Conceived and designed the experiments: MT CES TJM. Performed the experiments: MT JDO SK. Analyzed the data: MT TX SK. Contributed reagents/materials/analysis tools: JDO AEC SR CES. Wrote the paper: MT TX CES TJM.
The authors have declared that no competing interests exist.
Kinesin-5 (also known as Eg5, KSP and Kif11) is required for assembly of a bipolar mitotic spindle. Small molecule inhibitors of Kinesin-5, developed as potential anti-cancer drugs, arrest cell in mitosis and promote apoptosis of cancer cells. We performed a genome-wide siRNA screen for enhancers and suppressors of a Kinesin-5 inhibitor in human cells to elucidate cellular responses, and thus identify factors that might predict drug sensitivity in cancers. Because the drug's actions play out over several days, we developed an intermittent imaging screen. Live HeLa cells expressing GFP-tagged histone H2B were imaged at 0, 24 and 48 hours after drug addition, and images were analyzed using open-source software that incorporates machine learning. This screen effectively identified siRNAs that caused increased mitotic arrest at low drug concentrations (enhancers), and vice versa (suppressors), and we report siRNAs that caused both effects. We then classified the effect of siRNAs for 15 genes where 3 or 4 out of 4 siRNA oligos tested were suppressors as assessed by time lapse imaging, and by testing for suppression of mitotic arrest in taxol and nocodazole. This identified 4 phenotypic classes of drug suppressors, which included known and novel genes. Our methodology should be applicable to other screens, and the suppressor and enhancer genes we identified may open new lines of research into mitosis and checkpoint biology.
Kinesin-5 (also known as Kif-11, Eg5 and KSP), is a plus-end-directed, tetrameric motor protein required for establishing spindle bipolarity during mitosis
RNA interference (RNAi) technology provides an efficient strategy to systematically test the role of individual genes in the response of live cells or model organisms to drug treatments
HeLa H2B-GFP cells
Two Dharmacon siRNA SMARTpool libraries were used for primary screening. Both were arrayed such that each library well contained one pool of four siRNA duplexes directed against one gene. A smaller library of 509 SMARTpools that covered most of the kinases in the human genome (Dharmacon siARRAY siRNA Kinases Library Thermo Fisher Scientific, Lafayette, CO) was provided as a generous gift by Pfizer Inc. (Groton CT) and was originally obtained from Dharmacon in the late fall of 2004. This was used mainly to optimize procedures, though we did recover some kinases as enhancer hits. A full human genome library of 21,121 siRNA SMARTpools, (Dharmacon siARRAY siRNA Library, Human Genome, G-005000-05, Thermo Fisher Scientific, Lafayette, CO) was obtained by the Harvard Medical School ICCB-Longwood Screening Facility in 2006. This was used for the full genome screen.
All library stocks were made up in 1x Dharmacon Buffer (20 mM KCl, 6 mM HEPES-pH 7.5, 0.2 mM MgCl2), transferred to twin.tec 384-well PCR plates (Eppendorf), sealed with PlateLoc Thermal Seals (Velocity-11), and stored at −20C. Concentrated Master Stocks of the siRNA libraries are stored at 10 uM or 20 uM. Aliquots from the Master Stocks were diluted in 1x Dharmacon Buffer to make 1 uM Screening Stocks of the libraries, which were also stored in twin.tec 384-well plates at −20C.
Control siRNAs were plated in duplicate in each library plate. Polo-like kinase 1 (PLK1, Dharmacon M-003290-01) was used as a positive control for mitotic arrest and apoptosis. Negative controls included siCONTROL #2(Dharmacon D-001206-13-05) non-targeting siRNA and siGLO RISC-Free siRNA (Dharmacon D-001220-01-05) to control for non-sequence-specific effects. A diagram of the plate layout for a typical ICCB-Longwood siRNA library stock plate is shown in
(A) siRNA library pools were plated in twin.tec 384 well plates with controls in column 3 (see key). (B) A typical cherry pick source plate for the KI5 enhancer and suppressor screens. Individual duplexes were plated randomly (one duplex per well) in twin.tec 384 well plates so that duplexes targeting the same gene were unlikely to lie adjacent to each other. The edge wells (two outside rows and two outside columns) were left empty in each cherry pick source plate. Control siRNA duplexes were plated in columns 20–22 (see key). (C) Heat map of enhancer screen plate after 48 hours using ratios of monopolar spindles to interphase cells; wells with a higher ratio are displayed as warmer colors (yellow), allowing visual identification of potential hits (D) Heat map of suppressor screen plate using ratios of monopolar spindles to total object count; the cooler color (blue) indicates a positive well. (E) Heat map showing an extreme example of edge effects in a sample plate from the enhancer screen.
Re-testing of primary screening positives was carried out using the four individual siRNA duplexes that comprise each siRNA SMARTpool. For initial confirmation, these duplexes were obtained from the ICCB-Longwood human genome siRNA duplex library (Dharmacon Human Genome siRNA Library, Thermo Fisher Scientific, Lafayette, CO, obtained in 2006), which comprises the approximately 84,000 individual siRNA duplexes corresponding to the 21,121 human genome SMARTpools. Individual duplexes are stored in Master Stocks at 10 uM or 20 uM, and in Cherry Pick Stocks at 1 uM, in twin.tec 384-well plates. Duplexes were picked from Cherry Pick Stock plates using an EVO75 (Tecan) liquid handler served by a BenchCel plate stacker (Velocity11) and arrayed in cherry pick source plates for screening. A diagram of the cherry pick plate layout used in both the enhancer and suppressor screens is shown in
Day 1: HeLa H2B GFP cells are transfected with siRNAs. Day 2: 24 hours after transfection, cells are imaged using a screening microscope. Transfection mix is removed, and media containing 60 nM Kinesin 5 inhibitor (K5I; Enhancer screen) or 1 uM K5I (Suppressor screen) is added. Days 3 and 4: 48 and 72 hours after transfection, cells are imaged to monitor phenotypic changes.
Cells were reverse transfected with each siRNA SMARTpool at a final concentration of 24 nM using HiPerFect transfection reagent (Qiagen). Automated transfections were carried out in a Bioprotect II biosafety cabinet (Baker) using a Velocity11 Bravo liquid handler for pipeting. A Wellmate (Matrix) was used for plate filling. Each siRNA pool was diluted to 429 nM by mixing 3 ul of 1 uM siRNA Screening Stock with 4 ul of OptiMEM I Reduced Serum Medium 1x (Invitrogen) in an intermediate 384-well twin.tec plate. Next, 3 ul of each diluted siRNA pool was added to 9.6 ul of OptiMEM and 0.4 ul of HiPerFect transfection reagent previously plated in each well of in a black, clear bottom-384 well assay plate (Corning). Transfection reagent and siRNAs were allowed to complex at room temperature for 10 minutes. HeLa H2B GFP cells (2500 cells per well in 40 uL of DMEM) were then added to the HiPerFect + siRNA complexes and incubated at 37°C and 5% CO2 for 24 hours.
siRNAs targeting the genes KIF11 and TTK were used as positive controls in the enhancer and suppressor screens, respectively. Negative controls included Dharmacon si
Cells were transfected with individual siRNA duplexes at a final concentration of 18 nM using HiPerFect transfection reagent, following a reverse-transfection protocol. Control siRNAs were plated in sextuplicate in each assay plate in Columns 20–23 as shown in
EMD534085, a potent and specific small molecule Kinesin-5 inhibitor (K5I) that targets the mitotic kinesin Eg5/KIF11 was used in this study
Images were acquired on an ImageXpress Micro high content screening microscope (MDS Analytical Technologies, formerly Molecular Devices) with a Photometrics CoolSNAP ES digital CCD camera and a 20x Plan Fluor ELWD objective. Four single wavelength GFP images (488 nm excitation, FITC filter cube, Semrock) were taken in each well to capture 200–600 cells per well. HeLa H2B-GFP cells were imaged without fixation. No environmental chamber was used with the ImageXpress Micro. Assay plates were kept in the incubator until imaging time and transported to the microscope in an insulated bag. On average, each assay plate was imaged at room temperature for 20 minutes every 24 hours. Samples were illuminated for 100 ms per image to obtain the optimal signal to noise ratio for automated phenotyping. The 100 ms imaging time at 488 nm is not long enough to significantly damage cells, as cell damage typically occurs at high intensities of Ultraviolet radiation (350 nm)
We analyzed acquired images with CellProfiler, an open-source image-analysis package
Each nucleus was classified as interphase, monopolar, or “other” (neither interphase nor monopolar). For phenotypic classification, we took a machine learning approach using CellProfiler Analyst, an extension of CellProfiler
Heat maps were used to visualize the data for quality control purposes. For the enhancer screen, heat maps were plotted for the ratio of cells in monopolar arrest to cells in interphase. Wells with a higher ratio are displayed as warmer colors (yellow, in
Both interphase and monopolar cells were classified and counted at 48 hours and 72 hours after transfection. The phenotypic counts were used to calculate enhancer or suppressor scores for each well, which were then ranked to identify the siRNA pools that were positive in each assay. An evaluation of the total cell count in each well at 24 hours was also included in the score formula to eliminate wells with extremely low cell density.
For an siRNA to score as positive in the enhancer screen, it must cause a significant increase in the ratio of monopolar to interphase nuclei (color-coded as red and yellow objects in
All cells were grown in 384-well plates and treated according to the primary screen assay protocol outlined in
The nuclei from each time point are color-coded to illustrate their CellProfiler classifications: yellow = interphase, red = monopolar, blue = unclassified objects. As illustrated in
The enhancer score is defined as
For an siRNA to score as positive in the suppressor screen, there must be a significantly lower percentage of monopolar arrested nuclei at 48 and 72 hours after transfection. Due to the high ratio of drug-induced apoptosis in the suppressor screen, fragments of DNA from dead cells often float on top of healthy interphase nuclei. Hence, we used a monopole to total nuclei count ratio to calculate the suppressor score. All statistics were calculated relative to population average of cells imaged at the same time points from the same experiment.
The suppressor score is defined as
The
For re-testing potential hits in secondary screens, a new baseline was established to account for enrichment of positively scoring wells. Multiple non-targeting siRNA oligos were placed as negative controls on each cherrypick source plate (
HeLa H2B-GFP cells were reverse transfected with individual siRNA duplexes or pools (both at 15 nM final concentration) in 96 well glass bottom plates (MatTek P96G-0-5-F). 24 hours after transfection, K5I was added to the cells, so that the final inhibitor concentration was 1 uM. To prevent evaporation during imaging, 50 uL of mineral oil was added to each well. Starting immediately after drug and mineral oil addition, the cells were imaged at 10 minute intervals for 48 hours on an automated inverted fluorescence microscope (Nikon TE2000E) fitted with an incubation chamber and automated focus. Both GFP and phase contrast images were collected. The chamber was kept at 37°C with 5% CO2 throughout imaging. For each siRNA pool and duplex reagent tested, 50–100 cells were scored visually for phenotypic changes during imaging.
HeLa H2B-GFP cells were reverse transfected by hand with individual siRNA duplexes and pools (both at 10 nM final concentration) in 384 well plates (Corning Costar #3712). 24 hours after transfection, cells were imaged to determine initial cell density. Following removal of the transfection mix, media containing either 150 nM of taxol (Paclitaxel) or 300 nM of nocadazole was added to the transfected cells. Cells were then incubated for an additional 48 hours and imaged for phenotypic changes at 24 hour intervals. The images generated were scored by visual inspection.
Our goal was to find proteins that, when their levels were reduced, made cells either more, or less, sensitive to small molecule inhibition of Kinesin-5. We hoped to find proteins that modified cell responses to the drug at the level of mitotic arrest, and also at the level of apoptosis following mitotic arrest, though in practice our screen was more effective for finding the former. The concept was to pre-treat cells with siRNAs targeting one protein, then to add drug at either a low concentration, scoring for enhancement of the drug effect, or at a high concentration, scoring for suppression. We elected to omit a no drug arm to save on reagents, and distinguished true enhancers from siRNAs that scored without drug in follow-up assays. The drug response plays out over several days in unsynchronized cells. At ∼24 hours after adding drug there is a peak in mitotic arrest, and at ∼48 hours a peak in apoptosis
We chose high content cell imaging to read out our screening assay, an information rich method that facilitated monitoring of multiple phenotypes. We then considered whether to perform a fixed endpoint or time-lapse assay. Typically, an imaging screen for mitotic phenotypes would be performed as a fixed endpoint assay in microplates with immunofluorescence staining for DNA and indicators of mitotic arrest and perhaps also apoptosis
We developed a compromise strategy, intermittent live cell imaging. By imaging the same wells at several time points, we benefited from the ability to score the same wells immediately before drug, during peak mitotic arrest (∼24 hours in drug) and peak apoptosis (∼48 hours in drug), but can still use relatively standard image collection and analysis methods developed for fixed cells. This sparse time-lapse strategy should be useful for any assay where cell responses play out over a few days.
HeLa cells were chosen for the assay because they are easily transfected and a line expressing Histone H2B-GFP was available (Kanda et al., 1998, kind gift of Randy King, HMS). The cell phenotypes we needed to score have unique nuclear morphologies (interphase nuclei vs. mitotic chromosomes). Recent advances in computational methods have proven effective at using a single nuclear channel to quantify these phenotypes
The image analysis program CellProfiler Analyst
Interphase nuclei display relatively uniform intensities, a property that allows straightforward detection by the descriptors with 95% accuracy. While larger intensity variations are observed between monopolar cells due to their slightly elevated focal plane, the accuracy of the descriptors is between 75–90% depending on image quality. Although bipolar mitotic spindles were another major phenotype observed in the screen with highly accurate descriptors, its classification was omitted due to uncertainty introduced by the relatively short duration of this phenotype.
Some cell density variation was observed after cells were plated with the Matrix WellMate. The distribution followed a Gaussian distribution slightly skewed to the right. The cell count cut-off for the step function in our formula to calculate Enhancer/Suppressor scores was set to 100, which excluded 251 (1.2%) and 362 (1.7%) low cell count wells from the enhancer and suppressor arms of the whole genome respectively.
To classify the phenotypes produced by each siRNA and to rank genes for further follow-up, data from CellProfiler were used to calculate Enhancer or Suppressor scores for each siRNA as described in detail in Materials and
Twenty-four hours after transfection, 60 nM of Kinesin-5 inhibitor was added to the cells to screen for siRNA enhancers of K5I activity. Based on pilot experiments, K5I administered at 60 nM induced no obvious increase in monopolar spindles. As shown in
A primary screen of 21,121 siRNA pools targeting most genes in the human genome identified 336 (1.6%) siRNAs that caused a significant increase in monopolar mitotic arrest in 60 nM K5I. Each pool targeted one gene and contained 4 individual siRNA duplexes. To confirm the K5I enhancer phenotypes, all four individual siRNAs from each preliminary screening positive pool were re-tested in the original assay, in duplicate. We found that at least one of four of the siRNA duplexes recapitulated the primary screen phenotype in 150 (45%) of the 336 pools followed-up in this way. Sixty-one (18%) of the positive pools confirmed with at least 2 out of 4 duplexes scoring per pool (
The slices of each pie chart indicate the percent of genes for each screen for which 4/4, 3/4, 2/4, 1/4, or 0/4 of the individual siRNA duplexes reproduced the enhancer or suppressor phenotype observed for the corresponding siRNA pool. For the enhancer screen, 4 individual siRNA duplexes for each of 336 siRNA pools that scored as hits in the primary screen were tested in the same assay as used for the primary screen. For the suppressor screen, the individual duplexes for each of 337 siRNA pools were tested.
To identify and eliminate siRNAs that arrest cells in mitosis independent of low-dose K5I treatment and are thus not true K5I enhancers, we monitored siRNA-induced mitotic arrest in the absence of drug for 90 of the 150 genes for which at least one out of four individual duplexes were shown to enhance K5I activity. We chose these 90 siRNAs based on their relatively strong enhancer phenotypes. siRNAs for 38 of the 90 genes were found to cause arrest of cells in mitosis in the absence of K5I treatment (see
Among the genes targeted by siRNAs that score as K5I enhancers, we found numerous genes involved in regulation of mitosis and cytokinesis. Some of these, including PLK1, STK6 (Aurora A Kinase, AURKA), SGOL1 (shugoshin), CDCA5 (sororin), CKAP5 (cytoskeleton associated protein, CH-TOG), and KIF11 (the target of our test drug), caused some increase in the fraction of cells in monopolar mitotic arrest with siRNA knockdown alone. These are all proteins known to play essential roles in mitosis, which, if completely knocked down, should on their own cause very strong mitotic arrest. Thus, while they score as K5I enhancers in the screen, they are better thought of as essential mitotic proteins. We surmise that the siRNA knockdown is incomplete in most cases, and that low drug concentrations have an effect on mitotic phenotype that is additive with partial loss of protein. This consideration complicated analysis of the enhancer siRNAs, and we therefore prioritized K5I suppressor siRNAs for further analysis. A complete list of the K5I enhancer siRNAs identified in our screen is provided in
For the suppressor screen, transfected cultures were treated with 1 uM K5I, which causes 80–90% of cells to accumulate in monopolar mitosis after 24 hours of drug treatment (
Nuclei from each time point are color-coded to illustrate their CellProfiler classifications: yellow = interphase, red = monopolar, blue = unclassified objects. As illustrated in
Primary screening revealed 337 (1.6%) potential suppressor siRNA pools. In secondary screening of individual siRNA duplexes (4 duplexes/siRNA pool), 198 (59%) of the pools that scored as positive in the primary screen retested in duplicate with at least 1 duplex scoring for each pool, and 95 (28%) of the pools retested with at least 2 out of 4 duplexes scoring per pool (
False positives due to “off-target” effects of RNAi are a notorious problem in siRNA screens in human cells with current technology, and no completely satisfactory method is yet available to eliminate them. We chose the simple expedient strategy of prioritizing the 31 genes for which 4/4 or 3/4 individual siRNA duplexes demonstrated the K5I suppressor phenotype in the secondary screen. This should enrich for true positives, though it will no doubt eliminate many. We added one other gene, KATNB1 (a subunit of the microtubule severing enzyme Katanin) that scored with 2/4 duplexes, because of its interesting function.
The mitotic arrest and apoptosis phenotypes of high K5I could potentially be suppressed by multiple mechanisms, and we used time-lapse imaging to classify the 32 prioritized primary hits by phenotype. HeLa H2B GFP cells were transfected with the four individual duplexes, and also the pool of 4, at 15 nM total siRNA, a lower concentration than the primary screen to further select for on-target effects. 24 hours later 1 µM K5I was added, and images were collected every 10 minutes for 48 hours using an automated inverted microscope fitted with an incubation chamber. Movies were scored by visual inspection for three aspects of phenotype: entry into mitosis, time in mitosis, and whether cells divided after exit from mitosis, or simply flattened out due to failed cytokinesis. Visual inspection was used to identify genes for which at least 2 siRNAs had a strong phenotype (15 genes), and for preliminary classification. The data in
HeLa H2B-GFP cells were transfected with siRNA pools and individual siRNA duplexes directed against the indicated genes or with a non-targeting control siRNA. “Mock” cells were treated with HiPerFect transfection agent in the absence of siRNA. 24 hours after transfection, K5I was added to all of the samples, so that the final inhibitor concentration was 1 uM. Cells were imaged at 10 minute intervals for 48 hours on an automated inverted fluorescence microscope and automated focus. 50–100 cells from each condition were scored by visual inspection for (A) the average duration of mitosis in the population, and (B) the percent of cells going through cytokinesis upon exit from mitosis. For each of the genes shown, both the siRNA pool and the individual siRNA duplexes were observed to cause the same phenotypes, but only data for the siRNA pools is graphed here.
A: Drug specificity tests. HeLa H2B-GFP cells were transfected with siRNA pools as indicated, or “Mock” transfected in the absence of siRNA. 24 hours after transfection, medium containing either 150 nM of taxol or 300 nM of nocadazole was added to all of the samples. Cells were then incubated for an additional 24 hours and imaged for phenotypic changes. The pools were scored visually for cells in mitotic arrest, and this value was quantified for the pool by counting >100 cells. (B) Phenotypic categorization of hits on the basis of data in
By visual inspection, we were able to divide our hits into three phenotypic classes: I) failure to enter mitosis, II) brief mitotic arrest without cytokinesis, and III) brief mitotic arrest with cytokinesis. We will refer to category III siRNAs as “true suppressors.” The only strong hit in category I was RRM2, a subunit of ribonucleotide reductase that is essential for progression through S-phase. Blocking cells in S-phase was expected as a mechanism for lowering the mitotic index in K5I. We expected many such hits, but only RRM2 passed all our prioritization criteria, probably because cells would have had to arrest within ∼12 hours of addition of siRNA treatment, which would select for very unstable and/or dose-sensitive proteins. To measure the penetrance of the siRNA phenotypes for categories II and III, we scored the average duration of mitosis arrest compared to control siRNA (
To further categorize the genes that shortened mitotic arrest, we tested their ability to suppress mitotic arrest caused by Paclitaxel (which blocks microtubule dynamics) and nocodazole (which depolymerizes microtubules). Both drugs arrest cells in mitosis by activating the spindle assembly checkpoint (SAC), like K5I. We tested all the genes in
An important caveat in discussing any of the new genes in
In the SAC defect (all drugs) category, we recovered 4 known SAC genes: TTK (Mps1 kinase), BUB3, BUB1B, and MAD2L1. We missed a few other well-annotated SAC genes, including MAD1, presumably because the knockdown was insufficient in our time window. Two other genes scored with the same phenotype, NPC2 and PMAIP1. NPC2 is an endosome protein implicated in cholesterol transport
The SAC defect (K5I only) category was unexpected, and the gene annotations provide little in the way of mechanistic clues. DONSON is, by sequence, the probable ortholog of the Drosophila gene humpty dumpty, which has been implicated in control of DNA replication
Category III, the true suppressors, were particularly interesting from a spindle assembly perspective. This phenotype requires that the cell assemble a normal bipolar spindle even though Kinesin-5 is inhibited, or alternatively, this class might by due to increased drug efflux or metabolism. KIFC1 (also known as HSET), a mitotic kinesin that works opposite to Kinesin-5, was expected with this phenotype, since previous work has demonstrated that loss of KIFC1 function can rescue spindle bipolarity when Kinesin-5 function is compromised
Our intermittent imaging approach was effective for identifying siRNA enhancers and suppressors of K5Is that act at the level of spindle assembly and mitotic arrest, and should be useful for other cell cycle related screens. It was less effective at identifying genes that influence apoptotic responses, perhaps because we were not able to develop a specific score for apoptosis using H2B-GFP imaging. Characterization of siRNAs that score as K5I enhancers was complicated by the fact that many of those siRNAs had some mitotic phenotype on their own, in addition to acting as enhancers of drug action. However, the enhancer siRNA hit list is strongly enriched in genes with mitotic functions, and may reveal other genes not previously implicated in spindle bipolarization. Our K5I suppressor siRNAs broke fairly cleanly into 4 phenotypes (
(0.15 MB TIF)
List of genes with at least one out of four oligos confirmed in the same assay as used for the primary screen.
(0.06 MB XLS)
We thank the staff of ICCB-Longwood Screening Facility and members of the HMS mammalian RNAi screening community for helpful advice and encouragement. In particular we would like to thank Jade Shi for reagents and protocols, and Laura Selfors and Amanda Birmingham for bioinformatics advice.