|Home | About | Journals | Submit | Contact Us | Français|
Acquisition of invasive cell behavior underlies tumor progression and metastasis. To define in more molecular detail the mechanisms underlying invasive behavior, we developed a high throughput screening strategy to quantitate invadopodia; actin-rich membrane protrusions of cancer cells which contribute to tissue invasion and matrix remodeling. We developed a high content, imaged-based assay, and tested the LOPAC 1280 collection of pharmacologically active agents. We found compounds that potently inhibited invadopodia formation without overt toxicity, as well as compounds that increased invadopodia number. One of the two compounds that increased both invadopodia number and invasive behavior was the chemotherapeutic agent paclitaxel, which has potential clinical implications for its use in the neoadjuvant and resistance settings. Several of the invasion inhibitors were annotated as cyclin-dependent kinase (cdk) inhibitors. Loss-of-function experiments determined that Cdk5 was the relevant target. We further determined that the mechanism by which Cdk5 promotes both invadopodia formation and cancer invasion is by phosphorylation and down regulation of the actin regulatory protein caldesmon.
Most cancer deaths are caused by metastasis, a process in which cancer cells spread from the primary tumor and invade and grow in other organ sites. The transition of a cancer from benign to metastatic form requires the acquisition of a number of properties by the cell, including migratory ability and invasiveness (1). Invasive behavior facilitates tumor escape through the basement membrane barrier, as well as local invasion and remodeling of the tumor microenvironment in both the primary and metastatic sites (1). Both basement membrane degradation and local invasion are thought to require the activity of pericellular proteases.
In culture, invasive behavior is often monitored by evaluating the ability of cells to move through a layer of extracellular matrix (ECM), typically matrigel, in a transwell chamber. Degradation of the ECM can also be measured by plating cells on thin films of fluorescent matrix proteins, and evaluating its degradation after a set period of time. A third way takes advantage of the correlation between degradation of the ECM and the presence of actin-rich membrane protrusions known as invadopodia. Many human cancer cells, including breast cancers, melanoma, squamous cell carcinomas of the head and neck, and glioblastomas can form invadopodia (2). A large number of studies have correlated the ability of cells to form invadopodia with invasiveness in vitro and in vivo. As an example, cortactin, known to be required for correct functioning of invadopodia, has been shown to promote tumor invasion and metastasis using tumor xenograft studies (3, 4). Also, when cells lacking the invadopodia scaffold protein Tks5 were injected into mice, the growth of both primary tumors and lung metastases were severely inhibited (5).
We wanted to take a high throughput approach to identify regulators of cancer cell invasive behavior. Cell-based assays using a microscopy platform have become increasingly popular for such tasks. For example, a phenotypic assay was recently conducted to screen for genes that regulate epithelial cell migration (6). The goal of such assays is to provide direct measurement of changes in intracellular target expression or distribution, using reporter systems such as immunofluorescence. By applying a combination of automated fluorescent microscopy and computational analysis to cell based assays, it is possible to obtain multiple cellular readouts from a single experiment. Such High Content Screening (HCS) is particularly suitable to identify invasion regulators. Invasion is a complex, highly regulated and incompletely understood process, and a cell-based assay allows a comprehensive analysis. Here we describe a new high-content screening approach to identify previously unknown regulators of cancer cell invasion.
We first evaluated the suitability of the three types of invasion assay for high throughput screening. The transwell chamber assay is expensive, time consuming, difficult to quantitate, and is not available in a high throughput format. Quantitating degradation of films of matrix proteins is possible, but it is extremely challenging to reproducibly and evenly coat high throughput optical screening plates with ECM proteins. We therefore opted to establish the methodology to quantitate invadopodia in a high content format (see Materials and Methods). We chose not to use human cancer cells for the primary screen, since not all cells in the population elaborate invadopodia at any given time, they are fewer in number, and they are frequently obscured by the nucleus, the actin cytoskeleton, or both, making automated detection difficult. Instead we chose a more robust cell line for the initial screen; NIH-3T3 cells transformed with activated Src (referred to as Src-3T3). These cells, in which invadopodia were first discovered (7, 8), uniformly contain a large number of invadopodia, which are frequently arranged in characteristic rings or rosettes. Similar rosette structures have been visualized in both vascular smooth muscle cells (9) and human cancer cells (10). The observation that podosome formation in normal cells and invadopodia formation in human cancer cells both depend on Src (2), provides further justification for the choice of Src-3T3 cells as our model system. Briefly, Src-3T3 cells were plated and grown in 384-well optical plates for 3 days, incubated with compounds for a further 16 hours, then fixed and stained with DAPI to visualize nuclei, and phalloidin to stain F-actin (Fig. 1, A and C). In order to obtain the final assay read outs, the following steps were performed, as outlined in Fig. 1, A to H, and more extensively described in the Materials and Methods. We subtracted the image background for the red channel (F-actin) and applied a 3-by-3 median filter to smooth the actin images (Fig. 1D). Then, the nucleus of each cell was identified from the DAPI images (Fig. 1, B and E), and cellular regions were assigned to each nucleus using Cytoshop (Fig. 1G). Detection of the rosettes of invadopodia was performed on the smoothed F-actin images using the Cytoshop “Aggregate Segmentation” algorithm (Fig. 1F). The number of detected fluorescent “aggregates” represented invadopodia rosettes and was reported for each cell object as the “Number of Detected Rosettes per Cell” parameter. The distribution of the cell-by-cell values was bell-shaped (frequency plot in fig. S1A), and a D’Agostino & Pearson omnibus normality test indicated non-normal distribution (p<0.0001). However, there was a strongly statistically significant difference (p<0.0001) between the inhibited and uninhibited cell populations, independent of whether a parametric (Unpaired t-Test, assumes Gaussian distribution) or non-parametric (Mann-Whitney Test, does not assume Gaussian distribution) test was utilized. Additionally, the 95% confidence intervals were well separated (fig. S1B). The average number of detected rosettes of invadopodia per cell was calculated for each well (Fig. 1H) and the results for all screened compounds are plotted in fig. S1, C and D. Additionally, several other metrics were calculated on a per cell basis and reported as averages per well, including metrics representing nuclear area, nuclear shape, DNA content, and chromatin condensation (Fig. 1, A to H).
Having established the assay, a library of 1280 pharmacologically active compounds (LOPAC 1280) was screened in duplicate at a final compound concentration of 10µM (high compound concentrations are typically used in primary screens to capture all possible active compounds). Each 384-well plate contained 32 positive (10µM Src inhibitor PP2) and 32 negative (0.5% DMSO) control wells. To assess the robustness of the screen, the Z’-factor (defined in Materials and Methods) was determined for the screen plates. A Z’-factor of 1 is ideal, and a Z’-factor between 0 and 0.5 indicates acceptable robustness if the screen is performed in duplicate (11). The Z’-factors for the screen plates ranged from 0.2 to 0.6, resulting in an average Z’-value of 0.446, perfectly acceptable for cell based screens. For some plates, we noticed that there were outlier wells caused by plate preparation mis-dispensing; these wells were removed from the analysis.
To determine the active compounds in the primary assay, several metrics were evaluated for the duplicate wells for each compound, as summarized in Fig. 1I. First, wells with an increased number of dead/dying cells, and wells with cell counts <250 in the imaged area were identified from the DAPI images (fig. S1C, Supplementary Table 1), flagged “inconclusive/cytotoxic” and removed from the active compound list. Compounds were considered invadopodia inhibitors if the average number of detected rosettes per cell was less than 3 for both duplicate wells, and invadopodia activators if the average number of detected rosettes per cell was greater than 6 for both duplicate wells (Supplementary Table 2). The screen resulted in an initial active compound list of 20 inhibitors and 8 activators.
As is typical in such a screening assay, the images of the wells containing active compounds were next examined visually to exclude wells with technical artifacts such as cell clusters or weak F-actin staining. The final visually confirmed active compound list consisted of 5 inhibitors and 2 activators (Fig. 1I). This list contained two Src-inhibitors [SU6656 and the 7-Cyclopentyl-5-(4-phenoxy)phenyl-7H-pyrrolo[2,3-d]pyrimidin-4-ylamine (RBI)] and the MEK inhibitor UO126, which are known to inhibit invadopodia formation (2) and thus provided good internal positive controls for the assay and the analysis. The other two inhibitors identified were purvalanol A and GCP74514A. Both are annotated as cyclin-dependent kinase inhibitors. The LOPAC 1280 collection contains three additional cyclin-dependent kinase inhibitors. Two of them, roscovitine and kenpaullone, had also scored positive but had initially been discarded from the final active compound list due to cytotoxicity at the high concentration used or edge effects. The final cyclin-dependent kinase inhibitor, indirubin 3-oxime did not inhibit in the primary screen. However, when we retested this compound using fresh stock solution we observed good inhibitory activity, indicating that this compound was a false negative in the primary screen, likely because of compound instability. Therefore, 5 out of 5 Cdk inhibitors inhibited invadopodia rosette formation. The 2 activator compounds identified in the primary screen were the microtubule stabilizer paclitaxel and the serine/threonine phosphatase inhibitor cantharidin.
To confirm the results of the primary screen, we next tested the effect of the compounds on the ability of cells to degrade a fluorescent film of gelatin. We observed profound reduction of FITC-gelatin degradation using the 5 invadopodia inhibitors, while the two activators increased ECM degradation (Fig. 1J and fig. S2A). The invasion of Src-3T3 cells through a layer of matrigel in a transwell chamber was also inhibited by both purvalanol A and GCP74514A (fig. S2B). Finally, we measured the effective compound concentration required to decrease by 50% (EC50) the number of rosette-positive cells (Supplementary Table 3). The low EC50 of both purvalanol A and GCP74514A (0.2µM) suggested that invadopodia rosette inhibition was unlikely to be an off-target effect. The EC50 required to increase invadopodia number per cell for the two activators was calculated at approximately 2µM for each compound (Supplementary Table 3), which is consistent with their efficacy on their known targets.
To test the effects of active compounds on invasiveness of human cancer cells, we used SCC61 cells (squamous cell carcinoma of the head and neck cells), in which the relative absence of actin stress fibers allows for more facile quantification of invadopodia. Purvalanol A and GCP74514A significantly reduced both the fraction of cells positive for invadopodia and number of invadopodia per cell. Paclitaxel and cantharidin, on the other hand, increased the number of invadopodia per cell (Fig. 2 A and B). We also performed the FITC-gelatin degradation assay in SCC61, C8161.9 (melanoma), OVCAR8 (ovarian cancer), and NCI-H460 (non small cell lung cancer) cell lines. Purvalanol A and GCP74514A inhibited gelatin degradation in the invasive cell lines (SCC61, C8161.9, H460), while cantharidin and paclitaxel increased FITC-gelatin degradation in all tested cell lines (Fig. 2, A and C and fig. S2C). In the case of paclitaxel, degradation appeared to occur at cell edges. To rule out pulling effects on the ECM, we treated the coverslips with the general metalloprotease inhibitor GM6001 (figure S2, E and F). While some pulling was visible, the majority of the paclitaxel effect was due to gelatin degradation.
In sensitive cells, paclitaxel is a highly effective apoptosis-inducing agent, so we next asked if the increased invasive behavior was a consequence of apoptosis. To address this, we used the ovarian cancer cell line TR-SKOV3, which is resistant to paclitaxel-induced apoptosis (12). Even though no cell death was detectable, paclitaxel treatment of TR-SKOV3 increased invadopodia formation (Fig. 2D) (as confirmed by co-localization with Arp2/3 and cortactin), FITC-gelatin degradation (fig. S2D) and invasion through matrigel in a transwell chamber (Fig. 2E). Finally, we wanted to address what would be the effect of combining an inhibitor and an activator of invadopodia. We found that the combination of paclitaxel with purvalanol resulted in inhibition of invadopodia formation and function (figure S2, E and F).
Purvalanol A and GCP74514A are inhibitors of the Cdk family. Many members of this family are involved in cell cycle regulation. We compared the EC50 of purvalanol A and GCP74514A for invadopodia inhibition and cell cycle arrest. We confirmed that the EC50 for invadopodia rosette inhibition was 0.2µM, while the EC50 for cell cycle arrest was determined to be 5µM (Fig. 3A and fig. S3). These data demonstrate that invadopodia regulation was independent of cell cycle inhibition, and that the target of the drugs responsible for invadopodia control was unlikely to be involved in cell cycle regulation. We hypothesized that Cdk5 might be required for invadopodia formation and function, since it is not involved in cell cycle control, has previously been implicated in the migration of neuronal cells (13), and is a known target of purvalanol A (14). We first determined expression and localization of Cdk5 and its activator p35. We observed that both kinase and activator were expressed in Src-3T3 cells, at increased abundance compared with the parental non-invasive 3T3 cells (Fig. 3B and fig. S4A). Moreover, we detected specific Cdk5 and p35 staining in invadopodia rosettes (Fig. 3C and fig. S4B). We next determined the effect of Cdk5 knockdown with small interfering RNA (siRNA). Transfection of Src-3T3 cells with either a pool, or 4 individual, Cdk5 siRNAs significantly reduced the number of invadopodia rosettes (Fig. 3, D and fig. S5A) and FITC-gelatin degradation (Fig. 3, D and E). The specificity of the silencing was supported by the observation that there was no change in the mRNA abundance of two related kinases, Cdk1 and Cdk2 (fig. S5B). Moreover, transfection with a plasmid encoding human Cdk5 together with the smartpool siRNAs rescued formation of the rosettes (fig. S5, C and D). To test the effect of low concentrations of purvalanol A on Cdk5 activity, we measured the phosphorylation status of a known Cdk5 substrate, serine 727 on Stat3 (15), using tyrosine 418 on Src phosphorylation as a negative control (fig. S6, A and B). Stat3 phosphorylation was inhibited by the same concentrations of purvalanol that inhibited rosette formation. To further confirm that Cdk5 was the target of purvalanol A important for invadopodia formation, we measured the compound concentration required to decrease by 50% the number of rosettes in control cells compared with those engineered to overexpress Cdk5 and its associated protein p35. In this experiment, the EC50 for purvalanol A in cells transfected with empty vector was 0.15µM. Overexpression of Cdk5/p35 increased the EC50 to approximately 0.8µM (Fig. 3F). Finally, we also tested the effect of knockdown of Cdk5 in SCC61 with three different shRNAs, which reduced invadopodia number per cell and the percentage of invadopodia-positive cells, and decreased FITC-gelatin degradation (Fig. 4A), concomitant with a reduction in the level of the protein (Fig. 4B). The extent of Src phosphorylation was unchanged in a representative Cdk5 shRNA stable cell line (fig. S6C). A similar requirement for Cdk5 was also observed in a different squamous cell carcinoma of the head and neck cell line (SCC25) (fig. S6D), suggesting that Cdk5 also controls invadopodia formation and invasion in human cancer cells.
To elucidate the mechanism by which Cdk5 regulates invasion, we used a bioinformatic database (Phosphosite, http://www.phosphosite.org) to search for Cdk5 consensus sequences (16) (S/T−P−X−K/H/R) in candidate proteins. We found that caldesmon, a calmodulin and actin binding protein, has at least one such sequence - S(527)PTK. Knockdown of Cdk5 with siRNA decreased phosphorylation of caldesmon at this site (as judged by extent of immunoblotting with a phospho527 antibody), and effected an increase in the total amount of caldesmon protein (Fig. 5A and fig. S7A) without affecting the abundance of mRNA (Fig. S7B). In contrast, knockdown of MEK1 had little effect on phosphorylation status, protein or mRNA abundance (Fig. 5A, and fig. S7B). We next asked whether Cdk5 regulates proteasome-dependent degradation of caldesmon. Treatment of Src-3T3 cells with the proteasome inhibitor MG132 increased the amount of caldesmon at both short (4 hours) and long (24 hours) time points (Fig. 5B and fig. S7C), suggesting that its stability is regulated by ubiquitination. Co-expression of wild-type caldesmon together with Cdk5/p35 resulted in reduced abundance of caldesmon, compared with transfection of caldesmon together with an empty vector (Fig. 5C and fig. S7D). This effect was partially rescued with short term treatment with MG132 (Fig. 5C and fig. S7D) (longer treatments were toxic to the transfected cells). To further probe whether phosphorylation of caldesmon on Ser527 by Cdk5 caused proteasome-dependent degradation, we designed DNA constructs for the expression of phosphorylation-deficient rat CALD1S527A (SA) and phospho-mimic CALD1S527D (SD) rat caldesmon-GFP fusion proteins. Immunoblot analysis using phospho527 and total caldesmon antibodies confirmed that the SA mutant was indeed not phosphorylated (fig. S7E). MG132 treatment did not increase the level of the SA mutant protein, whereas the SD mutant protein did accumulate under the same treatment conditions (Fig. 5D and fig. S7F). Together, these data demonstrate that phosphorylation of caldesmon on Ser527 by Cdk5 serves to down regulate its stability.
It has previously been reported that caldesmon can negatively regulate podosome and invadopodia formation (17, 18). To determine whether regulation of caldesmon is the primary function of Cdk5, we evaluated purvalanol inhibition of invadopodia in cells with and without caldesmon. Knockdown of caldesmon increased the EC50 of purvalanol more than 10-fold (from 0.15µM to 1.7µM – Fig. 5E), suggesting that Cdk5 inhibition only inhibited rosette formation in cells expressing caldesmon. Similar conclusions were reached when the effect of combining Cdk5 and caldesmon knockdowns in both mouse and human cancer cells was tested (fig. S8). Moreover, we observed a decrease in the number of rosettes of invadopodia after transfection of the non-phosphorylatable SA caldesmon construct when compared with wild-type and SD constructs (Fig. 5F). Finally, we wanted to determine whether inhibition of proteasomal degradation affected invasive behavior. Long term treatment with MG132 was highly toxic. However, 4 hour treatment with MG132, which was not toxic, inhibited FITC-gelatin degradation by more than 60%. Interestingly, prior knockdown of caldesmon using siRNA negated the inhibitory effect of MG132 (Fig. 5G). Together, these data are consistent with the hypothesis that Cdk5 activity is required for invadopodia formation only in cells that express caldesmon, and that Cdk5-mediated phosphorylation of caldesmon at Ser 527 destabilizes the protein to allow invadopodia to form.
We describe here the identification of small molecules that regulate invasive behavior. The initial goal of this research was to generate a system to identify inhibitors of invasion. However, from the primary screen we noticed that cantharidin and paclitaxel increased invasive behavior. Cantharidin is a general protein phosphatase-1 (PP1) and protein phosphatase-2 (PP2A) inhibitor (19, 20), and likely promotes invadopodia formation by inhibiting the dephosphorylation of key invadopodia components such as ERK-1 and -2 (21). Analogues of cantharidin have been reported to increase xanthine oxidase activity which can increase intracellular reactive oxygen (ROS) (22). We have recently demonstrated that ROS are necessary for invadopodia formation (23), suggesting that cantharidin-mediated induction of intracellular ROS might also promotes invadopodia formation.
The cytoskeleton is the master regulator of several cellular functions, including migration and mitosis. Some of the most clinically successful cancer chemotherapies directly target one component of the cytoskeleton, microtubules (24). A good example is paclitaxel, which is used to treat patients with many forms of cancer (25). The anti-tumor effect of paclitaxel is based on its ability to bind and stabilize microtubules and consequently inhibit mitosis and induce apoptosis (26). However, formation of podosomes is dependent on intact microtubules (27), and it has recently been shown that mature invadopodia contain microtubules (28). We show here that paclitaxel treatment promotes the invasive behavior of a number of cancer cells through the stimulation of invadopodia formation. Of note, the concentration of paclitaxel required to promote invadopodia formation, 2µM, is an achievable clinical dose. These results raise the concern that continued treatment with paclitaxel in those cancer patients with refractory tumors, or neo-adjuvant treatment with the drug, may stimulate tumor cell extravasation and dissemination. In this regard, paclitaxel has been shown to increase metastasis formation in animals bearing paclitaxel-resistant leukemic cells (29). It has also been shown that paclitaxel increases the number of circulating tumor cells more than 1000-fold in breast cancer patients treated with paclitaxel in the neoadjuvant setting (30), which may be a result of extravasation of tumor cells from the primary tumor. Furthermore, in epithelial ovarian cancer, most patients initially respond to surgery and chemotherapy, but then relapse and become refractory to chemotherapy with more aggressive tumors associated with metastasis into the abdomen (31). Our observation that paclitaxel did not promote invasive behavior in cells treated with purvalanol suggests that combining paclitaxel with an invadopodia inhibitor might be explored as a means to limit tumor dissemination.
Several of the inhibitors detected in our screen were cyclin-dependent kinase inhibitors. This enzyme family has not been previously associated with control of either invadopodia or podosomes. Several members of the Cdk family act as checkpoint kinases to regulate cell cycle progression (32). But our EC50 analysis dissociated invadopodia inhibition from cell cycle inhibition. This led us to test Cdk5, which is also potently inhibited by purvalanol A and other Cdk inhibitors in the collection (14). Cdk5 is highly expressed in neurons and is involved in post-mitotic processes such as neuronal migration and neurite outgrowth (13). Cdk5 is expressed in many tissues besides neurons (33, 34). Our finding that Cdk5 regulates invasion is consistent with a previous report that expression of dominant negative Cdk5 in prostate carcinoma cells reduces their ability to metastasize (35). And Cdk5 expression in glioblastoma cells is higher than in normal astrocytes and appears to play a role in glioblastoma cell migration and invasion (36). We used an siRNA-mediated knockdown approach to show for the first time that Cdk5 is required for the formation and function of invadopodia, and for invasion, in mouse fibrosarcoma cells, and in several human cancer cells. Very little is known about the regulation of Cdk5 expression in non-neuronal tissues. In the future it will be interesting to investigate the mechanism by which it is frequently over-expressed in human cancers. Interestingly, it was recently shown that the Cdk5-mediated phosphorylation of talin prevents its ubiquitination and degradation (37), resulting in limited focal adhesion turnover and stabilization of lamellipodia at the leading edge, thus promoting cell migration. Taken together, these studies suggest that Cdk5 promotes cell migration and invasion by optimizing the formation of both focal adhesions and invadopodia.
Cdk5 has been reported to phosphorylate a large number of proteins in vitro and in vivo, many involved in cell morphology and motility, including ezrin (38) and dynamin (39), both invadopodia proteins (40–42). Here we provide new mechanistic insights into the role of Cdk5 in invadopodia formation, and propose that Cdk5 promotes invasive behavior by phosphorylating and down regulating the invadopodia suppressor caldesmon. In support of this hypothesis, we show that in the absence of caldesmon, inhibition of Cdk5 has little effect on invadopodia formation. We describe for the first time that Cdk5 controls caldesmon protein abundance, predominantly through phosphorylation of serine 527. Caldesmon is an actin-binding protein present in both smooth muscle, where the long form (h-caldesmon) predominates, and non-muscle cells, where only a short form (l-caldesmon) is detectable (43, 44). In vitro evidence implicates caldesmon in the regulation of smooth muscle and non-smooth muscle contraction and cell motility (45). Furthermore, caldesmon localizes to the invadopodia in Src-transformed cells (46), and reduction in the level of caldesmon facilitates podosome (17) and invadopodia formation and increases invasive behavior (18). Phosphorylation of caldesmon at S527 has previously been reported to correlate with ERK1/2, Cdc2 and p38 MAPK activity (47–49), although these studies used high concentrations of inhibitors, and did not address the consequence of caldesmon phosphorylation. In this study we have provided strong evidence that the primary S527 kinase in cancer cells is Cdk5, and that phosphorylation on this site destabilizes the protein, thus promoting invadopodia formation. This conclusion is supported by the observation that MG132-mediated inhibition of FITC-gelatin degradation only occurred in cells expressing caldesmon. These experiments also provide the first link between protein ubiquitination and ECM degradation.
Our studies demonstrate the feasibility of using high content invadopodia screens to identify invasion regulators, and highlight the importance of both the microtubule and actin-based cytoskeletons in controlling invasive behavior. Invadopodia-related structures, called podosomes, have been observed in monocyte-derived cells, such as macrophages and osteoclasts, in endothelial cells, and in vascular smooth muscle cells (2). Dysregulation of podosome formation has recently been associated with development of atherosclerosis (9). These results suggest that a strategy to identify regulators of invadopodia and podosomes might lead to the development of new therapeutic drugs to control metastatic cancer growth, as well as other diseases.
Src-3T3 cells were grown in DMEM, 10% FBS, and antibiotics, plated at a density of 200 cells/well in 25µl using a Matrix Wellmate liquid dispenser, in black 384 well plates (F-bottomed, µClear; Greiner Bio-One, GmbH), and cultured at 37°C with 10% CO2. After 72 hours, compounds were added to a final concentration of 10µM for a further 16 hours. Src-3T3 treated with the Src inhibitor PP2 was the positive control and Src-3T3 cells treated with vehicle (0.5% DMSO) was the negative control. Screening was performed in duplicate.
The LOPAC 1280 library (Sigma Aldrich) is a collection of 1280 pharmacologically active compounds from 56 pharmacological classes with well-characterized activities. Compounds were dissolved in DMSO, and transferred to 384-well microplates using the Biomek FX liquid handler (Beckman Coulter Inc.). The plates were stored at −70° C.
Cells were fixed after removal of the media with a wand aspirator (VP Scientific) with the Beckman Coulter Biomek FX liquid handler. 50µl fixation solution (PBS 1×, 3% PFA) was added for 15 minutes. Wells were aspirated and incubated for 20 minutes with 0.1M glycine; washed 3× with PBS, then incubated with 0.1% Triton-X 100 in PBS for 15 minutes. Cells were incubated in PBS with 5% BSA for 30 min, followed by staining with Alexa568-phalloidin (Invitrogen) 1:1000 for 1 hour and DAPI 1:1000 per well for 15 minutes. Wells were washed 3× with PBS and imaged.
Images with a bit-depth of 8 bits were acquired using an IC100 High-throughput Microscopy (HTM) System (Beckman Coulter) with a Hamamatsu Orca-ER scientific CCD camera. Images were acquired using a Nikon 20× 0.5 NA objective and 2×2 camera pixel binning. Each resulting field had a dimension of 672×512 pixels, with an effective pixel width/height of ~0.645µm. Six fields (2×3) were acquired throughout each well at two wavelengths (Chroma 86013 filter set) accounting for nuclear DAPI staining and actin phalloidin staining respectively. Exposure times were ~300 ms for phalloidin and ~10 ms for DAPI.
The algorithm to detect rosettes of invadopodia was designed using Matlab (Ver. 7.01, MathWorks, USA) and Cytoshop HTM software (Beckman Coulter, USA). First, using a custom-built algorithm in Matlab, a smoothing filter was applied to the actin-phalloidin images and image background subtracted to remove some background detail of the actin fibers (Fig. 1D). Then, the nucleus of each cell was identified from DAPI images using Cytoshop’s “Nuclear Segmentation” algorithm, which included an “open” morphological operation to separate touching nuclei. Objects with object heights/widths of <5 pixels or >300 pixels were discarded. Next, equidistance tessellation lines were drawn between the centroids of the identified nuclei, breaking up the images into cellular regions as shown in Fig. 1G. Errors introduced by approximating the cell boundaries did not significantly affect the primary assay read-out, because the read-out was based on the cell population average of the detected rosettes per cell. Rosette detection was performed on the actin-phalloidin images using Cytoshop’s “Aggregate Segmentation” algorithm (Fig. 1F). The object scale for the aggregate detection, “aggregates” representing rosettes of invadopodia in this application, was set to 20 pixels (average rosette size) and the minimum intensity peak height was set to 20 grayscale values (minimum difference in intensity between rosette object and neighboring image intensities). Detected objects with area <10 or >1000 pixels were discarded as artifacts. Moreover, because of significant edge effects, rows A and P of each plate were invalidated. The number of detected rosettes was then reported for each cell object and the average number of rosettes per cell calculated for each well. Additionally, several other metrics were calculated on a per-cell basis and reported as average per well, including nuclear area, nuclear shape, DNA content, chromatin condensation, and average rosette size. The Z’ factor was used for quality control (QC) of the screen plates and is defined as described in the following equation: 1−[(3SDposcontrol + 3SDnegcontrol)/(Meanposcontro−Meannegcontrol)]. Statistical analysis of the rosettes of invadopodia data, the screen data and plate QC assessment was performed using Excel (Microsoft), Spotfire (TIBCO), and GraphPad Prism software.
Mouse 3T3 and Src-3T3 cells, SCC61 and SCC25 human head and neck carcinoma cells (a gift from Dr. Alissa Weaver), C8161.9 human melanoma cells were grown as previously described (10, 23). OVCAR8, SKOV-3 and NCI-H460 cells were grown in RPMI (MediaTech), 10% FBS. The paclitaxel-resistant SKOV3 cell line was generously provided by Dr. Zhenfeng Duan. Drug treatment of Src-3T3 cells was conducted by plating cells at 10% confluence (16000 cells in 6 wells), waiting 72 hours, and then adding compounds for 4–16 hours.
Lentiviral shRNA constructs targeting hCdk5 were from Open Biosystems (TRCN00000021465, TRCN00000021466, TRCN00000021467). Lentiviruses were generated by the SBMRI Lentiviral Core Facility. The siRNA constructs for mCdk5 (M-040544 pool and individual sequences) and non-targeting siRNA#3 (D-001210-03-05) were from Thermo Scientific. mMEK1 siRNA pool (1177649) was from Sigma. siRNA constructs were transfected with Lipofectamine 2000 (Invitrogen). Efficiency of RNA interference was monitored by RT-PCR or by immunoblotting.
Src-3T3 cells were transfected on glass coverslips and processed for immunofluorescence 60 hours later. Lentivirus-infected cells were assayed 5 days after selection with 2µg/ml puromycin. Invadopodia quantification was performed on at least 15 randomly chosen fields representing approximately 150 cells per experimental point. Src-3T3 cells containing at least one complete rosette of invadopodia were scored positive. SCC61 cells containing at least two punctate F-actin–rich podosomes were scored positive. Total cell numbers were calculated by scoring number of nuclei in the same field. Invadopodia function assays were performed as previously described (10). Src-3T3 cells were processed after 16 to 18 hours and human cancer cells after 24 hours. Quantification of gelatin degradation activity was performed on at least 15 randomly chosen fields, representing a minimum of 200 total cells per experimental point. Quantification of degradation area per field used ImageJ software (NIH), and percent of degraded area per field was normalized to the number of cells in the field. Invasion assays were conducted in matrigel invasion chambers (10), except that 50×105 cells were used. Paclitaxel was added to the cells upon plating in the upper chamber. Where specified, MG132 was added at a concentration of 20µM for 4 or 16 hours.
RNA, protein, and migration values were compared using two-tailed analysis of variance (ANOVA) test adjusted for multiple comparisons. P < 0.05 was considered to be statistically significant.
We thank Drs. Duan, Mak, Tsai and Weaver for cell lines and reagents, the SBMRI lentiviral facility for preparation of lentiviruses, Joseph Russo for assistance with confocal microscopy, and Drs. Azimi and Vasile for assistance with algorithm and assay development in the early stages of the project. This research was supported by the National Cancer Institute of the National Institutes of Health.
We have determined that paclitaxel promotes invasive behavior of cancer cells, and that Cdk5 promotes invasion by phosphorylating and destabilizing caldesmon, a negative regulator of invadopodia formation.
*This manuscript has been accepted for publication in Science Signaling. This version has not undergone final editing. Please refer to the complete version of record at http://www.sciencesignaling.org/. The manuscript may not be reproduced or used in any manner that does not fall within the fair use provisions of the Copyright Act without the prior, written permission of AAAS.
The authors declare no competing financial interests.