Functional interplay between LIS1, NDE1 and NDEL1 in dynein-dependent organelle positioning

LIS1, NDE1 and NDEL1 modulate cytoplasmic dynein function in several cellular contexts. However, evidence that they regulate dynein-dependent organelle positioning is limited. Here, we show that depletion of NDE1 or NDEL1 alone profoundly affected the organisation of the Golgi complex but did not cause it to disperse, and slightly affected the position of endocytic compartments. However, striking dispersal of organelles was observed when both NDE1 and NDEL1 were depleted. A substantial portion of NDE1 and NDEL1 is membrane associated, and depletion of these proteins led to complete loss of dynein from membranes. Knockdown of LIS1 also caused the Golgi complex to fragment and disperse throughout the cell, and caused endocytic compartments to relocalise to the periphery. Depletion of LIS1, which is primarily cytosolic, led to partial loss of membrane-associated dynein, without affecting NDE1 and NDEL1. These data suggest that NDE1 and NDEL1 act upstream of LIS1 in dynein recruitment, and/or activation, on the membrane. Consistent with this hypothesis, expression of exogenous NDE1 or NDEL1 rescued the effects of LIS1 depletion on Golgi organisation, whereas LIS1 was only partially effective at rescuing the loss of NDE1 and NDEL1.

In contrast to these events, the involvement of LIS1 in the dyneindependent movement and positioning of cytoplasmic organelles is controversial. In several studies, interventions that inhibited other LIS1-dependent activities did not affect membrane organisation (Dujardin et al., 2003;Faulkner et al., 2000;Tai et al., 2002). Elsewhere, mild (Sasaki et al., 2000;Smith et al., 2000) or more severe (Ding et al., 2009;Kondratova et al., 2005;Rehberg et al., 2005) effects on Golgi positioning were seen upon LIS1 reduction, LIS1, NDE1 and NDEL1 in membrane positioning but these might have been secondary to the profound changes in microtubule architecture that were also reported. Drosophila LIS1 -/neurons display defects in axonal transport, although the phenotype differs somewhat from DHC mutants (Liu et al., 2000). Overexpression of LIS1 has been reported to leave the Golgi complex position unchanged (Faulkner et al., 2000), more compact  or dispersed (Kondratova et al., 2005). There is stronger evidence for an involvement of NDEL1 in membrane positioning, because overexpression of LIS1-or dynein-binding NDEL1 mutants alters the position and centripetal movement of a range of organelles and membrane cargo (Liang et al., 2004). Reduction in NDEL1 expression affects the positioning of the Golgi complex (Liang et al., 2004;Sasaki et al., 2005) and lysosomes (Guo et al., 2006), but it is possible that changes in cytoskeletal organisation also contribute to these phenotypes . In addition, the relationship between NDE1 and NDEL1 in controlling organelle positioning has not been explored. This paper uses a systematic, RNAi-based approach to identify the contributions of LIS1, NDE1 and NDEL1 to membrane positioning unambiguously and to explore the functional relationships between each of them.

LIS1 is essential for maintaining organelle positioning
The cellular positioning of the Golgi complex and endosomal compartments are dependent on dynein. However, the requirement for LIS1 in these functions is controversial, despite its involvement in other dynein-dependent cellular processes. We found that simple overexpression of GFP-LIS1 in HeLaM cells did not cause Golgi complex dispersal (a typical phenotype seen following disruption of dynein activity), in agreement with some studies (Dujardin et al., 2003;Faulkner et al., 2000;Tai et al., 2002), but in contrast to other work (Kondratova et al., 2005;Sasaki et al., 2000). Instead, the Golgi became somewhat more compact compared with control cells (Fig.  1A; see supplementary material Table S1 for statistical analysis), as reported before . An alternative way of looking at LIS1 function is to deplete the endogenous protein. LIS1 expression was prevented using a previously characterised siRNA oligonucleotide (Tsai et al., 2005), leading to almost complete removal of LIS1 protein (Fig. 1B). The time taken for LIS1-depleted cells to progress through mitosis increased to 260% compared with that in control cells (P<0.0001; Table 1). This is consistent with the delay in mitosis caused by microinjection of anti-LIS1 antibody and the twofold increase in mitotic index observed upon depletion of LIS1 (Faulkner et al., 2000). In addition, many LIS1-depleted interphase cells displayed micronuclei (supplementary material Fig.  S1A), indicating a high incidence of aberrant mitosis (Table 2), as observed previously (Faulkner et al., 2000). These data demonstrate a functional depletion of LIS1 protein.   or untreated (right), were fixed, and immunostained for golgin84 (transfected cells are indicated by asterisks). Scale bar: 10m. (B)Cells were mock treated or depleted of lamin A/C or LIS1, lysed, and equal amounts of protein from each lysate were immunoblotted accordingly. (C)Cells depleted of lamin A/C or LIS1 were immunostained for golgin84, LAMP-1 or EEA1. Scale bar: 10m. (D)Cells treated with siRNA were scored for Golgi phenotypes as indicated (see supplementary material Fig. S2 for examples of each phenotype). Values are expressed as mean percentages ± s.e.m. from three independent experiments. 500 cells were counted for each condition per experiment. Statistical analysis is provided in supplementary material Table S2.
Upon depletion of LIS1, the positioning of the Golgi-tethering factor, golgin84 (Diao et al., 2003), was severely affected. In controltreated cells (data not shown) or cells treated with siRNA against an unrelated protein, lamin A/C, (Fig. 1C), the Golgi complex was localised to the perinuclear region and was organised into a series of interconnected structures. In LIS1-depleted cells, these were replaced by small fragments of golgin84-positive membrane that were dispersed throughout the cell (Fig. 1C). Localisation of another golgin, GM130 (Puthenveedu et al., 2006), was similarly affected upon loss of LIS1 (data not shown). In addition, Golgi positioning was similarly affected in cells depleted of LIS1 using an independent, 'SmartPool' combination of oligonucleotides (supplementary material Fig. S1A). To provide a more quantitative assessment of the degree of Golgi dispersal, cells were selected into one of four categories based on their distribution of golgin84 staining (see supplementary material Fig. S2 for details). Of cells treated with LIS1 siRNA, 70% showed a severe defect in Golgi positioning, with highly scattered small fragments. This phenotype was never observed in untreated cells or cells treated with siRNA against lamin A/C ( Fig. 1D; see supplementary material Table S2 for statistical analysis). In addition, 17% of LIS1-depleted cells exhibited an intermediate defect in Golgi positioning (partial scatter), in which golgin84 staining was highly fragmented, but remained localised fairly centrally within the cell. This phenotype was rarely seen in control cells.
The importance of LIS1 for organelle positioning is not confined to the Golgi complex. LAMP-1 is an integral membrane marker for lysosomes and late endosomes. LAMP-1 redistributed from a mainly perinuclear location in control-treated (data not shown) or lamin-A/C-depleted cells (Fig. 1C) towards the cell periphery when LIS1 was depleted (Fig. 1C). In lamin-A/C-depleted cells, the early endosome marker, EEA1, varied somewhat in positioning, but in most cells some EEA1 concentrated in larger structures towards the cell centre (Fig. 1C). EEA1 staining became spread more evenly throughout the cell, or became dispersed towards the cell periphery upon loss of LIS1 (Fig. 1C). The distribution of transferrin receptor Journal of Cell Science 123 (2) (TfR), a marker for the recycling endosome, was also affected (supplementary material Fig. S3). In control HeLaM cells, TfR was most concentrated in the perinuclear region, although it was also found at the cell periphery. The perinuclear concentration of TfR was lost when LIS1 was depleted, and instead TfR labelling was somewhat stronger at the cell periphery.
To identify whether these effects resulted specifically from loss of LIS1, siRNA-treated cells were transiently transfected with human GFP-LIS1 incorporating silent mutations making it resistant to RNAi (GFP-LIS1*). This substantially restored the positioning of golgin84, LAMP1 and EEA1 towards normal positions ( Fig. 2A). Indeed, quantitative analysis showed that transfection of GFP-LIS1* into both siRNA-depleted or mock-depleted cells often generated a more tightly focussed and compact Golgi complex than seen in untransfected cells (Fig. 2B, see supplementary material Table S2 for statistical analysis), as was seen when LIS1 was overexpressed in control cells (Fig. 1A, supplementary material Table S1). Therefore, the structure and position of the Golgi complex correlates with levels of LIS1.

Both NDE1 and NDEL1 are important for maintaining organelle positioning
Previous studies have shown that RNAi-mediated depletion of NDEL1 causes scattering of the Golgi complex, but the interpretation of these experiments was complicated by the fact that NDE1 levels were also reduced under these conditions (Liang et al., 2004). Neurons with reduced NDEL1 expression displayed some defect in the distribution of the COPI coatomer subunit COP, which is used as a Golgi complex marker , although in this case NDE1 expression was maintained at normal levels. To determine whether both NDEL1 and NDE1 are essential for organelle organisation and positioning, or whether they might be redundant in these respects, cells were treated with siRNA specific for each protein, either individually or in combination. Treatment with NDEL1 or NDE1 siRNA reduced the levels of the respective protein specifically (Fig. 3A). The duration of mitosis increased by 223% after depletion of both NDE1 and NDEL1 (P<0.0001; Table  1). It increased by 197% in cells depleted of NDE1 alone (P<0.0001; *siRNA-treated cells were fixed after 72 hours and stained with DAPI. The indicated number of cells was counted from two independent experiments, and phenotypes scored as described. Values are expressed as mean percentages of the total cell number. The level of apoptosis was not affected by any treatment.  Table 1) and did not increase after depletion of NDEL1 alone (Table  1). These data are consistent with the conclusions of previous studies showing that NDE1 has an important role in mitotic progression in HeLa cells (Vergnolle and Taylor, 2007) and that NDE1-knockout mice display a reduction in mitosis in the ventricular zone (Feng and Walsh, 2004), but further indicate that loss of both NDE1 and NDEL1 might have a greater effect on mitosis than loss of NDE1 alone. Values are expressed as mean percentages ± s.e.m. from four independent experiments. 100 cells were counted for each transfection per experiment. Statistical analysis is provided in supplementary material Table S2.
Loss of NDE1 affected the structure and positioning of the Golgi complex, with many cells displaying more fragmented golgin84 labelling. This consisted either of golgin84 in threads that had separated from each other, but remained close to the cell centre, or somewhat smaller fragments that were also mainly centrally located ( Fig. 3B,C). Dispersal of golgin84 labelling into small, dispersed fragments, as found after LIS1 depletion, was rarely observed. Transfection with a separate siRNA oligonucleotide against NDE1 generated a similar phenotype (supplementary material Fig. S1B). This second oligonucleotide generated more pronounced Golgi scattering when used at higher concentrations, but this was accompanied by extensive cell death and loss of ~50% dynein intermediate chain (data not shown). In cells depleted of NDEL1 alone, although the Golgi phenotype ranged from short threads to widely scattered fragments, ~80% of cells displayed an intermediate phenotype of partially scattered fragments (Fig. 3B,C). Strikingly, when both NDEL1 and NDE1 were depleted, the Golgi complex became scattered throughout the cytoplasm in ~70% of cells, with a severity of defect very similar to that observed after LIS1 depletion ( ). Another dynein regulator, ZW10, is also required to maintain the structure of the Golgi complex (Hirose et al., 2004;Sun et al., 2007). In cells depleted of ZW10, the Golgi complex fragmented into very short tubules that remained in a patch close to the cell centre (supplementary material Fig. S1B), a phenotype that clearly differed from those observed upon depletion of LIS1, NDE1 or NDEL1.
Depletion of NDEL1 and NDE1 also had a combinatorial effect on the positioning of lysosomes and early endosomes. LAMP1 distribution was somewhat affected by loss of NDE1, or by loss of NDEL1, as previously reported (Guo et al., 2006). However, LAMP1 became highly dispersed to the cell periphery when both NDE1 and NDEL1 were depleted (Fig. 3B). EEA1 localisation was slightly altered compared with control cells upon NDEL1 depletion, or upon NDE1 depletion. However, it became substantially dispersed when both NDEL1 and NDE1 were depleted (Fig. 3B). TfR distribution was also affected in these cells, and appeared similar to that in cells depleted of LIS1 (supplementary material Fig. S3).

The functional relationship between LIS1, NDE1 and NDEL1 in maintenance of organelle positioning
Our data suggest that both NDE1 and NDEL1 participate in dyneindependent positioning of organelles, although their roles in other aspects of cell physiology might be distinct. To further address whether their functions in supporting organelle positioning are equivalent, cells depleted of NDEL1, NDE1, or a combination of the two, were transfected with RNAi-resistant mutants of GFP-NDEL1 (GFP-NDEL1*) or GFP-NDE1 (GFP-NDE1*). Since high levels of NDEL1 or NDE1 overexpression lead to Golgi scattering (data not shown and Liang et al., 2004) and disruption of the microtubule network (Feng and Walsh, 2004), we first carefully titrated the levels of GFP-NDEL1* and GFP-NDE1* plasmids used in transfections to avoid these dominant-negative effects. The Golgipositioning phenotypes of cells depleted of NDEL1, NDE1, or NDEL1 plus NDE1 were effectively reversed when either GFP-NDEL1* or GFP-NDE1* were expressed ( Fig. 4; see supplementary material Table S2 for statistical analysis), suggesting that the functions of these proteins are redundant in this respect. No further rescue was observed upon combined transfection with GFP-NDEL1* and Myc-NDE1* (data not shown). Interestingly, expression of low levels of GFP-NDEL1* or GFP-NDE1* led to a more compact Golgi organisation in 40-50% of cells, compared with ~10% of mockdepleted cells ( Fig. 4C; see supplementary material Table S1 for statistical analysis). The positioning of early endosomes and lysosomes, which was altered in NDE1/NDEL1-depleted cells, was rescued by expressing GFP-NDEL1* or GFP-NDE1* (Fig. 5).
When cells depleted of LIS1 were transfected with either GFP-NDEL1* or GFP-NDE1*, this resulted in a significant reduction Journal of Cell Science 123 (2) in the severity of the Golgi-dispersal phenotype compared with cells transfected with a GFP control ( Fig. 6; see supplementary material Table S1 for statistical analysis). Broadly similar findings were seen for distribution of both LAMP-1 and EEA1 when GFP-NDEL1* was expressed (supplementary material Fig. S5). By contrast, the phenotype displayed by cells depleted of both NDE1 and NDEL1 was only slightly reduced by overexpression of GFP-LIS1*, with (A)Cells were mock treated, depleted (kd) of NDE1, NDEL1 or both NDE1 and NDEL1, lysed, and immunoblotted. Note that a band migrating just above the position of NDE1 is sometimes seen in NDE1-depleted lysates. (B)Cells depleted as indicated were immunostained for golgin84, LAMP-1 or EEA1. Scale bar: 10m. (C)Cells treated with siRNA were scored for Golgi phenotypes, as indicated. Values are expressed as mean percentages ± s.e.m. from three independent experiments. 500 cells were counted for each condition per experiment. Note that the values for mock-treated control cells are also shown in Fig. 1. LIS1, NDE1 and NDEL1 in membrane positioning a shift from full to partial scattering, but little reassembly of Golgi ribbons (Fig. 6).

Loss of LIS1, or NDE1 and NDEL1, impairs dynein recruitment to membranes
The phenotypes generated by LIS1, or NDE1 and NDEL1, depletion suggest that dynein activity during membrane positioning is impaired in these cells. To test whether RNAi reduced the stability of dynein and/or dynactin, cells were analysed by western blot for dynein intermediate chain (DIC) and p150 Glued , which are components of the dynein and dynactin complexes, respectively. The levels of both proteins were unaffected (Fig. 7A). In addition, all DIC remained part of a high molecular weight dynein complex in siRNA-depleted cells, as assessed by sedimentation on sucrose gradients (Fig. 7B). However, loss of LIS1 substantially reduced the amount of DIC that co-fractionated with membranes, whereas no membrane-associated DIC was detected at all following depletion of both NDE1 and NDEL1 (Fig. 7C). Interestingly, although dynactin and LIS1 were predominantly cytosolic in all conditions, both NDE1 and NDEL1 were plentiful in the membrane fraction, and their levels were not altered by LIS1 depletion (Fig. 7D). Hence, Fig. 5. NDE1 or NDEL1 restore endosome and lysosome positioning to NDE1-and NDEL1-depleted cells. Cells depleted of NDE1 and NDEL1 were transiently transfected with GFP-NDE1* or GFP-NDEL1* and immunostained for LAMP1 or EEA1 (transfected cells are indicated by asterisks). Scale bar: 10m. Fig. 4. NDE1 or NDEL1 restore Golgi positioning to NDE1-and NDEL1depleted cells. Cells depleted of one or both of NDE1 and NDEL1, or subjected to mock depletion, were transiently transfected with GFP-NDE1* (A) or GFP-NDEL1* (B) and immunostained for golgin84 (transfected cells are indicated by asterisks). Transfected cells are indicated by asterisks. Scale bars: 10m. (C)Cells depleted as indicated were transfected with GFP, GFP-NDE1* or GFP-NDEL1*, and stained for golgin84. Cells were scored for Golgi phenotypes as indicated in Fig. 2. Values are expressed as mean percentages ± s.e.m. from four independent experiments. 100 cells were counted for each transfection per experiment. Statistical analysis is provided in supplementary material Table S2. LIS1 and NDE1 plus NDEL1 are essential for efficient dynein binding to membranes, whereas LIS1 is not required for NDE1 or NDEL1 membrane association. Note that a minor loss of LIS1 might occur in cells depleted of NDE1 or NDEL1 (Fig. 7A).
The effects on organelle positioning caused by loss of LIS1, NDE1 and NDEL1 cannot be attributed to changes in cytoskeleton structure or dynamics LIS1 and NDEL1 help to stabilise cortical microtubules via a mechanism that involves CLIP170, dynactin and dynein (Coquelle et al., 2002;Li et al., 2005a;Sheeman et al., 2003;Yingling et al., 2008). In addition, interfering with LIS1 or NDE1 and NDEL1 function in some systems has been shown to alter microtubule distribution profoundly (Ding et al., 2009;Guo et al., 2006;Rehberg et al., 2005;Sasaki et al., 2005;Sasaki et al., 2000;Smith et al., 2000). To address whether gross changes in microtubule organisation could account for the redistribution of organelles observed upon depletion of LIS1 or NDE1 and NDEL1, siRNAdepleted cells were stained for microtubules and for golgin84, and scored for a range of interphase microtubule organisation phenotypes (supplementary material Fig. S6) and for Golgi positioning. As shown in supplementary material Table S3, depletion of LIS1, NDE1 and NDEL1 had only minor effects on microtubule organisation within a small proportion of cells, giving Journal of Cell Science 123 (2) phenotypes that are broadly consistent with previous findings. Although the whorls of microtubules previously identified upon alteration of LIS1 levels (Rehberg et al., 2005;Smith et al., 2000) were observed in 7% of LIS1-depleted cells, this rearrangement was clearly independent of the changes in Golgi complex distribution seen in over 85% of LIS1-depleted cells (data not shown; see also Fig. 1D).
LIS1 and NDEL1 influence actin organisation and dynamics by indirectly regulating Rho GTPases (Kholmanskikh et al., 2003;Kholmanskikh et al., 2006;Rehberg et al., 2005;Shen et al., 2008). Although we observed alterations in actin organisation in LIS1 and NDE1-depleted cells, the phenotypes were not conserved between different LIS1 and NDE1 oligonucleotides (data not shown). Knockdown of NDEL1 using two distinct oligonucleotides caused the majority of cells to display prominent stress fibres, and they also lacked filopodia, whereas untreated HeLaM cells displayed a much wider range of actin distribution (data not shown). Hence, although recent evidence suggests a role for actin in the positioning of the Golgi complex (Campellone et al., 2008) and lysosomes (Loubery et al., 2008), our data show that there is no consistency between actin organisation and Golgi positioning, so that changes in actin distribution cannot account for the alterations in Golgi distribution. In support of this conclusion, overexpression of the dynactin subunit p50 caused substantial dispersal of the Golgi Equal protein samples from each cell preparation were western blotted for DIC, transferrin receptor (TfR) and GM130 as membrane markers, or Alix as a soluble marker. Loading for membranes was approximately 10 times higher than for cytosol in terms of cell equivalents. (D)As for C, except fractions were immunoblotted for p150, LIS1, NDE1 or NDEL1.  Table S2. (B)As for A, but cells depleted for NDE1, NDEL1 or a combination of NDE1 and NDEL1 were transfected with either GFP or GFP-LIS1*. complex, early endosomes and lysosomes, without affecting distribution of the actin cytoskeleton (data not shown).
Since LIS1 and NDEL1 might influence Cdc42-mediated processes through binding IQGAP1, Rac and Cdc42GAP (Kholmanskikh et al., 2003;Kholmanskikh et al., 2006;Rehberg et al., 2005;Shen et al., 2008), and because Cdc42 has also been suggested to regulate the recruitment of dynein to COPI-coated vesicles , it was possible that alterations in Cdc42 activity were playing an important role in disrupting Golgi complex morphology following loss of LIS1 or NDEL1. We therefore tested whether expressing wild-type or mutant Cdc42 could mimic the effects of RNAi-mediated depletion on Golgi morphology ( Fig. 8; see supplementary material Table S4 for statistical analysis). Expression of constitutively active (G12V) Cdc42 did not alter the  Table S4. distribution of the Golgi complex in most cells, but did lead to a partial fragmentation and scattering of the Golgi complex in some cells, which was mild compared with that observed in knockdown cells. This effect was nevertheless statistically significant ( Fig. 8; supplementary material Table S4). Although the distribution of fragments somewhat resembled that seen upon knockdown of NDE1 and NDEL1 (Fig. 3), the Golgi fragments were much more varied in size and shape (Fig. 8A). Wild-type or dominant-negative (T17N) Cdc42 expression had no effect on Golgi distribution or morphology ( Fig. 8; supplementary material Table S4). We also tested whether the Golgi-scattering phenotype induced by loss of LIS1, NDE1, NDEL1, or loss of both NDE1 and NDEL1 expression could be rescued by overexpressing Cdc42. No rescue was seen following expression of wild-type, dominant-negative or constitutively active Cdc42 (data not shown).

Discussion
Using a systematic approach, we have shown that LIS1 is essential for dynein-dependent organelle positioning and that, together, NDE1 and NDEL1 are also essential. In each case, the most severe phenotype is similar to that observed upon expression of the p50 subunit of dynamitin, which completely disrupts most dyneindependent processes (Burkhardt et al., 1997;Echeverri et al., 1996). It is also striking that transfection of LIS1, NDE1 or NDEL1 against a wild-type background causes the Golgi complex to become more compact within the perinuclear region, suggesting that it might lead to increased dynein activity on membranes. Hence, under normal circumstances, levels of LIS1, NDE1 and NDEL1 might be limiting for dynein activity, which would explain why disease in the developing brain is associated with haploinsufficiency (Lo Nigro et al., 1997). This work builds on previous studies in which a reduction of LIS1 levels was reported to alter Golgi complex localisation, albeit slightly Sasaki et al., 2000;Smith et al., 2000), whereas expression of a hypomorphic allele of LIS1 in Dictyostelium resulted in more profound scattering of the Golgi complex (Rehberg et al., 2005). However, in contrast to these studies, we have been able to rule out the possibility that the effects on Golgi complex positioning might be secondary to disruption of either actin or microtubule organisation. Similarly, a recent study using the sequestration of LIS1 by overexpression of its interactor plateletactivating factor acetylhydrolase Ib, found that this led to Golgi complex scattering independently of microtubule reorganisation (Ding et al., 2009). A more direct influence of LIS1 on dyneindriven organelle movement has also been suggested by the reduced yolk platelet migration in LIS1 mutant embryos (Cockell et al., 2004) and impairment of pronuclear migration in bovine embryos injected with anti-LIS1 antibodies (Payne et al., 2003).
Our findings have allowed us to dissect the relationship between NDE1 and NDEL1 in dynein function on membrane organelles. These proteins show differential expression, with NDEL1 being widely expressed, and NDE1 expressed at high levels in specific tissues and during a narrow developmental period (Feng and Walsh, 2004). Given this, it is possible that they could fulfil equivalent functions in different tissues. However, both are expressed in HeLa cells. The proteins share 54% sequence identity, with particularly high sequence conservation within an N-terminal coiled-coil domain, which binds LIS1 (Derewenda et al., 2007). The C-terminal portion is much less highly conserved between NDE1 and NDEL1, except for a stretch of approximately 30 amino acids, which is necessary for binding dynein (Sasaki et al., 2000). The C-terminal portions of NDE1 and NDEL1 bind a range of effectors (Toyo-oka et al., 2003). Some of these have been shown to bind both NDE1 and NDEL1, but in most cases their interactions with both proteins has not been tested. So far, the p60 subunit of katanin is the sole effector that has been shown to interact specifically with only one or the other (Toyooka et al., 2005). In addition, although several studies have assigned cellular functions to either NDE1 or NDEL1 (Feng et al., 2000;Feng and Walsh, 2004;Shu et al., 2004), few studies investigate how the proteins work together.
Our data suggest that NDE1 and NDEL1 are interchangeable with respect to their function in supporting dynein-dependent organelle positioning. This is consistent with the ability of both NDE1 and NDEL1 to bind both LIS1 (Derewenda et al., 2007) and dynein (Liang et al., 2004;Niethammer et al., 2000;Sasaki et al., 2000;Stehman et al., 2007), and would explain the mild organellepositioning phenotypes previously reported upon loss of NDEL1 only . However, in other respects, NDE1 and NDEL1 appear to have distinct functions. We have confirmed a previous report showing that mitotic functions of NDE1 and NDEL1 are not equivalent (Vergnolle and Taylor, 2007). Similarly, expression of NDEL1, but not NDE1, can rescue microtubule abnormalities and neuronal migration defects associated with NDEL1 deficiency . Consistent with that study, our data suggest that microtubule organisation is rather differently affected by the loss of each protein.
We have shown that depletion of NDEL1 and NDE1 leads to a dramatic loss of dynein from cellular membranes, explaining why the movement of multiple organelles is affected by this treatment. Interestingly, both NDE1 and NDEL1 are relatively enriched in membrane fractions in comparison to dynein, whereas LIS1 and dynactin are primarily cytosolic. NDE1 and NDEL1 might therefore help to recruit dynein to its various cargoes, perhaps by forming part of a scaffold. Studies on the kinetochore have suggested that ZW10 and NDE1 contribute to dynein targeting via separate pathways, which also influence each other (Kops et al., 2005;Starr et al., 1998;Varma et al., 2006;Vergnolle and Taylor, 2007;Yang et al., 2007). However, there are conflicting reports as to whether NDEL1 is needed for dynein localisation at the kinetochore (Liang et al., 2007;Vergnolle and Taylor, 2007).
Interestingly, we found that loss of LIS1 leads to a major reduction in membrane-associated dynein, although some remains. LIS1 has previously been shown to have a role in dynein localisation to the cell cortex (Cockell et al., 2004;Faulkner et al., 2000) and to the kinetochores in Drosophila (Siller et al., 2005). By contrast, other studies have suggested that LIS1 is recruited after dynein, as it is not found at the kinetochore when dynein and dynactin are absent (Coquelle et al., 2002;Faulkner et al., 2000;Tai et al., 2002), and its absence does not affect dynein or dynactin targeting in mammals (Faulkner et al., 2000;Tai et al., 2002). These data would be more consistent with LIS1 being a regulator of dynein function. In fact, our results support a role for LIS1 both in dynein cargo binding and regulation, because depletion of LIS1 leads to a very strong Golgiscattering phenotype, even though some dynein remains membrane associated, but presumably inactive. Indeed, LIS1 has been reported to stimulate dynein ATPase activity in vitro (Mesngon et al., 2006). However, a recent report surprisingly indicates that LIS1 might uncouple the mechanochemical cycle of dynein, and that this effect is released by NDEL1, even though NDEL1 on its own inhibits dynein ATPase activity (Yamada et al., 2008).
Our results shed new light on the hierarchy of interactions between LIS1, NDEL1, NDE1 and dynein on membranes. We have Journal of Cell Science 123 (2) shown that either NDEL1 or NDE1 can partially rescue the Golgi scattering caused by LIS1 knockdown, but that LIS1 allows only a minor rescue of the NDEL1-or NDE1-knockdown phenotype. Furthermore, we found that the large pool of membrane-associated NDE1 and NDEL1 is not reduced following LIS1 knockdown. These data are consistent with NDEL1 and NDE1 being crucial for dynein recruitment to (or retention on) membranes, whereas LIS1 is vital for motor activation. However, even though LIS1 levels on the membrane are very low, it seems to facilitate the dyneinmembrane interaction in some way. Altogether, these data fit with studies showing that NudE is a multicopy suppressor of NudF (LIS1) (Efimov and Morris, 2000), that LIS1 and NDEL1 mutations are synthetic in mice , and that NDEL1 overexpression rescues cortical microtubule capture and dynein localisation in LIS1 mutant cells (Yingling et al., 2008). By contrast, other studies have placed NDE1 downstream of LIS1 in dynein localisation at microtubule plus ends in yeast (Li et al., 2005b), and in neuronal migration (Shu et al., 2004).
Surprisingly, even though they have interchangeable roles in dynein-mediated membrane positioning, NDE1 and NDEL1 might also contribute to Golgi structure via a dynein-independent pathway in which their functions are not equivalent, or which is acutely sensitive to levels of NDE1 and NDEL1. Although knockdown of NDEL1 and NDE1 together generates Golgi complex fragments that are spread throughout the cell (a phenotype typical for dynein inactivation), loss of NDEL1 or NDE1 individually caused the Golgi ribbon to disassemble into a broken network of fragments that remained in the perinuclear region. This phenotype cannot be explained by a partial inhibition of dynein activity, because it is distinct from that seen in cells expressing low levels of p50 (in which the Golgi ribbon remains, albeit reduced in size, together with a population of small peripheral structures; data not shown).
Given that NDEL1 regulates Cdc42 activity by interacting with Cdc42GAP (Kholmanskikh et al., 2003;Kholmanskikh et al., 2006;Rehberg et al., 2005;Shen et al., 2008), and that active Cdc42 prevents the recruitment of dynein to COPI-coated vesicles , one possible explanation for this partial Golgi-scattering phenotype is that depletion of NDEL1 affects Golgi morphology indirectly by modulating Cdc42 activity. However, we think this is unlikely given that the Golgi morphology following the expression of constitutively active Cdc42 (Fig. 8) is different to that seen after NDEL1 loss (Fig. 3), and since the overexpression of wild-type, dominant-negative or constitutively active Cdc42 did not suppress the NDEL1 phenotype. Instead, the fragmented phenotype seen upon knockdown of NDEL1 or NDE1 is very similar to that seen following knockdown of Golgi-tethering proteins, such as golgin84 (Diao et al., 2003) or GM130 (Puthenveedu et al., 2006), suggesting that it reflects some structural defect in the Golgi complex, rather than simply loss of motor activity.
These results raise the possibility that NDEL1 and NDE1 interact with effector(s) that maintain Golgi architecture, as well as regulating dynein recruitment or activity on many different cargoes. A precedent for this is ZW10, a dynein and dynactin regulator that also interacts with the ER SNARE syntaxin 18 (Hirose et al., 2004;Sun et al., 2007). Loss of ZW10 also affects the Golgi complex (Hirose et al., 2004;Sun et al., 2007), but the phenotype of ZW10 depletion is clearly different to that seen in the absence of NDE1 or NDEL1 (Fig. S1), or when constitutively active Cdc42 is expressed (Fig. 8). Importantly, ZW10 regulates dynein-driven motility in both the secretory and endocytic pathways (Varma et al., 2006), as well as LIS1, NDE1 and NDEL1 in membrane positioning during mitosis (Karess, 2005), indicating that many different dynein functions are modulated by ZW10, LIS1, NDEL1 and NDE1. By investigating the latter three dynein regulators together, we have taken the next step towards defining the roles these proteins have in controlling dynein activity at the Golgi complex.

Cell culture and transfection
HeLaM cells were transfected with DNA constructs using JetPEI (QBiogene) and analysed after 16-24 hours. For GFP and LIS1-GFP transfections, 1 g DNA was used per ml of growth medium. For GFP-NDE1* and GFP-NDEL1*, 30 ng specific DNA was used per ml of growth medium and 1 g pBlueScript was used as carrier DNA in order to obtain low levels of protein expression. For RNAi, transfections used INTERFERin (QBiogene), with cells analysed 72 hours after transfection. All siRNA oligonucleotides were obtained from Dharmacon. For single transfections, siRNA oligonucleotides were used at a final concentration of 10 nM. For double or triple transfections, the concentration of each oligonucleotide was 5 nM. Lamin A/C siRNA was a control for all experiments, unless indicated. The following sequences were used: LIS1a, 5Ј-GAACAAGCGAUGCAUGAAGdTdT-3Ј (Tsai et al., 2005); LIS1b, a combination of sequences supplied as an ON-TARGETplus SMARTpool; NDE1a, 5Ј-GGACCCAGCUCAAGUUUAAUU-3Ј; NDE1b, 5Ј-GGAGGGAAG -CCGAGAAUAUUU-3Ј (Vergnolle and Taylor, 2007); NDEL1a, 5Ј-GGAC CA A -GCAUCACGAAAAUU-3Ј; NDEL1b, 5Ј-GGACCAAGCAUCACGAAAAUU-3Ј; ZW10, 5Ј-AAGGGUGAGGUGUGCAAUAUG-3Ј (Varma et al., 2006). Unless specifically stated, experiments were performed using the 'a' sets of oligonucleotides. For RNAi rescue experiments, cells were transfected after 48 hours with GFP-tagged, RNAi-resistant constructs or with GFP as a control. GFP was visualised directly in formaldehyde-fixed cells or via an anti-GFP antibody in methanol fixed cells. To quantify levels of RNAi rescue, cells labelled for GFP were counted for classifications of phenotypes as described, with a minimum of 100 cells counted for each experimental condition. No correlation was observed between the brightness of GFP fluorescence and phenotype at the expression levels used here (data not shown). Interexperimental errors are illustrated using s.e.m. Statistical analysis was performed using multinomial logistical regression via SPSS software (SPSS Inc.). All statistical analysis is provided in supplementary material Tables S1-S4.

Microscopy and image analysis
Images of fixed cells were generated using a 60ϫ 1.4 NA Plan Apo objective on an Olympus IX70 microscope equipped for optical sectioning microscopy (Deltavision; Applied Precision). For each cell, a z-series at 0.2 m intervals was captured using a Roper Scientific CoolSnap HQ camera, then processed using constrained iterative deconvolution. Deconvolved image stacks were generated and projected using SoftWorx (Applied Precision). For some experiments, indicated, images were captured using an Olympus BX-60 microscope fitted with a 60ϫ 1.40 NA Planapo objective or a 40ϫ U-Planapo-phase 1.0 NA objective and a Roper Scientific CoolSnap ES camera. Images were captured using MetaVue. All images were opened as 16bit grey-scale images and scaled using linear transformations in ImageJ, and then converted to 24-bit RGB files. PhotoShopCS and Adobe IllustratorCS were used to construct final figures.
For imaging of live cells, cells were grown in black polystyrene glass based sixwell IWAKI plates (Jencons). Images were acquired every 5 minutes for 15 hours on an AS MDW live-cell imaging system (Leica) using a 20ϫ 0.50 NA Plan Fluotar objective and collected using a Coolsnap HQ CCD (Photometrics) camera. Point visiting was used to allow multiple positions to be imaged within the same time course and cells were maintained at 37°C and 5% CO 2 . Mitosis was timed from the first sign of cell rounding to the first sign of cytokinesis. Statistical analysis was performed with a two-tailed Mann-Whitney test using SPSS software.

Preparation of cell extracts and western blotting
Cells were scraped and lysed by sonication in 10 mM HEPES, pH 7.4, 100 mM NaCl, 0.1% (w/v) Triton X-100, containing Protease Inhibitor Cocktail III (Sigma). Protein concentrations were determined by the BCA assay (Perbio) and equalised by dilution. Samples were separated by SDS PAGE, transferred onto Hybond-C Extra nitrocellulose membrane (GE Health Sciences) and immunoblotted with appropriate antibodies. For sucrose gradient analysis, cells were grown on 10 cm dishes and lysed in 30 mM potassium acetate, 3 mM magnesium acetate, 5 mM EGTA, 10 mM HEPES, pH 7.4, containing 0.1% Triton X-100, protease inhibitors, and 1 mM DTT. Insoluble material was removed by centrifugation and 60 l supernatants were applied to 650 l 5-20% (w/v) continuous sucrose gradients in lysis buffer lacking detergent, prepared in Beckman centrifuge tubes (344090). These were centrifuged in a SW55 rotor, using inserts, for 5 hours at 116,000 g. Fractions (60 l) were removed and analysed for dynein intermediate chain by western blotting. To prepare membrane and cytosol fractions, cells were grown on three 15 cm dishes, trypsinised and washed in complete medium followed by two washes in 250 mM sucrose, 3 mM magnesium acetate, 5 mM EGTA, 10 mM HEPES, pH 7.4. Cells were homogenised in this buffer containing protease inhibitors and 1 mM DTT, using a ball-bearing homogeniser (Isobiotec, Heidelberg). Post-nuclear supernatants were prepared by centrifuging twice at 2000 g for 10 minutes. These were applied to 1 ml cushions of 800 mM sucrose, 3 mM magnesium acetate, 5 mM EGTA, 10 mM HEPES, pH 7.4 and centrifuged for 30 minutes at 106,000 g in a TLS55 rotor. Supernatants were removed carefully. Pellets were washed carefully, resuspended in lysis buffer, and re-pelleted at 106,000 g for 10 minutes. This work was supported by the BBSRC (grant BB/C512929/1). We are grateful to Martin Lowe for commenting on the manuscript. We thank Richard Woodman, Flinders University, Adelaide, for advice on statistical analysis.