCR with primers IFT46-Q1 and IFT46-Q2 as a probe. The mutated region in the gene in the mutant was located by PCR using primer pairs 1 (IFT46-1/IFT46-2), 2 (IFT46-9/IFT46-10), and 3 (IFT46-3/IFT46-4). The open reading frame of MmIFT46 was PCR amplified from mouse testis cDNA using primers mIFT46-1 and -2. The PCR product was digested with BamHI, cloned into the BglII site of pJAF113.1, and called pJAF161.24. pJAF113.1 was derived from p3XFLAG-myc-CMV-26 (Sigma-Aldrich) by filling in the HindIII site to shift the polylinker by four nucleotides. pJAF161.24 encodes a fusion protein in which the 3× Flag tag is fused to the N-terminal end of IFT46. pJAF146.1 encoding GFP-Flag was constructed by moving the XbaI–EcoRI GFP-containing fragment from pEGFP N2 (CLONTECH Laboratories, Inc.) into pJAF113.1. pJAF134.3 encoding MmIFT20-Flag was described by . IMCD3 cells that had been transfected with JAF161.24 (MmIFT46-Flag), pJAF134.3 (MmIFT20-Flag), or pJAF146.1 (GFP-Flag) were lysed in Cell Lytic M (Sigma-Aldrich) containing 0.1% Tween 20 and 0.1% CHAPSO (Bio-Rad Laboratories). After incubation for 10 min at 4°C, the extract was clarified by centrifugation at 18,000 and treated for 10 min with Sepharose- 4B beads, which were removed by centrifugation through a Macro Spin Column (Harvard Apparatus). The treated extract was then incubated with anti-Flag M2-Agarose Affinity Gel (Sigma-Aldrich) for 1 h at 4°C. Unbound proteins were removed by washing the beads in Wash Buffer (Sigma-Aldrich) containing 1% Tween 20 and 150 mM NaCl, followed by washes in Wash Buffer alone. Bound proteins were eluted with 200 ng/μl 3× Flag Peptide (Sigma-Aldrich) and analyzed by Western blotting, as described in . Preparation of whole-cell extracts, isolated flagella, and the flagellar membrane plus matrix fraction, as well as PAGE and Western blotting, were performed as described in . Sucrose gradient analysis was carried out as described by . T8a4-11 and CC124 gametes were induced to mate by dibutyryl-cAMP and papaverine treatment, and the zygotes were selected by the freeze/thaw method (; ). mutant cells were cotransformed () with linearized plasmid pSP124S () and the cloned 4.8-kb fragment, which contains only the wild-type gene. cells were fixed in glutaraldehyde for EM () and processed as described in . cells were fixed and stained for immunofluorescence microscopy by the alternate protocol of , using Alexa Fluor 488– or 568–conjugated secondary antibodies (Invitrogen); images were acquired at room temperature with an AxioCam camera, AxioVision 3.1 software, and an Axioskop 2 plus microscope equipped with a 100×/1.4 NA oil DIC Plan-Apochromat objective (all from Carl Zeiss MicroImaging, Inc.) and epifluorescence. Mammalian cells were fixed in 2% paraformaldehyde and processed for immunofluorescence microscopy as described by . Images were prepared for final publication using Photoshop 6.0 (Adobe). Table S1 shows the antibodies used in this work. Table S2 shows the primer sequences used in this study. The online version of this article is available at .

Integrins are heterodimeric transmembrane adhesion receptors composed of α and β subunits, which serve, often in combination with receptor tyrosine kinases, to control cell adhesion, migration, proliferation, differentiation, and survival (; ; ; ). Together, these functions govern morphogenesis and tissue homeostasis and, when deregulated, contribute to tumorigenesis and cancers (; ; ; ; ; ; ; ). α3β1, α2β1, α5β1, αvβ6, and α6β4 are all found in skin epidermis, where they bind to their ligands in the basement membrane and serve a diverse array of functions (; ; ). Conditional targeting of the genes encoding the hemidesmosomal integrins α6 and β4 results in epidermal–dermal detachment, skin blistering, and defects in cell survival (; ; ), whereas targeted deletions of and result in microblistering and wound defects, respectively (). In contrast, ablation causes defects in epidermal–dermal attachment, basement membrane organization/assembly, hair follicle downgrowth/morphogenesis, and epidermal wound closure (; ; ; ). αβ1 integrins form focal adhesions (FAs), composed of a complex group of proteins and signaling molecules that associate with the actin and microtubule cytoskeletons and orchestrate these diverse functions. Primary -knockout (KO) epidermal keratinocytes can be cultured in vitro under conditions where they proliferate similarly to their wild-type (WT) counterparts (). These keratinocytes display alterations in cell-matrix adhesion, enlarged, peripheral FAs, and robust actin stress fibers, suggestive of properties that might contribute to the perturbations in hair follicles and epidermal wound closure seen in vivo. However, the molecular mechanisms by which αβ1 integrins control the diverse set of cell biological processes in epidermal keratinocytes remain largely unknown. Integrins lack endogenous enzymatic activity and are thus believed to depend on signal transducers such as the nonreceptor kinases FAK, its sequence homologue proline-rich tyrosine kinase 2 (PYK2), or integrin-linked kinase (ILK), as well as a variety of scaffolding proteins that link integrins to the actin cytoskeleton to unfold their functions (). Upon integrin engagement with the extracellular matrix, FAK becomes activated and physically interacts with β1's cytoplasmic tail as well as with various signaling molecules at FAs (; ). In cultured –null epidermal cells, FAK activity is markedly reduced, revealing FAK as a potential transducer of β1 integrin's diverse activities in skin (). The notion that FAK-mediated β1 integrin signaling controls these processes in the skin epidermis is suggested from conditional gene targetings, which have revealed defects in basement membrane assembly and/or remodeling in developing dorsal forebrain (). However, although -deficient fibroblasts exhibit restricted migration, endothelial cells and HeLa cells show increased motility (; ; ; ). Additionally, loss of FAK results in apoptosis in embryonic fibroblasts and endothelial cells, but not in meningeal fibroblasts (; ; ; ). Whether cellular context accounts for these seemingly opposing findings is not yet clear. Recently, was conditionally targeted in postnatal mouse skin epithelium, revealing defects in hair follicles and sebaceous glands () and resistance to tumorigenesis (). However, the molecular and cellular mechanisms responsible for these phenotypes have yet to be elucidated, and the differences in KO strategies and genetic backgrounds preclude comparisons between and conditional targetings. To unravel the relative importance of epidermal FAK in mediating downstream signaling functions of β1 integrins, we have mated ( mice with the same recombinase mice that we used previously to target . For these comparisons, we also derived primary epidermal keratinocytes from both -null and -null skins. We show that, in contrast to β1 and distinct from some prior functions attributed to FAK, FAK is largely dispensable for epidermal adhesion, basement membrane organization, proliferation, survival, and terminal differentiation. Rather, FAK controls cytoskeletal dynamics and FA disassembly, and without it, keratinocytes are perturbed in their motility and their ability to polarize migration out of skin explants. Mechanistically, our data suggest that these defects are in part due to decreased phosphorylation of Src and p190RhoGAP (GTPase-activating protein), yielding increased Rho/Rho kinase (ROCK) signaling and inefficient recruitment of the p95 paxillin kinase linker–PAK-interacting exchange factor–p21-activated kinase (PKL–PIX–PAK) complex to FAs. Finally, we substantiate the biological significance of these activities downstream of αβ1 integrin and FAK by inhibiting tension signaling with small molecule inhibitors and explore the consequences to FA dynamics. FAK activity has been found to depend on αβ1 integrin signaling in the skin epidermis (). To test the hypothesis that FAK functions as a key transducer of αβ1 signaling in epidermis, we mated mice () to mice where the second kinase domain of was flanked by loxP sites. Upon excision of floxed exons by Cre recombinase, an early stop codon is generated, ablating FAK protein expression (). Immunofluorescence microscopy documented specific loss of FAK in skin epithelium and not dermis, and this was further confirmed by immunoblot analyses (). Although, at birth, conditional KO (cKO) mice were phenotypically indistinguishable from WT littermates, their hair coat did not emerge on time (Fig. S1, A and B, available at ; ). Histological analysis further revealed a defect in follicle downgrowth (Fig. S1 C). This defect appeared to be rooted in a failure of follicles to migrate downward, rather than alterations in proliferation and/or apoptosis (Fig. S1, D and E). Detection of such early neonatal defects was uniquely possible with K14-Cre mice, which embryonically ablate floxed genes in epidermis (). To determine whether FAK is required for integrin expression or basement membrane formation in skin, we first conducted immunofluorescence microscopy using antibodies against β1 integrin, expressed in dermal fibroblasts and basal epidermal cells, and laminin 5, the primary ECM ligand for epidermal integrins and a major component of the basement membrane at the dermal–epidermal boundary. These labeling patterns were indistinguishable between WT and FAK-deficient skin (). Ultrastructural analyses further revealed an intact basement membrane, accompanied by structurally normal α6β4 integrin–containing hemidesmosomes of comparable length and density (). An explanation of the box-and-whisker diagram () used to graphically display the spread of these data can be found in Fig. S2 (available at ). We next used FACS analyses to quantify integrin surface levels on freshly isolated epidermal keratinocytes (). Normal surface levels of β1, β4, α2, α6, αv, and β6 integrins were observed, indicating that loss of FAK did not affect integrin expression. Moreover, as judged by FACS with an antibody specific for activated β1 integrin, no differences were noted in the activity of αβ1 integrins in the absence of FAK. Consistent with these observations, WT and null epidermal cells adhered similarly to a variety of purified ECM substrates of the underlying basement membrane (). Collectively, these studies suggest that FAK functions in the subset of αβ1 integrin functions required for hair follicle morphogenesis, but it is dispensable for basement membrane assembly, integrin expression, adhesion, and tissue homeostasis. To probe for molecular and biological functions of FAK in epidermal keratinocytes, we cultured primary keratinocytes on fibroblast feeder layers (). KO keratinocytes were propagated over multiple passages without major growth defects, and feeder cells could be withdrawn upon passaging. However, these cells appeared to be more sensitive to serum starvation than WT keratinocytes. Thus, on fibronectin (FN)-coated dishes, MAPK activity, as measured by p42/44 phosphorylation, was reduced in serum-starved KO, as compared with either starved WT or serum-stimulated KO or WT keratinocytes. In contrast, the phosphorylation of other signaling molecules such as AKT appeared unchanged. We next examined whether ablation affected the activity and/or subcellular localization of the two other β1 signal transducing, nonreceptor kinases, PYK2 and ILK (; ; ) In the absence of FAK, Pyk2 was slightly hyperphosphorylated, but phospho-Pyk2 still localized to Fak-deficient FAs. ILK levels and localization were similar between WT and KO keratinocytes (). In contrast, activated FAK and PYK2 were markedly reduced in -null keratinocytes, where some delocalization of ILK from FAs was also noted (). PYK2/ILK signaling was still at least partially maintained in KO cells; however, rhodamine-phalloidin labeling revealed marked defects in actin organization and cell spreading, resembling those seen in null keratinocytes. To investigate these defects in greater detail, we plated keratinocytes on FN-coated glass dishes, fixed and stained them at multiple intervals, and quantified the contact area between the cell and its underlying substratum (). In the absence of FAK, actin cables lined cell boundaries, while atypically dense bundles of actin stress fibers converged at FAs, often crossing intracellularly at multiple sites. This was in striking contrast to WT keratinocytes, which displayed organized actin networks, with relatively modest stress fibers converging in parallel arrays at peripheral FAK-positive FAs. These defects were accompanied by marked abnormalities in the rate and extent of spreading in FAK-deficient cells. As depicted in the box-and-whisker diagrams, the perturbations in cell shape and spreading became more pronounced over time, as did the range of effects observed ( and Fig. S2). To further investigate the root of FAK-dependent changes in cytoskeletal architecture and cell shape/spreading, we counterstained rhodamine-phalloidin labeled keratinocytes with antibodies against the FA molecules vinculin (VIN) and paxillin (PXN). VIN and PXN localized at the sharp, constricted cellular apexes that were closely associated with massive bundles of actin fibers lining the periphery of aberrantly shaped -deficient keratinocytes (). Although peripheral FAs exhibited markedly enhanced staining intensity and area in FAK-deficient cells, the smaller and more weakly stained central FAs were largely absent (). Defects in cell shape; actin organization; and FA size, distribution, and intensity were largely restored after reexpressing FAK in KO keratinocytes (). These data underscore the importance of FAK in FA regulation and actin organization. To examine actin–FA dynamics in real time, we first mated ) mice on the background of previously generated transgenic mice () and then transfected null and WT keratinocytes with an expression vector (; ). Video microscopy revealed a considerably more static actin cytoskeleton and associated FAs in null keratinocytes relative to their WT counterparts ( and Video 1, available at ). Most WT keratinocytes extended lamellipodia uniformly around their circumference, and as they formed, lamellipodia seeded numerous focal complexes connected by short, thin actin bundles (). A few of these focal complexes slowly developed into more robust FAs, which then served as convergence sites for larger stress fibers. In contrast, null keratinocytes formed lamellipodia primarily around existing robust FAs, with new focal complexes maturing rapidly into atypically large peripheral FAs associated with unusually thick actin cables ( and Videos 1 and 2). WT keratinocytes displayed dynamic FAs, which were continually drawn toward the cell center ( and Video 1). In contrast, the large FAK-deficient FAs accumulated at cell edges for extended times ( and Video 2). After prolonged periods of cellular tugging, retractive forces often caused abrupt release of associated cellular contacts with the substratum, resulting in rapid, uncoordinated changes in FAK KO morphology (Videos 1 and 2). To assess differences in the kinetics of FA dynamics, we monitored individual FAs in keratinocytes transfected with GFP-PXN and calculated the rate constants for both FA assembly and disassembly (). Although FA assembly rates were only slightly lower in KO versus WT keratinocytes, disassembly rates were significantly decreased (). This was further substantiated by FRAP experiments, where a significant increase was noted in the half-times of fluorescence recovery after photobleaching (). In contrast, the mobile fraction of GFP-PXN appeared unaffected by the status of FAK (). These data provide compelling evidence that the large FAs in FAK-deficient keratinocytes arise from a defect in FA disassembly. The observed defects in FA disassembly and cytoskeletal organization led us to posit that FAK may function in promoting directed cell migration. We tested this possibility by first generating tissue explants from , WT, and cKO skins placed on FN-coated glass dishes and then monitoring the outgrowth of interconnected epithelial sheets of GFPactin-expressing keratinocytes from these explants (). Quantification revealed a marked delay in the outgrowth from cKO explants compared with WT explants (), whereas lamellipodial activity appeared similar between WT and KO explants (Video 3, available at ). WT explants polarized their actin cytoskeleton, exhibiting parallel actin bundles that were oriented perpendicularly to the leading edge of the outgrowing cells (). Antibodies against E-cadherin, which mark stable cell–cell contacts, underscored the elongated, polarized character of these epidermal sheets. Actin fibers associated with PXN-labeled FAs were also distributed evenly and unidirectionally oriented toward the leading edges of the outgrowing WT explants. In contrast, null epidermal explants exhibited a seemingly random orientation of actin bundles, often crossing at multiple places, whereas enlarged FAs pointed in different directions in disoriented cells at the leading edge (). A priori, the differences in explant outgrowth could be due solely to a failure to coordinate cytoskeletal dynamics across the outgrowing epithelial sheet. Alternatively, it could be that the defects in outgrowth stem from intrinsic defects in the migration of individual cells, which subsequently fail to coordinate directed movements within the outgrowing sheet. To assess the motility of individual keratinocytes, equal numbers of WT and KO keratinocytes were seeded in the top compartment of Boyden chambers and assayed for the number of cells that migrated through the filter to the FN matrix located in the chamber below. Quantification revealed that the number of WT cells migrating through the filter was approximately five times higher than the number of KO cells (). These data suggest that FAK promotes FA dynamics and directed migration in keratinocytes and supports the polarized, unidirectional migration of epithelial sheets at the wound edge. The migration defects arising from loss-of-function FAK mutations pointed to a major role for FAK in regulating FA dynamics for the purpose of orchestrating cytoskeletal organization and cellular movements. Increasing evidence has implicated microtubule targeting to FAs as an initial step in FA turnover (; , ; ; ). To assess whether microtubule targeting to FAs might be altered by FAK deficiency, we used immunofluorescence and video microscopy to examine microtubule networks. In WT keratinocytes, microtubules formed a regular array around the nucleus and projected toward the cell periphery (Fig. S3, available at ). In null keratinocytes, this organization was perturbed, and the majority of microtubules pointed toward the robust FAs that were associated with massive stress fibers. Time-lapse analyses of keratinocytes transfected with either GFP-tubulin (Video 4) or a GFP-tagged plus-end microtubule binding protein (Eβ1; Video 5) further indicated that dynamic microtubules target robust, peripheral FAs, resulting in an altered distribution of the microtubule cytoskeleton in the absence of FAK. Collectively, our data suggest that FAK modulates FA dynamics in mouse epidermal keratinocytes but it is not essential for the targeting of dynamic microtubules to FAs. Additionally, as microtubule targeting is thought to promote FA turnover, the presence of microtubules at FAK-deficient FAs implies that, whatever the mechanism involved in microtubule-mediated FA turnover, it is dysfunctional in the absence of FAK. This result supports previous observations where FAK-deficient embryonic fibroblasts do not disassemble FAs when microtubules repolymerize in response to washout of the microtubule depolymerizing drug nocodazole (). The main characteristic of KO keratinocytes is robust FAs, which are tightly associated with prominent stress fibers. Elevated Rho activity can result in elevated stress fiber formation and stabilization of FAs (), whereas RhoA inactivation by p190RhoGAP can promote cell spreading and migration (). Furthermore, FAK has been associated with the transient suppression of Rho activity during cell spreading () and it has been speculated that FAK may cooperate with β1 integrin and Src tyrosine kinase to phosphorylate and activate p190RhoGAP at FAs (; ; ; ). To test this hypothesis, we first assessed Src activity in FN-stimulated keratinocytes grown in the presence or absence of serum. Immunoprecipitation of Src followed by anti–pY418-Src immunoblot analysis revealed a marked reduction of phosphorylated (active) Src in -null keratinocytes grown under serum-free conditions where Src activation is likely to emanate primarily from integrin activation (). As shown by its faster electrophoretic mobility, Src was mostly unphosphorylated (inactive). In contrast, no considerable changes in the status of Src were noted in cells cultured in the presence of serum, where Src activation can additionally occur through ligand-mediated engagement of transmembrane tyrosine kinase receptors. These results agree with the previously established requirement of FAK autophosphorylation for efficient recruitment of Src to FAs (). We assessed the status of p190RhoGAP by immunoprecipitating it from total cell lysates and conducting an anti-phosphotyrosine (p-Tyr) immunoblot analysis. p190RhoGAP was hypophosphorylated in KO cells grown without serum, whereas comparable amounts of tyrosine-phosphorylated p190RhoGAP were detected when serum was present (). The requirement for FAK in phosphorylating and activating both Src and p190RhoGAP at integrin activation sites provides an underlying mechanism for the convergence of robust actin fibers at sites of FAs in KO keratinocytes. RhoGTP can activate ROCK, resulting in the phosphorylation and inactivation of the regulatory subunit of myosin light chain (MLC) phosphatase (myosin phosphatase target [MYPT]), leading in turn to increased phosphorylation and activation of MLC and stress fiber formation (). To test whether this tension-signaling pathway is hyperactivated in null keratinocytes, we examined the levels and phosphorylation status of MYPT and MLC (). Although total MYPT levels appeared to be slightly reduced, MYPT was hyperphosphorylated (inactive) at its ROCK-sensitive threonine residue. The outcome of these changes appeared to be a small but statistically significant increase in the total levels of active pMLC, as judged not only by immunoblot analysis but also quantitative immunofluorescence microscopy (). Moreover, in WT cells, anti-pMLC decorated the central actin fibers as well as the cortical actin network, whereas in KO cells, anti-pMLC displayed a more robust decoration of the massive stress fiber network. To determine whether this increase in Rho/pMLC-mediated tension is attributable to the spreading defects of KO keratinocytes and/or their defective FA dynamics, we treated keratinocytes with the ROCK inhibitor Y-27632, the myosin inhibitor blebbistatin, and the MLC kinase (MLCK) inhibitor ML7, and then conducted immunofluorescence microscopy and quantitative relative integrated fluorescence intensity analyses to evaluate the effects on cell spreading, FA size, and FA localization. As shown in , the differences in WT versus KO keratinocyte spreading were largely ameliorated by Y-27632 and blebbistatin, and to a lesser extent by ML7. Quantitative analyses of FA staining intensity are shown in (see also Fig. S4, available at ). These data further revealed that the statistical distribution of FAs in KO keratinocytes, which was highly skewed toward larger, more intensely stained FAs (mock), was significantly restored to normal upon treatment with these RhoA/tension-relieving drugs. In contrast, treatment with nocodazole, which depolymerizes microtubules and enhances Rho activity, resulted in increased FA size and VIN staining intensity even in KO cells. Quantitative analyses further revealed a clear shift in the distribution of FAs to a more peripheral location (Fig. S4). MLCK can be phosphorylated and inhibited by PAK (), which has recently been proposed to form a complex with PKL, βPIX, and PXN at FAs (; ; ). To test whether the function of this complex might require FAK activity, we first examined the relative levels of PKL, βPIX, and PAK. As judged by immunoprecipitation and immunoblot analyses, PKL was markedly reduced in KO keratinocytes (). This was confirmed by immunofluorescence microscopy where anti- PKL labeled WT FAs but was only weakly detectable at KO FAs (). This difference was substantial, as KO FAs are considerably larger and thus would have otherwise been expected to label more strongly with PKL. By immunoblot analyses, overall levels of βPIX and PAK appeared to be largely unaffected by deficiency (). However, even though βPIX localization appeared normal, PAK localization was markedly altered in KO keratinocytes (). Rather than the strong anti-PAK labeling of FAs in WT keratinocytes, anti-PAK labeling was largely diffuse in FAK-deficient cells and appeared reduced in intensity over KO FAs. Moreover, when we examined the status of activated (phosphorylated) PAK, we found that anti–phospho-PAK (p-PAK) labeled WT FAs but was barely detected in -null FAs and largely diffuse throughout the cytoplasm (). Based on these data, we conclude that PAK was neither recruited nor activated efficiently at FAs in the absence of FAK. For the present study, we established a system that allowed us to directly compare the relative contributions of FAK to the many diverse functions of αβ1 integrins in the epidermis. Our studies provide compelling evidence that although αβ1 integrins are known to function broadly in cell adhesion, basement membrane assembly and organization, tissue homeostasis, and balancing growth and differentiation, FAK is selectively involved in efficient cell spreading, regulation of FA and cytoskeletal dynamics, and directed migration (Fig. S5 A, available at ). When our present data are coupled with our previous observation that FAK is hypophosphorylated in KO keratinocytes (), we could position FAK activation downstream of αβ1 signaling and upstream of this subset of cellular functions. PYK2 localized to the FAK-deficient FAs. However, in contrast to the -deficient FAs, PYK2 was still phosphorylated in the absence of FAK. Hence, we surmise that some of the more global alterations seen in KO cells may reflect the loss of overlapping but discrete subsets of signaling functions mediated through FAK and PYK2. Additionally, although ILK was still expressed comparably, its localization in -null keratinocytes was diffuse, suggesting that one or more of its functions () may be selectively compromised in the -null but not -null state. Our loss-of-function studies in keratinocytes revealed that Rho/ROCK signaling was hyperactive in the absence of FAK, as reflected by increased MLC phosphorylation and massive cortical stress fiber bundles, which converged on large, peripheral FAs. The enhanced Rho activity in null keratinocytes implies that FAK indirectly or directly must either activate RhoGEFs (guanine nucleotide exchange factors) or repress RhoGAPs (). In neural development and in fibroblasts, p190RhoGAP can be phosphorylated and activated in response to integrin and Src signaling to temper RhoA activation and promote cell spreading and migration (; ; ). Consistent with this notion, -null fibroblasts are not able to suppress Rho activity during cell spreading (). Our data unveil an essential role for FAK in activating Src and p190RhoGAP upon integrin engagement. Thus, in keratinocytes, FAK is required for integrin/Src-dependent p190RhoGAP phosphorylation, suggesting a model where FAK/Src-induced phosphorylation of p190RhoGAP at FAs might locally promote RhoGTP hydrolysis to suppress Rho-induced stress fiber formation and FA stabilization (Fig. S5 B). Although we were unable to establish a physical association between endogenous p190RhoGAP and FAK (unpublished data), FAK has been shown to bind p190RhoGAP and phosphorylate it in vitro (). When coupled with the biochemical interactions between these molecules, the FAK, Src, and integrin loss-of-function analyses provide a tight molecular link between Fak/Src-mediated integrin signaling and the regulation of Rho activity via p190RhoGAP. Rho activity can result in elevated stress fiber formation and stabilization of FAs (), whereas RhoA inactivation by p190RhoGAP promotes cell spreading and migration (). Consistent with the notion that elevated Rho is responsible for the defects in FA dynamics and cell spreading seen in our -null keratinocytes, these defects could be largely suppressed by inhibitors of ROCK or myosin II. MLCK inhibition also had an effect, suggesting that multiple pathways are involved in regulating FA-mediated tension. In fibroblasts, MLCK can be inactivated by PAK (), and a constitutively active form of PAK causes dissolution of stress fibers and FA reorganization (). Our finding that endogenous phosphorylated (active) PAK is diminished at FAs in null keratinocytes extends these earlier inverse connections between activated PAK and MLCK-mediated tension. Moreover, when considered with the knowledge that PKL, βPIX, and PAK form a complex that binds to PXN at FAs (; ; ; ; ; ; ), our data provide compelling evidence that inactivation of the PXN–PAK–MLCK pathway in the absence of FAK activity might lead to increased tension and larger, less dynamic FAs. Our findings point to a function for FAK in localizing PKL–PIX–PAK to FAs to ease MLCK activity. How FAK regulates PKL–PIX–PAK localization remains unclear, but we do see a reduction in PKL at null FAs, which could account for the markedly reduced PAK activity at these sites. It was recently shown that PKL can be cooperatively phosphorylated on multiple tyrosine residues by FAK and Src to promote PKL localization to FAs (). Overall, these data support a model whereby FAK promotes the recruitment and activation of PAK at FAs to suppress MLCK activity. This adds an additional way in which activation of FAK tips the balance in favor of the unphosphorylated (inactive) form of MLC to reduce tensile stress fibers associated with FA dynamics (Fig. S5 B). The migratory behavior of cells is complex and can be altered by a variety of parameters. Providing a structural link between the matrix and the cytoskeleton are integrins and their associated scaffolding and signaling molecules. This machinery couples matrix stiffness to cellular tension and is a crucial determinant of cell motility (). Our studies provide functional evidence to reinforce the notion that FAK is an essential component of FA distribution and dynamics within cells. In the absence of FAK, the small central FAs diminished dramatically in number and were replaced by large peripheral focal contacts. Kinetic studies on FA assembly and disassembly indicate that FAK's major requirement resides in promoting FA disassembly, although moderate alterations in assembly rates were observed (this study; ). The defects in directed migration of cells exiting from null skin explants were striking and graphically revealed the physiological consequences to tension-induced perturbations in actin organization. The reduced FA dynamics also appeared to contribute, as robust FAs were misoriented in FAK-deficient explants. The pronounced targeting of microtubules to these FAs is consistent with prior studies showing that microtubules grow along actin stress fibers (; ). This finding suggests that despite their robust appearance, these FAs are still able to undergo the microtubule-mediated turnover first described by (for review see ). Future studies will be necessary to explore these connections in depth. The grossly perturbed epidermal migration that we observed is consistent with both the abnormalities seen in hair follicle morphogenesis and the resistance of null skin to tumorigenesis and metastasis (; ). Although postnatal ablation of FAK in the skin epithelium resulted in no significant wound healing defects (; ; unpublished data), delays in epidermal wound closure were noted in null skin (). We surmise that our enhanced ability to detect and monitor migration differences in skin explants arises from an increase in keratinocyte mobilization in vitro where growth conditions are optimal and hemidesmosomal adhesion to ECM is reduced (). It is also noteworthy that FAK's specific functions may vary both with changes in microenvironment and inherent differences in cell types. The importance of these contexts is underscored by null–mediated basement membrane assembly defects, which were observed in cerebellum () but not in embryonic fibroblasts () or skin epidermis (this study). Furthermore, although both null primary keratinocytes and immortalized embryonic fibroblasts exhibit elaborate arrays of stress fibers and FAs that impair migration, FAK-deficient HeLa cells display elevated Rac rather than Rho activity and show enhanced rather than impaired migration (; ; ; ). In contrast, null endothelial cells spread poorly and display aberrant lamellipodial extensions and altered actin cytoskeleton, and yet surprisingly, they exhibit no obvious perturbations in polarized migration during vascularigenesis in tissue explants (). Finally, mutant fruit flies, which do not express a like gene, are surprisingly viable and fertile and show no defects in either integrin function or cell migration (). In closing, our analysis of FA dynamics in primary keratinocytes lacking FAK function has provided new insights and strengthened prior notions as to how FAK activity converges on the constellation of pathways that regulate cytoskeletal dynamics and balance FA assembly and disassembly in cells. Our findings are particularly interesting in the context of recent in vivo studies, which showed that oncogenic transformation is severely reduced in skin with reduced or abrogated FAK function (, ). In this regard, FA signaling has recently been implicated in oncogenic transformation and activated FAK is a well-established marker for both transformation and metastasis. The identification of FAK-mediated alterations in several specific FA-associated proteins, cytoskeletal changes, and FA dynamics now illuminates several potential molecular pathways by which FA signaling might result in oncogenic transformation. These new findings provide fertile ground for future investigations in this arena. mice (), mice (), and mice () have been described. Homozygous FAK floxed; and ; and , animals were generated by mating, and their genotypes were determined by PCR. Primary mouse keratinocytes (MKs) were isolated from the epidermis of newborn mice using trypsin, after prior separation of the epidermis from the dermis by a 2-h dispase treatment. MKs were plated on either mitomycin C–treated 3T3 fibroblast feeder cells or onto different ECM components. Cells were cultured in E-media supplemented with 15% serum and a final concentration of 0.05 mM Ca (). Cell adhesion assays were performed as described previously (). Wells were coated using 10 μg/ml FN, 10 μg/ml laminin-1, 10 μg/ml collagen-4, and poly--lysine. Cell spreading assays were performed by plating MKs at low density on FN-coated coverslips. Cells were fixed 2, 12, 24, or 36 h after plating and stained with TRITC-phalloidin and DAPI. Suppression of tension signaling during spreading was analyzed by plating MKs at low density on FN-coated coverslips for 12 h in E-media containing 0.05 mM Ca. After 12 h, the medium was exchanged for medium containing 5 μM ML7 (BIOMOL Research Laboratories, Inc.) or 5 μM Y27632 (Calbiochem) and incubated for an additional 12 or 24 h before fixation. For ectopic expression and rescue experiments, MKs were transfected with RFPzyxin (M. Beckerle, University of Utah, Salt Lake City, UT), pEGFP-PXN (E. Marcantonio, Columbia University Medical Center, New York, NY), and pBABE-FAK (Y. Pylayeva and F. Giancotti, Memorial Sloan-Kettering Cancer Center, New York, NY) using Fugene6 (Roche Applied Science). Kinetics of FA assembly and disassembly were performed as previously described (). MKs were plated on FN-coated dishes (MatTek) in media containing 0.05 mM Ca and transfected with GFP-PXN. Time series were acquired on a spinning-disc confocal microscope equipped with a 100× α-plane fluar (1.45 oil) lens and an EM charge- coupled device camera (Hamamatsu). The rate constants for FA assembly and disassembly were obtained by calculating the slope of relative fluorescence intensity increases or decreases of individual FAs on a semilogarithmic scale against time. For FRAP experiments, MKs were plated on FN-coated 3.5-cm dishes in media containing 0.05 mM Ca and transfected with GFP-PXN. FRAP experiments were performed on a microscope (DeltaVision; Stress Photonics) equipped with a 60× Plan APO N (1.42 oil) objective. The refractive index of the oil was 1.514 (Applied Precision). Five prebleach events were acquired followed by a 1-s bleach event. Fluorescence recovery was recorded for 120 s after photobleaching (71 frames), and data from these photokinetic experiments were analyzed using DeltaVision software. Explant outgrowth migration assays were performed as described previously () with minor modifications. In brief, explants were cut using a 3-mm dermal biopsy punch (Miltex), placed on FN- coated 35-mm, glass-bottomed plates (MatTek), and submerged in E-media containing 0.6 mM Ca. For video microscopy, explant cultures were incubated with E-media containing 0.6 mM Ca and 50 mM Hepes buffer, pH 7. Transwell migration assays were performed on 24-well plates. The underside of each well was coated with 10 μg/ml FN and placed atop fibroblast-conditioned E-media containing 0.05 mM Ca. Primary MKs were freshly isolated, and a total of 50,000 cells/well were plated in 100 μl E-medium containing 0.05 mM Ca. 12 h later, cells were washed off the top membrane and fixed on the bottom membrane. Cells were stained using hemotoxylin and eosin and counted under the microscope. Proteins were separated by electrophoresis on 6–10% PAGE gels or 4–12% gradient gels (Invitrogen), transferred to nitrocellulose membrane, and subjected to immunoblotting. Membranes were blocked for 30 min with 5% nonfat milk in PBS containing 0.1% Tween 20, except for phosphoantibodies and MLC, where membranes were blocked for 30 min with 3% BSA in PBS containing 0.1% Tween 20. Primary antibodies were generally used at a concentration of 1:1,000, and HRP-coupled secondary antibodies were used at 1:3,000. Immunoblots were developed using standard ECL (GE Healthcare) or Super Signal West Pico substrate (Pierce Chemical Co.). Freshly isolated primary MKs were washed twice with PBS blocking solution containing 2% FCS at 4°C. Cells were resuspended and incubated in primary antibody (1:50) in blocking solution for 30 min at 4°C, washed three times with blocking solution, and incubated with PE-conjugated secondary antibodies (1:100) for 30 min at 4°C. Cells were washed three times in blocking solution and resuspended in blocking solution containing propidium iodide. FACS analyses were performed on a FACScan (Becton Dickinson). BrdU incorporation was quantified by FACS using a FITC BrdU FACS Flow kit (BD Biosciences) following the manufacturer's instructions. Tissues were fixed for >1 h in 2% glutaraldehyde, 4% formaldehyde, and 2 mM CaCl in 0.05 M sodium cacodylate buffer and then processed for Epon embedding. Samples were visualized with a transmission electron microscope (Tecnai 12-G2; FEI). For hemidesmosome quantitation, EM images were taken with a digital camera (model XR60; Advanced Microscopy Techniques Corp.) at a magnification of 49,000×. A total of 150 images were randomly taken at sites of the dermal–epidermal boundary for each experimental model. Total continuous membrane length and individual hemidesmosomes' lengths along the plasma membrane were measured using ImageJ (NIH). Statistical analysis was performed using OriginLab 7.5 software. Box-and-whisker blots are used to describe the entire population without assumptions on the statistical distribution. A test was used to assess the statistical significance of differences between two experimental conditions, and analysis of variance in combination with a Tukey post hoc test was used to compare multiple experimental conditions. Fig. S1 describes defects in hair follicle morphogenesis in mice conditionally mutant for FAK in the skin epithelium. Fig. S2 is a schematic representation of a box-and-whisker diagram. Fig. S3 shows the convergence of stress fibers and microtubules at FAs. Fig. S4 illustrates the distribution, relative VIN staining intensity, and size of FAs after treatment with small molecule inhibitors of tension-signaling components. Fig. S5 summarizes the presented data in a schematic model. Videos 1 and 2 are time-lapse videos illustrating the dynamics of FAs (RFPzyxin) and the actin cytoskeleton (GFPactin) in WT and KO keratinocytes. Video 3 is a time-lapse video of GFPactin in keratinocytes migrating out of WT and KO epidermal explants. Video 4 is a time-lapse video of GFPtubulin. Video 5 is a time-lapse video of EB1-GFP in WT and KO keratinocytes. Online supplemental material is available at .

Monocyte/macrophage desensitization is characteristic for late-phase immune responses (). Confined proinflammatory cytokine expression and mediator synthesis is important to avoid pathological settings, such as sepsis or atherosclerosis (; ). Down-regulating proinflammatory cytokine expression (TNF-α, interleukin [IL]-1β, and IFNγ) or proinflammatory mediator release (nitric oxide and reactive oxygen species [ROS]) concomitantly switches the proinflammatory phenotype toward an antiinflammatory one. The latter is characterized by the synthesis of antiinflammatory cytokines, such as TGF-β or IL-10, and is often accompanied by cellular desensitization upon secondary proinflammatory stimulation (; ). Therefore, the identification of molecular mechanisms contributing to cellular desensitization attracted growing interest (; ). One factor attenuating proinflammatory gene expression is peroxisome proliferator–activated receptor (PPARγ). PPARγ is a nuclear hormone receptor that, upon agonist binding, transactivates gene expression as a heterodimer bound to retinoic acid receptor-α (). Its role in blocking proinflammatory gene expression comprises several options, mainly antagonizing signaling cascades. Specifically, PPARγ negatively regulates transcription factors by scavenging transcriptional coactivators, such as the cAMP-response element–binding protein or the steroid receptor coactivator-1 (). However, a direct association with the transcription factors NF-κB, NF of activated T cells, signal transducer, and activator of transcription or NF-E2–related factor 2 (; , ; ) blocks their recruitment to responsive elements in promoter structures of target genes. Recently, it has been shown that PPARγ is targeted to nuclear receptor corepressor–histone deacetylase-3 complexes in response to ligand-dependent SUMOylation (), protecting these complexes from proteosomal degradation. Normally, histone deacetylase-3 removes a corepressor complex, provoking expression of proinflammatory genes. Additionally, PPARγ represses activation of a mitogen-activated protein kinase, which keeps downstream transcription factors unphosphorylated and, consequently, inactive (). Moreover, PPARγ influences the cell cycle by up- regulating p21 expression, which is an established cell cycle inhibitor (), or down-regulating phosphatase PPA2, which is known to adjust E2F/DP DNA-binding activity, which is necessary for the G to S-phase transition (). In response to proinflammatory stimulation, PPARγ-dependent gene transcription also contributes to cellular desensitization. PPARγ agonists inhibit diacylglycerol (DAG)–PKC signaling by inducing DAG kinase-α (DGKα) expression (). This enzyme lowers the amount of DAG, which is an established PKC activator. Normally, DAG is released from membrane lipids and activates classical PKCs (). Based on gene induction of DGKα as the underlying mechanism, this type of desensitization demands at least 6–15 h. Thus, it appears that PPARγ transrepresses proinflammatory gene expression, often in a DNA-unbound state, by provoking direct protein–protein interactions. We provide evidence for a new PPARγ-dependent mechanism in blocking PKCα signaling. Depletion of PKCα is attenuated by PPARγ1 activation in RAW 264.7 cells or human primary monocyte–derived macrophages. Cytosolic localization of PPARγ1 interferes with PKCα cytosol to membrane translocation, which is a prerequisite for its activation-dependent depletion. Translocation is restored in cells transfected with a dominant-negative PPARγ1 mutant. Coimmunoprecipitation studies and a mammalian two-hybrid system revealed a direct PPARγ1–PKCα interaction as the underlying mechanism. PPARγ1 deletion constructs support the idea that ligand-dependent PPARγ activation is necessary for PKCα binding, which is mediated by the helix 1 of the PPARγ1 hinge domain. Our data suggest a new mechanism for how activation of PPARγ1 blocks PKCα translocation, thereby achieving cellular desensitization. Recent data demonstrate that monocyte/macrophage desensitization in response to phagocytosis of apoptotic cells is achieved by attenuating PKCα signaling, which blocks NADPH oxidase–dependent formation of ROS (). Therefore, we were interested in identifying molecular mechanisms interfering with PKCα depletion. A potential candidate known to affect the pro- versus antiinflammatory phenotype in monocytes/macrophages is PPARγ. Because controversial data exist concerning its expression in monocytic and macrophage cell lines, as well as in primary human monocytes and macrophages, we performed a first set of experiments determining PPARγ expression in the monocytic cell lines and primary cells under investigation. As shown in , PPARγ is constitutively expressed in murine RAW 264.7 macrophages. In contrast, in THP-1 cells, PPARγ is only fractionally expressed, but differentiation toward macrophages with 100 nM PMA for 24 h provoked up-regulation of PPARγ (, lane 2 vs. 3). A similar expression pattern is observed in primary monocytes and macrophages, respectively. PPARγ is only marginally expressed in monocytes, but induced upon differentiation toward macrophages (). To identify the expressed PPARγ isoform 1 or 2, we performed a Western blot using human PPARγ1-transfected human embryonic kidney (HEK) cells as a positive control. Taking into consideration that murine and human PPARγ1 are identical in size (475 aa), we conclude that PPARγ1 is expressed in RAW 264.7 macrophages, differentiated THP-1 cells, and primary macrophages (unpublished data). Based on these results, we choose RAW 264.7 cells, differentiated human THP-1 cells, and primary monocyte–derived macrophages as experimental cell models. To analyze the role of PPARγ in macrophages in affecting PKCα activation, we pretreated RAW 264.7 macrophages for 1 h with the PPARγ agonists ciglitazone and rosiglitazone, followed by the addition of 100 nM PMA, which is a DAG homologue and established activator of PKCα. As expected, PKCα depletion was observed in control cells in response to 100 nM PMA (, lane 2). Depletion of PKCα was attenuated in cells prestimulated with a PPARγ agonist, such as ciglitazone (, lanes 3 and 4) or rosiglitazone (, lanes 5 and 6), in a concentration-dependent manner. However, 1 μM PMA-mediated PKCα depletion was not blocked (unpublished data). From these data, we conclude that PPARγ agonists attenuate activation-dependent PKCα depletion, in part controlled by the magnitude of the PKCα–activating stimulus. In PPARγ1 activating function (AF) 2 mutant overexpressing RAW 264.7 macrophages (), pretreatment with 10 μM rosiglitazone or 10 μM ciglitazone did not inhibit PKCα depletion in response to PMA (). Because a 1-h prestimulation period is short for gene expression and protein synthesis, we hypothesized that preserved PKCα expression did not require protein synthesis. To prove this assumption, we added the established translation inhibitor cyclohexamide (CHX) 1 h before PPARγ agonist stimulation (). As expected, blocking translation with CHX did not interfere with the ability of PPARγ agonists to block PKCα depletion, suggesting a translation-independent mode of action. The physiological significance of these results obtained in murine RAW 264.7 macrophages was verified in primary human monocyte–derived macrophages isolated from peripheral blood. Similar to RAW 264.7 cells, in primary macrophages, pretreatment with ciglitazone and rosiglitazone preserved PKCα expression upon PMA addition (). To elucidate whether the PPARγ1–PKCα interaction shows an impact on PKCα signaling in inflammatory gene expression in macrophages, we analyzed two proinflammatory markers of macrophage activation, i.e., NF-κB DNA binding and TNF-α expression in response to PMA in RAW 264.7 macrophages. To determine activation of the proinflammatory transcription factor NF-κB, we performed a set of electrophoretic mobility shift assays (EMSAs), demonstrating the DNA-binding capability of the transcription factor. As shown in , 100 nM PMA supplied for 3 h significantly induced NF-κB activation (, second lane) compared with the untreated control (, first lane). To elucidate the composition of the transcription factor complex, we used antibodies against the p50 (, left) and p65 subunits (, right) of NF-κB. As shown in (left), the lower and the upper NF-κB shifts contained the p50 subunit. Therefore, the two bands were significantly reduced when an α-p50 antibody was included in the binding reaction and a new band, the p50 supershift, occurred. Only the upper NF-κB shift included the p65-subunit, as indicated by the addition of the α-p65 antibody, which provoked the reduction of the upper NF-κB shift, but did not alter the lower NF-κB shift (, right). As expected, a new band was detectable (the p65 supershift). Thus, we conclude that the lower NF-κB shift is formed by a p50 homodimer, whereas the upper NF-κB shift consists of a p50/p65 heterodimer. To identify whether activation of NF-κB complexes is influenced by PPARγ activation, we treated RAW 264.7 cells with the natural PPARγ agonist 15-deoxy-Δ-prostaglandin J (15d-PGJ; ; ). Taking into consideration that 15d-PGJ may also act PPARγ independently on NF-κB activation (), we included the PPARγ antagonist GW9662 in this experiment (). This allowed us to discover to what extent 15d-PGJ affected PMA-mediated NF-κB activation PPARγ dependently. As depicted in , pretreatment of RAW 264.7 cells with 10 μM 15d-PGJ for 1 h reduced the p50/p65 heterodimer formation in response to PMA (, second lane) compared with PMA-treated controls (, first lane). Preincubation of the cells for 1 h with 10 μM GW9662 completely eliminated the influence of 15d-PGJ on NF-κB activation (, right lane). To show that these results are not restricted to our cell line model, we performed a similar EMSA using nuclear extracts isolated from primary human macrophages. In primary cells, 10 μM of the natural PPARγ agonist 15d-PGJ inhibits 100 nM PMA-mediated NF-κB activation (, middle lane), which is restored after 10 μM GW9662 pretreatment for 1 h (, right lane). However, in human macrophages, only one NF-κB shift in response to PMA, which is formed by a p50/p65 heterodimer (unpublished data), is observed. From these results, we reasoned that PPARγ activation reduced the NF-κB DNA- binding ability in response to PMA by ∼50% compared with PMA-treated controls. To determine whether reduced NF-κB activation modulates expression of proinflammatory cytokines, we finally examined TNF-α expression of RAW 264.7 macrophages in response to PMA. TNF-α expression was determined by the cytometric bead array using a FACSCanto flowcytometer. As shown in , pretreatment of RAW 264.7 macrophages for 1 h with 10 μM rosiglitazone before addition of 100 nM PMA for 6 h reduced PMA-mediated TNF-α expression to ∼70%. These results suggest that activated PPARγ1 inhibits PKCα-dependent signaling in macrophages, thereby provoking, at least in part, an attenuated proinflammatory gene expression profile in association with cellular desensitization. Considering that activation of PKCα, followed by its translocation to the cell membrane, is a prerequisite for its depletion, we were interested to determine whether PPARγ blocks PKCα translocation. To follow PPARγ and PKCα distribution in RAW 264.7 cells, we stained for PPARγ and PKCα in paraformaldehyde-fixed cells (). As shown in (third panel), PPARγ localizes in the cytosol and the nucleus in untreated cells, whereas PKCα is localized in the cytosol (, second panel). The nucleus is counterstained, using DAPI (, first panel), and an overlay is provided in (fourth panel). To prove specificity of the secondary antibodies used, which were labeled with either Alexa Fluor 488 or 546, we used these antibodies alone without a first antibody. In both cases, no signal is observed (unpublished data). Activation of the cells with 100 nM PMA for 50 min provokes PKCα translocation (, second panel), whereas localization of PPARγ is not altered (, third panel). Pretreatment of RAW 264.7 macrophages with 10 μM of the synthetic PPARγ agonist rosiglitazone for 1 h prevents PKCα translocation in response to 100 nM PMA stimulation for 50 min (, second panel). Localization of PPARγ remains unaltered (, third panel). To prove a PPARγ-dependent effect, we used the PPARγ-specific antagonist GW9662. Preincubation of the cells for 1 h with 10 μM GW9662, followed by rosiglitazone treatment (1 h, 10 μM), restores PKCα translocation after 100 nM PMA addition for 50 min (, second panel). PPARγ localization was not affected (, third panel). From these data, we conclude that activated cytosolic PPARγ in RAW 264.7 macrophages inhibits PKCα translocation in response to 100 nM PMA. Based on the aforementioned Western blot results, RAW 264.7 cells express isoform 1, which is partially located in the cytosol. To verify the impact of PPARγ1 activation on PKCα translocation, we used HEK293 cells. Cells were transiently transfected with a PPARγ1 wild-type–encoding vector, tagged with DsRed-monomer or a DsRed-monomer–tagged PPARγ1 AF2 mutant–encoding vector in combination with a PKCα-EGFP–encoding vector. The PPARγ1 AF2 mutant contains two amino acid exchanges (L468A/E471A), thus preventing ligand binding and concomitant PPARγ1 activation (). To follow PKCα translocation, 100 nM PMA was added to rosiglitazone-pretreated and control cells. Changes in PKCα localization were documented 1 h after rosiglitazone stimulation and 50 min after 100 nM PMA addition. PMA provokes PKCα-EGFP translocation to the cell membrane in DsRed-tagged PPARγ1 wild type, as well as DsRed-tagged PPARγ AF2 mutant–expressing cells, as expected (, second row, second panel vs. fourth row, second panel). Localization of PPARγ does not change (, first row, third panel vs. second row, third panel; and third row, third panel vs. fourth row, third panel). In cells transfected with the DsRed-tagged PPARγ1 wild-type construct, rosiglitazone pretreatment inhibited PKCα-EGFP translocation to the cell membrane in response to PMA (, second row, second panel), whereas in cells transfected with the DsRed-tagged PPARγ AF2 mutant, rosiglitazone preincubation does not prevent PKCα-EGFP translocation (, fourth row, second panel). However, PPARγ localization remains unaltered in all analyzed samples (, first through fourth row, third panel). As shown in , preincubation of the cells with the PPARγ antagonist GW9662 (10 μM) for 1 h, completely abolished the PPARγ-dependent inhibition of PKCα translocation in response to PMA (bottom row, second panel). Inline pretreatment of the cells with the PPARα agonist WY14643 (10 μM) for 1 h did not inhibit PMA-mediated PKCα translocation (, bottom row, second panel), which further approved a PPARγ-dependent effect. In corroboration with (A and B), PPARγ localization was unaffected in response to GW9662 or WY14643 and PMA treatment (, first and second row, third panel). To elucidate whether PPARγ1 inhibits PKCα translocation by a direct PPARγ1–PKCα interaction, we performed a set of coimmunoprecipitation experiments. Immunoprecipitation of PKCα from lysates of differentiated THP-1 cells, which had been stimulated for 1 h with rosiglitazone or left untreated, was conducted. As shown in , immunoprecipitation of PKCα resulted in coimmunoprecipitation of PPARγ1 in THP-1 cells that had been challenged with a PPARγ agonist (, lane 2). In the flowthrough, PPARγ1 was only detected when agonist stimulation was omitted (, lane 1). After PPARγ1 activation, PPARγ1 was almost completely retarded in the immunoprecipitation column. To verify a PPARγ1-dependent mechanism, we transfected COS-7 cells with PPARγ1 wild-type or AF2-encoding plasmids and a PKCα-EGFP expression plasmid. Immunoprecipitation was performed using μMacs anti-GFP beads. In cells transfected with the PPARγ1 AF2 mutant, little if any PPARγ1 coimmunoprecipitated with PKCα-EGFP in response to 10 μM rosiglitazone (, lane 4). In cells transfected with the PPARγ1 wild-type plasmid, rosiglitazone treatment allowed to coimmunoprecipitate PPARγ1 with PKCα-EGFP (, lane 2), pointing to the importance of agonist activation to promote PKCα binding. To provide further evidence for a direct PPARγ1–PKCα interaction, we used the mammalian two-hybrid system. In COS-7 cells transiently transfected by electroporation with a combination of pCMV-AD-PPARγ1, pCMV-BD-PKCα, and the Gal4 reporter vector pFR-luc, addition of rosiglitazone or ciglitazone provoked induction of luciferase expression as determined by a luciferase assay. As shown in , addition of both PPARγ agonists induce luciferase expression roughly threefold compared with untreated controls. A PPARγ-dependent effect was verified because addition of the PPARα agonist WY14643 left basal luciferase activity unaltered. With this two-hybrid model, direct binding of target (PPARγ1) to bait protein (PKCα) is required to induce luciferase expression. Therefore, our data suggest that PPARγ1 directly binds PKCα upon agonist activation. This interaction inhibits PKCα translocation to the cell membrane, and thus, PKCα activation. To identify PPARγ1 domains that promote binding to PKCα, we first generated a set of point mutations, each substituting one aa in helix 4 of the ligand-binding domain (LBD), taking into consideration that this region is important in binding transcriptional coactivators (; ), and therefore might be responsible for binding to PKCα as well. We generated six clones, with L309, N310, G312, V313, L316A, or K317 being individually substituted by an alanine (). In addition, we generated the construct PPARγ1 Δaa309-319, with helix 4 (aa309-319) being completely removed (). To prove the functionality of these constructs, we first verified their expression by Western blotting. As a control, the DsRed-PPARγ1 wild-type–encoding vector was included in the experiment. Because of a single aa exchange, or the 12 aa deletion, the molecular mass of proteins originating from the constructs remained unaltered compared with DsRed-PPARγ1 wild type when transfected into HEK293 cells (unpublished data). To finally analyze the impact of the various mutations and the deletion on PKCα translocation, HEK293 cells were transiently cotransfected with the mutated/deleted PPARγ1 constructs tagged with DsRed-monomer, in combination with a PKCα-EGFP–encoding vector. PKCα localization was documented in cells that were untreated (, first rows), treated for 50 min with PMA (, second rows), treated for 1 h with rosiglitazone (, third rows), or preincubated for 1 h with rosiglitazone, followed by the addition of PMA for 50 min (, fourth rows). In cells transfected with one of the six constructs of the DsRed-tagged PPARγ1 mutations (L309A, N310A, and G312A []; V313A, L316A, and K317A []), PKCα-EGFP did not translocate to the cell membrane. A similar result was obtained in cells transfected with DsRed-PPARγ1 Δaa309-319 (, right), showing no PMA-mediated PKCα-EGFP translocation in rosiglitazone-pretreated cells. From these data, we conclude that helix 4 of the LBD is not involved in PPARγ1 binding to PKCα. Based on these results, we decided to generate three PPARγ1 deletion constructs (DsRed-PPARγ1 aaΔ32-198, DsRed- PPARγ1 Δaa32-250, and DsRed-PPARγ1 Δaa51-406) with the belief that ligand binding is necessary for PPARγ1–PKCα interactions. As shown in , all deletions lack the DNA-binding domain (DBD) of PPARγ1. Furthermore, to characterize the role of the hinge domain in PKCα binding, it was eliminated to variable extents. In the DsRed-PPARγ1 Δaa32-198 construct, the first 26 aa of the hinge domain were deleted, and in the DsRed-PPARγ1 Δaa32-250 construct, 78 aa of the hinge domain were deleted. The hinge domain was completely removed in the DsRed-PPARγ1 Δaa51-406 construct. In this construct, a part of the LBD/AF2 domain was deleted as well (aa288-406). All constructs lack a part of the AF1 domain. Expression of the cloned constructs was verified by Western blotting. As controls, the DsRed-PPARγ1 wild-type– and AF2 mutant–encoding vectors were included in the experiment. Estimated molecular mass of deletion construct proteins, transfected into HEK293 cells, were verified using an anti–red fluorescent protein antibody (). Taking into account that the DBD was removed, DNA binding and concomitant transactivation by corresponding PPARγ1 deletion constructs should be abolished. Therefore, we performed a set of reporter experiments, cotransfecting DsRed-PPARγ deletion constructs in combination with a PPRE-reporter plasmid into HEK293 cells. As expected, adding 10 μM rosiglitazone for 6 h to cells transfected with the PPARγ1 deletion constructs did not alter basal transactivation. In contrast, the DsRed PPARγ1 wild-type–encoding plasmid provoked a twofold induction of luciferase expression, whereas the DsRed PPARγ1 AF2 dominant-negative mutant blocked transactivation even below basal values, mediated by endogenous PPARγ in HEK293 cells (unpublished data). To elucidate the role of these deletions on PKCα translocation, HEK293 cells were transiently cotransfected with the shortened DsRed-monomer–tagged PPARγ1 constructs in combination with a PKCα-EGFP–encoding vector. To follow PKCα translocation, 100 nM PMA was added to (1 h, 10 μM) rosiglitazone-pretreated cells. PKCα localization was documented in untreated cells (, first row), cells treated for 50 min with PMA (, second row), for 1 h with rosiglitazone (, third row), or preincubated for 1 h with rosiglitazone, followed by the addition of PMA for 50 min (, fourth row). In cells transfected with the DsRed-tagged PPARγ1 Δaa32-198 construct, PKCα-EGFP did not translocate to the cell membrane. However, in cells expressing the DsRed-tagged PPARγ1 Δaa32-250 or Δaa51-406 construct, PKCα translocated to the cell membrane in response to 100 nM PMA. From these data, we conclude that for PKCα, binding a part of the hinge domain of PPARγ1 is indispensable. To further narrow the involved region of PPARγ1, we finally created the construct DsRed-PPARγ1 Δaa206-224 (), containing a deletion of helix 1 (aa206-224) of PPARγ1, which is located in the hinge domain (aa173-288). Helix 1 has already been identified to mediate the protein–protein interaction of PPARγ with ERK5 (). Expression of the construct results as expected in protein, demonstrating a slightly reduced protein mass (, lane 2) because of the aa206-224 deletion compared with the DsRed-PPARγ1 wild type (, lane 1). We transiently cotransfected HEK cells with the PPARγ1 Δaa206-224 construct tagged with DsRed-monomer in combination with a PKCα-EGFP–encoding vector. In cells expressing the DsRed-tagged PPARγ1 Δaa206-224 (), PKCα translocated to the cell membrane in response to 100 nM PMA. Recently, we demonstrated that monocyte/macrophage desensitization at least partially attenuates PKCα signaling (; ). We provide evidence that PPARγ agonists block PKCα translocation to the cell membrane and concomitant protein depletion, which normally occurs after cell activation. In monocytic cell lines, PPARγ expression has been previously described (; ; ), and it was verified using primary human monocyte–derived macrophages. These data corroborate the work of and , showing PPARγ expression in differentiated macrophages. However, even if PPARγ is expressed, PPARγ agonists are known to mediate PPARγ-dependent and -independent effects (). To this end, 15d-PGJ has been described to directly modify H-ras, provoking a constitutively active enzyme () or inhibiting I-κB kinase, and thus suppressing NF-κB signaling (). Our approach, using cells expressing PPARγ1 wild type or the PPARγ1 agonist-binding mutant AF2, substantiates the need of PPARγ activation in our system. Only in cells expressing PPARγ1 wild type was translocation of PKCα blocked by PPARγ activation. The PPARγ1 AF2 mutant did not prevent PMA-mediated PKCα translocation. These data support the notion of a PPARγ-dependent mechanism. PPARγ-mediated inhibition of classical PKCs has been previously described (). In their case, PKCβ translocation was blocked by PPARγ agonists via DGKα up-regulation. DGKα metabolizes DAG, which is an established activator of classical and novel PKC isoforms. Therefore, its induction/activation will remove the potential PKC activator, causing desensitization as seen in our experiments. However, in our experiments, a role of DGKα up-regulation must be excluded because the protein-synthesis inhibitor CHX did not restore PKCα translocation. In line with this, our PPARγ1 Δaa32-198 construct, where the PPARγ1 DBD was removed, still inhibits PKCα translocation. Further support for our hypothesis, suggesting a direct PPARγ1–PKCα interaction in preventing PKCα translocation, came from previous studies (). In this case, PPARγ was activated in response to apoptotic cells, attenuating PKCα translocation and concomitant ROS production. In this study, the role of PPARγ was verified using a PPARγ d/n cell line. In these cells, pretreatment with apoptotic cells left PMA-mediated PKCα translocation and subsequent ROS production unaltered. A premise for this assumption is that PPARγ is expressed at least partially in the cytosol. Generally, the nuclear hormone receptor PPARγ is described to be exclusively localized in the nucleus (; ). In support of our hypothesis, suggesting cytoplasmatic localization as well, we noticed a minor amount of PPARγ1 to remain in the cytosol. This is based on results using DsRed-PPARγ1–transfected cells, as well as immunohistochemical detection of endogenous PPARγ1 located in the cytosol of RAW 264.7 macrophages besides its major nuclear localization. It should be noted that cytoplasmatic distribution of PPARγ is in line with the work of . In their study, an approach similar to our experiments was used, with EGFP-tagged PPARγ used to characterize intracellular distribution of PPARγ. Results indicated that PPARγ is not exclusively located in the nucleus. Furthermore, localization of PPARγ in the cytoplasma in the promonocytic cell lines HL-60 and K-562 has been observed, especially in response to the PPARγ agonist troglitazone (). This work was done using immunohistochemical detection of endogenous PPARγ. Therefore, side effects, such as unphysiological high expression or a modified protein behavior as a result of a tag or label (), can be excluded. In addition, recently provided evidence that PPARγ is actively exported from the nucleus into the cytosol in a MEK1-dependent manner, further supporting our observed PPARγ localization pattern. Furthermore, described cytoplasmatic localization of a different PPAR isoform, PPARα, when coexpressed with CAP350, which is a putative centrosome-associated protein of unknown function. Therefore, we propose that members of the PPAR family may localize in the cytoplasm, possibly after activation, when bound to cytoplasmic proteins such as PKCα. Immunoprecipitation of PKCα from lysates of differentiated THP-1 cells coimmunoprecipitated PPARγ. Remarkably, PPARγ1 coimmunoprecipitation was only seen once PPARγ1 became activated. The requirement of PPARγ1 activation was verified using an agonist-binding mutant of PPARγ1, which did not block PKCα translocation in response to PMA stimulation. A direct PPARγ1–PKCα interaction was further supported by a mammalian two-hybrid system with PPARγ1 as the target and PKCα as the bait construct, provoking luciferase reporter gene expression when target and bait proteins interact. To avoid autocrine activation of the reporter system, PPARγ has to be cloned as a target protein linked to the NF-κB transactivation domain, not allowing this hybrid protein to bind to the promoter of the reporter. However, DNA binding of PPARγ1 to PPREs, and concomitant scavenging the NF-κB-AD-PPARγ1 hybrid protein from the two-hybrid assay, cannot be excluded. Based on the well-established role of helix 4 of the PPARγ LBD in mediating protein–protein interaction of PPARγ with coactivators, such as CBP and SRC-2, or repressors, such as the nuclear receptor corepressor and the silencing mediator for retinoic acid receptor and thyroid-hormone receptor (; ; ; ), we first generated 6 PPARγ1 constructs in which only 1 aa was exchanged and 1 construct in which helix 4 was completely removed. Unexpectedly, these constructs did not alter rosiglitazone-dependent inhibition of PKCα translocation. Taking into account that PPARγ binding to other factors, such as adipocyte-type fatty acid–binding protein or extracellular signal-related kinase 5, which do not belong to the family of transcriptional coactivators, can be mediated by other PPARγ domains, such as A/B/C and D/E/F () or the hinge domain (domain D; ), we created three PPARγ1 deletion constructs. All of them lack the entire DBD (domain C). In addition, different parts of the A/B and D domains have been removed, and one construct contained the C-terminal third of the E/F domains only. Based on our collective results, we provide evidence that a part of the hinge domain probably confers the PPARγ1–PKCα interaction, which is present in the PPARγ1 Δaa32-198 construct but absent in the Δaa32-250 construct, when PPARγ1 is activated by an agonist, thus requiring the LBD/AF2 domains. One known region of PPARγ1 located in aa198-250 is the hinge helix 1 (aa 206–224). Therefore, we cloned a PPARγ1 construct with helix 1deleted (DsRed- PPARγ1 Δaa206-224). In cells transfected with this construct, PKCα translocated even after rosiglitazone pretreatment in response to PMA. From these results, we conclude that PPARγ1 binds to PKCα via the hinge helix 1 domain, after PPARγ1 has been activated by a ligand. The proposed mechanism of PPARγ1–PKCα binding proceeds fast. 1 h of prestimulation with PPARγ agonists is sufficient to inhibit PKCα translocation in response to 100 nM PMA. However, PKCα translocation by 1 μM PMA was not blocked. These results support the assumption that the capacity of cytoplasmatic PPARγ to bind PKCα correlates with the strength of PKCα activation. Likely, very strong activation signals, such as 1 μM PMA, exceed the inhibitory impact of PPARγ. Thus, the role of PPARγ in blocking PKCα signaling might be only transient, allowing PKCα activation by a more stringent activator. This makes the mechanism more interesting for the development of new therapy strategies. Prolonged periods of PPARγ activation, which provoke transcriptional control to target members of the NADPH oxidase system, have already been described (p22, p47, and gp91; ; ; ). Consequently, in these cells PPARγ contributes to an antiinflammatory phenotype by blocking NADPH oxidase-dependent ROS production. An involvement of PPARγ in attenuating inflammatory reactions to improve the clinical picture of sepsis has previously been shown (for review see ). In line with this, our results add to this data. In our system, PMA-mediated NF-κB activation was inhibited in response to PPARγ agonist pretreatment to 50% in RAW 264.7 cells, as well as primary human macrophages. In accordance, PMA-induced TNF-α expression was PPARγ dependently reduced to 70%. It has been observed that PPARγ activation inhibits multiple organ failure in an animal model (), although the underlying mechanism remains unclear. The option to adjust a pro- versus antiinflammatory monocyte/macrophage phenotype will provide new possibilities for the development of therapies to control systemic inflammation. Our data add a new antiinflammatory role for PPARγ based on the ability to scavenge PKCα in the cytosol, thus, blocking membrane translocation and downstream signaling. We analyzed human cells from peripheral blood of healthy donors. For monocyte enrichment, we isolated PBMCs from donors using Ficoll-Hypaque gradients (PAA Laboratories). Cells were left to adhere on culture dishes (Primaria 3072; Becton Dickinson) for 60 min at 37°C. Nonadherent cells were removed. Afterward, cells were differentiated to macrophages by culturing them in complete RPMI containing 10% AB-positive human serum. Flow cytometry confirmed that the monocyte-like population was 90–95% pure (CD14 vs. CD14). We cultivated RAW 264.7 and THP-1 in RPMI 1640 (PAA Laboratories). HEK293 and COS-7 cells were cultured in DME high glucose (PAA Laboratories). Both media were supplemented with 100 U/ml penicillin (PAA Laboratories), 100 μg/ml streptomycin (PAA Laboratories), and 10% heat-inactivated fetal calf serum (PAA Laboratories). Ciglitazone (Biomol), rosiglitazone (Biomol), WY14643 (Biomol), and CHX (Sigma-Aldrich) were dissolved in DMSO. Appropriate vehicle controls were performed. To determine intracellular PPARγ localization, we seeded RAW 264.7 macrophages directly on a slide. After 24 h, cells were treated as indicated and fixed on the slides by 1-h incubation in 4% paraformaldehyde at 4°C. Thereafter, cells were permeabilized in PBS containing 0.2% Triton X-100 for 15 min. After a washing step in PBS, cells were incubated for 2 h with a 1:250 dilution of a rabbit α-PPARγ antibody (Calbiochem) at 4°C. After three 5-min washing steps with PBS, cells were incubated with a secondary goat α-rabbit antibody (1:250) labeled with Alexa Fluor 546 (Invitrogen) for 2 h at 4°C. Cells were incubated for 2 h with a 1:250 dilution of a mouse α-PKCα antibody (BD Biosciences) at 4°C. After three 5-min washing steps with PBS, cells were incubated with a secondary goat α-mouse antibody (1:250) labeled with Alexa Fluor 488 (Invitrogen) for 2 h at 4°C. Again, cells were washed three times with PBS and counterstained with DAPI (1 μg/ml in PBS for 15 min). After a final 5-min washing step in PBS, cells were covered with Vectashield mounting medium (Linaris) and a coverslip. PPARγ and PKCα localization were determined using an AxioScope fluorescence microscope with the ApoTome upgrade (Carl Zeiss MicroImaging, Inc.; lens 63×/0.6 NA; ocular 10×) at room temperature, documented by a charge-coupled device camera (Carl Zeiss MicroImaging, Inc.) and AxioVision Software (Carl Zeiss MicroImaging, Inc.). To examine cellular PPARγ localization, we subcloned human PPARγ1 into the DsRed-monomer–encoding vector pDsRed-Monomer-C1 (CLONTECH Laboratories, Inc.) using the infusion ligation kit (CLONTECH Laboratories, Inc.). To allow integration of the PPARγ1 fragment, the vector was cut within the multicloning site (MCS) by BamHI and XhoI. To insert PPARγ1 (provided by V.K.K. Chatterjee, University of Cambridge, Cambridge, UK), we used the pcDNA3-PPARγ1 wild-type and AF2 vectors for PPARγ1 amplification by PCR, using the following sequences based on the infusion ligation requirements (changed nucleotides are underlined): wild type, 5′-GGACTCAGATCTCGAATGGTTGACACAGAGATC GCATTCTG-3′ and 3′-AGGACGTCCTCTAGATGTTCCTGAACATGCTAGGTGGCCT AGA T-5′; AF2 mutant, 5′-GGACTCAGATCTCGAATGGTTGACACAGAGATCGCAT- TCTG-3′ and 3′-GACGTCCCTAGATGTTCCTGAACATGCTAGGTGGCCT AGAT-5′. Annealing temperatures were 62°C for the first cycle and 72°C for the later ones and calculated using the Oligo software (MBI). Infusion reaction of the cleaved vector with the amplified PPARγ1 wild-type or AF2 fragment was performed according to the distributor's instructions. Site-directed mutagenesis to generate single aa exchanges (L309A, N310A, G312A, V313A, L316A, K317A) and deletion of helix 1 (aa206-224) or 4 (aa309-319) of PPARγ1 were performed using the QuikChange XLII kit (Stratagene). The following primers were used (changed nucleotides are underlined): L309A, 5′-CCTGGTTTTGTAAATCTTGACAACGACCAAGTAACTCTCCTC-3′ and 5′-GAGGAGAGTTACTTGGTCGTTGTCAAGATTTACTTTTCCAGG-3′; N310A, 5′-CC TGGTTTTGTAAATCTTGACTTGGACCAAGTAACTCTCCTC-3′ and 5′-GAGGAGAG TTACTTGGTCCAAGTCAAGATTTACTTTTCCAGG-3′; G312A, 5′-GTAAATCTTG ACTTGAACGACGTAACTCTCCTCAAA- TATGG-3′ and 5′-CCATATTTGAGGAGAGT TACGTCGTTCAAGTC- AAGATTTAC-3′; V313A, 5′-GTAAATCTTGACTTGAACGA CCAAACTCTCCTCAAATATGG-3′ and 5′-CCATATTTGAGGAGAGTTTGGTCG- TTCAAGTCAAGATTTAC-3′; L316A, 5′-CTTGAACGACCAAGTAACTCTC- AAAT ATGGAGTCCACGAG-3′ and 5′-CTCGTGGACTCCATATTT- GAGAGTTACTTGGTCG TTCAAG-3′; K317A, 5′-CTTGAACGACCAAGTAACTCTCCTCTATGGAGTCCAC GAG-3′ and 5′-CTCGTGGACTCCATAGAGGAGAGTTACTTGGTCGTTCAAG-3′; Δaa309-319, 5′-CCTGGTTTTGTAAATCTTGACCCGCTGACCAAAGCAAAG-3′ and 5′-CTTT GCTTTGGTCAGCGGGTCAAGATTTACAAAACCAGG-3′. The pcDNA3-PPARγ1 wild-type vector was used as a template. An initial denaturation step was performed at 95°C for 1 min, followed by 18 cycles at 95°C for 50 s, annealing at 60°C for 50 s, and extension at 68°C for 7 min. A final extension phase was performed at 68°C for 7 min. To follow PKCα translocation and PPARγ distribution, HEK293 cells were seeded directly onto a slide, and then transiently transfected by CaPO-precipitation with combinations of pDsRed-Monomer-C1 PPARγ1 wild type/pPKCα-EGFP, pDsRed-Monomer-C1 PPARγ1 AF2/pPKCα-EGFP, or the generated deletion and mutation constructs together with pPKCα-EGFP. 24 h after transfection, cells were used for experiments. Cells were treated as indicated. Afterward, cells were fixed on the slides by 1-h incubation in 4% paraformaldehyde at 4°C. Cells were washed three times with PBS and counterstained with DAPI (1 μg/ml in PBS for 15 min). After a final 5-min washing step in PBS, cells were covered with Vectashield mounting medium and a coverslip. Translocation of PKCα-EGFP and DsRed-PPARγ1 wild type/AF2 distribution was analyzed using an AxioScope fluorescence microscope with the ApoTome upgrade (lens 63×/0.6 NA; ocular 10×) at room temperature, documented by a charge-coupled device camera and the AxioVision Software. For reporter analysis, HEK293 cells were transiently transfected by CaPO-precipitation with pDsRed-Monomer-C1 PPARγ1 wild-type, -AF2, Δaa32-198, Δ32-250, Δ51-406 constructs, or the empty DsRed vector in combination with the PPRE-containing p(AOX)-TK-luc reporter plasmid. Transfection efficiency was normalized by cotransfecting a pRL-TK control vector encoding for luciferase. Transfections were performed in duplicate, and each experiment was repeated at least three times. After THP-1, cells were differentiated for 24 h with 50 nM PMA, PMA was removed, and cells were incubated for an additional 48 h in complete medium. Afterward, cells were stimulated for 1 h with 10 μM rosiglitazone or remained as controls. Eventually, cells were harvested and lysed in lysis buffer (50 mM Tris, 5 mM EDTA, 150 mM NaCl, 0.5% Nonidet-40, and 1 mM PMSF, pH 8.0). To assure cell lysis, cells were sheared 10 times with a 16-gauge needle, followed by a brief 10-s sonication (Sonifier; Branson; duty cycle 100%, output control 60%). Cell debris was removed by centrifugation (10,000 for 5 min), and 1 mg of protein was used for immunoprecipitation. Sample volume was adjusted with lysis buffer to 1 ml. 2 μg anti-PKCα antibody (BD Biosciences) was added and incubated at 4°C overnight. Thereafter, 50 μl μMACS protein A microbeads (Miltenyi Biotech) were added and incubated for 6 h. Lysate was applied onto an equilibrated μ column, which was already placed in the magnetic field of a μMACS separator. The flowthrough was collected and saved for further analysis. The column was rinsed 4 times with 200 μl wash buffer (150 mM NaCl, 1% Igepal CA-630, 0.5% sodium deoxycholate, 0.1% SDS, and 50 mM Tris HCl, pH 8.0), followed by 2 washes with low ionic buffer (20 mM TrisHCl, pH 7.5). Afterward, the column was removed from the magnetic field and the remaining proteins were eluted using 50 μl of lysis buffer. To use PPARγ1 and PKCα in the mammalian two-hybrid system (Stratagene), PPARγ1 was cloned into the BamHI–HindIII site of the pCMV-AD MCS, and PKCα was cloned into the BamHI–HindIII site of the pCMV-BD MCS. PPARγ was amplified from the pcDNA3-PPARγ1 wild-type vector and PKCα from the vector pPKCα-EGFP. The following primers were used: pCMV-BD-PPARγ1, 5′-GCCGGAA TTGGGATCCATGGTTGACACAGAGATGCCATTCTG-3′ and 5′-ACGCGGCCGCAAGC TCTAGTACAAGTCCTTGTAGATCTCCTGCAGG-3′; pCMV-AD-PKCα, 5′-CAGCGGCC AAGGAT- CCATGGCTGACGTTTTCCCGGG-3′ and 5′-ACGCGGCCGCAAGC- TTCATA CTGCACTCTGTAAGATGGGGTGC-3′. Annealing temperatures were 62°C for the first cycle and 72°C for the later ones, and were calculated using the Oligo software (MBI). Infusion reaction of the BamHI–HindIII–cleaved vectors with the amplified PPARγ1 wild-type- or PKCα- fragment was performed according to the distributor's instructions. Correct orientation and sequence of the generated vectors was verified by restriction analyses and sequencing. COS-7 cells were transiently transfected by electroporation using a combination of the two constructed vectors, as well as the pFR-luciferase reporter vector (Stratagene). Afterward, cells were incubated for 24 h, and then stimulated for 6 h with 10 μM ciglitazone, 10 μM rosiglitazone, or 10 μM WY14643, or they remained as controls. Thereafter, cells were lysed and assayed for firefly luciferase activity by a luciferase assay (Promega). Cell lysis was achieved with lysis buffer (50 mM Tris, 5 mM EDTA, 150 mM NaCl, 0.5% Nonidet-40, and 1 mM PMSF, pH 8.0) and 20-s sonication (Sonifier; duty cycle 100%, output control 60%). Whole-cell lysates were cleared by centrifugation (10,000 for 5 min), and protein concentration was determined with the Lowry method. 80 μg of protein was resolved on 10% polyacrylamide gels and blotted onto nitrocellulose sheets, basically following standard methodology. Equal loading and correct protein transfer to nitrocellulose was routinely quantitated by Ponceau S staining. Filters were incubated with the anti-PKCα antibody (1:500; BD Biosciences), anti-PPARγ antibody (1:500; Santa Cruz Biotechnology, Inc.), anti-RFP antibody (1:1,000; MBL), or anti-actin antibody (1:2,000; GE Healthcare) overnight at 4°C. Horseradish peroxidase–conjugated polyclonal antibodies (1:5,000; GE Healthcare) were used for enhanced chemiluminescence detection. Supernatants from RAW 264.7 macrophages treated as indicated were harvested after the indicated times. Content of TNF-α was quantified using the BD Cytometric Bead Array TNF-α Flex Set (BD Biosciences) according to the supplier's instructions using a FACSCanto flowcytometer. Interpretation of the results was performed with the FCAP Array software (Soft Flow, Inc./BD Biosciences). Nuclear extracts were prepared as previously described (). An established EMSA method, with slight modifications, was used (). Nuclear protein (20 μg) was incubated for 30 min at room temperature with 2 μg poly(dI-dC) from GE Healthcare, 2 μl buffer D (20 mM Hepes/KOH, 20% glycerol, 100 mM KCl, 0.5 mM EDTA, 0.25% Nonidet P-40, 2 mM DTT, and 0.5 mM PMSF, pH 7.9), 4 μl buffer F (20% Ficoll-400, 100 mM Hepes/KOH, 300 mM KCl, 10 mM DTT, and 0.5 mM PMSF, pH 7.9), and 250 fmol 5′-IRD700–labeled oligonucleotide (Metabion) in a final volume of 20 μl. Specific p65 and p50 supershift antibodies (2 μg; Santa Cruz Biotechnology, Heidelberg, Germany) were added as indicated. DNA–protein complexes were resolved at 80 V for 1 h in a native 6% polyacrylamide gel, and visualized with the Odyssey infrared imaging system (LI-COR). Oligonucleotides with the consensus NF-κB site (bold letters) were used (): 5′-GCCAGTTGA CAGGC-3′; 3′-CGGTCAA GGTCCG-5′. Each experiment was performed at least three times. Statistical analysis was performed using the paired test. We considered P values ≤ 0.05 as significant. Otherwise, representative data are shown.

TGF-β inhibits cell growth and acts as a tumor suppressor (). TGF-β signals via receptor serine/threonine kinases that phosphorylate Smad proteins, which move to the nucleus and regulate gene transcription (). During epithelial cytostasis (growth arrest), Smads induce cell cycle inhibitors and and repress and inhibitors of differentiation , , and (). TGF-β up-regulates rapidly and maintains prolonged mRNA and protein levels, which is critical for epithelial cytostasis (; ). The mechanism of sustained p21 maintenance is not clear, and we hypothesized that it could be achieved by a secondary wave of TGF-β signaling that activates new factors capable of maintaining p21 levels. A candidate pathway for involvement in such a scenario is Notch, a major regulator of cell fate (). Four distinct mammalian receptors (Notch1–4) interact extracellularly with transmembrane ligands Jagged1, 2, and Deltalike1–3 (DLL1–3), which are expressed by adjacent cells (). Such an interaction leads to the proteolytic cleavage of Notch by the γ-secretase activity of presenilin, thus releasing the Notch intracellular domain, which enters the nucleus and regulates transcription after binding to the transcription factor CSL (). Retroviral insertions in mice and chromosomal translocations in human leukemias cause oncogenic truncations or fusions of Notch (). The skin- or liver-specific knockout of Notch1 leads to tumorigenesis, classifying Notch1 as a tumor suppressor (; ). Notch1 inhibits epidermal, endothelial, and hepatic cell growth (; ; ). Notch arrests the keratinocyte cell cycle by transcriptionally inducing via CSL or calcineurin–nuclear factor of activated T cells pathway activation (; ). Notch and TGF-β pathways cross talk, as TGF-β induces Jagged1 expression, leading to epithelial-mesenchymal transition (). During heart organogenesis, Notch uses TGF-β signaling to cause the epithelial-mesenchymal transition (). Alternatively, Notch induces nodal, a TGF-β family regulator of embryogenesis (). The Notch intracellular domain directly binds to Smads, leading to the coregulation of gene expression in neuronal and endothelial cells (; ). Based on these facts, we investigated cross talk between TGF-β and Notch during epithelial cytostasis. We demonstrate that the TGF-β cytostatic response at least partly requires Notch signaling. A novel mechanism based on transcriptional induction of the Notch ligand Jagged1, involvement of the Notch effector CSL, and sustained p21 induction explains the interdependent roles of TGF-β and Notch during cytostasis. To study cross talk between Notch and TGF-β, we ectopically expressed the human Notch1 intracellular domain (N1ICD; ). Usually, 70–80% of cells expressed N1ICD at roughly endogenous levels, which induced a classic target of this pathway (transcription factor ; unpublished data). In mock-infected (Ad-GFP) mouse mammary epithelial NMuMG cells, TGF-β1 suppressed S-phase entry by 60–70% (). Ectopic N1ICD did not have much effect on its own, but N1ICD plus TGF-β1 suppressed S-phase entry by 80–95% (). This effect was dependent on TGF-β1 dose (Fig. S1 A, available at ) and was also confirmed in human mammary MCF-10A cells (see ). TGF-β1 stimulation in the presence of a γ-secretase inhibitor (GSI), which blocks endogenous Notch signaling (), led to a substantial but not complete restoration of S-phase entry (), which was confirmed in the mammary epithelial MCF-10A cells (see ) and in immortalized human mammary epithelial cells (HMECs; Fig. S1 B). In contrast, in mink lung epithelial cells, ectopic N1ICD inhibited the suppressive effect of TGF-β1 (Fig. S1 C). This highlights the cell context dependency of the cytostatic response and confirms a recent study that shows c-Myc up-regulation by Notch signaling, which counteracts cytostasis by TGF-β (). In human HaCaT keratinocytes, Ad-N1ICD alone suppressed S-phase entry almost to the same extent as 2 ng/ml TGF-β1 (). Ad-N1ICD combined with TGF-β1 led to >95% growth suppression, and up to 80% of the cells were arrested in G1 phase of the cell cycle (). This showed strong Notch1–TGF-β1 cooperativity that was blocked by TGF-β receptor kinase inhibitors (Fig. S1 D), suggesting the interdependence of the two pathways. GSI also blocked cytostasis by TGF-β1 in HaCaT cells ( and S1 E) and shifted the cell cycle profile to that of mock-treated cells (). We conclude that Notch and TGF-β cooperatively induce growth arrest in human and mouse epithelial cells of mammary and skin origin. Endogenous Notch signaling is partly necessary for growth arrest by TGF-β. To further understand the Notch–TGF-β cross talk, we performed a transcriptomic screen in HaCaT cells stimulated with TGF-β1 in the absence or presence of GSI. We measured gene expression after cycloheximide pretreatment after 2, 6, and 48 h of TGF-β1 stimulation, aiming at immediate/early, intermediate, and sustained gene responses. Several hundred TGF-β–responsive genes were measured ( and Table S1, available at ), which is in accordance with previous microarray analyses in the same cell line (; ; ). GSI decreased the number of TGF-β–regulated genes by 36% ( and Table S1). At 2 h, only immediate/early TGF-β gene targets were measured, and GSI had no effect. At 6 h, we observed 85% inhibition. At 48 h, we did not observe dramatic effects on total gene numbers, but effects were seen on individual gene profiles. A comparison of the two gene lists (minus and plus GSI) showed 198 genes whose response to TGF-β was unaffected by GSI (). Examples are (TGF-β–inducible early growth response protein 1), a zinc finger transcription and proapoptotic factor; (TNF ligand superfamily member 10), a proapoptotic secreted protein; (neurofibromin 2), a cytoskeletal regulator; and (interferon-induced transmembrane protein 1), a cell surface antigen. The expression of 394 TGF-β1–responsive genes (roughly 50% of the regulated genes) was neutralized by GSI, demonstrating a strong dependency on Notch (). Examples are (serine-threonine kinase receptor-associated protein), an adaptor that binds to TGF-β receptor and inhibitory Smad7 to mediate the termination of TGF-β signaling; (Smad ubiquitylation regulatory factor 1), an E3 ubiquitin ligase that causes TGF-β receptor and Smad degradation; , a calcium-binding protein that mediates epithelial cytostasis by TGF-β as it transcriptionally induces the cell cycle inhibitor ; and (involucrin), a keratinocyte differentiation marker that cross-links to the keratin cytoskeleton. Finally, 179 genes were not previously recognized as TGF-β targets, as their regulation is revealed only after GSI treatment (), suggesting that Notch signaling may repress genes in a manner that prohibits responses to TGF-β. This experimental design did not test for adverse effects of GSI on gene expression in general, which is formally possible. However, GSI both inhibited and induced specific gene expression when combined with TGF-β, and we never observed adverse effects of GSI in the absence of TGF-β in RT-PCR assays. For the first time, we uncovered large gene sets that are coregulated by TGF-β and Notch positively or negatively ( and Table S1). Notch seemed to counteract the regulation of many genes by TGF-β1. This suggests that to a large extent, the transcriptomic response to TGF-β incorporates regulation by Notch signaling. Among the genes identified, two were members of the Notch pathway: TGF-β1 induced () and repressed (). We examined whether TGF-β1 regulates the expression of all Notch ligands and receptors in HaCaT () and NMuMG cells (Fig. S2, available at ). TGF-β1 considerably induced mRNA and protein and mRNA, weakly induced mRNA at 24 h, and did not appreciably affect or mRNA in HaCaT cells. In NMuMG cells, TGF-β1 induced mRNA and protein and mRNA but did not appreciably regulate mRNA levels (Fig. S2, A and B). On the other hand, TGF-β1 repressed mRNA and protein in HaCaT cells (). Even more dramatic was repression by TGF-β1 in the same cells (). and expression was not appreciably affected by TGF-β1 in HaCaT cells (). Although similar HaCaT expression profiles were measured for the receptor in NMuMG cells responding to TGF-β1, mRNA and protein were considerably induced in NMuMG cells (Fig. S2, C and D). In HaCaT cells, GSI primarily perturbed the expression profile of and, to a lesser extent, that of (), weakly induced at 24 h, and repressed the weak induction of at 24 h but did not affect the other regulated ligands or receptors. This agrees with the effect of cycloheximide that blocks the induction of , , and by TGF-β1 (unpublished data), which represents indirect Notch-mediated responses to TGF-β. The lack of effect of GSI on Notch receptor profiles is also seen at the protein levels of TβRI, which is slowly down-regulated during the time course but is not affected by GSI (). We conclude that TGF-β1 induces the expression of endogenous Notch ligands in keratinocytes and mammary epithelial cells, whereas the regulation of Notch receptors is complex and tissue type dependent. Between the two regulated ligands JAG1 and DLL4, we could only verify the regulation of JAG1 protein (), as our DLL4 antibody showed poor efficacy (unpublished data). Thus, during the stimulation of epithelial cells with TGF-β, the initial induction of various Notch ligands may activate this pathway, whereas the delayed repression of Notch receptors may reflect a negative loop of Notch receptor down-regulation. 11 genes with known links to the cytostatic and apoptotic programs of TGF-β were identified in the transcriptomic screen: the cell cycle inhibitors p21 () and p15 (), cyclins B2 (), D1 (), and D2 (), the transcriptional regulators c-Myc () and Id2 (), the signal transducer S100A11 (), and the apoptotic/survival regulators GADD45β (), GADD45γ (), and TIEG (; ). Prolonged up- or down-regulation of many of these genes was neutralized by GSI () as verified by quantitative RT-PCR analysis (). The gene, a well studied transcriptional target of TGF-β/Smad signaling that plays major regulatory roles in the epithelial cytostatic program of TGF-β (), exhibited its characteristic repression phase followed by the recovery of basal mRNA levels after 24 h of TGF-β stimulation (). GSI did not affect the expression profile, suggesting that endogenous Notch signaling is not involved in the response of keratinocytes, which is in contrast to what was previously reported for mink lung epithelial cells that overexpressed N1ICD (). Similar to , () also exhibited relative insensitivity to GSI throughout the time course. Among the 11 genes of the cytostatic/apoptotic program, GSI most prominently affected , , , , and expression profiles from 6 h onwards (; , , and ). The immediate/early response of all the genes measured after 2 h of stimulation with TGF-β1 in the presence of cycloheximide was not substantially affected by GSI (). GSI quantitatively reduced the amplitude of the mRNA profiles of the aforementioned genes (, and ) but preserved the dynamic changes in the overall profile of mRNA expression. In the case of the p21 cell cycle inhibitor, GSI not only reduced the amplitude of the response but also distorted the expression profile beyond 2 h dramatically (). Although the immediate/early response of to TGF-β1 was unaffected by GSI, long-term mRNA induction was substantially blocked by GSI, suggesting that Notch signaling was critical for this response, acting as a secondary signal to the primary TGF-β stimulus. The effect of GSI was also considerable at the protein level because sustained (6–24 h) p21 protein induction by TGF-β1 was converted to an early response (1.5–3 h) in the presence of GSI (). This prompted us to further analyze the profile of p21 expression and also test the functional relevance of this profile. The present data demonstrate that although endogenous Notch signaling contributes to regulation of a substantial subset of the TGF-β cytostatic gene program, Notch is not involved in the regulation of every gene in this program. The evidence so far has led to a working model in which TGF-β signaling induces Jagged1 production, which then leads to Notch receptor activation and further signaling via CSL, leading to regulation of the cell cycle inhibitors p15 and p21 and, thus, mediating epithelial cell cycle arrest (). To examine the functional relevance of induction by TGF-β during cytostasis, we depleted endogenous by siRNA (). The three- to fourfold induction of mRNA throughout the 24-h time course in response to TGF-β was reduced to a mere 1.3–1.6-fold induction in the presence of siRNA. Under the same conditions of endogenous depletion, JAG1 protein accumulation in the 6–24-h interval of the time course was severely lost to essentially undetectable levels (). The specificity of siRNA–mediated depletion was verified by demonstrating that three unrelated proteins, Smad2, Smad3, and α-tubulin, were not affected by the same siRNA. In addition to the total Smad2 and Smad3 levels, TGF-β–inducible phospho-Smad2 and -Smad3 levels were not appreciably affected by siRNA during the 24-h time course (). Notably, the knockdown of mRNA and protein resulted in a concomitant decrease in the TGF-β–inducible levels of mRNA and protein (). This decrease was evident throughout the time course and was more robust during the 6–24-h interval when endogenous JAG1 protein accumulated at maximal levels. Finally, we demonstrated that knockdown reverted the 70% growth inhibition by TGF-β1 to a mere 25% inhibition (), suggesting that endogenous JAG1 participates in the TGF-β cytostatic response. These data strongly suggest that transcriptional induction of the gene by TGF-β is intimately linked to the robust transcriptional induction of the p21 cell cycle inhibitor and to the growth inhibitory response of HaCaT keratinocytes (). Knockdown of endogenous Notch1 via siRNA was effective but failed to inactivate Notch signaling (unpublished data), as epithelial cells express other Notch receptors whose expression is regulated by TGF-β ( and S2). Therefore, we depleted CSL, which is the only known common mediator of all Notch signaling pathways. siRNA reduced mRNA expression by 85% () and reduced protein to undetectable levels (), whereas mock siRNA had no effect in HaCaT cells. The knockdown was specific as verified by demonstrating that three unrelated proteins (Smad2, Smad3, and α-tubulin) were not affected by the same siRNA. In addition, the TGF-β–inducible phospho-Smad2 and -Smad3 levels were not appreciably affected by the siRNA during the 48-h time course (). A comparable 65–75% knockdown of endogenous was achieved when HaCaT cells were simultaneously infected with mock (Ad-GFP) or specific (Ad-N1ICD) adenoviruses (). During the concomitant stimulation of HaCaT cells with TGF-β1, we observed a minor trend for the induction of endogenous mRNA levels (), which we could not reproduce at the protein level (). As an additional confirmation of the specificity of siRNA, ectopic mRNA levels obtained after adenoviral infection of HaCaT cells were not affected by knocking down endogenous (). Notably, under the same conditions of the combined knockdown of endogenous and ectopic N1ICD expression, endogenous mRNA induction was dramatically reduced (). Under mock infection conditions, the 2.5–3-fold induction of endogenous mRNA by TGF-β1 was reduced to a weak 1.3-fold induction (), which was correspondingly reflected at the p21 protein level (). Furthermore, the synergistic induction by TGF-β1 and N1ICD also depended on proper endogenous levels because knockdown of the latter considerably reduced the inducible mRNA levels () and even more dramatically reduced the corresponding p21 protein levels (). Finally, knockdown substantially reverted cytostasis by TGF-β1; in mock-transfected cells, TGF-β1 stimulation caused a suppression of thymidine incorporation to 35% of unstimulated cells, whereas after CSL knockdown, the suppression was only to 64% of unstimulated cells (). In addition, reverted the cytostatic effect of N1ICD alone to control levels and strongly blocked synergistic cytostasis by TGF-β1 plus N1ICD (). These experiments with knockdown demonstrate a similar phenotype to knockdown () or the inhibition of γ-secretase activity by GSI ( and ) and collectively prove that endogenous Notch/CSL signaling is critical, at least in part, for the antiproliferative response of HaCaT cells to TGF-β. The γ-secretase activity of presenilin regulates Notch, Wnt/ β-catenin, CD44, ErbB signaling, and β-amyloid processing and deposition in Alzheimer's disease (). TGF-β did not appreciably affect the expression or activation of CD44 and ErbB2 in our cell models nor did ligands for these receptors show cooperation with TGF-β–induced cytostasis (unpublished data). More convincingly, the similarity of cellular phenotypes with respect to gene regulation and keratinocyte proliferation arrest obtained after the use of GSI and knockdown of endogenous and after siRNA transfection strongly enforces the model that Notch signaling operates downstream of TGF-β during epithelial cell growth inhibition (). During the course of all of the previous experiments, we also monitored the influence of Notch pathway inhibition on the primary activation step of Smad signaling, namely the TGF-β receptor–mediated phosphorylation of Smad2 and Smad3. As previously presented, the knockdown of JAG1 or CSL did not appreciably perturb the normal flow of TGF-β receptor signaling as monitored by phospho-Smad protein levels in extensive time course experiments ( and ). In contrast, when the same experiment was repeated after the stimulation of HaCaT cells with TGF-β1 in the presence of GSI, we could observe a partial but considerable inhibition of both phospho-Smad2 and -Smad3 levels (). The negative and adverse effects of GSI on phospho-Smad levels was evident throughout extensive time course experiments and was more prominent after 2 h of stimulation with TGF-β1. To test whether activated Notch signaling led to the opposite effect, namely the induction of phospho-Smad levels in HaCaT cells, we infected cells with mock (Ad-GFP) or specific (Ad-N1ICD) adenoviruses and measured phospho-Smad2/3 (). Although TGF-β1 induced robust phospho-Smad2 and -Smad3 levels in HaCaT cells, in the presence of control (Ad-GFP) or Ad-N1ICD adenovirus, N1ICD by itself failed to induce phospho-Smad levels. Furthermore, TGF-β1 stimulation of cells expressing ectopic N1ICD did not lead to any further increase of phospho-Smad levels compared with TGF-β1 stimulation alone (). Therefore, we conclude that Notch signaling does not seem to contribute to R-Smad phosphorylation by TGF-β receptors in HaCaT cells. However, the use of GSI demonstrates that γ-secretase activity is linked to the process of R-Smad phosphorylation by TGF-β receptors via as yet unknown mechanisms. The aforementioned result on a potential role of γ-secretase during R-Smad activation obliged us to test even more rigorously the specificity of the observed effects of GSI on gene induction and epithelial cytostasis downstream of TGF-β. If the GSI effect was primarily caused by the reduction on phospho-Smad levels, Notch should not be able to rescue such effects when provided ectopically. Upon control Ad-GFP infection, TGF-β1 induced endogenous p21 protein levels, and GSI partially blocked this response () as described for uninfected cells (). Under such conditions, we also verified that p21 protein induction by TGF-β1 could be enhanced by ectopic N1ICD (), confirming a cooperative role of TGF-β1 and Notch1 signaling in maintaining high p21 protein levels. Furthermore, the rescue of p21 expression could be achieved by ectopic N1ICD in a dose-dependent manner (). The 3.8-fold inhibition elicited by GSI became 1.3-fold when GSI was combined with a high dose of N1ICD, confirming that GSI primarily blocks endogenous Notch signaling during p21 regulation. Similar to these rescue experiments of p21 induction, thymidine incorporation assays confirmed our conclusion about a major role of GSI as a Notch pathway inhibitor (). Accordingly, upon control Ad-GFP infection, TGF-β1 inhibited S-phase entry, and GSI reversed this effect (). Dose-dependent Ad-N1ICD infection reduced cell growth and, combined with TGF-β1, suppressed growth by 98% (). Under these conditions, GSI could weakly restore cell growth, and the higher the N1ICD dose, the less effective GSI was. Thus, N1ICD can antagonize GSI, suggesting that the Notch pathway is a primary target of GSI in the cell model used. It follows from the model we present in that if p21 induction and epithelial cytostasis by TGF-β requires downstream activation of Notch signaling and because JAG1 protein levels accumulate after 6 h of stimulation with TGF-β1 (), the addition of GSI in HaCaT cells that are prestimulated with TGF-β1 should effectively block the cytostatic response. In all previous experiments, GSI was added 0.5–1 h before TGF-β1 (). However, also when GSI was added 12–18 h after TGF-β1 stimulation, it was effective in blocking the cytostatic response of TGF-β1 weakly by 20–30%, whereas addition between 24 and 48 h gradually enhanced the potency of TGF-β1 in causing cytostasis (). Thus, a time window for TGF-β1 cytostasis spans the first 24 h. On the other hand, the addition of GSI 12–48 h after TGF-β1 could not considerably block p21 protein induction (). Thus, robust levels of p21 correspond to the even weak (20%) suppression of thymidine incorporation observed when GSI is added 12 h after TGF-β1 (; GSI 48 h). This implies that the early period of 0–12 h of TGF-β stimulation represents a critical window during which both p21 induction and suppression of S-phase entry is sensitive to GSI. The small difference observed between the effect of GSI on p21 expression () and thymidine incorporation () when added after TGF-β stimulation emphasizes the role of additional cytostatic regulators such as p15, S100A11, and Id2, which are coregulated by TGF-β and Notch, or c-Myc, which is not regulated by Notch (). Thus, we conclude that GSI primarily acts as an inhibitor of the Notch pathway, and its adverse effect on the accumulation of phosphorylated R-Smads cannot fully explain the cellular phenotypes under investigation. The data support a model whereby TGF-β induces Notch ligands that activate signaling. This supports the duration of TGF-β signaling at sufficiently high and long levels for cell cycle arrest to occur (). The induction of JAG1 (and possibly DLL4) by TGF-β leads to the activation of endogenous Notch/CSL signaling, which is required for sustained p21 induction and is important for epithelial cytostasis by TGF-β. The latter conclusion was verified after ectopic p21 expression (Fig. S3, available at ), which antagonized the reversion in TGF-β1–mediated growth arrest elicited by GSI and led to robust cytostasis (Fig. S3 A). Interestingly, very high levels of ectopic p21 protein (6–10-fold relative to the endogenous TGF-β–induced p21 level) were required to bypass the neutralizing effect of GSI (Fig. S3 B). This suggests that TGF-β in the presence of GSI might induce target genes that permit sustained cell proliferation even in the presence of high levels of potent cell cycle inhibitors such as p21. This finding plus the previous result on p21 induction by TGF-β1 in the presence of GSI that was added several hours after TGF-β1 () raised the possibility that although clearly is a responsive gene to the TGF-β–Notch pathways, the physiological relevance of p21 to epithelial cytostasis induced by the same pathways remains to be determined. To rigorously test the role of p21 on epithelial cytostasis downstream of TGF-β–Notch signaling, we attempted to knockdown p21 expression in HaCaT cells after siRNA transfection. Such attempts always led to a partial reduction of p21 mRNA and protein levels by 60–70%, which correlated with a partial defect in the cytostatic response to TGF-β (unpublished data). To obtain definitive evidence for a role of p21 in the TGF-β cytostatic program, we made use of two individual cell clones of human mammary epithelial MCF-10A cells, whose endogenous gene was deleted after homologous recombination ( and Fig. S4, available at ; ). Similar to the effect on HaCaT, NMuMG, and HMEC cells ( and S1), TGF-β1 suppressed thymidine incorporation in control uninfected MCF-10A cells (unpublished data) or in MCF-10A cells transiently infected with control Ad-GFP (). Ad-N1ICD infection led to a substantial suppression of S-phase entry, which was comparable with that obtained by 2 ng/ml TGF-β1 (). The combination of TGF-β1 and N1ICD led to a very strong cytostatic response (). In contrast, infection of the p21 knockout clones of MCF-10A cells and stimulation with TGF-β1 failed to show any measurable suppression of thymidine incorporation ( and S4 A). The reciprocal experiment using GSI as a means of blocking endogenous Notch signaling corroborated the results with ectopic N1ICD expression. Thus, GSI effectively blocked the suppression of thymidine incorporation by TGF-β1 in wild-type MCF-10A (), whereas in the p21 knockout clones, thymidine levels remained high in the absence or presence of GSI ( and S4 B). It is worth noting that the two p21 knockout clones incorporated substantially higher levels of thymidine compared with wild-type cells ( and S4). This correlated well with the absence and presence of endogenous p21 protein expression, respectively (, bottom; and Fig. S4 C). Therefore, these experiments strongly implicate a functional role of p21 in the cytostatic response of epithelial cells downstream of TGF-β and Notch signaling. The impetus for this study was the realization that TGF-β and Notch pathways act as tumor suppressor and prometastatic or oncogenic pathways during carcinogenesis (; ). We establish that Notch and TGF-β cooperatively suppress epithelial cell growth when both pathways are simultaneously activated. On the other hand, TGF-β induces Jagged1 ligand synthesis, which then activates Notch signaling in the same cell population, thus rendering TGF-β partially dependent on Notch signaling during the establishment of cytostasis (, , and ). We observed the same type of interdependent relationship between the two pathways when large-scale gene expression analysis was performed ( and ). Additionally, however, we measured many genes that were uniquely regulated by the combined input of TGF-β1 and Notch1 both positively and negatively. Finally, we demonstrate that TGF-β–induced cytostasis requires the durable expression of factors such as p21, which is achieved by an initial TGF-β input followed by a secondary but indispensable Notch input ( and ). While p21 is not the only gene of the cytostatic program of TGF-β that is affected by Notch signaling inhibition, evidence derived from p21 knockout epithelial cells strongly links this cell cycle regulator to the cytostatic response of the cells. Analyzing in detail the functional roles of other genes uncovered in this study and the detailed mechanisms of their regulation by TGF-β and Notch may shed light on even more novel facets of the binary roles these two pathways play during the control of epithelial proliferation and tumor development. In analyzing the transcriptomic response of HaCaT keratinocytes to TGF-β in the presence of GSI ( and ), we uncovered an extensive dependence of gene expression regulation on endogenous Notch pathway activation. Based on careful control experiments in which we examined the role of GSI on phospho-Smad accumulation in response to TGF-β (), two possible working models can explain the transcriptomic results. First, an adverse negative effect of GSI on phospho-Smad accumulation may be the main reason behind the substantial decrease in the number of TGF-β–responsive genes we measured, especially at the 6-h time point (). In simple terms, GSI lowers the active levels of Smads in the epithelial cell, thus reducing the downstream output of these signal transducers as measured by gene expression readouts. However, the majority of the evidence presented here argues against this model. Two examples are illustrative: a large number of new target genes of the TGF-β pathway were uncovered in our screen whose expression is regulated only when cells are treated with GSI ( and Table S1). This suggests that the inhibition of γ-secretase activity in the cell redirects the specificity of gene expression regulation by TGF-β toward new targets. This phenomenon is hard to reconcile based on a model in which Smad activation is gradually diminishing as a result of GSI. Specific gene targets of TGF-β/Smad signaling such as TIEG () or c-Myc () are not affected at all by the presence of GSI. If the inhibitor were simply reducing phospho-Smad levels, these two and other genes should have shown new expression profiles in the presence of GSI, which was never observed. The second working model, which is corroborated by the majority of the data presented here, is outlined in and essentially favors a sequential mode of signaling starting with TGF-β and later followed by Notch. This pathway targets critical mediators of the cytostatic response of epithelial cells, namely the cell cycle inhibitors p15 and p21. Depletion of endogenous Jagged1 and CSL proteins supports this model, and no evidence for a contribution of these classic Notch pathway components to the process of Smad activation could be gained ( and ). This model of sequential signaling suggests that gene targets like p15 and p21 are directly regulated by the incoming Smad pathway as previously established (; , ; , ; ), and subsequent onset of Notch signaling after the accumulation of ligands of this pathway, such as Jagged1, contributes to a sustained and robust transcriptional induction of the same genes. From this perspective, it would be interesting to examine in deeper detail the transcriptional mechanisms that mediate the regulation of p15 and p21 gene expression by the combined TGF-β/Smad and Notch/CSL signaling inputs. In this respect, it is interesting that Jagged1 clusters together with p15 and p21 as genes of the same synexpression group downstream of TGF-β, as all of these genes seem to require the activity of Smad signaling and the cooperation of transcription factors of the FoxO family (). The mechanistic details of how Smads, FoxO members, and additional cofactors orchestrate the time-dependent induction of Jagged1 remain to be elucidated. An interesting question remaining open at this stage is the mechanism by which the inhibition of γ-secretase affects the accumulation of phosphorylated R-Smads downstream of the TGF-β receptor. Presently, we examine three alternative possibilities: γ-secretase may be involved in the activation process of the TGF-β receptor, thus playing a critical role in the phosphorylation of R-Smads by the type I receptor; γ-secretase positively contributes to the stability of phosphorylated R-Smads, possibly by down-regulating an ubiquitin ligase involved in phosphorylated R-Smad turnover; or γ-secretase negatively regulates the phosphatases that remove the C-terminal phosphates from phospho–R-Smads. Ongoing work aims at addressing these alternative mechanisms. Among all components of the Notch pathway whose expression is regulated by TGF-β signaling, our evidence favors more prominent roles for the ligands of these pathways such as Jagged1 and DLL4 in keratinocytes () or Jagged1 and DLL1 in mammary epithelial cells (Fig. S2). The observed regulation of receptors of the Notch pathway appeared to be indirect (unpublished data) and possibly the result of an autogenous negative feedback pathway whereby the activation of Notch signaling itself leads to the down-regulation of its receptor genes. Although our evidence favors this model, TGF-β was found to up-regulate the expression of Notch4 concomitantly to the down-regulation of Notch1, at least in mammary epithelial cells (Fig. S2). The functional relevance of such a reciprocal regulation of Notch receptors during cytostasis of mammary epithelial cells remains unknown. Alternatively, TGF-β may instruct for this switch of Notch receptor expression as it promotes epithelial-mesenchymal transition of the mammary cells, a physiological response in which the cross talk between TGF-β and Notch signaling has already been established at least in keratinocytes (). In establishing the sequential signaling pathway of TGF-β followed by Notch as a critical regulator of epithelial cytostasis (), we primarily focused on regulation of the cell cycle inhibitor p21. This was prompted by the characteristic expression profile measured for p21 during our experiments (). However, regulation of additional factors such as p15, Id2, or S100A11 seems to also be integrated in the same physiological response. Thus, in emphasizing a role of p21 as a major target gene of the sequential signaling cascade outlined here, one should strongly consider the legitimate and equipotent contribution of the other regulators of this multigenic response to TGF-β. This point is underscored by the experiments using p21 knockout MCF-10A cells ( and S4). Our evidence fully recapitulates the original findings of and further demonstrates the role of p21 downstream of Notch signaling in mammary epithelial cells. However, it should be kept in mind that MCF-10A cells represent relatively normal immortalized human epithelial cells that have spontaneously lost the expression of their endogenous p15 cell cycle inhibitor gene (). Thus, the p21 knockout MCF-10A clones represent a double knockout for p15 and p21 expression, and this is the main reason why TGF-β completely fails to elicit proliferation arrest in these cell clones. Our attempts to deplete p15 or p21 individually from HaCaT or other epithelial cell models in which TGF-β–mediated cytostasis is well understood always led to partial and relatively weak phenotypes, presumably because of the compensation provided by the other genes of the cytostatic program that remained intact (unpublished data). In summary, this study establishes a relay mechanism of signal transduction that plays critical roles for the establishment of epithelial cell cycle arrest. This mechanism fits well with the established tumor suppressor roles of TGF-β and Notch signaling. Additionally, this mechanism opens the exciting possibility whereby the two signaling pathways may be misregulated in an interdependent manner during human tumor progression, thus offering a promising territory for future studies in cancer cell biology. Human HaCaT keratinocytes, human MCF-10A mammary epithelial cells, human embryonic kidney 293 cells, mouse NMuMG mammary epithelial cells, and their derivative clone NMe have been described previously (). Mink lung epithelial cells (Mv1Lu) were purchased from the American Type Culture Collection, and HMECs were obtained from R.A. Weinberg (Whitehead Institute for Biomedical Research/Massachusetts Institute of Technology, Cambridge, MA). MCF-10A clones 1 and 2 deficient in the endogenous gene were obtained from B.H. Park (The Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University, Baltimore, MD; ). Recombinant mature TGF-β1 was purchased from PeproTech. The TGF-β type I receptor kinase inhibitor LY580276 and TGF-β types I and II receptor kinase dual inhibitor LY364947 were obtained from J.M. Yingling (Eli Lilly, Inc., Indianapolis, IN; ). The inhibitor X against γ-secretase activity (GSI) was purchased from Merck Biosciences/Calbiochem. Adenoviruses expressing GFP were based on the bicistronic Adeasy vector obtained from B. Vogelstein (The Johns Hopkins Medical Institutions, Baltimore, MD). Adenoviruses expressing N1ICD were based on Adeasy, which was obtained from G.P. Dotto (Harvard Medical School, Boston, MA) and F. Radtke (Ludwig Institute for Cancer Research [LICR], Lausanne, Switzerland). Adenoviruses expressing wild-type human p21 were obtained from K. Walsh (Boston University School of Medicine, Boston, MA). Adenoviruses were amplified and titrated in human embryonic kidney 293 cells, and transient infections were performed as described previously (). Under standardized conditions, epithelial cells were infected at a rate of 75–85% without any signs of cytotoxicity as assessed by live GFP autofluorescence and immunofluorescence microscopy. The human -specific (GenBank/EMBL/DDBJ accession no. ; reagent number M-007772; human RBPSUH) and human JAG1-specific (GenBank/EMBL/DDBJ accession no. ; reagent number L-011060; human Jag1) siRNAs were pools of four RNA oligonucleotides termed On-Target Plus SMARTpools that minimize off-target effects; siRNA against the reporter vector pGL2 (GenBank/EMBL/DDBJ accession no. ) served as a control. All siRNAs were purchased from Dharmacon. HaCaT cells were transiently transfected with 20 nM siRNA using siLentFect (Bio-Rad Laboratories) according to the manufacturer's protocol. Cells were transfected 1 d after seeding, remained with transfection cocktail for 24 h, were switched to fresh medium plus TGF-β1, and were retransfected with siRNA for another 24 h before cell analysis. Total proteins from NMuMG or HaCaT cells were extracted, subjected to SDS-PAGE, and analyzed by Western blotting as described previously (). Mouse monoclonal anti–β-tubulin (T8535) antibody was obtained from Sigma-Aldrich; mouse monoclonal anti-Cip1/WAF1 (clone 70) was purchased from BD Transduction Laboratories; rabbit polyclonal anti-Notch1 (ab8925) was purchased from Abcam; mouse monoclonal anti-Smad1/2/3 (H2), rabbit polyclonal anti-Notch4 (H-225), rabbit polyclonal anti-Notch3 (M-134), rabbit polyclonal anti-Jagged1 (H-66), rabbit polyclonal anti-TGFβRI (V-22), rabbit anti-CSL/RBP-Jκ, rabbit anti-DLL4/Delta-4 (H-70), and mouse anti–α-tubulin (TU-02) were obtained from Santa Cruz Biotechnology, Inc. Secondary anti–mouse IgG and anti–rabbit IgG coupled to HRP were obtained from GE Healthcare. The ECL detection system was prepared in house, and immunoblots were scanned on a CCD camera (LAS-1000; Fuji). Densitometry was performed using the AIDA program of the scanner. Cells were cultured, stimulated with growth factors, and labeled metabolically with [H]thymidine as described previously (). The data are plotted as mean values with SEMs of triplicate repeats per independent experiment. Each independent experiment was repeated at least three times. Cell monolayers were washed with PBS, trypsinized, and stained with trypan blue (Sigma-Aldrich), and viable cell numbers were calculated using a counter (Z1; Beckman Coulter). Cell numbers are plotted as means from triplicate determinations with SEMs per experiment. Statistical analysis of thymidine incorporation assays was performed by two-tailed paired tests. Significance was considered at a p-value of <0.05. HaCaT cells were cultured in the presence of 3% FBS and stimulated with 2 ng/ml TGF-β1 for 2, 6, and 48 h in the absence or presence of 4 μM GSI. Cells for the 2-h time point were also incubated with 10 μg/ml cycloheximide to block protein synthesis. Total RNA extraction and cDNA probe labeling was performed as described previously (). Equal amounts of labeled cDNA probes per pair were hybridized to cDNA microarray chips (Hver2.1.1) from the Sanger/LICR/Cancer Research UK Consortium (see for details and hybridization protocols). The glass chips contained 14,633 single-stranded cDNA elements of 1.5-kb mean length, which represent 10,252 unique human genes. The human IMAGE cDNA clone collection was obtained from the Medical Research Council Human Genome Microarray Platform Resource Centre. cDNA clone resequencing was performed by Team 56 at the Sanger Institute. Hybridizations were performed in triplicate using RNAs from three independent cultures and including the dye swap control. Microarray scanning, image analysis, and primary spot intensity statistical analysis were performed as described previously (). Regulated genes were selected based on the mean ratio value of ≥1.7 for up-regulated genes and ≤0.55 for down-regulated genes. In addition, regulated genes had to be expressed on three arrays out of three and with a test value for the ratios within replicates corresponding to a probability of <0.05. Statistically significant genes (P < 0.05) were clustered based on their expression values using the K-means statistical algorithm that is incorporated into GeneSpring 7.2 data mining software (Silicon Genetics/Agilent Technologies). For all time points, we considered as a reference a duplicate cell culture in which TGF-β1 was replaced by vehicle. Functional classification of regulated genes was performed manually based on exhaustive PubMed searches (). Total RNA from NMuMG or HaCaT cells was analyzed by semiquantitative RT-PCR as described previously () using specific primers (). Primers for mouse ′ () were used to ascertain that an equivalent amount of cDNA was synthesized. Specificity controls included reactions in which reverse transcriptase was omitted (−RT) and in which cDNAs were replaced with water. DNase RQI–digested RNA from NMuMG and HaCaT cells was analyzed by quantitative real-time RT-PCR as described previously (). Primers () were designed with Primer Express (Applied Biosystems). Reactions were performed in a sequence detector (ABI-Prism 7000; Applied Biosystems) in triplicate, and, for each condition, the ground condition (minus TGF-β1 and/or mock infected with Ad-GFP) was set as 1; expression data are presented as bar graphs of mean values plus SD. Fig. S1 shows thymidine incorporation and cell counting assays in various epithelial cell types. Fig. S2 shows semiquantitative RT-PCR assays and corresponding immunoblot assays for Notch family member expression in NMuMG cells. Fig. S3 shows thymidine incorporation and immunoblot assays in HaCaT cells expressing ectopic p21. Fig. S4 shows thymidine incorporation and immunoblot data from clone 1 of the p21 knockout MCF-10A cells. Table S1 provides information about transcriptomic analysis of the TGF-β1 response after Notch inhibition.Online supplemental material is available at .

In addition to its participation in growth factor regulation of proliferation and apoptosis, the extracellular signal–regulated kinase (ERK)–MAPK pathway has important functions in the differentiation of postmitotic cells. For example, in embryos, inhibition of ERK activation prevents animal caps from differentiating into mesenchymal tissue (), whereas mice harboring deletions in ERK2 exhibit severe defects in primary mesenchyme formation without major changes in cell proliferation or apoptosis (). In addition, this pathway can regulate the activity of several lineage-specific transcription factors, including MyoD (muscle; ), Sox9 (cartilage; , ), and PPARγ (adipose tissue; ). In bone, the ERK–MAPK pathway is a major conduit for conveying information about the extracellular environment to the nucleus, and has been implicated in the response of bone to a variety of signals, including hormone/growth factor stimulation (; ; ), extracellular matrix–integrin binding (; ), and mechanical loading (). Despite intensive investigation, the physiological role of the MAPK pathway in osteoblasts remains controversial, with some studies supporting a stimulatory role in osteoblast differentiation and others proposing that this pathway is inhibitory (for review see ). For example, activation of α2β1 integrins by the type I collagen matrix of bone was reported to stimulate in vitro osteoblast differentiation via focal adhesion kinase activation of MAPK (; ). Also, short-term pharmacological inhibition of MAPK signaling blocked osteoblast-specific gene expression in mature osteoblasts, whereas a constitutively active form of the MAPK intermediate, MEK1, was stimulatory (). In related studies, we showed that RUNX2, which is an essential transcription factor for bone formation (), is required for cells to respond to MAPK in vitro. Both phosphorylation and transcriptional activity of RUNX2 were stimulated by the ERK–MAPK pathway, suggesting that this factor is a MAPK substrate and an important mediator of the MAPK response (; ). Also, treatment of MC3T3-E1 cells or primary cultures of marrow stromal cells with FGF2 induced RUNX2 phosphorylation and osteocalcin expression by a process requiring MAPK activation (). In contrast, other in vitro studies reached an opposite conclusion, showing that the ERK–MAPK pathway can antagonize osteoblast functions. Thus, epidermal growth factor stimulation of ERK prevented SMAD1 activation by bone morphogenetic proteins in epithelial cells (). This inhibition was explained by ERK-dependent phosphorylation of a distinct site in the linker region of SMAD1 that led to the exclusion of this molecule from the nucleus (). Also, chronic treatment of osteoblast cultures with MAPK inhibitors was reported to actually stimulate osteoblast differentiation, whereas ERK activation was inhibitory (; ). The conflicting results of in vitro studies emphasize the need to address the role of the ERK–MAPK pathway in osteoblast function in vivo. To this end, a unique transgenic strategy was developed involving selective expression of constitutively active and dominant-negative forms of the MAPK intermediate MEK1 in osteoblasts using the osteocalcin promoter. With this approach, ERK–MAPK activation was found to stimulate osteoblast differentiation and skeletal development through a pathway involving RUNX2. Transgenic mice were developed using a 647-bp mouse osteocalcin () gene 2 (mOG2) promoter to drive osteoblast-specific expression of constitutively active (MEK-SP) or dominant-negative (MEK-DN) forms of the MAPK intermediate MEK1 (; ). transcription is low in proliferating osteoprogenitor cells, and it does not become elevated until the later stages of differentiation, when cells are largely postmitotic (; ). By using a promoter that is selectively expressed in postmitotic cells, we reasoned that it should be possible to target MAPK functions related to osteoblast differentiation, rather than earlier functions associated with cell proliferation. A schematic representation of the transgene construct is shown in . This mOG2 promoter region contains sufficient information to selectively express a lac Z reporter in osteoblasts with no detectable expression in cartilage or joints (). A total of five transgenic founders were obtained for each construct. Of these, three lines having approximately equivalent transgene expression were retained for further study. Mice were viable and bred normally. As shown in , strong transgene expression was detected in skeletal tissues, first appearing at E14.5 and persisting in newborn animals. No transgene expression was seen in soft tissues, such as muscle, brain, or liver (unpublished data). Similar expression levels and tissue distribution were seen with all lines examined. For this reason, subsequent experiments used lines SP221 and DN288, with certain experiments repeated with SP413 and DN315. In situ hybridization conducted in newborn mice showed clear localization of the transgene to osteoblasts on endosteal and select trabecular surfaces of long bones and the conspicuous absence of expression in growth plates (). Similar localization to osteoblast layers of the calvarium was also observed (unpublished data). Initial studies examined the effects of transgene expression on the in vitro growth and the differentiation of calvarial osteoblasts. Cells were isolated from 4-wk-old wild-type, TgMek-sp, and -dn mice and grown under differentiating conditions. Transgene expression was first detected after 7 d in culture, and continued to increase throughout the experiment (). Western blotting with a specific anti–phospho-ERK antibody was used to verify that transgenes altered ERK–MAPK signaling (). TgMek-sp cells had phospho-ERK levels that were nearly twice those of their wild-type littermates, whereas phospho-ERK in TgMek-dn cells was reduced by 50%. In contrast, total ERK levels were not affected by transgene expression. To confirm that the increased phospho-ERK levels in TgMek-sp cells are explained by constitutive activation of MEK, cells were treated with the Raf inhibitor ZM336372, which blocks the MAPK pathway upstream of MEK1/2 (). As expected, this inhibitor dramatically reduced phospho-ERK levels in cells from wild-type and TgMek-dn mice, while having no effect on levels in TgMek-sp cells. As a component in growth factor signaling, the ERK–MAPK pathway can stimulate cell proliferation (). However, growth curves for wild-type, TgMek-sp, and TgMek-dn cells were identical, with all groups growing to the same saturation density (). This is likely explained by the delayed expression of transgenes from the mOG2 promoter that does not become active until culture day 7, when cells are close to confluence and largely postmitotic (compare ). This late transgene expression facilitated the interpretation of subsequent experiments by removing proliferation and cell density as possible confounding variables. In contrast to the results of proliferation studies, Mek-dn and -sp transgenes dramatically altered calvarial cell differentiation and gene expression (). Differentiation of TgMek-dn cells was reduced relative to wild type, as measured by von Kossa staining of mineralized nodules at day 14 (), alkaline phosphatase activity (), calcium accumulation as mineral () and expression of OCN, bone sialoprotein (BSP, and alkaline phosphatase (ALP) mRNAs (). Time course studies revealed that the earliest effects of transgenes were seen at day 10, immediately after initial transgene expression was detected at day 7 (). Runx mRNA was not affected at day 7, but was reduced by 40% on day 10 and 14. Opposite results were obtained with TgMek-sp cells that accumulated twice the mineral of wild-type cells at d 14 and had higher levels of alkaline phosphatase and osteoblast marker mRNAs. Runx2 mRNA levels were only slightly stimulated on days 10 and 14. The (ALP), and genes shown to be regulated by Mek-sp and -dn in are all known to be directly or indirectly controlled by RUNX2. For , and itself, this regulation involves direct binding of RUNX2 to regulatory elements in the proximal promoter region (; ; ). In previous cell culture studies, we showed that MAPK activation by either transient transfection of the MC3T3-E1 osteoblast cell line with a MEK-SP expression vector or activation of endogenous ERK phosphorylation by treatment with FGF2 stimulated phosphorylation and transcriptional activity of RUNX2 (, ). Two types of experiments were conducted to determine if RUNX2 phosphorylation and transcriptional activity is modified in cells from TgMek-sp and -dn mice, and if changes in phosphorylation could explain differences in osteoblast differentiation. In the first experiment, cells were grown under differentiating conditions for 7 d and metabolically labeled with [P]orthophosphate, and then total RUNX2 was immunoprecipitated from cell extracts using a specific polyclonal antibody. A replicate culture was labeled with [S]Met/Cys and processed in the same way to normalize P incorporation to total RUNX2. As shown in , steady-state RUNX2 phosphorylation was increased twofold in TgMek-sp cells versus wild-type controls, while phosphorylation in TgMek-dn cells was reduced by 50%. In the second study, cells were transfected with 1.3-kb mOG2-luc or 6OSE2-luc reporter genes, and luciferase activity was measured after 3 and 7 d. A previous work showed that both of these promoter constructs are highly responsive to RUNX2 (). The 1.3-mOG2 fragments, like 0.647-kb mOG2, contains all known elements necessary for osteoblast-specific expression of Ocn, including RUNX2 binding sites (osteoblast-specific element 2 [OSE2]) at −608 and 137, whereas 6OSE2-luc contains six copies of the OSE2 site in front of a minimal mOG2 promoter (from −47 to 13 bp; ; ). After 7 d in culture, luciferase activity from both reporter genes was higher in TgMek-sp cells than wild-type cells, whereas activity in TgMek-dn cells was reduced (). In contrast, no differences in activity were seen at day 3, which is before Mek-sp and -dn transgenes become active. These results cannot be explained by differences in total RUNX2 levels or overall cell differentiation that were unaffected by transgenic status at the 7-d time point (). Rather, the ability of RUNX2 to stimulate OSE2-dependent transcriptional activity was increased. Collectively, these studies show that modest perturbations in ERK–MAPK activity via transgenic expression of MEK-SP or -DN in postmitotic osteoblasts dramatically affect in vitro RUNX2 phosphorylation, RUNX2-dependent transcriptional activity, and osteoblast differentiation. Changes in RUNX2 activity preceded differentiation changes, which is consistent with the concept that Mek-dependent regulation of this transcription factor accounts, at least in part, for subsequent changes in differentiation. Consistent with results obtained in osteoblast cultures, skeletal development was accelerated in TgMek-sp and delayed in TgMek-dn embryos (). This was reflected by differences in skeletal size, body weight, and mineralization. Skeletons from TgMek-sp embryos were significantly larger (based on femur lengths at embryonic day (E) 15.5 [] and body weight []) than wild-type littermates, and weights and embryo size were reduced in TgMek-dn mice. Moreover, differences in calvarial mineralization were apparent with mineralized area of calvarial preparations at E15.5 being reduced by 20% in TgMek-dn and stimulated by an equivalent amount in TgMek-sp embryos (). Striking differences in skeletal maturation of long bones were also seen when histological sections of E15.5 embryos were examined (). At this age, the initial stages of endochondral bone formation have normally already taken place, with vascular invasion and metaphyseal bone formation already well underway (). However, this process was significantly delayed in TgMek-dn embryos. Only early bone collar formation was apparent, with cartilage still occupying the inner metaphyseal regions of all long bones examined. In contrast, TgMek-sp embryos exhibited accelerated trabecular bone formation, with the metaphyseal region already having expanded toward the diaphyses. These results indicate that ERK–MAPK activity is important for the early stages of intramembraneous and endochondral bone development. The cell culture studies shown in , as well as previous work from this laboratory (), suggest that RUNX2 is an important target for regulation by the ERK–MAPK pathway in osteoblasts. To test this concept in vivo, we examined genetic interactions between and Mek-sp or -dn transgenes. mice are known to phenocopy many aspects of the human genetic disease cleidocranial dysplasia (CCD), including the characteristic clavicular hypoplasia and open fontanelles of the cranium with lesser involvement of other skeletal sites (). Because clavicles and cranial bones are particularly sensitive to the effects of haploinsufficiency, we predicted that these two tissues should be preferentially sensitive to transgenic modification of the ERK–MAPK pathway under conditions where RUNX2 is limiting. TgMek-sp or -dn mice were crossed with mice, and skeletal phenotypes were determined using whole-mount skeletal preparations (). show the analysis of E18.5 animals from TgMek-sp breedings, and show results of the TgMek-dn cross. mice exhibited the expected clavicular hypertrophy and calvarial hypomineralization of CCD, with a resultant reduction of clavicle cross-sectional area of 75 (), and 28, and 18%, respectively, in mineralized areas of calvaria (). In contrast, femur length was not significantly affected (). The Mek-sp transgene was able to partially rescue both the clavicular hypertrophy and calvarial hypomineralization of mice. Transgene expression increased the cross-sectional area of clavicles by 53% in animals, although it only increased this parameter by 12% in wild- type mice (). Restoration of calvarial mineralization was also observed with TgMek-sp, increasing this parameter in animals to 90% of the wild-type control (a 23% increase relative to mice; ). In contrast, Mek-sp had no effect on femur length in mice, but it did increase this parameter by 18% in wild-type littermates (). Results with the TgMek-dn cross were even more striking. First, after four consecutive breedings, we failed to obtain viable newborn pups with the TgMek-dn genotype. This is in contrast to the TgMek-sp genotype that was present in the expected Mendelian ratio. When embryos were harvested by Caesarean section at E18.5, TgMek-dn embryos were obtained in the expected ratios, indicating that this genotype cannot survive the birth process. Embryos were smaller and exhibited selective worsening of CCD-associated features. Specifically, Mek-dn reduced clavicular cross-sectional area by 67% in embryos, but only reduced this value by 17% in +/+ mice (). Also, Mek-dn reduced calvarial mineralization by 31% in embryos, but only reduced this value by 15% in wild type (). Collectively, these results show that in mice, the clavicles, and calvaria are selectively sensitive to manipulation of ERK–MAPK activity in osteoblasts, which is consistent with the concept that RUNX2 is an in vivo target of this pathway. In this study, we show that the ERK–MAPK pathway can positively regulate bone development in vivo through a mechanism involving RUNX2. Phosphorylation and transcriptional activity of RUNX2 were stimulated in calvarial osteoblasts from Tg Mek-sp mice and inhibited in TgMek-dn cells. Furthermore, haploinsufficiency selectively enhanced the effects of Mek-sp and -dn transgenes on clavicles and calvaria, two tissues known to be preferentially sensitive to gene dosage. On the basis of this work, we conclude that ERK–MAPK signaling is critical for in vivo osteoblast activity, and propose that this response is explained, at least in part, by phosphorylation and activation of RUNX2. These results are consistent with previous cell culture studies from this and other laboratories. For example, we reported that MAPK activation, either by transfection of osteoblasts with constitutively active MEK1 or treatment with FGF2, rapidly increased RUNX2 phosphorylation and transcriptional activity through a process that was blocked by MAPK inhibition, whereas acute inhibition of MAPK in differentiated cells blocked osteoblast-specific gene expression (, ,). In related studies, showed that expression of a dominant-negative ERK1 blocked differentiation of human osteoblasts. Similarly, differentiation of human marrow stromal cells to osteoblasts is accompanied by sustained ERK1/2 phosphorylation; pharmacological MAPK inhibition or dominant-negative ERK blocked osteoblast formation in this system and stimulated adipogenesis (). Also, several reports support our hypothesis that MAPK actions in osteoblasts are, at least in part, mediated by RUNX2. During osteoblast differentiation of human marrow stromal cells, RUNX2 type II levels remain relatively unchanged, but binding to OSE2 DNA increased, as did RUNX2 phosphorylation (). Also, mechanical loading of osteoblasts, which is known to be mediated, in part, through integrin activation, induces MAPK (; ); loading of periodontal ligament cells (which are osteoprogenitor-like cells associated with tooth cementum) increased RUNX2 phosphorylation and binding to OSE2 DNA via an ERK–MAPK-dependent process (). In osteoblast-like prostate cancer cells, differentiation is accompanied by ERK1/2 activation, increased RUNX2-OSE2 binding, and expression, responses that are all blocked by MAPK inhibition (). Lastly, IGF-1, which activates P13K and, subsequently, the ERK–MAPK pathway, stimulates RUNX2-OSE2 binding and phosphorylation in vascular endothelial cells (), as well as differentiation of marrow stromal cells (; ). This work is to be contrasted with other in vitro studies that reported antagonism between the ERK–MAPK pathway and BMPs in epithelial cells (, ) and inhibitory effects of this pathway on osteoblast differentiation (; ; ). Possible explanations for these disparate results may be related to differences in the magnitude and duration of MAPK activation and/or stage of cell maturation that are both known to affect MAPK responsiveness. For example, treatment of PC12 neuronal progenitor cells with EGF transiently stimulated ERK1/2 phosphorylation and cell proliferation, whereas NGF led to sustained ERK activation and neuronal differentiation, a result explained by the formation of transient versus long-lived complexes of MAPK-activating factors (). Similarly, in studies with MC3T3-E1 preosteoblasts, EGF suppressed BMP-dependent Smad 1 activation, whereas sustained ERK–MAPK activation by extracellular matrix synthesis or constitutively active Ras stimulated BMP activity (). In contrast, the aforementioned studies, indicating an inhibitory role for MAPK in osteoblast differentiation, generally either treated cells with MAPK inhibitors for extended times or stably transfected cells with expression vectors that produced high levels of constitutively active or dominant-negative MAPK intermediates during all stages of growth and differentiation. To examine the in vivo actions of the ERK–MAPK pathway in osteoblasts, this study used the mOG2 promoter to drive expression of constitutively active/dominant-negative MEK1 in developing mice. Because this promoter is only active in mature osteoblasts that are largely postmitotic (), we were able to discriminate between effects of MAPK on cell proliferation versus differentiation. Furthermore, transgenic modification led to sustained, but quite modest, changes in ERK1/2 phosphorylation (∼50% increased/decreased levels). Using this approach, ERK–MAPK activation was shown to stimulate osteoblast differentiation of calvarial osteoblasts from transgenic animals and to accelerate bone accumulation in developing embryos without affecting cell proliferation. In contrast, dominant-negative MAPK inhibition slowed osteoblast differentiation in cell culture and delayed bone development. In vivo effects of transgenes were most striking in calvaria and long bones. Of particular interest, long bones of TgMek-dn mice displayed delayed bone collar formation, whereas TgMek-sp was associated with an increase in metaphyseal bone without any obvious effects on cartilage. This is consistent with an in situ analysis of transgene expression that was unable to detect transgene expression in growth plates (). In agreement with our in vitro studies, we also obtained evidence that MAPK regulates RUNX2 transcriptional activity. Calvarial osteoblasts from transgenic mice exhibited the predicted changes in RUNX2 phosphorylation and transcriptional activity, with these parameters being stimulated in TgMek-sp cells and inhibited with Mek-dn. Of note, changes in RUNX2 were coincident with initial activation of Mek-sp and -dn transgenes and preceded any changes in osteoblast differentiation markers or RUNX2 mRNA that were all equivalent in wild-type, TgMek-sp, and -dn groups at the 7-d time point examined ( and ). Thus, it is unlikely that changes in RUNX2 phosphorylation/activity are secondary to changes in osteoblast differentiation induced by Mek-sp/dn transgenes. Studies that examined genetic interactions between Runx2 and Mek-sp and -dn transgenes provided further evidence for in vivo regulation of RUNX2 by MAPK (). Because clavicles and calvaria are preferentially sensitive to Runx2 haploinsufficiency, they are good markers for changes in RUNX2 transcriptional activity. The observation that TgMek-sp is able to rescue the CCD phenotype of embryos, whereas TgMek-dn exacerbated clavicular and calvarial defects provides compelling in vivo evidence for our hypothesis that the ERK–MAPK pathway regulates RUNX2. Studies in other differentiating systems, such as muscle, fat, and cartilage, suggest that the ERK–MAPK pathway, acting through tissue-specific transcription factors, may generally control progenitor/stem cell lineage decisions in mesenchymal tissues. For example, in muscle differentiation, MAPK increases levels of XMyoD, possibly by decreasing protein turnover (). In fat, phosphorylation of PPARγ by ERK1 inhibits transcriptional activity of this factor to block adipogenesis (). Lastly, MAPK activation by FGFs, constitutively active FGFR3, or MEK1 mutants in cartilage increases levels of Sox9 to keep chondrocytes in a prehypertrophic state (, ). Studies in cartilage are of particular interest in that they used the promoter to express the same constitutively active MEK1 mutant used in this study. In contrast to our results with osteoblasts, -mediated expression led to a chondrodysplasia phenotype that is characterized by shortened limbs and inhibition of hypertrophy without affecting chondrocyte proliferation. The observed limb shortening was explained by a reduction in the size of chondrocytes in the hypertrophic zone. Because the ERK–MAPK pathway is known to activate (), the authors postulated that this transcription factor prevents hypertrophy, which normally requires down-regulation. In contrast, our use of the mOG2 promoter to express constitutively active MEK1 resulted in expression in osteoblasts rather than cartilage, acceleration of bone development and skeletal mineralization, and no obvious changes in cartilage. As was the case for stimulation of in cartilage, we find that ERK–MAPK can activate RUNX2-dependent transcriptional activity in osteoblasts. Although our studies focused on RUNX2 as the primary mediator of MAPK signals, other transcription factors may also participate in this response, functioning together with RUNX2 to control gene expression. Of primary interest, ERK is known to phosphorylate and activate a second kinase, RSK2, in the cytoplasm of many cells (). RSK2 phosphorylates ATF4, a transcription factor that, together with RUNX2, controls the osteoblast-specific expression of the osteocalcin gene (). Mice deficient in either RSK2 or ATF4 have similar bone phenotypes characterized by delayed cranial and long-bone development (). Furthermore, RUNX2 and ATF4 physically and functionally interact in osteoblasts, although it is not known whether phosphorylation of RUNX2 and/or ATF4 affects this interaction (). While this work was under review, reported that the high bone-mass phenotype of mice harboring an osteoblast-specific deletion of the Ras GTPase-activating protein (GAP), Nf1, involves activation of the ERK–MAPK pathway. Although these authors attributed the phenotype of −/− mice exclusively to activation of ATF4 via RSK2, the involvement of Runx2 was not directly examined. To address the possible involvement of ATF4 in TgMek-sp and -dn mice, we transfected calvarial osteoblasts with an ATF4 reporter-luciferase construct containing oligomerized copies of the ATF4-responsive enhancer OSE1 (; ). Unlike results with the RUNX2 reporters (), this construct was not affected by transgene activity (not depicted). Based on this result, the involvement of ATF4 in the osteogenic response seen with transgenic ERK–MAPK activation appears to be secondary to that of RUNX2. However, this issue still needs to be examined in greater detail. This study establishes an important role for the ERK–MAPK pathway in RUNX2 regulation, osteoblast differentiation, and fetal bone development. Furthermore, the animal models developed will be extremely useful for assessing roles of ERK–MAPK signaling during bone remodeling in adult and ageing mice, as well as for evaluating the possible therapeutic value of MAPK manipulation as a means of regulating bone mass in diseased states. The plasmid pGL647 contains a region of the mOG2 gene from −647 to +13 bp, which was previously shown to contain sufficient information to drive osteoblast-specific gene expression in vivo (). Two mutated forms of MEK1 cDNA were individually subcloned into pGL647. MEK(SP) includes S218/222E mutations and deletion of residues 32–51 to produce a constitutively active MEK1 mutant (), whereas the dominant-negative mutant MEK(DN) contains S218/222A mutations (). Plasmids were linearized by digestion with ClaI, and DNA fragments for microinjection were purified using a Nucleospin Extraction kit (CLONTECH Laboratories, Inc.). Transgene constructs are quantified and microinjected into (C57BL/6 X SJL) F2 mouse oocytes (Charles River Laboratories) and surgically transferred to pseudopregnant C57BL/6 dams. The founders were screened by PCR using mouse tail genomic DNA. Transgenic expression in tissues was assessed by RT-PCR using two sets of transgene-specific primers. Primers used for MEK(SP) were 5′-CGGAGACCAACTTGGAGGC-3′ and 5′-CGAATTCGTTGGCCATTTC-3′; primers used for MEK(DN) were 5′-CGGAGACCAACTTGGAGGC-3′ and 5′-CGAAGGCGTTGGCCATGGC-3′. The last six nucleic acids in the ends of downstream primers harbored mutations that were used to discriminate between transgene and endogeneous MEK1 expression. Transgenic founder animals were bred into C57BL/6 mice for at least six generations to obtain a defined genetic background. Previously described heterozygous Runx2+/− mice () were obtained from P. Ducy (Baylor College of Medicine, Houston, TX). All studies with mice were performed in compliance with the University of Michigan Committee for the Use and Care of Animals. Tissues were fixed in 10% formalin and embedded in paraffin, and 7-μm-thick sections were cut and stained with hematoxylin and eosin. RNA in situ hybridization was done using a digoxigenin-labeled riboprobe specific for SV40 sequences in the transgene (Roche). Sections were viewed with a microscope (Eclipse 50i; Nikon) using a 10× objective. Images were captured with a digital camera (Stereo TLRC; Leica) using Image-Pro Plus 5.1 (Media Cybernetics) acquisition software without further manipulation. Skeletal morphology was analyzed by alizarin red and alcian blue staining, followed by tissue clarification with KOH, as previously described (). To measure bone length, isolated bones were laid next to a ruler and photographed using a digital camera and stereo dissection scope. Bone area was calculated using ImagePro Plus software. Primary calvarial osteoblasts were isolated from 4-wk-old transgenic mice and littermates, as previously described (). Cells were grown in α-MEM/10% FCS until confluent. Cells were then replated at a density of 5 × 10 cells/cm in 35-mm dishes and grown for up to 2 wk in differentiation medium (α-MEM, 10% FCS, 50 μg/ml ascorbic acid, and 5 mM β-glycerol phosphate). Quantitative RT-PCR was performed using an ABI PRISM 7700 Sequence Detection System. Optimized primers and probes for , , , , and were purchased from Applied Biosystems. Total and phospho-ERK1/2 were measured on Western blots using specific antibodies at a dilution of 1:2,000 (Cell Signaling Technology). Primary calvaria osteoblasts were isolated from newborn mice and cultured in 75-cm flasks for 3 d until confluent. Cells were then trypsinized and replated on 35-mm dishes at a density of 2.5 × 10 cells/cm. After 24 h, cells were transfected with Fugene6 (Roche) according to the manufacturer's instructions. Each transfection contained 0.5 μg of p1.3mOG2-luc or p6OSE2-luc () and 0.05 μg of pRL-SV40 (encodes luciferase to control for transfection efficiency). Cells were then cultured in differentation medium for an additional 6 d before harvesting and assaying for luciferase activity using a Dual Luciferase Assay kit (Promega) on a luminometer (Monolight 2010; BD Biosciences). For cell labeling, calvarial cells were cultured for 6 d, as described in the previous paragraph, and transferred to phosphate or cysteine and methionine-free medium containing 0.1% dialyzed FCS (Sigma-Aldrich). After an overnight preincubation, cells were labeled for 5 h in the same media containing either 150 μCi/ml of [P]orthophosphate or [S]cysteine/methionine (MP Biomedical). Preparation of nuclear extracts and immunoprecipitation of Runx2 were described previously (). P/S incorporation was measured using an A2024 InstantImager (Packard). Statistical analyses were performed using Instat 3.0 (GraphPAD Software, Inc.). For cell culture studies ( and ), values are the means ± the SD of triplicate independent samples. In studies with mice, = 8 mice/group. Analysis of variance followed by Tukey-Kramer multiple comparisons test was used to assess significance between groups as indicated.

Atherosclerosis involves the progressive accumulation of lipids, immune cells, and ECM in the vessel wall, which can decrease blood flow or rupture to cause acute thrombosis. Endothelial cell dysfunction is the key initiating event in atherogenesis, resulting in decreased flow-induced dilation and inflammatory gene expression (). Activated endothelium recruits monocytes, which differentiate into macrophages. Elevated permeability of the endothelium is believed to allow entry of lipoproteins into the vessel wall, which become oxidized and propagate endothelial dysfunction (). Macrophages engulf low-density lipoprotein (LDL) and other lipoproteins and become foam cells, which can be visualized as fatty streaks in the vessel wall. In the continued presence of high LDL cholesterol and oxidant stress, fatty streaks progress to advanced atherosclerotic plaques (). Despite the systemic nature of most atherogenic stimuli, atherosclerosis is a focal disease affecting discrete regions of the vasculature, such as vessel curvatures and bifurcations. These regions are characterized by complex flow patterns. including flow reversal, flow gradients, secondary flows with rapid changes in flow direction, and, in some regions, turbulence (). We group all of these flow patterns under the rubric of disturbed flow. Endothelial cells sense the force of flowing blood, termed shear stress, and different blood flow patterns regulate endothelial behavior. Regions of blood vessels exposed to undisturbed, unidirectional laminar flow (henceforth termed laminar flow) are protected from atherosclerosis, and in vitro prolonged laminar flow stimulates expression of athero-protective genes (; ). By contrast, disturbed flow patterns stimulate proatherosclerotic events, including increased monolayer permeability; decreased antioxidant capacity; and enhanced expression of proinflammatory genes, such as ICAM-1, VCAM-1, and monocyte chemotactic protein-1 (MCP-1; ; ; ; ). The correlation between flow patterns and endothelial monolayer permeability has recently been demonstrated in vivo, where vascular permeability is inversely proportional to time-average shear stress and correlated with increased flow oscillation and flow gradients (; ). Interestingly, onset of laminar shear stimulates many of the same responses as disturbed shear; however, in laminar shear, these events are down-regulated as cells adapt, whereas in disturbed shear, they are sustained. Thus, failure to adapt is thought to be critical for responses to disturbed shear (). The molecular mechanisms involved in flow-induced endothelial permeability are unknown. Although vesicular transport and transcellular channels may contribute to endothelial permeability, paracellular pore formation is most likely the major pathway for macromolecule transport across arterial endothelium (). Paracellular permeability is limited by cell–cell interactions, especially those in tight junctions (TJs). Multiple molecular mechanisms implicated in regulation of endothelial paracellular permeability include changes in gene expression, phosphorylation of junctional components, myosin-dependent contractility, and stability of cortical actin (). Many signaling pathways regulate permeability, most of which affect cortical actin or myosin (). Actin remodeling is regulated by the Rho family of small GTPases, including Rho, Rac, and Cdc42 (). The p21-activated kinase (PAK) family of Ser/Thr kinases is important for Rac and Cdc42-induced cytoskeletal remodeling, affecting both actomyosin contractility and the stability of actin filaments (). Recently, PAK was shown to stimulate paracellular pore formation and increased endothelial cell permeability in response to a wide range of cellular stimuli (). PAK-mediated permeability responses require the localization of active PAK to cell–cell junctions, where PAK stimulates the phosphorylation of myosin light chain to induce contractility (). In addition, PAK can also promote paracellular pore formation by phosphorylating VE-cadherin, which results in its arrestin-dependent internalization (). PAK contains multiple domains that bind scaffolding proteins, such as Nck and Grb2, capable of regulating PAK localization (; ). Interestingly, both PAK localization to cell–cell junctions and PAK-mediated permeability were inhibited with a cell-permeable peptide corresponding to the Nck-binding sequence of PAK (). Shear stress activates the integrin family of ECM receptors, and new integrin ligation mediates effects of flow on Rac, Cdc42, and Rho activity (; , , ). Flow-induced GTPase regulation mediates cell alignment in the direction of flow and stimulates the transcription factor NF-κB, which is important for expression of inflammatory genes in the endothelium (). The idea that integrin ligation mediates these effects suggested that alterations in the subendothelial matrix composition would affect which integrins become ligated, resulting in differential signaling in response to flow. Indeed, shear stress activates NF-κB when endothelial cells are plated on a fibronectin (FN) or fibrinogen matrix, but not when cells are plated on collagen or laminin. Furthermore, FN and fibrinogen were deposited at sites of disturbed flow in vivo, which correlated with expression of NF-κB target genes (). These results suggest that matrix remodeling plays a causal role in atherogenesis. In this work, we investigate the role of flow and ECM in endothelial permeability in atherogenesis. The N terminus of PAK contains a Rac/Cdc42 binding domain that overlaps an autoinhibitory domain (AID) such that binding of active GTPases alleviates an inhibitory interaction between the AID and the C-terminal kinase domain (). PAK activation results in autophosphorylation at multiple sites (; ), including Ser141 at the end of the AID. Phosphorylation of this residue prevents the interaction of the AID with the kinase domain to maintain the active conformation. Flow-induced integrin signaling activates both Rac and Cdc42 (, ), suggesting that PAK might be activated. Using PAK Ser141 phosphorylation as a marker, bovine aortic endothelial (BAE) cells were examined. Cells were plated for 4 h on coverslips coated with either FN or diluted matrigel (MG), which, under these conditions, adsorbs to the glass as a thin layer similar to FN. MG was used as a model for normal basement membrane proteins. We found that flow stimulated biphasic PAK activation on FN; however, no significant activation occurred in cells on MG (). Collagen also failed to support PAK activation under these conditions (unpublished data). Immunofluorescence staining showed that activated PAK localized to cell–cell borders (). No major changes in cell–cell junctions themselves were noted on this time scale (see ). Consistent with previous results (), this localization was abrogated by the addition of a cell-permeant peptide that blocks the binding of PAK to Nck (unpublished data). Responses to the onset of laminar shear are transient but otherwise resemble events triggered by disturbed flow (). We therefore determined PAK activity in endothelial cells exposed to different flow patterns for longer times. BAE cells plated on FN were stimulated for 24 h with flow profiles derived from the athero-protective common carotid artery (CCA) or the athero-prone internal carotid sinus (ICS; ; ). Matrix specificity was not determined in this assay because cell-derived matrices deposited over this extended time course could affect signaling responses. Consistent with the adaptation to flow hypothesis, cells stimulated with ICS flow show elevated PAK phosphorylation compared with cells stimulated with CCA flow (). Because PAK regulates permeability of endothelial monolayers (; ), we tested whether matrix-specific PAK activation correlates with permeability. To assay flow-mediated endothelial cell permeability, we developed a novel transwell assay that used a modified cone and plate device adapted to 75-cm transwell chambers (). Using this system, we applied shear to endothelial cell monolayers and assessed the movement of a tracer across the filter. Membranes were then fixed and stained to ensure that no cell loss occurred during the assay. Consistent with previous results, we found that laminar flow transiently increased endothelial cell permeability, which returned to baseline by 4 h (). To determine whether these effects are matrix specific, BAE cells were plated on either FN or diluted MG for 4 h. Endothelial cells formed a complete monolayer with both adherens junctions and TJs as assessed by β-catenin and ZO-1 staining but deposited very little endogenous matrix (Fig. S1, available at ). Onset of flow triggered a greater increase in permeability in cells on FN compared with MG or collagen IV (). In addition, the low level of permeability in cells on MG was enhanced in a dose-dependent manner when overlaid with FN (). Matrix proteins alone without cells did not differentially affect permeability (Fig. S2). To test whether PAK is involved in flow-induced permeability, cells were either transfected with a construct encoding the PAK AID or treated with a cell-permeant peptide that contains the Nck-binding sequence from PAK. This peptide was previously shown to mimic the dominant-negative effects of kinase-dead PAK, including inhibition of endothelial permeability (; ). The peptide blocked the flow-induced increase in permeability by ∼80%, whereas an inactive control peptide containing mutations in key proline residues involved in Nck binding () had no effect (). Though transfection efficiency with the PAK AID was ∼50%, the decrease in flow-induced permeability approached 50%, indicating that it is also highly effective (). In addition to HRP, Alexa 488–labeled BSA was also used to determine flow-induced permeability. Absolute permeability to both BSA and HRP were similar (), and both showed sensitivity to PAK inhibition (). Disturbed flow is known to increase permeability compared with steady or arterial flow patterns (). To confirm these results in our system, BAE cells on FN were exposed to CCA or ICS flow for 4 h, and permeability was assessed. ICS flow increased monolayer permeability nearly twofold compared with CCA flow (). Immunofluorescence revealed that active PAK was localized to cell– cell junctions after 4 h of ICS flow but not after CCA flow (unpublished data). The blocking peptide also inhibited permeability induced by ICS flow () as well as junctional phospho-PAK staining (not depicted). Taken together, these data show that matrix-specific PAK activation triggered by onset of flow or prolonged disturbed flow mediates enhanced endothelial monolayer permeability. Multiple growth factors and other bioactive substances use a pathway in which PAK regulates phosphorylation of MLCK to increase cellular contractility, thereby inducing endothelial cell permeability through formation of paracellular pores (). To test whether flow induces PAK-dependent paracellular pores, BAE cells were treated with either the control or PAK-Nck inhibitory peptide, sheared for 30 min, and assayed for the presence of paracellular pores by staining for the adherens junction protein β-catenin. Flow induced the formation of paracellular pores, which was strongly reduced by the pretreatment with the PAK-Nck inhibitory peptide (). Although flow patterns regulate susceptibility to atherosclerosis, a number of soluble factors also promote atherosclerotic plaque development and likely contribute to endothelial permeability in atherosclerosis. OxLDL stimulates endothelial cell permeability through a Rho-dependent pathway (; ). In early atherogenesis, activated endothelial cells and macrophages produce MCP-1, which also stimulates endothelial permeability (), as do the macrophage-derived cytokines TNFα and IL-1β (; ). Furthermore, mice deficient in either MCP-1 or TNFα show reduced atherosclerosis (; ). We previously showed that TNFα-induced endothelial permeability was reduced by the PAK-Nck inhibitory peptide (). To analyze the matrix dependence of these factors, PAK phosphorylation was assessed in endothelial cells plated on FN or MG. Though the time courses were distinct, MCP-1, TNFα, and oxLDL stimulated PAK phosphorylation in cells on FN but not on MG (). In all cases, phosphorylated PAK localized to cell–cell junctions, and this localization was inhibited by the Pak-Nck peptide (Fig. S3, available at ). We next examined monolayer permeability. All of these factors triggered matrix-dependent increases in permeability () that were inhibited by the PAK-Nck blocking peptide () and by expression of the PAK AID (). Thus, effects of a number of atherogenic soluble factors on PAK- dependent permeability are strongly modulated by the ECM. Areas of disturbed flow in vivo show elevated endothelial cell permeability (; ). These regions also show deposition of FN in the subendothelial ECM and expression of intercellular adhesion molecule 1 (ICAM-1) and vascular cell adhesion molecule (VCAM-1; ). We therefore tested whether permeability, PAK activity, and FN correlate in vivo. The carotid arteries from young (20-wk-old) ApoE mice fed either a chow or Western diet were isolated and processed for immunohistochemistry. In mice on a chow diet, the carotid sinus displayed some monocyte infiltration but no foam cell formation. PAK phosphorylation was observed specifically in the atherosclerosis-prone region of these vessels but not nearby athero-resistant regions (). Nearby sections showed FN in the subendothelial matrix in the same regions of the artery (). Furthermore, enhanced expression of VCAM-1 was detected in these regions, indicating endothelial activation. The opposite side of the carotid sinus can develop atherosclerosis in some cases but in this mouse shows no FN, VCAM-1, or phospho-PAK staining (). PAK phosphorylation, VCAM-1, and FN were all enhanced by the Western diet within the carotid sinus () but not in athero-resistant regions of the CCA (). Monocytes recruited to athero-prone regions of arteries from these mice also stained positively for phospho-PAK. Thus, PAK activation correlates with subendothelial FN and inflammatory markers. To determine whether active PAK localizes to cell–cell junctions in atherosclerosis in vivo, aortas from ApoE mice on a chow diet were fixed, excised, and examined en face. Staining for platelet-endothelial cell adhesion molecule 1 (PECAM-1) confirmed the ability to visualize endothelial cell junctions in vivo and illustrated endothelial cell alignment in an athero-protected region of the ascending aortic arch (unpublished data). Although athero-resistant regions of the aorta showed no staining, athero-prone regions of the lesser curvature of the arch showed focal areas of high phospho-PAK staining at cell–cell junctions (). To determine whether PAK is responsible for the increased permeability during development of atherosclerosis, 32-wk-old ApoE mice (chow diet) were given intraperitoneal injections of the PAK-Nck blocking peptide or a control peptide. Mice under these conditions are reported to develop moderate atherosclerotic lesions, though plaque development is slower than in animals on a high-fat Western diet (). Vascular permeability within the aorta was then assessed by measuring leakage of Evans blue dye into the vascular wall. Aortas from C57Bl/6 mice were used as a source for healthy, atherosclerosis-free vessels. Each mouse received 1 mg of peptide at 24 h and 1 h before Evans blue injection via the tail vein. After 30 min, leakage of dye into the aorta was assessed. Although little Evans blue accumulated in the aorta of C57Bl/6 mice, in ApoE mice, dye was apparent at the lesser curvature of the arch and at branch points for major arteries in both the nontreated and control peptide–treated animals (), consistent with known athero-prone regions. The Pak-Nck peptide inhibited 67% of the increase in permeability, relative to healthy vessels. These data suggest that PAK makes an important contribution to permeability in atherogenesis. These data support the concept that remodeling of the subendothelial ECM plays a crucial role in atherogenesis. Previous work demonstrated a correlation between enhanced vascular permeability and atherosclerosis (). In this work, we present evidence for ECM-specific activation of PAK by atherogenic stimuli, leading to increased permeability. PAK activation may be initiated by disturbed flow, though as atherosclerosis develops, soluble factors such as oxLDL and cytokines produced by immune cells and activated endothelium most likely make major contributions. Importantly, PAK activation at athero-prone sites in vivo correlates with areas of FN deposition. Finally, inhibiting PAK function in vivo reduced permeability in athero-prone regions. The mechanisms regulating the matrix specificity of PAK activation are presently unclear. Flow-induced Rac activation is equivalent on all matrices (unpublished data), suggesting that there may be matrix-specific signals that inhibit PAK activation. Known mechanisms limiting PAK activation include binding of PAK to Nischarin or hPIP1 and dephosphorylation by the phosphatases PP2A and POPX1/2 (; ; ; ). Phosphorylation of PAK by protein kinase A also inhibits PAK activation (). Further examination of matrix-specific PAK activation will be an interesting avenue for future work. The current data suggest that reducing either PAK activation or localization to cell–cell junctions should reduce the permeability of the endothelial cell layer. Recently, the Ser/Thr kinases Akt and protein kinase G (PKG) were found to phosphorylate PAK at Ser21 within the Nck-binding sequence, inhibiting the interaction between PAK and Nck (; ). Because blocking the PAK–Nck interaction inhibits localization of PAK to cell–cell borders and decreases endothelial permeability, these kinases might decrease permeability in a similar manner. Indeed, both Akt and cyclic GMP/PKG can decrease vascular permeability (; ; ). Whether PAK is the relevant target for these effects remains to be explored. The mechanisms by which permeability is elevated in the plaque endothelium are not well understood. Dissolution of intercellular interactions during endothelial cell division and apoptosis, both of which are elevated at athero-prone sites in vivo (; ), has been suggested as a possible mechanism. However, the correlation between endothelial cell turnover and enhanced permeability in vivo is weak (; ). A more likely mechanism involves TJs in athero-prone regions, which are discontinuous compared with athero-resistant regions (). Changes in TJ protein expression, phosphorylation, and reorganization could all contribute to decreased barrier function (). Both flow and cytokines induce permeability too rapidly for changes in gene expression to be an attractive mechanism. Shear stress stimulates occludin phosphorylation on Ser/Thr residues, which could alter occludin localization to TJs or function (; ). VEGF stimulates PAK-dependent VE-cadherin phosphorylation, resulting in its arrestin- dependent internalization and the formation of paracellular pores (). Myosin light chain phosphorylation triggers cell contraction and the formation of paracellular pores (), and contractility appears to be a common pathway for endothelial cell permeability by multiple atherogenic stimuli (; ; ; ; ). PAK inhibition decreases myosin phosphorylation and contractility in endothelial cells (; ). Thus, effects of PAK on the cytoskeleton appear to be involved in regulation of permeability, though other events, such as VE-cadherin and occludin phosphorylation, are likely to contribute. PAK regulates cytoskeletal organization, proliferation, and movement in many cell types, making PAK activity by itself an unlikely target for long term therapy. For example, PAK inhibition in mice with a cell-permeable peptide was recently shown to mimic Alzheimer's disease (). However, specific interactions, such as Nck, may offer more attractive therapeutic targets. The ECM dependence of PAK activity may provide an especially attractive means for therapeutic intervention that would be less perturbing than global inhibition of kinase activity. BAE cells (a gift from H. Sage, Hope Heart Institute, Seattle, WA) were maintained in low-glucose DME containing 10% calf serum (CS), 10 U/ml penicillin, and 10 μg/ml streptomycin (Invitrogen). Cells were plated for 4–24 h on 38- × 75-mm glass slides (Corning) precoated with collagen IV (20 μg/ml in PBS; Sigma-Aldrich), MG (1:100 dilution in serum-free media; Calbiochem), or FN (10 μg/ml in PBS). After 4 h, cells were fully attached and spread and formed a confluent monolayer. Slides were then loaded onto a parallel plate flow chamber in 0.5% CS, and 12 dynes/cm shear stress was applied for varying times as previously described (). To stimulate BAE cells with athero-prone (ICS) or athero-protective (CCA) shear stress profiles, BAE cells were plated as described except in a custom Petri dish and stimulated as previously described (). Human hemodynamic shear stress profiles were developed from MRI-generated near-wall velocity profiles of normal carotid arteries (). Transient transfection of HA-tagged PAK AID was accomplished by Effectene (QIAGEN) using the manufacturer's protocols. Cell lysis and immunoblotting were performed as previously described (). Antibodies used include rabbit anti–phospho-PAK (Ser141; 1:5,000; Biosource International) and rabbit anti-PAK (1:1,000; Cell Signaling Technologies). For immunocytochemistry, cells were fixed with PBS containing 2% formaldehyde, permeabilized with 0.2% Triton X-100, and blocked for 1 h in PBS containing 1% BSA and 10% goat serum. Primary antibodies were incubated with cells in blocking buffer as follows: rabbit anti–phospho-PAK (Ser141; 1:500 overnight), rabbit anti–β-catenin (1:200 overnight; Santa Cruz Biotechnology, Inc.), and mouse anti–ZO-1 (1:500 overnight). Cells were then incubated in 1 μg/ml Alexa 488– conjugated goat anti–rabbit IgG or goat anti–mouse IgG (Invitrogen). Slides were mounted with Fluoromount G, and images were taken using the 60× oil-immersion objective on a microscope (DiaPhot; Nikon) equipped with a video camera (CoolSnap; Photometrics) using the Inovision ISEE software program. A novel transwell well-flow device was developed to assay macromolecule permeability across an intact endothelial monolayer using previously established methods (). In brief, a previously developed cone-and-plate flow device was adapted to accept a 75-mm chamber transwell insert (). Custom flanges mounted on the lip of the Petri dish hold inlet and outlet tubing for the top and lower chambers, respectively, to inject and remove HRP without interrupting flow. Transwell chambers (3.0-μm pore size; Costar) were coated with either MG or FN, and BAE cells were allowed to attach for 4–24 h. Some transwells were coated with a fixed concentration of MG followed by increasing concentrations of FN. For flow experiments, cells on 75-mm chambers were serum deprived for 4 h in phenol red-free DME containing 0.5% CS and 2% dextran (wt/vol) and loaded onto the flow device stage, and shear stress was applied using the modified cone-and-plate device. At desired times, the medium was replaced with fresh medium containing 60 μg/ml HRP (Sigma-Aldrich) or Alexa 488–conjugated BSA (Invitrogen). After 1 h, medium was removed from the lower chamber, and cells were fixed in 2% formaldehyde and stained with Coomassie blue to detect cell loss or examined by immunocytochemistry for Ser141 phosphorylated PAK. For cytokine and LDL-induced permeability assays, cells grown on 6.5-mm filters were serum deprived for 4 h in phenol red–free DME containing 0.5% CS and transferred to fresh medium containing soluble factors for 90 min. HRP was then added to the top well to give a final concentration of 60 μg/ml. After 30 min, medium from the bottom well was removed, incubated with 0.5 mM guaiacol, 50 mM NaHPO, and 0.6 mM HO, and formation of -phenylenediamine was determined by measure of absorbance at 470 nm. Alexa 488–conjugated BSA was measured using a spectrofluorometer (FluoroLog; Jobin Yvon). Results are shown as a fold increase in HRP activity or in absolute solute permeability. Solute permeability coefficients for the endothelial monolayer were calculated as P = ΔCV/ΔCΔtS, where ΔC is the final concentration in the lower well, V is the volume of the bottom well (ml), ΔC is the concentration in the top well, Δt is the sampling interval (s), and S is the surface area of the transwell (cm; ). Nine male ApoE-deficient mice on a C57Bl/6 background from The Jackson Laboratory, 8–12 wk of age and weighing 18–20 g, were used in these experiments. Four mice were fed a Western-type atherogenic diet (TD 88137 [Harlan-Teklad]; containing 21% fat by weight, 0.15% by weight cholesterol, and 19.5% by weight casein without sodium cholate) for 10 wk before sacrifice. Control mice were fed a chow diet during this time. At 20 wk of age (10 wk on diet), mice were perfused with 4% paraformaldehyde, and the aortic arch, left carotid sinus, and right carotid sinus were processed for paraffin embedding. For Evans blue assays, six male C57Bl/6 and nine male ApoE-deficient mice (The Jackson Laboratory) were maintained on chow diets for 8 or 32 wk, respectively. 5-μm paraffin sections were obtained for immunohistochemistry. Immunohistochemistry for adhesion molecules VCAM-1 (Santa Cruz Biotechnology, Inc.) was performed as previously described (). After microwave antigen retrieval with antigen unmasking solution (Vector Laboratories), rabbit anti-FN (1:400; Sigma-Aldrich) and rabbit anti-Ser141 phosphorylated PAK (1:250) were applied. Detection of antibodies was with Vetastain Elite kit (Vector Laboratories). Visualization was with diaminobenzidine (DakoCytomation). For en face staining, the aortic arch was cut into rings and stained for either PECAM-1 or Ser141 phosphorylated PAK using Alexa 488–conjugated goat anti–rabbit secondary antibodies to detect localization. Rings were then cut, opened, and mounted between two coverslips for en face viewing by fluorescence microscopy. Images were acquired using the 10× or 40× objective on a microscope (BX51; Olympus) equipped with a digital camera (DP70; Olympus) using ImagePro Plus software (Media Cybernetics). Mice were injected intraperitoneally with 0.1 ml of either control peptide or the PAK-Nck inhibitory peptide (10 mg/ml) at 24 h and at 1 h before Evans blue injection. Evans blue (0.1 ml of 1% dye in PBS) was injected into the tail vein. After 30 min, mice were killed with ketamine/xylazine and perfused through the left ventricle with 10 ml of 4% formaldehyde in PBS, and the aorta was excised from the cusp to the renal artery branches. Bright field microscopy of excised aortas was performed using the 0.5 and 1.2× objectives on a microscope (SZX12; Olympus) equipped with a DP70 digital camera using ImagePro Plus. Aortas were dried and weighed, Evans blue was extracted by incubation in formamide for 24 h at 60°C, and absorbance at 620 nm was determined. Concentration curves for pure Evans blue were used to calculate the total amount of dye extracted, and this value was normalized to the weight of the isolated aortas. Independent of matrix composition, the 4-h plating time is sufficient to allow both adherens and TJ formation, as assessed by staining cells for β-catenin and ZO-1, respectively (Fig. S1). Matrix-specific effects on monolayer permeability are not due to differences in matrix permeability, which shows no difference between MG and FN (Fig. S2). Localization to cell–cell junctions is required for PAK-dependent permeability (), and TNFα, MCP-1, and oxLDL all stimulate active PAK localization to cell–cell junctions, which was abrogated by the addition of the PAK-Nck inhibitory peptide (Fig. S3). Online supplemental material is available at .

Meiosis is the process of producing haploid gametes from diploid germ cells by one round of DNA replication followed by two rounds of cell division. Recombination takes place in the prophase of the first of the two cell divisions and is dependent on the formation of double strand breaks (DSBs) by SPO11 (; ). After homology search and strand invasion, the homologous chromosomes pair and, in many organisms, form the synaptonemal complex (SC). This is a structure with two axes, each joining together the pair of sister chromatids of each homologue (for review see ). These axes contain several specific proteins in addition to the cohesins that are generally responsible for the maintenance of chromosome structure (). In mammals, this includes the coiled-coil domain proteins SYCP2 and -3. In mice with a deletion of the gene, which removes the part of the protein interacting with SYCP3, assembly of SYCP3 into the axes is impaired. Males suffer a block in meiosis that leads to apoptosis and infertility, and females have a severely reduced litter size (). Likewise, mice engineered to be null for SYCP3 exhibit a dimorphic phenotype. Although males are infertile, females show a reduced litter size caused by the death of aneuploid embryos produced from aneuploid oocytes (). The repair of DSBs and recombination is also affected in these mutants (). Transverse filaments extend from and meet between axes in a structure called the central element (CE). This ultrastructural feature is common to many organisms and, until recently, was thought to consist entirely of the overlap between the N termini of SYCP1 molecules originating from the paired axes (). In agreement with this, overexpression of SYCP1 (or ZIP1 in yeast) produces structures termed polycomplexes, which have dimensions similar to SCs (; ). Targeted mutation of the gene in mouse results in infertility in both sexes, leading to the failure of synapsis and the absence of completed crossover in males (). The female cytology in this mutant has not yet been described. We have recently defined two proteins, SYCE1 and -2 (previously known as CESC1), with localization confined to the CE of mouse SCs (). Homologous genes exist in other vertebrate genomes, and structural homologues may exist more widely. The role of these proteins is suggested by their location and biochemical interactions. They colocalize with SYCP1 to synapsed axes at the light microscope level and are confined to the CE at electron microscope resolution. They are both capable of interacting with themselves, with each other, and with the N terminus of SYCP1. We have postulated that they provide reinforcement to the N-terminal SYCP1 interactions. A third CE protein, TEX12, which interacts with SYCE2, was recently described (; ). To test the dependence of SC formation on SYCE1 and -2, we are generating mice that lack these proteins. In this paper, we report mice derived from an embryonic stem cell line in which the gene was disrupted by insertion of a gene trap vector (). We have analyzed these mice immunocytochemically to look not only at the structural effect of the mutation but also at its effect on DSB processing and crossing over. Searches of the Sanger Center gene trap database () revealed a gene trap line, S8-7E, in which the gene was disrupted by insertion of a ROSAFARY vector into the locus (; ; Fig. S1, available at ). , were fertile and were intercrossed to give homozygous () animals. The gene-trapped allele was transmitted in Mendelian ratio to offspring. animals of both sexes were infertile when crossed to wild-type animals but showed no other overt phenotype. Ovaries of adult females were minute, and testes weights of adult males averaged 20% of wild-type littermates. Adult testis sections showed a stage IV arrest, with all spermatocytes undergoing apoptosis (Fig. S2). From the site of insertion and the nature of the gene trap vector, we predicted that the allele was likely to represent a null mutation, as only 10 amino acids remain from the original protein (Fig. S1). To confirm that aberrant splicing events had not rescued the expression of the trapped allele, we used Northern blots to check the level of wild-type RNA in testes. Normal mRNA was not detectable using this technique, nor by RT-PCR (Fig. S1). Accordingly, we could not detect the SYCE2 protein in testis cell spreads (unpublished data). spermatocytes were analyzed using spread preparations. These were initially stained for axial element (AE) components, showing that mutant spermatocytes have AEs of normal morphology and composition at light microscope level. Cohesins SMC3 and the meiosis-specific REC8 and STAG3 are all present, together with SC proteins SYCP2 (not depicted) and SYCP3 (Fig. S2 A), in both male and female animals. In adult males, the AE appear to align homologously, at least in the majority of the cells, but a minority show alignment of only some chromosome pairs or no alignment at all (Fig. S2 B). It is possible that the last two classes of cells are entering apoptosis. In females (embryonic days 16.5–18.5), a lower percentage of AEs are in close alignment, suggesting that this stage may be of shorter duration in females than in males. In wild-type animals, entry into the zygotene stage is characterized by initiation of synapsis between homologous chromosomes. This can be visualized by staining with anti-SYCP1, -SYCE1, or -SYCE2 antibodies (; ). In the males, however, synapsis fails to develop between homologues, except for some small regions, varying in number and extent, of closer association that stain for both SYCE1 and SYCP1 (, A2, A3, B2, and B3). In these regions, SYCE1 largely colocalizes with SYCP1 (). SYCP1 was not associated with male AEs, suggesting that SYCE2 is necessary for the C terminus of SYCP1 to bind the axes. females, when immunostained with SYCP1/SYCE1 antibodies, showed no signs of synapsis, but SYCP1 and SYCE1 were detected on dispersed univalents as bright foci coating the AE even in the absence of synapsis (). These sites were not always coincident. This is not observed when homologous chromosomes are aligned. To test whether the regions of synapsis in males were as short as was observed by immunocytochemistry, or if this was an artifact of the spreading technique, we performed electron microscopy on fixed and sectioned material from adult testes (). Again, we found multiple unpaired AEs that were, however, thicker and showed a less regular surface than wild-type AEs/lateral elements (, E2). Occasional regions of synapsis were also found (, E3). Traces of a CE were present where this occurred. The overall width of these regions of synapsis was similar to that of a wild-type SC, although AEs are thicker and the central region thinner than wild type. This data is consistent with the immunocytochemistry, suggesting that the short regions of synapsis are not an artifact of the spreading technique. We propose that synapsis is initiated but is not, or is only minimally, extended. The dimensions of the polycomplex formed when SYCP1 is overexpressed in mitotic cells, in the absence of other meiotic chromosome components, are very similar to those of the SC (; ). SYCP1 has C- and N-terminal globular domains separated by a coiled-coil region. When the length of this coiled-coil region is varied experimentally, the spacing of the arrays in polycomplexes generated is increased or reduced accordingly (). This supports the concept that the SYCP1 molecule determines the spacing of the axes. Our new data shows that in vivo the assembly and/or stability of the system are dependent on more than SYCP1 alone. In the absence of SYCP1, both SYCE1 and -2 are delocalized from the axes of the chromosomes, and at the biochemical level, interactions occur between the N terminus of SYCP1 and SYCE1/2 (). In addition, SYCE1 and -2 can interact with themselves and with each other (). We suggested a model in which the CE proteins provided a structural role, perhaps associated with the postulated need to resist compression forces as a mechanism to produce interference (). Here, we revise this model. Based on the observation that short points of synapsis are detectable in the male as sites of colocalization of SYCP1 and SYCE1, we suggest that synapsis can initiate in the absence of SYCE2 but cannot propagate along the AE. One testable prediction is that, in the absence of both SYCE1 and -2, these short stretches of synapsis will not occur. We have not seen regions of synapsis in female meiosis in the animals. This may reflect a real difference in mechanism, but as prophase1 of female meiosis takes place in a compressed time scale compared with male, it is possible that we have not detected limited synapsis because it is more transient (). The known interactions between these proteins suggest a process of polymerization that would result in the self-assembly of the SC. Dimers of SYCP1 form head-to-head associations via their N termini to set the basic spacing between the lateral elements. This association alone would not be stable, but the interaction with SYCE1, probably in a multimeric form, could cause its stability to increase. The short regions of synapsis we observe would represent such sites of stable association. Extension of this would require the association of a dimer or tetramer of SYCE2 with the SYCE1–SYCP1 complex through a SYCE1–SYCE2 interaction. SYCE2 would then interact with an SYCP1 dimer and, through a repetition of the process, polymerize the SC. This model is represented in . animals should be phenotypically similar to animals in having stabilized points of axial contact. The sites of limited synapsis we observe in the males could represent a mammalian equivalent of the synaptic initiation complex in yeast (; ). This seems unlikely for two reasons. First, we do not see a preferential association of recombination proteins such as MSH4 with these sites of SYCP1 and SYCE1 localization. Second, the distribution does not match that of recombination events, with many chromosomes lacking these sites of synapsis and some chromosomes having multiple sites. At the leptotene stage of meiosis, DSBs are formed by the topoisomerase-like protein SPO11 and are then resected, leading to invasion of the homologous chromosome. Damage to the genome, including DSBs, is marked by the presence of the phosphorylated form of histone H2AX (γH2AX; ). The phosphorylation of H2AX is mediated by the kinase ATR, which in turn is recruited by BRCA1 (). γH2AX first appears during premeiotic S-phase, but it is most abundant in leptotene and early zygotene spermatocytes, before synapsis initiation (). As synapsis progresses, γH2AX-positive domains decrease and, by late zygotene through to pachytene, only the unpaired sex chromatin shows positive staining. When SYCE2 is absent, γH2AX shows a different dynamic. Although in early stages no difference between wild type and mutant is visible, later stages show only a moderate decrease in γH2AX (). Pachytene-like spermatocytes that show alignment of all the chromosomal complement display a patchy distribution of γH2AX over the aligned AEs. The distribution of γH2AX in mutant females is subtly different from that in males. In females, some cells with unaligned chromosomes display a very close association of the γH2AX-positive domains with the axes (). Analysis of an earlier step of the pathway, namely, BRCA1 distribution, revealed a slightly different picture. In wild-type spermatocytes, BRCA1 staining is first observed in leptotene spermatocytes as a punctate signal on the forming AEs (). By pachytene, the staining becomes continuous, covering the asynapsed axes of the sex chromosomes and rare autosomes that did not synapse (). In the male mutant, however, BRCA1 punctate staining remains strongly associated with the chromosome axes (). Only in cells with little or no chromosome alignment could we see decreased or absent BRCA1 staining (). In female mutant cells, BRCA1 staining shows the same distribution when chromosomes are aligned, but it seems to cover more contiguous regions of the AEs when the alignment is lost. This suggests that DSBs are being formed but do not appear to be processed efficiently, if at all. To achieve a better understanding of the extent of DSB processing, we studied the distribution of components of meiotic recombination nodules. These included RAD51 and DMC1, two recombinases that are involved in the formation of the nucleoprotein filament and strand invasion; RPA, a single-strand DNA binding protein that localizes to DSBs soon after RAD51; and MSH4, a component of recombination nodules that are found after synapsis is established (for review see ). In contrast to our observations on γH2AX and BRCA1, mutant male and female meiosis appear to be very similar with respect to the recombination proteins studied. Cells that have RAD51, DMC1, RPA, or MSH4 have closely aligned chromosomes ( and not depicted). Although RAD51 and DMC1 seem to disappear from these cells, RPA and MSH4 are unable to follow on the natural progression of meiosis and remain between the AEs until the alignment is lost. This suggests that the recombination process is being halted at some point after RAD51/DMC1 removal and MSH4 loading on to the AE. Also, late recombination foci components whose distribution pattern closely resembles that of chiasmata, MLH1 and -3, were not detected (). Either DSBs are being processed up to the loading of MSH4 but not later or cells do not survive beyond this point. As the X and Y are not true homologues, homology is only found along the distal pseudoautosomal region (). They are the last pair of chromosomes to align and synapse in late zygotene and the first to desynapse from early pachytene. In mutant spermatocytes, contrary to autosomes that show a high degree of alignment, the X and Y chromosomes are only found aligned in ∼10% of cells. Like the autosomes, AE composition is normal, except for REC8. REC8 is normally present in reduced levels in the asynapsed regions of the gonosomes (), but in mutant spermatocytes the levels of REC8 in the X and Y are comparable to those found in autosomes (). BRCA1 also has an unexpected distribution. When staining is present on the Y, it is limited to a single focus on the tip of the chromosome, which we confirmed to be the pseudoautosomal region (). Similarly, γH2AX shows patchy staining over the autosomes and the X chromosome but is absent from the Y or only present in a small distal region of that chromosome (). In summary, mice can make DSBs, align homologous chromosomes normally, and initiate recombination processes, but cannot complete SC formation and, in the males, do not form an XY body. A detailed study of the male phenotype of a null mutation in the transverse filament protein SYCP1 () describes effects of this mutation that are very similar to those seen in male mice in terms of the localization of recombination and repair proteins. The differences we do see could be due to differences in genetic background or technique, rather than basic biology of the system. The marking of chromosome axes by SYCP1 in zygotene/pachytene cells has been regarded as a hallmark of synapsis (). In males, SYCP1 is not associated with the AE except at the short regions of synapsis, even though this protein is present in the cell. This is the most striking difference we see between the mice and the knockout animals, as it leads to our model for SC assembly. The nature of the interactions that in wild-type animals result in the C termini of SYCP1 molecules localizing to the AE is not known, although based on protein motifs, this region of the protein has been suggested to have DNA binding activity. Importantly, whatever the molecular basis of these interactions, our observations show that they critically depend on protein–protein interactions at the opposite end of the SYCP1 molecule, which involve SYCE1. The otherwise high degree of similarity is unsurprising, given that both proteins are essential for SC formation. The mutual dependence means that we cannot say whether delocalization of SYCP1 or lack of SYCE2 in the mice causes the incomplete DSB repair, lack of crossing over, and failure of XY body formation. mice, delocalization of SYCE2 could be responsible for the phenotype. Both proteins, and probably others such as TEX12, are needed to form the functional SC necessary to complete recombination and meiosis. Embryonic stem cell line S8-7E was purchased from the laboratory of P. Soriano (Fred Hutchinson Cancer Research Center, Seattle, WA) as a sequence-verified clone and injected into F1 C57BL6/CBA blastocysts using standard methods. Chimeric males were mated to C57BL6 females and progeny genotyped by PCR (Fig. S1). Animals were intercrossed to generate homozygous mice. Timed mating was used to generate embryonic material, with the plug date set to 0.5 d postcoitum. Spread chromosomes from males and females were prepared and stained and previously described (, ). Images were captured using a system comprising a charge-coupled device camera (Orca-AG; Hamamatsu), a fluorescence microscope (Axioplan II; Carl Zeiss MicroImaging, Inc.) with Plan-neofluar objectives (100× NA 1.3), a 100-W Hg source (Carl Zeiss MicroImaging, Inc.), and quadruple band-pass filter set (model 86000; Chroma Technology Corp.), with the single excitation and emission filters installed in motorized filter wheels (Prior Scientific Instruments). Image capture and analysis were performed using in-house scripts written for IPLab Spectrum (Scanalytics). Images were imported into Photoshop (Adobe), and the curves of individual channels were adjusted for reproduction. Electron microscopy was performed using ultra thin sections of testis tissue fixed in 2.5% glutaraldehyde and 1% OsO as described previously (). In testis sections, the stages of the cycle of the seminiferous epithelium were distinguished as described by . In the absence of spermatids in the mice, epithelial stage IV was identified by the presence of intermediate spermatogonia in late phases of the cell cycle or in mitosis and very early B spermatogonia (). Images were captured as described in the previous section. Antibodies used were directed against SYCE1 and -2, SMC3 (), and STAG3 (). REC8 (), SYCP1 (rabbit and guinea pig; ), SYCP2 (), and SYCP3 antibodies were as described previously (; ; ), and ab12452 was obtained from Abcam. Antibodies directed against DNA damage and recombination proteins were γH2AX (Upstate Biotechnology), BRCA1 (), RAD51 (Abcam), DMC1 (), MSH4 (), MLH1 (BD Biosciences), and MLH3 (). Antibodies were provided by M.A. Handel (The Jackson Laboratory, Bar Harbor, ME), R. Jessberger (Technische Universität Dresden, Dresden, Germany), C. Heyting (Wageningen Agricultural University, Wageningen, Netherlands), C. Hoog (Karolinska Institutet, Stockholm, Sweden), P. Cohen (Cornell University, Ithaca, NY), and P. Moens (York University, Toronto, Canada). Fig. S1 shows details of gene trap characterization and the histological effect of the knockout in testis. Fig. S2 shows the distribution of cohesins in wild-type and knockout male and female meiotic chromosomes. Online supplemental material is available at .