Register for Mechanobiology 2016!


β1 integrins constitute a large group of widely distributed adhesion receptors, which regulate the ability of cells to interact with their surroundings. This regulation of the expression and activity of integrins is crucial for tissue homeostasis and development and contributes to inflammation and cancer. We report an RNA interference screen to uncover genes involved in the regulation of β1-integrin activity using cell spot microarray technology in cancer cell lines. Altogether, ten cancer and two normal cell lines were used to identify regulators of β1 integrin activity. Cell biological analysis of the identified β1-integrin regulatory genes revealed that modulation of integrin activity can influence cell invasion in a three-dimensional matrix. We demonstrate with loss-of-function and rescue experiments that CD9 activates and MMP8 inactivates β1 integrins and that both proteins associate with β1 integrins in cells. Furthermore, CD9 and MMP8 regulate cancer cell extravasation in vivo. Our discovery of new regulators of β1-integrin activity highlight the complexity of integrin activity regulation and provide a set of new genes involved in regulation of integrin function.


Integrins are a family of cell surface adhesion receptors that regulate the ability of cells to interact with their surroundings (Gahmberg et al., 2009; Legate et al., 2009). In humans, there are 24 different integrins and 12 of these are α and β1 heterodimers mediating adhesion to various extracellular matrix (ECM) components. Integrin ligand binding leads to clustering of various adaptor proteins and kinases that together regulate actin polymerization and downstream signalling to MAP kinases (Gahmberg et al., 2009; Guo and Giancotti, 2004; Schwartz and Assoian, 2001). As integrins are important regulators of cell proliferation and motility, their dysregulation can be associated with inflammation or tumour progression (Desgrosellier and Cheresh, 2010; Shattil et al., 2010). For example, studies with conditional knock-out mouse models of breast carcinomas suggest that β1 integrins have an important role in primary tumour growth (White et al., 2004). Also, high expression levels of β1 integrins have been shown to correlate with poor prognosis in breast cancer (Yao et al., 2007), and signalling by β1 integrin might induce dormant cancer cells to form proliferative metastasis in vivo (Barkan et al., 2010). Furthermore, a recently described tumour suppressor protein SCAI suppresses cell invasion by repressing β1-integrin transcription and expression (Brandt et al., 2009).

Integrin activity (i.e. the ability to bind ligands) is strictly controlled by intracellular signals and this has broad biological importance in adhesion-dependent events in cancer (Desgrosellier and Cheresh, 2010; Shattil et al., 2010). In leukaemia for example, β2 integrin and β1 integrin can be constitutively activated, which contributes to altered leukocyte trafficking and adhesion to stromal components (Chen et al., 2008; Fierro et al., 2008).

The best-characterized regulatory step of integrin activation involves binding of cytoplasmic regulators, such as talin-1 and -2 and kindlin-1, -2 and -3 (also known as FERMT1, 2 and 3), to the cytoplasmic domains of integrins (Moser et al., 2009; Shattil et al., 2010). Talins are known to stabilize the active open conformation of integrins and also strengthen the integrin connection to the actin cytoskeleton, which brings about the clustering of multiple actin-binding proteins to adhesion sites (Moser et al., 2009; Shattil et al., 2010). These focal contacts can mature into focal adhesions, which can constitute an interactome of more than 150 components [ (Zaidel-Bar et al., 2007)]. Given the complexity of cell adhesion machinery and the fact that there is even some controversy about the well-characterized final steps of integrin activation (binding of cytosolic proteins to the β-tail) (Shattil et al., 2010), we set out to search for β1-integrin activity regulators. We employed our recently developed cell spot microarray (CSMA) technique (Rantala et al., 2010; Rantala et al., 2011) to perform an unbiased druggable genome-wide RNA interference (RNAi) screen. Our study identified new regulators of β1-integrin activity in 12 different cell lines. In addition, we identified a set of previously unknown integrin regulators, which might contribute to cancer cell invasion in vitro and in vivo. Thus, these data underscore the role of β1 integrins and their upstream regulators as possible therapeutic targets.

Fig. 1.

Results of the primary screens for β1-integrin activity regulators. (A) In high-density cell spot microarrays (CSMA) cells were silenced on Matrigel spots containing siRNA and transfection reagent during 48 hours incubation. The example image shows VCaP cells on arrays with antibody stainings for α2 integrin (ITGA2), active β1-integrin epitope (9EG7) and F-actin (phalloidin). The array was scanned with a Tecan laser scanner (large panel) and further imaged using fluorescence microscopy (enlarged images). (B) The druggable genome of 4910 genes with two siRNAs per target (Qiagen), together with controls (AllStars negative, n=118; GFP, n=117) was silenced in VCaP cell line as above and stained for α2 integrin, active β1 integrin (9EG7) and F-actin (phalloidin). Shown are z-scores for 9EG7 staining intensities normalized to total α2 integrin signal (dark blue for druggable genome siRNAs, yellow for AllStars negative and orange for GFP siRNAs) and phalloidin also normalized to α2 integrin (light blue) from each spot. Druggable genome siRNAs with z-scores of <−2 or >+2 are indicated with green and red, respectively. (C) The genes with z-scores of <−2 or >+2 from the screens where either 9EG7 or 12G10 antibodies (both for active β1 integrin) were subjected to ingenuity pathway analysis (IPA). The top portion of the canonical pathways are displayed as a bar chart. The canonical pathways that are involved in this analysis are shown along the y-axis. As the default, the x-axis shows the −log of P-value, which was calculated using a right-tailed Fisher's exact test. (D) The z-score median for each gene as analyzed from all the knockdowns in the 9EG7 and 12G10 primary screens (n=4). The 50 genes listed (25 up and 25 down) were chosen for secondary validation.


A druggable genome-wide RNAi screen for regulators of integrin activity

Specific monoclonal antibodies (12G10 and 9EG7), which distinguish active forms of β1 integrins (Byron et al., 2009), were used in combination with a neutral integrin antibody (anti-α2 integrin; this was chosen because α2β1 integrin is an abundantly expressed integrin in these cell lines, and using an α-subunit-recognizing antibody circumvents possible issues with overlapping binding sites between different β1 antibodies) to detect RNAi-induced changes in integrin activity in cancer cells grown over a Matrigel matrix (Fig. 1A). Because the β1-integrin antibodies could influence integrin conformation in live cells, it was important to use fixed cells when using the antibodies as indicators of siRNA effects on receptor activity. To validate our experimental approach we assayed the effect of silencing the genes encoding talin-1 and -2 in CSMA spots (supplementary material Fig. S1A,B). Silencing of the talin-2 gene (using the CSMA platform) significantly (P<0.05) reduced integrin activity (9EG7 staining) in vertebral-cancer of the prostate (VCaP) cells (supplementary material Fig. S1A,B). The silencing efficiency and staining specificity on the CSMA were further confirmed by silencing β1-integrin and α2-integrin subunits (supplementary material Fig. S1C).

We assayed the VCaP prostate cancer cell line with the druggable genome RNAi library (Qiagen v1.0) targeting 4910 human genes with two individual siRNA constructs per gene. Fixed cells were analyzed for 9EG7 and 12G10 antibody staining in separate experiments, each consisting of 9820 cell spots on a single plate with each position containing a single individual siRNA. Of these, significant changes with z-score standardized values of <−2 or >+2 were found in 4.5 and 4.4% of the siRNAs for VCaP cells (9EG7 and 12G10, respectively; Fig. 1B; supplementary material Fig. S1D). Importantly, z-scores for all replicate control siRNA (n=235) transfections were between −2 and +2 (Fig. 1B; supplementary material Fig. S1D, yellow and orange spots). Furthermore, RNAi effects on active integrin and phalloidin intensities showed high correlation co-efficiency (supplementary material Fig. S1D,E), which was anticipated because integrins are known to signal to the actin cytoskeleton and stimulate de novo actin polymerization or actin stress-fibre bundling (DeMali et al., 2003).

Ingenuity pathway analysis of the genes involved in the regulation of β1-integrin activity indicated G-protein-coupled receptor (GPCR) signalling as the major pathway influencing β1-integrin activity in the screen (Fig. 1C). This was not surprising because the GPCR activation by various cytokines and peptide hormones is one of the most well-characterized pathways in integrin activity regulation (Laudanna and Alon, 2006; Li et al., 2010). Interestingly, many signalling pathways that are involved in the regulation of inflammation (e.g. GPCR, NF-kB, RANK, interleukin, arthritis, glucocorticoid, PI3K–AKT and retinoid acid receptor) were identified as important regulators of integrin activity in the screen. This suggests that inflammatory signalling could be coupled to regulation of β1-integrin activity in several different ways.

Secondary screen validation of the integrin regulators

For validation and follow-up studies we selected 50 genes that significantly either activated (25) or inactivated (25) β1 integrins [median of the 9EG7 and 12G10 z-scores (n=4) less than −1.0 or more than +1.0; Fig. 1D; supplementary material Table S1 for z-scores and Table S2 for gene annotations]. Four new siRNAs for each gene were applied for the secondary validation (supplementary material Table S3) in order to ascertain the specificity of the effects. Each siRNA was printed in duplicate on the secondary CSMAs and screened for 9EG7 and 12G10 binding. Together these resulted in 16 technical replicates and/or observations per gene. The VCaP secondary screen showed that 34% of the primary screen hits could be consistently reproduced when a more stringent criterion was applied (two additional siRNA oligonucleotides scoring for each gene in duplicate; Fig. 2A; supplementary material Table S4). Silencing of five out of the 50 genes (ERCC1, CDK5R1, HDAC4, LCK, COL9A1) in the secondary screen resulted in an opposite effect on integrin activity to that in the primary screen, highlighting the importance of the secondary validation screen with more siRNAs.

To investigate the generality of our findings from VCaP cells, we silenced the 50 genes in seven other prostatic cell lines (PC3, ALVA31, 22RV1, MDAPCA2a, RWPE1, primary epithelial prostate cells and primary stromal prostate cells) and in four non-prostatic cell lines (lung cancers NCI-H460, A549 and colon cancers HCT-116, SW-480; z-scores in supplementary material Table S4). Importantly, most of the reproducible VCaP hit siRNAs induced similar effects on integrin activity in many of the other cell types. The best correlation between cell types was seen between prostate cancer cell lines PC3 and ALVA31 (R2=0.48). As a comparison, the lung carcinoma cell lines NCI-H460 and A549 had a correlation of R2=0.36. Some of the poor correlation between VCaP and other prostate cancer cell lines (PC3, ALVA31, 22RV1, MDAPCA2a) could be due to the fact that VCaPs are androgen sensitive whereas the rest are androgen insensitive (van Bokhoven et al., 2003).

From integration of these data, consistent regulators of integrin activity could be identified. Altogether, silencing of 13 genes resulted in downregulation and silencing of 10 genes involved in upregulation of β1-integrin activity in at least four cell lines (gene IDs highlighted in green for downregulation or red for upregulation in Fig. 2A). Silencing of EPS15 resulted in the most frequent integrin inactivation among the different cell lines. Silencing of MAST2, by contrast, induced integrin activation most often in the cell lines analyzed. Other high-scoring integrin-activity-regulating siRNAs targeted ATRN, ERCC1 and DHRS4 (all inactivating) and HDAC4, INPP1 and MSX1 (all activating). Many of these 23 regulators of integrin activity (including EPS15 and MAST2) have not, to our knowledge, been identified previously as being involved in the regulation of integrin activity.

Fig. 2.

Secondary screen validation of integrin activity regulators in multiple cell lines. (A) The colour map shows the results from 50 selected genes with their corresponding median z-scores from the primary VCaP screens and the secondary validation screens with 11 additional other cell lines. The numbers in the left column indicate the median z-score of four siRNAs (2×12G10+2×9EG7) in the primary screen (see Fig. 1D). siRNAs inducing downregulation or upregulation of β1-integrin activity in at least four cell lines are indicated with coloured gene IDs in the right column together with the total number of siRNAs affecting 9EG7 or 12G10 binding (z-score <−1 or >+1) in all of the cell lines studied (numbers in black on the right column). Top-scoring genes were also ranked on the frequency of siRNA effects on integrin activity across the 12 cell lines (RANK column: green, inactivating; red, activating). (B) Confocal microscopy images of VCaP cells on array spots, showing maximum intensity z-projections of α2 integrin (ITGA2), 12G10, phalloidin and DAPI staining. Intensity histograms were drawn from α2 (green) and 12G10 (red) stainings (right). Scale bars: 20 μm. (C) Higher magnification of control-siRNA-silenced VCaP cells with 12G10 and phalloidin stainings. Scale bar: 20 μm.

The integrin activity state can also be analyzed by measuring the integrin–ligand binding of living cells in suspension, although this situation can be very different from the cell spot microarray approach, in which cells are adhering to the ligand over a period of 2 days. To compare our CSMA screen results with a conventional ligand-binding assay, we silenced the hits of the PC3 cells and measured the binding to the soluble fibronectin fragment (repeats 7–10). Importantly, most of the siRNAs affecting 12G10 or 9EG7 staining in PC3 cells also influenced binding of the monomeric fibronectin fragment to live cells in a flow cytometric analysis, further confirming the effects of the siRNAs on integrin activity (supplementary material Fig. S2A,B). In addition to fibronectin binding, we analyzed the levels of total cell surface β1 integrin and 9EG7 epitope in suspended and parformaldehyde-fixed PC3 cells after treating the cells with a set of integrin-inactivating and integrin-activating siRNAs (supplementary material Fig. S2C). Silencing of CD9, EPS15 and the talin-1 and -2 genes all reduced the surface-exposed 9EG7 epitope, whereas silencing of MASTL, MAST2 and MMP8 increased it. Measurement of the total surface β1 integrin showed that knockdown of talin-1 and -2 also reduce the total β1 integrin in PC3 cells, but of the other knockdowns only siMASTL resulted in changes in surface β1 integrin by reducing it slightly. However, this approach did not show changes in 9EG7 binding after silencing of the AKT3 or HDAC4.

The expression of the 50 genes was also analyzed from Affymetrix HG-U133A gene expression microarray analyses of four prostatic cell lines (VCaP, 22-RV1, MDA-PCA2a, PC3) and A549 lung carcinoma cells (supplementary material Table S5). From these data it is obvious that the expression levels of the genes identified as regulators of integrin activity can vary substantially between cell lines and this is likely to influence the effects that specific siRNAs have in each given cell line. It is important to note that some known integrin regulators, such as C1B1, kindlins and filamins (Shattil et al., 2010), were not included in the predefined commercial library used in this study. Therefore, these genes were not identified in the screen. In addition, redundancy between protein family members (such as talin-1 and -2) and gene expression profiles in the cell lines used might have resulted in the absence of some previously described candidates from the highest scoring hits.

The phenotypes and subcellular localization of active β1 integrin in siRNA-treated cells were further analyzed with confocal microscopy. In VCaPs, silencing of ALDH4A1 and PRKG2 decreased 12G10 staining, whereas silencing of MAST2 and HDAC4 prominently increased it (Fig. 2B). A higher magnification of VCaP cells on CSMA spots revealed that the active integrin was also localized to small focal contacts on cell edges (Fig. 2C). Increased active-β1-integrin staining at cell edges and protrusions was also detected in 22RV1 cells after treatment with activating siRNAs (HIP1, IFNA1, MSX1), and the staining was reduced upon silencing of ATRN, EPS15 and CACNA1I (supplementary material Fig. S3A). Images of PC3 cells on ECM coating (supplementary material Fig. S3B) or cells grown on plastic (supplementary material Fig. S3C) revealed that in these cells 12G10 staining was highly vesicular with no apparent focal contacts or focal adhesions. However, also in these cells β1-integrin inactivating siRNAs (PFKFB2, AKT3, CD9, DHRS4, PFTK1, EZH2) and activating siRNAs (MYCL1, MSX1, NCAM1, IRS2, MAST2, LCK) resulted in marked changes in 12G10 staining, thus verifying the screening data from the cell spot arrays. We also confirmed the silencing efficiency of siRNAs for 20 different genes in the PC3 cell line by quantitative Taqman RT-PCR (supplementary material Fig. S4A).

CD9 and MMP8 associate with β1 integrins and regulate integrin activity

Next we chose one activating (CD9) and one inactivating (MMP8) gene for further validation. We found that CD9 protein is present in PC3 cells and the individual siRNAs used in the screen readily silenced CD9 (Fig. 3A). In addition, silencing of CD9 was verified using immunofluorescence staining of CD9 and confocal microscopy imaging (supplementary material Fig. S4B,C). Next we analyzed soluble ligand (fibronectin fragment) binding to PC3 cells subjected to sequential transfection of a CD9 siRNA that targets the 3′ UTR (siCD9_7) or control siRNA followed by cDNA encoding GFP–CD9. Silencing of CD9 inhibited β1-integrin ligand binding by 31±2%. Conversely, expression of GFP–CD9 in these cells as well as control-siRNA-treated cells strongly induced binding of soluble ligand to the cells (Fig. 3B,C). In addition, we found using reciprocal immunoprecipitations that CD9 and β1 integrin associate in PC3 cells (Fig. 3D). This is consistent with earlier reports of integrin–CD9 complexes in other cell types (Berditchevski et al., 1996).

MMP8 protein was also expressed in PC3 cells and was detected both in the conditioned medium of the cells as well as in full-cell lysates (Fig. 3E), and MMP8 protein levels were downregulated in cells transfected with MMP8 siRNA (Fig. 3E). Interestingly, we found in reciprocal immunoprecipitations that MMP8 also associates with β1 integrin (Fig. 3F). Because MMP8 is secreted into the medium (Tanaka et al., 2007), we hypothesized that MMP8 could inactivate integrins by binding to the receptor on the plasma membrane. To test this we analysed soluble ligand binding of MMP8 siRNA-treated or control-siRNA-treated PC3 cells exposed to purified recombinant MMP8 for 1 hour. Silencing of MMP8 significantly (P=0.002) induced β1-integrin ligand binding and this was fully reversed by ectopic recombinant MMP8 in the medium (Fig. 3G). Taken together, these data confirm the regulatory effects of CD9 and MMP8 on integrin activity identified in our screen and demonstrate that both of these proteins associate with integrins in prostate cancer cells.

Fig. 3.

CD9 and MMP8 associate with β1 integrin. (A) PC3 cells were silenced with two different siRNAs against CD9, and CD9 protein was blotted from lysates. (B) siRNA targeting the 3′-UTR of CD9 (CD9_7; Qiagen) and cDNA encoding GFP–CD9 were used for rescue of fibronectin-647 binding in PC3 cells, and some of the rescue experiment cells were used to analyse the protein levels by western blotting (C). (D) Immunoprecipitations (IPs) from PC3 lysates showed co-precipitation of endogenous β1 integrin with CD9. (E) Western blotting was used to analyse the expression of MMP8 in PC3 cell lysate and in 2-day-conditioned PC3 medium, and the knockdown effect of MMP8 was analyzed from cell lysates. (F) β1 integrin co-precipitated with CD9, but not control (Ctrl) antibody in immunoprecipitations using PC3 cell lysates. (G) Recombinant purified MMP8 (rMMP8) or vehicle was added to the growth medium of PC3 cells to rescue the MMP8 knockdown effect in soluble fibronectin-647 binding as measured by FACS analysis. Values are means ± s.e.m., n=3; **P<0.01, ***P<0.005.

Integrin activity positively associates with cell invasion in vitro

Invasion into matrix is integrin dependent in most cell lines (Friedl and Wolf, 2003). However, limited data exist on the role of integrin activity in cell invasion. To investigate this we silenced 32 genes with replicates and analyzed by time-lapse microscopy the invasive growth of PC3 cells in Matrigel. These transfections included the identified β1-integrin inactivating and activating siRNAs together with siRNAs with no apparent effect on integrin activity, as well as talin-1 and -2 as a positive control (Fig. 4A). Silencing of 19 genes significantly (P<0.05) affected the invasive growth in Matrigel compared with the effect of the control siRNA (Fig. 4A–C). Out of these 19 genes, ten were our newly characterized β1-integrin regulators, and for eight of these ten genes there was a positive correlation between integrin activity and invasion. These effects were not detected in cells cultured on plastic, suggesting that regulation of invasive growth is distinct from regulation of cell proliferation in two-dimensional conditions (Fig. 4D). Talin-1 and -2 siRNAs and five of our inactivating siRNAs inhibited invasiveness, whereas siRNAs for three activating candidates increased invasive growth. Intriguingly, two genes (MSX1, EPS15) out of the ten newly found regulators in our assay perturbed the correlation between the β1-integrin activity and invasion. Of these two genes, the product of MSX1, a homeodomain transcription factor, regulates various developmental genes and, for example, in neuroblastoma can stimulate the expression of various canonical Wnt signalling inhibitor genes (Revet et al., 2010). EPS15, however, is an adaptor protein for clathrin and is involved in the endocytosis and downregulation of different receptor tyrosine kinases (Puri et al., 2005). The perturbation of the correlation between integrin activity and invasion for MSX1 and EPS15 could be due to these other functions elicited by the gene products.

Fig. 4.

β1-integrin activity associates with cancer cell invasion. (A) PC3 cells were treated with the indicated siRNAs (24 hours; AllStars negative=ctrl) and were overlaid with 25% Matrigel and cell growth and motility was followed for 160 hours with Incucyte live microscopy. Shown are invasive growth area (surface area occupied by the cells) values after 160 hours relative to starting point (means ± s.e.m., n=6; *P<0.05). Inactivating siRNAs are indicated in green and activating siRNAs are in red. (B) Examples of the invasive growth area curves of knockdowns with the indicated siRNAs from 4A (fold change). (C) Images of invasive growth of the cells in A. (D) PC3 cells were treated with the indicated siRNAs (24 hours; AllStars negative=ctrl) growing on plastic were followed for 160 hours with Incucyte microscopy. Blue columns shown the cell growth area after 160 hours on plastic relative to the starting point (values are means ± s.e.m., n=6). The dashed line indicates the invasive three-dimensional growth of PC3 cells transfected with the corresponding siRNAs shown in A.

Fig. 5.

β1-integrin regulators alter cell invasion in Matrigel. (A) Maximum z-projections of representative PC3 invasion assays with 5 μg/ml of the indicated antibodies (Mab13 is an β1-integrin-blocking antibody). z=100 μm; xy=775×775 μm. Values are mean invasive areas ± s.e.m., n=3; *P<0.05. (B) PC3 invasion assay of cells treated with the indicated siRNAs for 24 hours prior to the start of the assay. (C) Quantification of the invasion assays in B (mean invasive areas ± s.e.m., n=6; *P<0.05). (D) Invasiveness of ALVA31 cells after treatment of cells with the indicated siRNAs. (Top panels) Cell morphology in the bottom confocal plane after 5 days of invasion (phalloidin staining). (Bottom panels) Invasive areas as in A. The arrow on the left indicates direction of invasion (mean invasive areas ± s.e.m., n=3; *P=0.009). (E) Invasiveness of CD9-silenced 22RV1 cells compared with the control cells (mean invasive areas ± s.e.m., n=3; *P=0.003). Arrow indicates direction of invasion. (F) Still images from time-lapse movies (supplementary material Movies 1, 2) of control and CD9-silenced PC3 cells in 25% Matrigel. Right hand panels show similarly treated cells fixed and stained for 12G10 and DAPI. Scale bars: 5 μm.

Fig. 6.

MMP8 and CD9 regulate β1-integrin activity and lung invasion of breast cancer cells. (A,B) MMP8-silenced, CD9-silenced or control-siRNA-treated MDA-MB-231 cells plated on fibronectin and stained as indicated. Shown are representative images and quantification of 9EG7 staining intensities (means ± s.e.m., 48–52 cells, *P<0.05). Scale bars: 10 μm. (C) Cells were transfected with the indicated siRNAs and mRNA levels were analysed using Taqman quantitative PCR with specific primers and probes. (D) CD9-silenced (green), MMP8-silenced (green) or control-siRNA-treated (red) MDA-MB-231 cells were pre-labelled with fluorescent cell trackers and intravenously injected (5×105 green and 5×105 red cells together) into mice. After 48 hours the cells remaining in the vasculature were flushed out with PBS perfusion. Cells from one lung per mouse were isolated and the fluorescence quantified, and the other lung was processed for sectioning and counterstained with DAPI (representative sections from both experiments are shown). Results are expressed as means ± s.e.m. percentage of specified cells of all cells isolated (n=10 mice; *P=0.05). Scale bars: 50 μm.

The data of invasive growth were further validated with three-dimensional Matrigel invasion assays, in which cells invade the gel in a β1-integrin-dependent manner (Fig. 5A). The silencing of the genes encoding talin-1 and -2, CD9 and AKT3 again inhibited the invasiveness (Fig. 5B,C; supplementary material Fig. S5), and another β1-integrin activity inhibiting siRNA against COL9A1, which did not show an effect on invasive growth, decreased the invasiveness of PC3 cells in three-dimensional matrix (supplementary material Fig. S5). Conversely, β1-integrin stimulatory siRNAs against LCK and MMP8 induced invasiveness correlating well with the invasive growth assays (Fig. 5B,C). Silencing of CD9 inhibited invasion of ALVA31 and 22-RV1 cells also (Fig. 5D,E). CD9 is a tetraspanin family protein, which has been shown to associate with integrins on the membrane (Berditchevski, 2001). Using higher-resolution time-lapse imaging we observed that the cells in which CD9 was silenced were unable to protrude into Matrigel, but instead circulated in a non-polarized fashion (Fig. 5F; supplementary material Movies 1, 2). These cells also contained lower levels of active β1 integrin, as determined by immunofluorescence imaging of the same cells with 12G10 antibody (Fig. 5F). These results further highlight the fact that β1-integrin activity regulators identified in this screen, such as CD9, can function as regulators of cell invasion in vitro.

MMP8 and CD9 regulate cancer cell extravasation in vivo

The observed positive association between integrin activity and invasion in vitro prompted us to investigate whether some of the identified integrin regulators could also regulate the invasion of tumour cells from the blood stream into the lung parenchyma. Our attention was drawn to two genes (MMP8; an integrin-activating siRNA, and CD9; an integrin-inactivating siRNA) with apparently opposing effects on integrin activity and in vitro invasiveness. MDA-MB-231 (ATCC) breast cancer cells extravasate from the blood stream efficiently (Bos et al., 2009; Chabottaux et al., 2009; Gupta et al., 2007). Because this cell line was not included in the screens, we silenced CD9 and MMP8 in these cells and analyzed changes in integrin activity by staining with 9EG7 (Fig. 6A,B). Consistent with data from the other cell lines, silencing of CD9 (77% silencing, as measured by qRT-PCR) inhibited, and silencing of MMP8 (92% silencing) increased the intensity of staining for the active epitope of β1 integrin.

To visualize the early events of lung metastasis, we inoculated mice with an equal number of control-siRNA-silenced and MMP8- or CD9-silenced cells (Fig. 6C). Within 2 days of entering the circulation, tumour cells could be detected in the lungs. Importantly, CD9-silenced cells (green in Fig. 6) extravasated from the lung vasculature significantly (P=0.03) less than control cells (red), whereas MMP8-silenced cells (green) entered the lungs significantly (P=0.04) more efficiently than control cells (Fig. 6D). This was not due to differences in proliferation, as proliferation rates of CD9-silenced and control cells were equal and silencing of MMP8, in fact, slightly reduced the proliferation (supplementary material Fig. S6). These data demonstrate that genes identified in this screen, such as CD9 and MMP8, are involved in biologically important processes, for example, the control of cancer cell extravasation from the blood stream, in vivo.


In this study we performed a cell-microarray-based functional genetic screen with 12 different cell lines to identify regulators of β1 integrin in human cancer cells. We found a substantial number of previously unidentified proteins and pathways that contribute directly or indirectly to modulation of cell-matrix interactions.

Many of the hits, such as the highest scoring β1-integrin-positive (EPS15, ATRN and ERCC1) and β1-integrin-negative (MAST2, HDAC4 and INPP1) regulators have not been previously linked with integrin activity modulation. This is interesting because EPS15 is a well-characterized coat adaptor protein functioning in endocytosis of receptor tyrosine kinases (RTKs), such as EGFR, through clathrin-coated pits (Puri et al., 2005). It is possible that silencing of EPS15 influences integrin activity by altering the cross-talk between integrins and RTKs or by influencing clathrin-mediated endocytosis of integrins themselves. Alternatively, EPS15 could physically associate with integrins and influence their active conformation through this mechanism, as EPS15 was recently identified, using proteomics, as a putative integrin-binding protein (Humphries et al., 2009). MAST2, by contrast, is a poorly studied microtubule-associated serine-threonine kinase. Interestingly, it contains a PDZ domain (Lumeng et al., 1999; Prehaud et al., 2010), which could interact with integrins and influence their function similarly to ZO-1 binding to α5 integrin through its PDZ domains (Tuomi et al., 2009). Furthermore, MAST2 has been shown to phosphorylate and stabilize the tumour suppressor PTEN, which could influence integrin activity by upregulation of the PI3K–AKT pathway (Valiente et al., 2005). However, some of the 23 consistent regulators of β1-integrin activity in our screens have been linked to integrins or cell migration. For example, PFTK1, which was a positive regulator of β1-integrin activity in our assays, has also been shown to increase migration and perturb focal adhesion dynamics in two other RNAi screens (Simpson et al., 2008; Winograd-Katz et al., 2009). In addition, HDAC4 and NCBP2 were also recently identified in a proteomic screen as components of the integrin interactome (Humphries et al., 2009).

We demonstrate here that, in prostate cancer cells, CD9 associates with β1 integrins. We also verify with loss-of-function and overexpression experiments that CD9 regulates integrin binding to a soluble ligand in these cells. The integrin-binding CD9 tetraspanin protein has been shown to increase cell motility and β1-integrin activity when ectopically expressed in Chinese hamster ovarian (CHO) cells (Kotha et al., 2008). In agreement with our findings, CD9 has been shown to be positively associated with cancer cell invasion in vivo, especially during the process of extravasation through the endothelium (Hori et al., 2004; Sauer et al., 2003).

Taken together, our data demonstrate that regulation of integrin activity is highly complex and could be regulated not only by integrin binding proteins such as talins and kindlins but also by numerous other signalling pathways that can influence integrin function by indirect mechanisms or directly by physical interactions with integrins.

An interesting finding of this study was that the ingenuity pathway analysis of the genes involved in the regulation of β1-integrin activity yielded signalling pathways generally implicated in inflammation. In line with this, one of the negative regulators of integrin activity, MMP8 has been shown to have a protective role against inflammation in arthritis and lung fibrosis by suppressing the expression or availability of inflammatory players such as IL-1β and IL-10 (Garcia et al., 2010; Garcia-Prieto et al., 2010). In addition, a reduction of the plasma levels of MMP8 has been associated with higher lymph node and distant metastasis in inflammatory breast cancer patients (Decock et al., 2008). Thus, the mechanistic connection of inflammation and integrin activity regulation in cancer would be an interesting topic for future studies. These studies are in accordance with our findings where we show that MMP8 is a negative regulator of cancer cell invasion in vitro and in vivo, and that this could be due to its role in the suppression of β1-integrin activity. Interestingly, we find that MMP8 associates with β1 integrins in prostate cancer cells and that exposure of cells to recombinant MMP8 protein is sufficient to rescue the increased β1-integrin activity in MMP8-silenced cells. These data suggest a previously unknown mechanism, by which MMP8 binding to β1 integrins on the plasma membrane negatively regulates their activity.

The siRNA effects were most similar among the hormone-insensitive TP53 mutant prostate cancer cell lines PC3, ALVA-31, 22-RV1 and MDA-PCA-2a. Several siRNAs with prominent effects on integrin activity in prostatic cell lines had limited effects in the two lung cancer cell lines and in HCT-116 colon cancer cells with wild-type TP53 (Fig. 2A). This is interesting because these differences could be linked to the accelerated endosomal trafficking of integrins, which drives cell invasion and metastasis downstream of mutant TP53 (Muller et al., 2009; Selivanova and Ivaska, 2009). On the basis of the plausible functional link between integrin activity and trafficking (Caswell et al., 2009), the genes identified in this screen could influence both processes and thus be crucial for invasion of cancer cells, depending on their TP53 status. In conclusion, these data suggest that regulation of adhesion receptor function can be highly cell-context dependent.

β1 integrins have been investigated for more than two decades, and so far only relatively few activators and inhibitors of their function have been found. Here, in a functional screen, we identified 13 activators and 10 inhibitors of β1-integrin regulation in four or more of the cell lines tested. We employed two different in vitro invasion assays and showed that there is a positive link between β1-integrin activity and cell invasion. Furthermore, we demonstrated that the in vitro invasion effects could be reproduced in vivo in a lung extravasation assay by silencing of two different targets, CD9 and MMP8. Because our screen was performed with the druggable genome library, these pathways and proteins might be relevant in the development of new approaches to control dysregulated integrin activity in cancer.

Materials and Methods


The cell spot microarrays (CSMA) used for the analysis were as follows. An siRNA library with two individual siRNAs for 4910 target genes (Qiagen druggable genome library v1.0) and 220 replicate negative control samples of two control siRNAs were used for CSMA printing. The library was prepared with a Hamilton STAR liquid handling robot (Hamilton Robotics) by mixing (for each sample) 5 μl 1.67 μM siRNA (Qiagen) diluted in OptiMEM-I (Gibco) with 0.8 μl siLentFect (Bio-Rad) transfection reagent and 0.2 μl OptiMEM I. Arrays were printed on untreated polystyrene microplates with four rectangular wells (Nunc) using a Genetix Qarray2 (Genetix Ltd) microarray printer with 200 μm solid tip pins (PointTechnologies). For the CSMA experiments with all the described cell lines, cells were grown to 80% confluency on 10 cm culture dishes and dissociated with HyQtase (HyClone) treatment for 5 minutes, suspended in culture medium and dispersed on the arrays as a uniform cell suspension. To each array well 3×106 cells in 4.5 ml of medium were added and allowed to adhere at 37°C for 20 minutes. Non-adhered cells were washed off and 4.5 ml of fresh culture medium was added. Cells were then transfected by taking up the siRNA from the spot on arrays for 48 hours. Cells were fixed with 2% paraformaldehyde (PFAH), permeabilized with 0.15% Triton X-100 in PBS for 15 minutes and stained with antibodies for integrin β1, integrin α2 and for F-actin with phalloidin–Alexa-Fluor-555 and Syto60. Primary array analysis was performed using laser microarray scanning (Tecan LS400). Cumulative fluorescence intensities for measured channels per spot were normalized using pin normalization (Array ProAnalyzer v4.5) and used to calculate the ratio of integrin-β1 and F-actin staining against integrin-α2 counter staining. A z-score standardization was used to evaluate the significance of the measured signal ratios against the global array mean and standard deviation. Microscopy of the arrays was used to validate the results.


Integrin-β1 antibodies 12G10 and 9EG7 were from Abcam and BD Pharmingen, respectively. K20 total integrin-β1 antibody was from Beckman Coulter. Inhibitory integrin-β1 antibody Mab13 was from BD Biosciences. The neutral integrin-α2-recognizing antibody, AB1936, was from Millipore. Anti-talin-1/2 (clone 8d4) was from Sigma. Anti-CD9 was a generous gift from Francisco Sánchez-Madrid (CNIC, Madrid). Anti-MMP8 antibody MAB3316 was from Chemicon.

Western blotting and immunoprecipitations

Cells were lysed in lysis buffer (40 mM Hepes–NaOH, 75 mM NaCl, 2 mM EDTA, 1% NP-40 supplemented with protease and phosphatase inhibitors). Lysates were subjected to immunoprecipitation using the indicated antibodies at +4°C overnight. Protein-G beads (GE Healthcare) in lysis buffer were added and incubated for 1 h at 4°C. Beads were washed with wash buffer [20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% NP-40] and suspended into loading buffer. Samples were separated in SDS-PAGE and analyzed using western blotting.

Fluorescence-activated cell sorting with labelled fibronectin fragment and K20 or 9EG7 antibodies

PC3 cells were transfected with 20 nM siRNAs using siLentFect (Bio-Rad) in 48-well plates for 48 hours. The binding of Alexa-Fluor-647-labeled fibronectin repeat 7–10 (FN) to cells (Montanez et al., 2008) was determined by fluorescence-activated cell sorting (FACS). Cells were incubated in Tyrodes buffer at 37°C for 20 minutes with 100 μg/ml Alexa-Fluor-647-FN, washed with cold Tyrodes buffer and analyzed using a FACSarray (BD Biosciences) high-throughput flow cytometer. A total of 5000 cells from each sample well were measured and a geometric median value from each sample population was used for analysis of fibronectin binding. The binding of K20 and 9EG7 antibodies (1 μg/ml each 1 hour +4°C) to cells was carried out in Tyrodes buffer after fixation with 4% PFAH–PBS. Surface-bound antibodies were probed with Alexa-Fluor-conjugated secondary antibodies (Invitrogen; 1:400) for 1 hour and analyzed by FACS. For the rescue experiments, PC3 cells were sequentially transfected with CD9 or control siRNA and 24 hours later with cDNA encoding GFP–CD9 [a kind gift from Francisco Sanchez-Madrid (Barreiro et al., 2008)] or empty plasmid and analyzed for fibronectin binding as described above. Alternatively, PC3 cells were transfected with MMP8 siRNA or control siRNA and 48 hours later the medium was changed to fresh medium to remove MMP8. 16 hours later the cells were exposed to 5 μg/ml recombinant MMP8 protein (R&D) or buffer alone and analyzed for fibronectin binding.

Taqman qPCR

Taqman quantitative real-time PCR (qRT-PCR) analysis was performed with an Applied Biosystems 7900HT instrument using specific primers designed by the Universal Probe Library Assay Design Center (Roche Applied Biosciences). The expression of specific mRNAs was determined relative to GAPDH mRNA levels.


Images were acquired either directly from the CSMA platform or from microscopic cell chambers (Ibidi, Germany). Confocal images were taken using a Zeiss Axiovert 200M microscope with a Yokogawa CSU22 spinning disc confocal unit and a Zeiss Plan-Neofluar 63× oil/1.4 NA objective. Z-stacks with 1 Airy unit optical slices were acquired with a step size of 0.3 μm. The maximum intensity projections were created with SlideBook software and NIH ImageJ. Time-lapse phase-contrast images were taken with a Zeiss inverted wide-field microscope equipped with a heated chamber (37°C) and CO2 controller (4.8%) and with an LD Plan-Neofluar 40×/0.6 NA Korr objective (6 frames/hour). For invasive growth assays, Incucyte HD (Essen Bioscience, UK) imaging was used. QuickTime movies from time-lapse experiments were created with NIH ImageJ software.

Invasion assays

For invasive growth assays 0.5 μl siRNA (5 μM stock), 0.8 μl Hiperfect (Qiagen), 10 μl Optimem (Gibco) and 79 μl PC3 cell medium (see above) were mixed in the wells of 96-well plates. After 15 minutes incubation, 1500 cells were added to each well. Cells were grown for 24 hours, medium was changed to Matrigel-medium mixture (25% Matrigel) or cell were supplemented with medium and allowed to proliferate in two dimensions. Images were acquired each hour for 160 hours using Incucyte HD (Essen Bioscience, UK). Images were analyzed using ImageJ software.

For three-dimensional invasion assays, 7500 cells were applied onto 15-well μ-Slide angiogenesis chambers (Ibidi, Germany) and left to adhere for 6 hours, The medium was then replaced with 10 μl of Matrigel-medium mixture (33% Matrigel, serum-free medium) and overlaid with 50 μl 2% serum-containing medium after polymerization. Cells were allowed to invade upwards for 5 days by replacing 5 μl of the medium with fresh 10% serum-containing medium each day. Cells were fixed with 30 μl 3.7% PFAH (15 minutes), permeabilized with 0.3% Triton X-100 in PBS (15 minutes) and stained with Hoechst 33432 or with phalloidin-555 in PBS with 1% BSA. Confocal images were taken by using a Zeiss Axiovert 200M microscope with a Yokogawa CSU22 spinning disc confocal unit and a Zeiss 20× objective. Z-stacks of 2 μm optical slices were acquired with a step size of 2 μm, and the maximum intensity projections were created with SlideBook software and NIH ImageJ.

Lung extravasation assay

Female athymic nude mice (Hsd:Athymic Nude-nu; Harlan Scandinavia, Allerod, Denmark), aged between 4 and 6 weeks, were used for the xenograft studies. The experimental procedures were approved by the local ethical committees. Mice were anesthetized with ketamine (Pfizer) and xylazine (Bayer). Transiently siRNA-silenced MDA-MB-231 cells were stained with live cell dyes [AllStars negative control siRNA (Qiagen) stained red with CMTPX (Invitrogen); siMMP8 or siCD9 stained green with CMFDA (Invitrogen)] according to manufacturer's instructions. Cells were harvested, suspended in 50 μl PBS (5×105 each), mixed and injected (control and siCD9 or control and siMMP8) into the lateral tail vein of mice (n=10). The mice were anesthetised 48 hours post-injection and the pulmonary vasculature was perfused with PBS through the right ventricle (2 minutes) and blood was allowed to escape by a small incision in the left atrium. Animals were killed and cells were harvested from one lung per animal with collagenase XI treatment for 1 hour at 37°C, washed with PBS, and the fluorescence analyzed using a ScanR automated microscope. The other lung was processed as frozen sections and stained with DAPI.

Statistical analysis

All statistical analyses were performed using Student's t-test. P<0.05 was considered significant.


We thank H. Marttila, J. Siivonen, L. Lahtinen and P. Laasola for their excellent technical assistance. This study has been supported by Academy of Finland, ERC Starting Grant, Sigrid Juselius Foundation, EMBO YIP and Finnish Cancer Organizations. A. Arjonen and T. Pellinen have been supported by Turku Graduate School of Biomedical Sciences. The authors declare no conflict of interest.


  • * These authors contributed equally to this work

  • Funding

    This study was supported by the Academy of Finland [J.I. and O.K.]; a European Research Council Starting Grant [to J.I.]; the Sigrid Juselius Foundation [J.I.]; the European Molecular Biology Organisation Young Investigator Programme [J.I.]; and Finnish Cancer Organizations [J.I. and O.K.]; and Turku Graduate School of Biomedical Sciences [A.A. and T.P.].

  • Supplementary material available online at

  • Accepted September 29, 2011.


View Abstract