Supervillin sequences corresponding to those in gelsolin domains G2 through G6 (Burtnick et al. 1997
; Pope et al. 1998
) lie mostly within structural elements responsible for the folding of these five gelsolin domains (). No alignment could be generated between supervillin and gelsolin repeat 1 (G1). The architecture of gelsolin repeats G2-G6 is predicted to be conserved in supervillin, with central five-stranded mixed β-sheets sandwiched between two alpha-helices. The structural similarity is apparent in repeats G2 and G3 () and pronounced for repeats G4 through G6 (). In all 5 repeats, the divergent, supervillin-specific sequences lie mostly within surface loops (, magenta), suggesting supervillin-specific functions and interaction partners.
Fig. 1 Structural models of (A) supervillin amino acids 1019–1306 based on gelsolin repeats G2–G3 (residues 189–444) and (B) supervillin amino acids 1326–1699 based on gelsolin repeats G4–G6 (residues 445–765). (more ...)
To identify interactors for the supervillin C-terminus, we undertook two undirected yeast two-hybrid screens of a HeLa cell cDNA library with overlapping baits. The bait sequence encoding amino acids 834–1291 contained sequences unique to supervillin and regions of homology with gelsolin G2 and G3 repeats. The second bait encoded supervillin residues 1008–1791, which contained sequences homologous to gelsolin G2–G6 repeats and the C-terminal supervillin headpiece.
We found 33 candidate interacting sequences, representing 27 prey proteins, that passed specificity tests in subsequent directed yeast two-hybrid screens (, columns 1–4). Colonies grew faster and thus appeared earlier with the SV834-1291 bait, probably due to its smaller size (, column 5). Most prey sequences initially captured with SV834-1291 were specific for this bait, failing to cross-interact with SV1008-1791 in targeted screens (, columns 6, 7). The exception was MPHOSPH9, which was obtained as overlapping sequences [amino acids (aa) 809–858] from both screens. Nearly all the SV1008-1791 preys also interacted with SV834-1291. The exception was the most abundant prey, EPLIN/LIMA1, which was represented by two sets of clones with an overlapping sequence (EPLINβ-672–759) that emerged a total of 12 times (, column 5). These observations suggest that MPHOSPH9 aa 809–858 and most of the SV1008-1791 preys interact with sites within SV1008-1291 and that a binding site for EPLIN aa 672–759 lies within SV1291-1791.
Potential binding partners for the supervillin C-terminus
Supervillin aa 844–864 (21-aa window) and 947–976 (28-aa window) may form coiled-coils, as predicted by the COILS program (Lupas et al. 1991
). Because several of the preys obtained with SV834-1291 also contain predicted coiled-coils, we re-screened the SV834-1291 preys with bait vectors containing mutations designed to disrupt each potential supervillin coiled-coil (, columns 8, 9). The double mutation of supervillin aa Leu-952 and Leu-953 to His and Pro, respectively, did not affect any interactions (not shown). By contrast, doubly mutating supervillin Phe-851 and Leu-852 to Ser and Pro (F851S, L852P) interrupted the interactions with keratin 18, ART-27, BUB1, HAX1, lamins A/C and B2, PAN3, and TNFAIP1 in directed yeast two-hybrid screens, while interactions with the other 11 preys were maintained (, column 8). Because the prey sequences obtained for keratin 18, lamin A/C, and lamin B2 are predicted to be mostly coiled-coils, these interactions with supervillin may be incidental. However, over-expressed supervillin does mis-localize endogenous lamin A/C within the nucleus (Wulfkuhle et al. 1999
). The prey sequences for ART-27, BUB1, HAX1, PAN3, and TNFAIP1 lack predicted coiled-coils, suggesting different mechanisms of association (, column 9).
To further interrogate the likelihood of specific interactions with supervillin, we embarked on classifying screens to determine which of the yeast two-hybrid prey sequences could affect supervillin function and localization in mammalian cells (). We first tested 23 preys plus 2 full-length cDNA constructs (ITGB3BP, RHAMM) for co-localization with EGFP-tagged SV1-1792 (EGFP-SV). Because of expected effects on assembly of the endogenous proteins (Parry et al. 2007
), the coiled-coil sequences from keratin 18, lamin A/C, and lamin B were not tested. We found that 17 of the prey sequences (, asterisks) plus full-length wild-type (WT)-RHAMM co-localized with supervillin after co-expression in COS7-2 cells, supporting the existence of interactions between supervillin and these proteins ().
Fig. 2 Flow chart for screening of supervillin-interacting sequences in mammalian cells. Prey sequences were cloned into mammalian expression vectors and screened for effects on HeLa cell spreading onto fibronectin, for changes to COS7-2 cell morphology, and (more ...)
Hypothesizing that the short supervillin-interacting sequences might behave in a constitutively active or dominant-negative manner, we screened the co-localizing prey constructs for effects on cell spreading (), a process inhibited by supervillin expression (Takizawa et al. 2007
). As a sentinel for a supervillin-binding sequence with a demonstrated role in cell motility (Lai et al. 2005
; Takizawa et al. 2006
; Yi et al. 2002
), we also included the myc-tagged TRIP6 C-terminus, aa 265–476, in all screens. We found that the TRIP6 C-terminus and four of the new candidate supervillin interactors significantly inhibited HeLa cell spreading on fibronectin (). The strongest effect was observed for FLNA aa 987–1186 (filamin A repeats 8–10) (, panel F), a sequence that contains mutations associated with human skeletal disorders (Robertson 2005
; Stossel et al. 2001
). Cell spreading also was decreased by transfection with BUB1 aa 4–313, MIF4GD aa 122–256, or HAX1 aa 144–279 (, panels B, C, D). By contrast, cell spreading was unaffected by the other prey sequences tested (), including FLNA aa 2169–2414 (repeats 20–22) (, panel G), which shares a conserved binding site for several ligands with FLNA repeats 8–10 (Ithychanda et al. 2009
). Thus, the effect of filamin repeats 8–10 on cell spreading is probably independent of binding to the shared ligands. These results suggest effects by the TRIP6 C-terminus and four of the prey sequences on either the function of the endogenous parent protein or on its interaction with supervillin.
Fig. 3 Prey constructs with effects on the initial spreading of HeLa cells on fibronectin. Transfected HeLa cells expressing a tagged control pCMV vector (A) or this vector with sequences encoding (B) BUB1(4–313), (C) MIF4GD(122–256), (D) HAX1(144–279), (more ...)
During the secondary screening for co-localization with EGFP-supervillin, we noticed striking shifts by many co-localizing preys on supervillin distribution, which normally includes both peripheral and internal staining (, panels a–d; ). Of 23 co-expressed preys (n ≥ 20 cells each), 9 increased the percentages of cells with an EGFP-supervillin concentration in the cell center from 47.7% to >65.0% (). The prey sequence in TNFAIP1 (TNFAIP1-119-316) caused the largest shift towards the cell interior, with ~88% of the cells containing centrally located, supervillin-associated, vesicle-like structures that were lacking actin (, panels e-h). HAX1-144-279, BUB1-4-313, Tks5-318-655, STARD9-2528-2663, MIF4GD-122-256, ODF2-133-269, FLNA-2169-2414, and KIF14-1522-1648) also increased the percentages of cells with a primarily central localization, as did the supervillin-binding sequence in TRIP6 (aa 265–476) () (Takizawa et al. 2006
Fig. 4 Prey constructs altering the amount of internal EGFP-supervillin signal. (A) COS7-2 cells expressing control myc vector (a–d) or myc-tagged TNFAIP1(119–316) (e–h) were imaged for EGFP-supervillin (a, e), myc-staining (b, f), and (more ...)
Two prey sequences shifted EGFP-supervillin towards the cell edge (). EPLIN-650–759 () and KIFC3-237-366 (, panel b) increased the percentages of cells with a primarily peripheral supervillin signal from 1.5% in control cells to 38.1% and 31.8%, respectively (). Because of difficulties in cloning the supervillin-interacting sequence in RHAMM, we also tested a dominant-negative murine RHAMM variant 4 construct (DN-RHAMM) (, panel a). This construct contains mutations in 6 highly conserved charged residues outside the supervillin-interacting site in the RHAMM C-terminus. We found that DN-RHAMM caused 36.0% of the cells to have a primarily peripheral distribution of supervillin (). Including this result, 12 of the new candidate supervillin-interacting sequences both co-localize with and cause mis-localization of full-length supervillin ().
Fig. 5 Prey constructs targeting EGFP-supervillin to the cell edge. (A) COS7-2 cells expressing control myc vector (a–d) or myc-tagged EPLIN(650–759) (e–h) were imaged for EGFP-supervillin (a, e), myc-staining (b, f), and F-actin (c, (more ...)
Some of the prey sequences also affected the appearance of EGFP-supervillin (Supplementary Fig. S1
). In 98% of control COS7-2 cells, supervillin appears as both linear structures associated with actin fibers and with F-actin as punctae in the cell center (Crowley et al. 2009
; Wulfkuhle et al. 1999
). We re-scored the cells imaged for supervillin distribution for EGFP-supervillin appearance. KIF14-1522-1648 caused the largest increase in fibrous-only appearance (to 34.8% of transfected cells vs.
0% of controls) (Supplementary Fig. S1A
, panels e-h; Supplementary Fig. S1B
), with increases in fibrous staining also observed for KIFC3-237-366 (, panel b; Supplementary Fig. S1B
). While Bub1-4-313 also slightly increased fibrous-only staining, it strikingly increased punctate-only staining from 0.8% in controls to 25% (Supplementary Fig. S1A
, panels i-l; Supplementary Fig. S1B
). Therefore, 3 out of the 12 tested prey sequences that induced mis-localization of supervillin also caused changes in supervillin appearance; 2 of these sequences are non-motor regions of kinesins.
Collectively, these results identified 12 new proteins as likely functional interactors with supervillin in mammalian cells (, column 1, bold). Eleven prey constructs and WT-RHAMM co-localized with EGFP-supervillin (, column 2). Supervillin co-localizations with sequences from EPLIN, KIF14, KIFC3, and TNFAIP1 were especially strong. Also, the BUB1-, HAX1-, STARD9-, and Tks5-interacting sites had as much signal overlap as the sequence from the demonstrated supervillin binding partner, TRIP6. The 11 prey constructs and dominant-negative RHAMM caused mis-localization of EGFP-supervillin and/or other effects, suggesting effects on supervillin interactions in vivo (, column 3). Alignments of the human prey sequences with known and predicted sequences from protein databases showed a high degree of identity across species, consistent with an evolutionarily conserved function (, column 4). The exceptions were the prey sequences from KIF14 and STARD9, which to our knowledge has not yet been cloned. While the KIF14 supervillin-interacting sequence is only 52% identical overall between human and mouse, there are several regions of high local homology that could constitute a conserved binding site. Many of the new candidate interactors regulate aspects of cell motility, membrane trafficking, ERK signaling, and extracellular matrix degradation and invasion (), similar to the established functions of supervillin (Crowley et al. 2009
; Fang et al. In press
; Gangopadhyay et al. 2009
; Gangopadhyay et al. 2004
; Takizawa et al. 2007
; Takizawa et al. 2006
Probable functional supervillin (SV) interactors
We validated the predicted interactions between the supervillin C-terminus and 2 of the new candidate interactors (EPLIN, KIF14) by in vivo co-localizations of tagged full-length proteins and in vitro pull-downs with GST-supervillin fusion proteins (). Supervillin tagged with mRFP co-localizes almost perfectly with Flag-tagged EPLINα at intracellular punctae and apparent stress fibers (, panels a–c); mRFP-supervillin also exhibits near-perfect co-localization with EGFP-tagged KIF14 at punctae (, panels d–f). As expected from the selectivity shown by EPLIN for SV1008-1791 (), we recovered EPLIN from HeLa lysates with GST-tagged SV1398-1792 (, lane 4) but not with GST alone or GST-SV1009-1398 (, lanes 2 and 3). Similarly, as expected from the interaction of the KIF14 prey sequence with both SV834-1291 and SV1008-1791 (), endogenous KIF14 was recovered from HeLa cell lysates with GST-tagged SV1009-1398 (, lane 3), but not with GST alone or GST-tagged SV1398-1792 (). These results support interactions between supervillin and both KIF14 and EPLIN/LIMA1 in mammalian cells.
Fig. 6 Co-localization and binding of SV with EPLIN and KIF14. (A) Co-expressed mRFP-SV (a, d; red in c, f) co-localizes almost perfectly with Flag-EPLINα (b; green in c) and with EGFP-KIF14 (e; green in f) in COS7-2 cells; overlaps appear yellow (c, (more ...)
Because 5 of the new supervillin interactors, including EPLIN and KIF14, function during cell division () (Carleton et al. 2006
; Chircop et al. 2009
; Logarinho and Bousbaa 2008
; Maxwell et al. 2008
; Soung et al. 2009
), we explored the possibility that supervillin itself might play a role in this process (). To minimize potential redundancy with gelsolin (Crowley et al. 2009
), we used the HeLa S3 Tet-off cells, which contain little gelsolin (not shown). We reduced the expression of supervillin using either a stably expressed shRNA or transient treatments with dsRNAs against different supervillin sequences (). All 3 supervillin-specific dsRNAs () and the shRNA increased the numbers of HeLa cells with 2 or more nuclei (, arrows). Supervillin knockdowns with shRNA or dsRNA also increased the time required for cell doubling (). Though all 4 reagents were effective, the largest effect was observed with a dsRNA directed against nucleotides within the supervillin 3’-UTR (, 6016). These results support a role for supervillin during cell division, as predicted by the functions of its new interactors ().
Fig. 7 Supervillin (SV) is required for normal cell division. (A) Immunoblots showing the percentages (%) of endogenous supervillin remaining in HeLa cells with a stably incorporated control (lane 1) or supervillin-specific (lane 2) shRNA or after treatment (more ...)
To determine when supervillin functions in cell division, we imaged living, unsynchronized HeLa cells and found that supervillin-knockdown cells fail cytokinesis primarily at the stage of furrow ingression (; Movies M1
). For cells treated with a control dsRNA (), the first ingression of the plasma membrane (, 75′, arrow) occurred 22.0 min ± 13.9 min (mean ± s.d., n = 172) after the last frame containing an intact metaphase plate (, 54′, arrowhead; Movie M1
). This mean time to ingression was not significantly different from the 24.9 min ± 14.2 min (mean ± s.d., n = 187) observed for cells treated with the 6016 dsRNA (, Movie M2
). While 3.1% (6/193) of the monitored supervillin-knockdown cells vs.
1.6% (3/190) of the controls did not try to divide, most (84%; 36/43) of the supervillin-knockdown cells that failed cytokinesis did so after furrow initiation but before establishment of an intracellular bridge (). These cells underwent multiple contractions for 55 ± 20 min (n = 12) before flattening out and migrating away as bi-nucleated cells (, 180′). When supervillin-knockdown cells successfully completed cytokinesis, no significant difference was observed in the time between the initial furrow ingression and the final breakage of the cytoplasmic bridge (). No more than 1% of either control or supervillin-knockdown cells underwent apoptosis during their first mitoses after 48 hr of RNAi treatment (not shown).
Fig. 8 Cytokinesis failure in supervillin-knockdown cells occurs primarily during furrow ingression. Unsynchronized HeLa cells were treated with (A) control or (B) the supervillin-specific 6016 RNAi for 48 hr and then imaged every 3 min for 16–24 hr. (more ...)
Consistent with a role during furrow formation, supervillin was concentrated at cleavage furrows (). To monitor the intracellular localization of supervillin during cell division, we stably transfected HeLa cells with EGFP-tagged human supervillin and amplified this signal in fixed cells using anti-GFP and Alexa Fluor 488-labeled secondary antibody. The total supervillin level in these cells was 3.3 ± 2.2 (means ± s.d., n = 4) times greater than endogenous supervillin levels in the parental HeLa cells (). During metaphase and anaphase, supervillin was distributed throughout the cytoplasm with only a modest enrichment at the cell periphery (), but became strongly concentrated at the invaginating plasma membrane during late anaphase and telophase (, arrows). Supervillin also was enriched at both sides of the microtubule bundles within the intracellular bridge during cytokinesis, concentrating at the midbody (, arrows) and at the presumed minus ends of the microtubule bundles at the junction between the bridge and the daughter cell cytoplasm (, arrowheads).
Fig. 9 Supervillin concentrates at the cleavage furrows and midbodies of dividing cells. (A) Upper panel, immunoblots stained with antibody against supervillin (lanes 1, 2) or GFP (lanes 3, 4) of whole cell lysates from untransfected HeLa cells (lanes 1, 3) (more ...)
The supervillin distribution at the membranes of dividing cells overlaps with that of its interactors, myosin IIA, EPLIN, and KIF14 (). During the early stages of cytokinesis, supervillin was concentrated at the cleavage furrow with myosin IIA and EPLIN (; ; arrows). Supervillin was associated even more prominently than myosin IIA or EPLIN at the midbody (; ; arrowheads) and overlapped at the light level with the signal for KIF14 at the midbody (, arrowheads) and future abscission site (, arrows), consistent with an association during cytokinesis. These localizations of supervillin and its interactors, coupled with supervillin’s contribution to efficient cytokinesis, support the prediction from that supervillin interactions play a role in cell division.
Fig. 10 Supervillin overlaps with myosin IIA and EPLIN at the cleavage furrow and with KIF14 at the midbody. Wide-field immunofluorescence micrographs of synchronized HeLa cells during mitosis were stained for stably expressed EGFP-supervillin with anti-EGFP, (more ...)