Parallel RNAi screens reveal cell type-specific phenotypes

We have designed and constructed a dsRNA library targeting 265 Drosophila kinases and kinase regulatory subunits (Figure ​(Figure1b;1b; Additional data file 1), which could be used to carry out a comparable functional analysis across cell types. Each kinome screen was carried out in duplicate, in 384-well plates using the bathing method [15] (in the absence of transfection reagent). In each case, following plating, cells were incubated for 5 days to allow for protein turnover, before being fixed and stained to visualize actin filaments, microtubules and DNA. Images were then acquired using an automated microscope (Figure ​(Figure1b).1b). For each cell line, dsRNAs causing defects in cell morphology were identified by eye and classified using a controlled vocabulary. All cell images and hit annotations are available online through the online FLIGHT database [16].

Genes yielding a similar RNAi phenotype when targeted using multiple non-overlapping dsRNAs are likely to represent true hits, based on the low chances of different dsRNAs sharing the same off-target effects. However, in cases in which one out of two dsRNAs targeting the same transcript(s) elicit a phenotype, one of the two must be either a false positive or a false negative [17]. As with classical genetic screens, our major concern was to identify the false positives amongst this set [17]. False positives can arise as the result of problems with the gene annotation or because of experimental artifacts. In addition, false positives can arise as result of sequence-specific off-target effects, due to short regions of homology to unintended secondary transcripts in long dsRNAs [18], even though the use of long dsRNAs is thought to minimize this problem through the generation of a diverse pool of small interfering RNAs [19]. To estimate false positive and false negative rates in this study, we focused our attention on genes in the screen yielding a phenotype when targeted with one out of two dsRNAs. Each of these genes was then targeted with a third independent dsRNA. This analysis identified 4 false positives out of a total of 22 hits in S2R+ cells, and a single false positive out of 15 hits in the BG2 cell screen (Additional data file 2). Based on these data, we estimate a false positive rate for our experiment of 7-18%, and a false negative rate of 13-27%, depending on the cell line. Importantly, two-thirds of the false negative results could be attributed to defects with the library RNAi plates or dsRNA quality, as assessed by agarose gel electrophoresis during library construction (Additional data file 2).

After elimination of false positives, 17.3% (46 out of 265) of the kinases screened yielded a visible phenotype in at least one of six cell lines (Additional data file 3). This hit rate was similar to that determined in a related screen [14], but varied considerably across lines (Figure ​(Figure2a;2a; Additional data file 3). Much of the variation in hit rates across cell lines is likely to reflect variation in the ease of identifying defects in cell morphology in each line, since all the phenotypes identified in the BG3-c1 cell line, which is prone to grow in clumps, were also seen in at least one of the better spread central nervous system (CNS) lines (Figure ​(Figure2b).2b). Similarly, there were only two genes that yielded an RNAi phenotype in S2 or Kc167 cells that did not show up as a hit in the screen in large, well-spread S2R+ cells (Figure ​(Figure2b).2b). By contrast, there were significant differences in the kinase requirements of hemocyte and CNS-derived lines (Figure 2c,d) as expected based on the differences in the form and gene expression profiles that separate these two sets of lines (Figure ​(Figure1a).1a). This indicates that both gene expression and function can be used as indicators of a common origin. Using these data, it was possible to identify a set of cell type-specific hits (Figure ​(Figure2d).2d). However, there was no detectable bias in the number of hits in each kinase class between the two different tissues of origin (Table ​(Table22).

Open in a separate window

Figure 2

Parallel RNAi screens reveal cell line-specific phenotypes. (a) Different cell lines exhibited different hit rates in RNAi screens (Additional file 3). (b) Venn diagrams depict the segregation of screen hits between related cell lines. (c) A Venn diagram depicts the classification of hits into three distinct classes: those that are hits in both CNS and hemocyte cell lines; those that are hits in neuronal cell lines only; and those that are hits in hemocyte cell lines only. (d) Hierarchical clustering of hits across cell lines (depicted in the form of a tree) was used to give a more detailed picture of the three hit classes. Two hits of particular interest, CG7236 and minibrain (mnb), are highlighted. Note that the relationships defined by the functional analysis (depicted in the form of a tree at the top of figure) mirror the relationships defined by the microarray analysis (see Table 1 for the Pearson correlation coefficients in each case).

Since our goal was the identification of cell type-specific differences in the regulation of cell morphology, we clustered hits across all six cell lines. This revealed three strong clusters (Figure ​(Figure2d).2d). The first (C1) contains genes that have a strong phenotype in almost all cell lines tested, and is enriched in genes that participate in fundamental cell biological processes such as cell cycle control (for example, cdc2, polo and ial). The second cluster (C2) contains genes that elicit a phenotype in cell lines of CNS origin, and the third cluster (C3) identified hits specific to hemocyte-derived cell lines. We focused our subsequent analysis on genes with morphological phenotypes specific to one tissue type of origin.

CG7236 displays a hemocyte-specific phenotype

The C3 cluster identified a cyclin-dependent kinase, CG7236 [20], which elicited an RNAi phenotype only when targeted in hemocyte cell lines. Cyclin-dependent kinases are known to regulate cell cycle-dependent changes in cell organization together with a host of other processes, such as RNA Polymerase II activity [21]. In hemocyte cell lines RNAi-mediated silencing of CG7236 led to the accumulation of large cells with multiple or enlarged nuclei (Figure ​(Figure3a),3a), as verified using independent dsRNAs and confocal imaging (Figure ​(Figure3a,3a, bottom panels). This suggests a role for CG7236 in the regulation of the cell division cycle. However, RNAi-mediated silencing of CG7236 caused no detectable change in the appearance of neuronal cell lines such as BG3-c2 (Figure ​(Figure3a),3a), even though a quantitative PCR (Q-PCR) analysis revealed that CG7236 is both expressed and effectively silenced by RNAi in both S2R+ and BG3-c2 cells (Figure ​(Figure3d).3d). CG7236 has not been studied in detail before, but was previously identified as a cell cycle kinase in an RNAi screen in S2 cells [22], and as having a cytokinesis defect in RNAi screens in Drosophila hemocyte cell lines [14,23,24]. By analyzing its function across cell types, our analysis suggests that CG7236 differs from many other kinases involved in cell cycle control in performing a cell type-specific function.

Open in a separate window

Figure 3

CG7236 and minibrain show cell line-specific phenotypes. (a) Silencing of the cdc2-related kinase CG7236 in S2R+ cells gives rise to large cells that frequently contain multiple nuclei or a single large nucleus, whereas silencing in BG3-c2 cells has no discernable phenotype. (b) Silencing of the DYRK family kinase minibrain in BG3-c2 cells causes an increase in peripheral actin and an increase in the number of protrusions per cell, whereas silencing in S2R+ cells has no phenotype. Also, the BG3-c2 cells forced to spread by plating on concanavalin A (ConA) exhibit large lamellipodia when in the presence of a non-targeting dsRNA, but not in the presence of mnb dsRNA. (c) Quantification of the mnb RNAi phenotype shows a significant twofold increase in the number of long finger-like protrusions around the cell body. (d) Q-PCR analysis reveals that CG7236 and minibrain are effectively silenced by RNAi reagents in both S2R+ and BG3-c2 cells. Error bars indicate the standard error of the mean.

Mnb modulates actin-based protrusions in CNS-derived celllines

The C2 cluster identified minibrain (mnb) as a gene that has a strong morphological RNAi phenotype in all neuronal cell lines tested, without eliciting a visible RNAi phenotype in hemocyte cell lines (Figure ​(Figure3b;3b; Additional data file 4). As before, the specificity of the RNAi phenotype was confirmed using two sequence-independent dsRNAs in BG3-c2 cells to minimize the chances of sequence-specific off-target effects [25]. mnb encodes an evolutionarily conserved member of the DYRK (dual specificity tyrosine-phosphorylation-regulated kinase) family of serine/threonine protein kinases [26]. It was first identified in Drosophila as a gene involved in post-embryonic neurogenesis, since all strong loss-of-function mnb mutants generate animals with behavioral defects and adult flies with a specific and marked reduction in the size of optic lobes and central brain hemispheres [27]. Furthermore, DYRK1A, a human homolog of mnb, has been mapped within the Down's syndrome critical region of chromosome 21 and is over-expressed in Down's syndrome embryonic brain [28]. These data support a specific role for mnb in the regulation of neuronal cell morphology.

The mnb phenotype was similar across all three neuronal cell lines tested (Figure ​(Figure3b;3b; Additional data file 4), with silencing of mnb expression leading to a significant (>2-fold) increase in the number of long finger-like protrusions around the cell body (quantified in Figure ​Figure3c),3c), and an increase in cortical F-actin levels, and reduced cell numbers (Additional data file 5). Significantly, such filopodia are absent from hemocyte-derived cell lines, but are seen in all CNS-derived Drosophila cell lines tested [4,7,29]. Moreover, they are superficially similar to actin-based protrusive structures seen embedded in the growth cones of migrating neurons [30,31], where such finger-like processes are thought to sense local cues to guide the migrating neuron to its target [32], whilst the large mesh-like lamellipodium in which they are embedded generates the forces required to drive the growth cone or cell forwards [33-35]. Given this role for mnb in shaping actin-based protrusions, we considered two explanations for its neuronal-specific phenotype. First, it is possible that mnb is not expressed in hemocyte-derived cells or that the dsRNA failed to silence the mnb expression in these cell lines. Q-PCR analysis revealed that mnb is expressed and effectively silenced by RNAi in both S2R+ and BG3-c2 cells (Figure ​(Figure3d),3d), ruling out this explanation. Second, we considered the possibility that the ability to visualize a morphological phenotype associated with the loss of mnb was dependent on the shape of the cells used in the analysis. To test whether this might be the case, we forced BG3-c2 cells to spread on a conconavalin A coated substrate (Figure ​(Figure3b,3b, right-hand panels). Although this led to the formation of broad lamellipodia in control (lacZ RNAi treated) BG3-c2 cells, it was unable to suppress the induction of ectopic filopodia induced by mnb depletion. Thus, we cannot attribute the failure of mnb dsRNA to elicit an RNAi phenotype in Kc, S2 an S2R+ cells to differences in their form. Instead, these results suggest that mnb specifically acts to inhibit the transition between filopodia and lamellipodia in CNS-derived cells. Since mnb plays a conserved role in neurogenesis [27], and has a strong cell morphological RNAi phenotype in CNS-derived cell lines, it seems likely that it represents a cell type-specific regulator of cell morphology and behavior.

Tissue culture

Six Drosophila cell lines were used in this study. Kc167 and S2R+ cells were grown in Schneider's medium (Invitrogen, Carlsbad, California, USA) with 10% heat-inactivated fetal bovine serum (JRH Biosciences, Kansas, USA) and 1% penicillin-streptomycin (Sigma-Aldrich, St Louis, Missouri, USA) at 24°C in treated culture flasks (Falcon from BD Biosciences, San Jose, California, USA). S2R+ cells were removed from culture flasks using Trypsin-EDTA (Invitrogen). S2 cells were grown in InsectExpress media with L-Glutamine (PAA Laboratories, Pasching, Austria). The BG2-c2, BG3-c2, and BG3-c1 cell lines were cultured with Shields and Lang M3 insect medium (Sigma) with fetal bovine serum and antibiotics. M3 medium was supplemented with insulin for BG3-c2 (10 μg/ml) and BG3-c1 (5 μg/ml) cells.

RNAi screening and automated image acquisition

For RNAi screens, cells in serum-free medium were plated into 384-well assay plates containing dsRNA (20 μg/ml final concentration) by the Thermo Scientific Matrix WellMate mutlidrop machine (Thermo Fisher Scientific, Hudson, New Hampshire, USA), centrifuged briefly, then incubated at 24°C for 15 minutes before addition of complete medium. Cells were grown for 5 days at 24°C. In each experiment, positive and negative controls (pebble/thread/SCAR/LacZ RNAi) were included. Cells were fixed for 10 minutes in 4% formaldehyde (Polyscience, Niles, Illinois, USA). After fixation, cells were permeabilized by washing with phosphate-buffered saline (PBS) containing 0.1% Triton-X-100 (PBS-T), then blocked with 5% bovine serum albumin in PBS-T for 15 minutes. Cells were incubated with primary antibody (α-Tubulin) in PBS containing 1% bovine serum albumin overnight at 4°C. Cells were then washed and incubated with secondary antibody (FITC anti-mouse IgG) combined with TRITC-Phalloidin and DAPI for 2 hours. After staining, cells were washed and stored in 0.1% sodium azide in PBS-T at 4°C sealed with Costar6570 Thermowell sealing tape.

Fluorescent images were acquired using an automated Nikon TE2000 microscope with a 20× objective and HTS MetaMorph software (Universal Imaging, Molecular Devices, Downingtown, Pennsylvania, USA) running an automated stage, filter wheel and shutter, and a cooled-coupled device camera (Hamamatsu, Nishi Ward, Hamamatsu City, Japan). Automated wide focusing was performed on the DAPI channel first. Images were acquired in three channels at three sites per well. All image data and annotations are available online through the FLIGHT database [16].

Two step reverse transcriptase Q-PCR

Cytoplasmic RNA was harvested from BG3-c2 and S2R+ cells using the RNeasy miniprep kit (Qiagen) according to the manufacturer's guidelines. SuperScript II Reverse Transcriptase kit (Invitrogen) was used to synthesize the first-strand cDNA according to the manufacturer's guidelines. Escherichia coli RNase H was used to remove RNA complementary to the cDNA. Q-PCR was performed using SYBR green (Invitrogen Molecular Probes) and an MX4000 real-time PCR machine (Stratagene, La Jolla, California, USA). SYBR green fluorescence was quantified using a serial dilution of template containing PCR products of known concentration. Relative abundance of transcript was normalized against control (rp49) RNA levels. Primer sequences (Eurogentec, Southampton, Hampshire, UK) for all genes can be found in Additional data file 8.

Cell number measurement

Cell number counts were used to gain a quantitative assessment of the mnb phenotype in S2, BG2, BG3-c1 and BG3-c2 cells. In each case, 1.5 × 10⁶ cells were treated with mnb or lacZ dsRNA in a 4-well-dish. On the fifth day, cells were counted in triplicate using a Beckman Z2 Coulter Counter. The average cell number and standard deviation are presented for each.