We used the results of a genome-wide RNAi morphological screen in Drosophila melanogaster S2R+ cells 8 to select genes that altered the shape of S2R+ cells, but for which the function in Drosophila was not previously known. A bioinformatic approach was then used to select a subset of 16 of these genes for which there was little or no evidence of a role in regulating cell shape, but for which there was some other information on their function (Additional file 1, Table S1). We used functional information from studies of their homologues in other organisms (yeast, C. elegans, mammals), on whether they were hits in any other genetic screens, had known interaction partners in humans or other species (for example, Saccharomyces cerevisiae) or putative interaction partners based on large scale screens, and known or predicted protein domains (Table S1; see Materials and methods). For example, CG12505 was selected because the mammalian homologue ARC is known to bind to actin, whereas CG3542 was selected because a human homologue HYPC interacts with Huntingtin. The human homologues of these 16 genes were identified (Table 1), yielding a group of 26 human genes. Nine of the Drosophila genes have a single human homologue, while the remaining genes have two or three human paralogues. These genes were named 'putative motility modifiers' (PMMs).

To identify roles for these genes in regulating cell morphology in human cells, each of the 26 human PMM genes was knocked down in PC3 prostate carcinoma cells, which grow and migrate as single cells without strong cell-cell adhesions, using pools of four siRNAs. In this analysis, 25 of the 26 PMMs had a discernable effect on cytoskeletal organization and/or cell shape (Figure 1; Table 2; Additional file 2, Figure S1). In order to identify genes that induced similar changes and, hence, could act on a similar pathway, cells were classified for 14 different parameters: 7 distinct F-actin distributions, presence of microtubule-containing processes, 3 different cell shapes, spread area, and the presence of multinucleate cells (Table 2, examples of phenotypes are shown in Figure 1 and Additional file 3, Figure S2). These were compared to the phenotypes induced by knockdown of known regulators of the actin cytoskeleton, including the Rho GTPases RhoA, Rac1, Cdc42 and RhoU/Wrch1 and the actin polymerization regulators CAPZB, cofilin (CFL1), mDia1 (DIAPH1) and WAVE3.

This analysis of parameters revealed seven major groups of genes based on their predominant phenotype in the siRNA screen (Table 2), which could reflect functional interaction between members of each group. For comparison, the percent of control cells demonstrating each parameter is shown.

The first PMM group, ARC and ZRANB1 (also known as TRABID 9), both induced a strong increase in stress fibres, which are normally only rarely observed in PC3 cells (Figure 1; Additional file 3, Figure S2). They also increased the spread area and the percent of multinucleate cells (Table 2). ZRANB1/TRABID is a deubiquitinase previously linked to Wnt signalling 9, whereas ARC binds to actin filaments 10. The similar phenotype induced by these two genes suggests that ZRANB1 might regulate ARC. Of the known actin regulators, only Rac1 depletion increased stress fibres, although to a lesser extent than ZRANB1 or ARC.

PMM group 2 increased the level of cortical actin filaments (Additional file 3, Figure S2) and often altered the membrane morphology. This included FAM40A and FMNL1, which showed an increase in F-actin-rich lamellae around the cell periphery. A subset also made the cells larger, presumably by promoting cell spreading through the cortical F-actin changes. PMM group 3 also affected cortical F-actin but were most striking for the presence of long thin microtubule-containing processes as their most dominant phenotype. The cell shape was not altered, but at least 25% of the cells had one narrow straight process that often extended over multiple neighbouring cells (Additional file 3, Figure S2). Occasionally cells had two processes. Note that depletion of several PMMs in groups 4 and 5 also induced an increase in cells with microtubule-containing processes (Table 2). This suggests that all these proteins are linked in normally suppressing microtubule-based protrusions. This could reflect the known role of the actin cytoskeleton in limiting the generation of microtubule-based protrusions as previously described in Drosophila cells in culture 11. Knockdown of PMM group 4 genes, including FAM40B and FMNL2, induced an elongated phenotype (Additional file 3, Figure S2). Group 5 genes were characterized by a very elongated bipolar phenotype (Additional file 3, Figure S2), and included RhoU and the formin mDia1 (DIAPH1). RhoU has previously been reported to affect integrin-mediated adhesions 12, and thus this very elongated phenotype could reflect altered adhesion, and/or changes to localized actin polymeri

z

z

HYPC was unique in leading to a reduction in F-actin levels (Figure 1; Additional file 2, Figure S1), and did not fit into any of the groups. Finally, EPB41L4A did not induce any discernable phenotype in PC3 cells, although it was selected on the basis of its morphological phenotype in Drosophila S2R+ cells. It is possible that EPB41L4A is not expressed in PC3 cells.

Interestingly, closely related homologues were generally in different groups; for example, the formins FMNL1, FMNL2 and FMNL3, and FAM40A and FAM40B (Table 2).

To determine whether the morphological changes induced by PMM depletion led to changes in migration, we tested the effects of knocking down expression of each PMM on cell migration using scratch wound assays. Migration of cells was monitored by time-lapse microscopy (Figure 2A), and the change in wound size over time was quantified using an automated method of image analysis (Figure 2B). These results were used to generate graphs showing the relative velocity of wound closure (Figure 2C). Of the PMMs, depletion of FMNL3, FNBP3, ARC, ZRANB1 and LIMD1 led to the strongest inhibition of wound closure. Notably, ARC and ZRANB1 form PMM group 1, and have an increase in stress fibres and larger spread area (Table 2). Stress fibres can often inhibit cell migration 17. FMNL3, FNBP3 and LIMD1 are all part of PMM group 4, and have an elongated shape. Interestingly, the other members of this group also reduced migration (SH3KBP1, RhoA, Rac1, FAM40B). By contrast, closely related homologues of FMNL3 (FMNL1 and FMNL2) and FNBP3 (HYPC) were in different groups and did not significantly affect migration (Figure 2B). Opposite to group 4 PMMs, the group 5 PMMs WAVE3, DIAPH1, RhoU and LARP4 (very elongated cells) tended to increase rather than decrease cell migration (Figure 2). Taken together, these results imply that either an increase in stress fibres or an elongated phenotype inhibits migration of PC3 cells into scratch wounds, whereas very elongated bipolar cells can migrate more rapidly.

From the 26 PMMs, a subset of 7 was chosen for further analysis, based on their phenotypes in the morphology and migration screens. These genes included the five that most strongly inhibited migration in scratch wounds when knocked down, ARC, ZRANB1, FMNL3, FNBP3 (also known as FBP11/HYPA) and LIMD1 (Figure 2B, C). We also included two closely related genes, FAM40A and FAM40B, which had only a small effect on migration in scratch wound assays (Figure 2B, C), but which each had profound effects on cell shape (Figure 1): FAM40A depletion resulted in flatter cells, whereas FAM40B depletion increased elongation and is in the same PMM group as FMNL3, FNBP3 and LIMD1. To validate these hits, we began by deconvoluting each siRNA pool to test the phenotypic effects of using the four single siRNAs individually. At least three of the four single oligonucleotide siRNAs induced a similar phenotype to that seen using the pool in PC3 cells and/or HeLa cells (Additional file 4, Table S2; for HeLa cells, see Figure 6), implying that the henotypes arise as the result of RNAi-mediated silencing of the targeted gene rather than off-target effects.

To analyse further how the seven selected PMMs affect cell migration, we imaged cells migrating into the wound. Most control siRNA-transfected cells detached from their neighbours and migrated directionally to fill the gap exposed as the result of the scratch wound (Figure 3, Additional file 5, Movie 1). As expected from the quantification of scratch wound migration (Figure 2C), ARC, ZRANB1, FMNL3, FNBP3 and LIMD1 depletion strongly inhibited migration into the wound (Figure 3; Additional file 6, Movie 2; Additional file 7, Movie 3; Additional file 8, Movie 4; Additional file 9, Movie 5, Additional file 10, Movie 6). Most cells did not detach from neighbouring cells, in contrast to the control siRNA-transfected cells. FAM40B is in the same PMM group 4 as FMNL3, FNBP3 and LIMD1, but did not affect scratch wound migration as strongly (Figure 2C). Depletion of each of these PMM group 4 genes induced an elongated phenotype in the movies, but FAM40B-depleted cells were able to detach from neighbours and could move further into the wound (Figure 3; Additional file 11, Movie 7), FAM40A-depleted cells were also able to migrate into the wound but migrated with a very different phenotype to FAM40B-depleted cells. They had broad lamellipodia extending into the wound edge, consistent with the broad lamellar phenotype observed in the morphology screen (Figure 3; Additional file 12, Movie 8).

The seven selected PMMs were further analysed for their effects on morphology and migration of sub-confluent PC3 cells (Figures 4 and 5). PC3 cells normally have very few stress fibres, but ARC and ZRANB1 depletion led to the formation of stress fibres and strong cortical F-actin bundles at the periphery of cells (Figure 4), as already observed in the screen analysis (Figure 1, Table 2). LIMD1-depleted cells also often exhibited strong cortical F-actin bundles but no stress fibres, and in many cells microtubules were disorganized without any clear association with an MTOC. FMNL3 and FNBP1 knockdown cells were elongated and had long protrusions that extended over neighbouring cells, often with multiple short actin filament bundles at their tips. Microtubules extended right along these protrusions into the tips. FAM40B-depleted cells also often had very long thin protrusions containing microtubules, but showed no change in F-actin bundling compared to control cells (Figure 5A), and thus were clearly distinct from FMNL3 and FNBP1 knockdown cells. This correlates with the scratch wound analysis, which indicated that the FAM40B morphology was different from FMNL3 or FNBP1 (Figure 3). FAM40B-depleted cells were more elongated than control cells (Figure 5A), as already observed from the screen analysis (Table 2). FAM40A knockdown cells were less spread than controls and often had a broad lamellipodium (Figure 5A, B). Some cells had an intense region of F-actin staining at the opposite side to the lamellipodium, resembling the uropod of leukocytes 18.

To determine the effect of each PMM on the random migration of single PC3 cells, time-lapse movies of PMM-depleted cells were used to generate cell tracks (Figures 4 and 5, right panels) and determine cell migration speed (Additional file 13, Figure S3). FNBP3-depleted cells showed the least displacement from their point of origin and lowest migration speed, and LIMD1, ARC, FMNL3 and ZRANB1 knockdown cells all showed a strong decrease in migration speed, correlating with their migratory behaviour in the scratch wound assay (Figure 2). Only FAM40A-depleted cells had a similar migration speed to control cells (Additional file 13, Figure S3).

To determine whether depletion of each of the seven selected PMMs induced a similar phenotype in a different cell type, we tested their effects in HeLa cells (Figure 6). ARC, LIMD1 and ZRANB1 depletion had qualitatively similar effects in HeLa cells to PC3 cells. ARC depletion led to a strong increase in thick actin filament bundles at the edges of cells, and in some cells to an increase in stress fibres, similar to the response of PC3 cells. LIMD1 depletion increased stress fibres and/or cortical F-actin bundles in a proportion of cells, somewhat similar to ARC depletion. ZRANB1 knockdown led to a strong increase in F-actin bundles along cell edges and some increase in stress fibres, although these were not as thick as those in ARC or LIMD-depleted cells. ZRANB1 knockdown cells also had fewer lamellipodia.

The effects of FAM40A, FAM40B, FMNL3 and FNBP3 depletion were different between the two cell types, probably reflecting the different intrinsic phenotypes of HeLa versus PC3 cells. In contrast to PC3 cells, HeLa cells exhibit cell-cell adhesions. In addition, within the PC3 cell population there are some elongated cells, whereas the HeLa cell populations are more regular in shape with very few elongated cells under normal conditions (Figures 1 and 6). PC3 cells frequently became very elongated with long protrusions in response to PMM knockdown (Figure 1; Additional file 2, Figure S1), and FAM40B, FMNL3 and FNBP3 depletion induced elongation of PC3 cells (Figure 1) but not in HeLa cells (Figure 6). FAM40B-depleted HeLa cells had very bright regions of F-actin staining around the periphery associated with lamellipodia and filopodia, but very few thick cortical F-actin bundles. In addition, FAM40B-depleted HeLa cells appeared to detach from each other and had reduced cell-cell adhesion (Figure 7A). A reduction in cell-cell adhesion was also apparent in the scratch wound images from FAM40B-depleted PC3 cells (Figure 3; Additional file 11, movie 7). FMNL3 depletion did not have a strong effect on HeLa cell morphology, which might reflect differences in the relative levels of FMNL3 in HeLa and PC3 cells compared to FMNL1 and FMNL2. Finally, knockdown of FNBP3 in HeLa cells appeared to inhibit cytokinesis, since many cells were binucleate (Figure 7C). In addition, cells had multiple filopodia/microspikes, both linking between cells and at cell edges away from cell-cell interactions (Figure 7B). Like FAM40B knockdown, this could indicate a reduction in cell-cell adhesion.

To investigate how overexpression of selected PMMs affected cell morphology and cytoskeletal organisation and to determine their localization, cDNAs for ARC, FAM40A, FAM40B, FMNL3 and ZRANB1 were cloned downstream of GFP and transfected into PC3 cells (Figure 8A). GFP-ARC, FAM40A, FAM40B and FMNL3 proteins showed cytoplasmic localization with some enrichment on the plasma membrane compared to GFP alone (Figure 8B). In particular, FMNL3 and FAM40B were localized in lamellipodia. In contrast, ZRANB1 (also known as TRABID) showed punctate localization in the cytoplasm, similar to a previous report 9. Expression of FAM40A reduced PC3 cell area compared with control cells (Figure 8A, C), similar to knockdown of FAM40A, suggesting that FAM40A levels are critical for regulating cell spreading. In HeLa cells, expression of FAM40B increased cortical F-actin bundles (Figure 8D), consistent with the decrease of thick cortical F-actin bundles observed in FAM40B-depleted HeLa cells (Figure 6). The other PMMs had no consistent effect on cell shape or actin filament distribution.

A database containing images from a Drosophila melanogaster S2R+ cell genome-wide RNAi morphology screen, FLIGHT 46, was used to screen images manually to identify genes that affected actin filament or microtubule distribution and/or cell shape. Genes without a well-described function in Drosophila were chosen for further analysis. A bioinformatic approach was used to gather information on their potential functions, using similarity of protein domains and data available from orthologues and/or paralogues in other species. This included information from protein-protein interaction screens obtained from multiple databases including Fly General Repository for Interaction Datasets (Fly GRID), Human Protein Reference Database (HPRD), PubMed Gene, and Genecards. This led to the selection of 16 conserved Drosophila genes, for which 26 human genes were identified (26 PMMs) based on homology.

The prostate cancer-derived cell line PC3 (a gift from Prof J Masters, University College London) was grown in RPMI containing 25 mM Hepes and 2 mM glutamine supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin and 100 μg/ml streptomycin. HeLa (Kyoto strain) and NIH3T3 cells were cultured in Dulbecco's Modified Eagle Medium (Invitrogen, Paisley, UK[) supplemented with 10% FBS, 50 U/ml penicillin and 50 μg/ml streptomycin.

Four individual siRNAs (Dharmacon siGENOME duplexes, ThermoFisher Scientific, Loughborough, UK) were purchased for each gene, and then pooled (Additional file 14, Table S3). The control siRNA was a SMARTpool of four non-targeting siRNAs (Dharmacon D-001206-13). For the PMM screens, PC3 cells were reverse-transfected in black, thin-bottomed 384-well tissue-culture plates suitable for confocal imaging (2,500 cells per well) (Greiner Bio-One, Stonehouse, UK) with 25 to 100 nM of siRNA pools using Lipofectamine-2000 (Invitrogen, Paisley, UK) according to the manufacturer's instructions. For high-resolution confocal imaging, PC3 cells were transfected on Matrigel-coated (100 μg/ml) eight-well chambers (Becton Dickinson, Oxford, UK) or glass coverslips. HeLa cells (2,000 cells per well) were reverse-transfected in black 384-well tissue-culture plates at a final concentration of 25 nM of siRNA pools. PC3 and HeLa cells were analysed 72 h after transfection.

The open reading frames for ARC, FAM40A, FAM40B, FMNL3 and ZRANB were PCR amplified either from cDNA clones or primary cDNA and subsequently cloned by Gateway™ (Invitrogen, Paisley, UK) BP reaction into the Gateway™ vector pDONR221 as described 47. The resulting entry clones were completely sequence verified and used for the generation of GFP-tagged PMMs. In brief, 150 ng of each entry clone was recombined with 150 ng of destination vector, pcDNA-DEST53 or pcDNA-DEST47, to generate N-terminal or C-terminal GFP-tagged constructs, respectively. Positive clones growing on ampicillin were selected and were verified by enzymatic restriction using BsrGI. N-terminal GFP-tagged PMMs were generally expressed at a higher level and were used for further experiments. For transfection, cells were grown on Matrigel-coated (100 μg/ml) eight-well chambers. They were transfected with plasmid DNA (0.6 μg) using Lipofectamine-2000 according to the manufacturer's instructions, and fixed and stained after 18 h.

For fluorescence staining, cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% or 0.2% Triton X-100 in PBS and blocked with 1% BSA or 5% FBS (PC3 cells) or 5% BSA (HeLa cells) in PBS. Cells were stained with DAPI (1 μg/ml, Sigma-Aldrich, Gillingham, UK) for nuclear staining, Alexa Fluor-conjugated phalloidin (1 μM, wavelengths 480 nm, 543 nm, 546 nm or 633 nm, Molecular Probes/Invitrogen, Paisley, UK) or TRITC-conjugated phalloidin (0.125 μg/ml, Sigma) for F-actin visualization, and 1:400 FITC-labelled α-tubulin antibody (DM1A clone; Sigma) for microtubules. Images were acquired by automated microscopy on a Nikon TE2000 microscope (Kingston, UK) with a 20× objective. Confocal images were acquired with a Zeiss LSM 510 (Welwyn Garden City, UK) or a Leica SP5 (Milton Keynes, UK) inverted confocal microscope.

Morphology analysis was carried out using Image J software http://rsbweb.nih.gov/ij/index.html. Circularity was determined with the Circularity Plugin. The area of PC3 cells transfected with constructs encoding GFP-fusion proteins was determined from F-actin-stained fluorescence images. For stress fibre quantification, images of NIH3T3 cells expressing GFP-fusion proteins and stained for F-actin were magnified electronically, and cells were classified visually based on their content of stress fibres.

For time-lapse microscopy, a fully motorized, multi-field Nikon TE2000 microscope was used. PC3 cells were reverse transfected in 96-well plates (17,000 cells/well) with 50 nM siRNA pools using Lipofectamine-2000. For scratch wound migration, scratch wounds were created in the centre of the cell monolayer 72 h after transfection, using a micropipette tip. After washing twice with culture medium, phase contrast cell images were acquired every 30 minutes for 16 h (10× objective). Wound healing quantification was carried out using custom-written Metamorph journals. The area covered by cells on the two sides of the wound was detected by locating the plasma membrane (using a gradient filter) and filling the space surrounded by plasma membrane (using a morphological dilation transformation). The surface areas covered by the cells in the successive frames were exported into Excel to measure wound healing speed. The speed was estimated by applying a linear fitting to the area series using the "SLOPE" function (Excel, Microsoft, Reading, UK). The resulting wound healing speed is expressed in μm/min.

For random migration, cells were seeded on Matrigel-coated 24-well plates 48 h after transfection. After 24 h images were acquired every 10 minutes for 14 h. Cells were tracked using ImageJ Manual tracking and Chemotaxis tool plugins.