LIMK2 is a crucial regulator and effector of Aurora-A-kinase-mediated malignancy

Aurora A, a serine/threonine kinase, is primarily expressed during G2 and M phases of the cell cycle. Its depletion results in the failure of cells to enter mitosis. During the G2 phase, Aurora A is required for centrosomal duplication and separation, and in metaphase, it regulates mitotic spindle formation (Hirota et al., 2003). Aurora A is degraded in late mitosis or early G1 by the Cdh1–APC complex. Aurora A localizes on the centrosome during G2 and spreads to the mitotic spindle poles and mid-zone microtubules during metaphase.

Aurora A is overexpressed in several types of cancer including prostate, breast, ovarian, colorectal, gastric, pancreatic, hepatocellular, gliomas, nonendometriod and aggressive non-Hodgkin’s lymphoma (Mountzios et al., 2008). In breast tumor cells, Aurora A is expressed in all phases of the cell cycle and shows predominantly cytoplasmic localization (Nadler et al., 2008). Thus, Aurora A is believed to promote tumorigenesis through aberrant phosphorylation of cytoplasmic proteins, as well as of proteins that it encounters in phases of the cell cycle during which it is not normally expressed.

Aurora A is overexpressed in a high proportion of breast carcinomas (Miyoshi et al., 2001; Nadler et al., 2008; Tanaka et al., 1999). Aurora A is one of the genes in the Oncotype Dx assay for predicting the likelihood of breast cancer recurrence in early-stage, node-negative, estrogen-receptor-positive breast cancer (Cronin et al., 2007). High Aurora A expression is strongly associated with node status and decreased survival (Miyoshi et al., 2001). Polymorphism in the Aurora A gene is associated with increased risk of invasive breast carcinoma (Cox et al., 2006), and works synergistically with prolonged estrogen exposure (Dai et al., 2004). In animal models, Aurora A overexpression induced tumor formation and its inhibition reduced tumor multiplicity and size (Wang et al., 2006). MLN8237, an orally available, highly potent and selective Aurora A inhibitor is in Phase II clinical trials for advanced solid tumors.

Over a dozen Aurora A substrates are known, but few have been identified as potential cancer targets. We recently identified PHLDA1 as a direct target of Aurora A in breast cancer cells using a modified chemical genetic approach (Johnson et al., 2011). In the present study, LIMK2 was followed up as a direct target of Aurora A, which revealed a new mechanism by which Aurora A overexpression might not only promote breast malignancy, but other types of cancer as well.

We recently developed a tailored chemical genetic approach for identifying the direct substrates of Aurora family of kinases. The approach uses an analog-sensitive kinase and an orthogonal ATP analog (Johnson et al., 2011). In most cases, the analog-sensitive kinase is generated by the replacement of a conserved bulky residue (gatekeeper residue) in the kinase active site. A complementary substituent on ATP is created by attaching a bulky substituent at the N6 position of ATP. Because the ATP analog is not accepted by wild-type kinases, this strategy allows for unbiased identification of direct substrates of any kinase in a global environment (Shah et al., 1997; Shah and Shokat, 2002; Shah and Shokat, 2003; Shah and Vincent, 2005; Kim and Shah, 2007; Sun et al., 2008a; Sun et al., 2008b; Sun et al., 2009; Chang et al., 2011).

Our previous study revealed that Aurora A requires an additional mutation (L194V), along with the gatekeeper mutation (L210G, human Aurora A numbering) to generate the analog-sensitive allele (Aurora-A-as7 or AA-as7), capable of accepting orthogonal ATP analogs and inhibitors (Johnson et al., 2011). N-6-Phenethyl ATP was identified as the most optimal ATP analog. 4-amino-1-tert-butyl-3-(1′-naphthylmethyl)pyrazolo[3,4-d]pyrimidine (1-NM-PP1) was identified as a highly potent and selective inhibitor for AA-as7 kinase (IC₅₀=1.7 nM). The sensitized allele created by these mutations has identical substrate specificity to that of the wild-type kinase.

As our goal was to specifically identify oncogenic effectors of Aurora A, we chose MDA-MB-231 (MDA) cells, which are highly malignant and express high levels of Aurora A. Furthermore, in MDA cells, Aurora A is overexpressed in all phases of the cell cycle and shows diffuse cytoplasmic localization (Johnson et al., 2011), similar to the Aurora A expression pattern and localization in breast tumor cells (Nadler et al., 2008). Therefore, an in vitro kinase reaction was performed using unsynchronized MDA cell lysate, [γ-³²P]phenethyl-ATP and AA-as7–TPX2 for identifying Aurora A substrates. Radiolabeled proteins were separated on 2D gels, excised, and subsequent to trypsin digestion, the peptide cleavage products were identified by tandem mass spectrometry. This study revealed several Aurora A substrates including PHLDA1, p53 and vimentin (Johnson et al., 2011). In this study, we followed LIMK2 as a putative Aurora A substrate in MDA cells. LIMK2, a serine/threonine kinase, is involved in regulating actin dynamics (Sumi et al., 1999; Sumi et al., 2001a; Sumi et al., 2001b).

Proteomics screening can often lead to false positives, thus LIMK2 phosphorylation was determined using recombinant Aurora-A–TPX2 and 6x-His–LIMK2 in an in vitro kinase assay. TPX2 is an activator of Aurora A. Our results showed that Aurora A directly phosphorylated LIMK2 (Fig. 1A, lane 2).

We next investigated whether Aurora A and LIMK2 associate in MDA cells. LIMK2 immune complex was isolated and Aurora A binding analyzed, which revealed that Aurora A associated with LIMK2 (Fig. 1B, lane 3). Similar results were obtained when Aurora A immune complex was isolated and LIMK2 binding analyzed (Fig. 1C, lane 2).

Aurora A and LIMK2 proteins were examined in unsynchronized MDA cells, which revealed predominantly cytoplasmic localization for both; however, some LIMK2 was also nuclear (Fig. 1D, top panel). Aurora A was inhibited using MLN8237 for 12 hours, which resulted in perinuclear localization of both Aurora A and LIMK2 (Fig. 1D, bottom panel). These findings show that Aurora A and LIMK2 not only colocalize, but Aurora A also regulates LIMK2 subcellular localization.

Aurora-A-mediated phosphorylation of its substrates often promotes their degradation (e.g. p53, PHLDA1) (Mao et al., 2007; Johnson et al., 2011) or stabilization (ASAP1) (Venoux et al., 2008). Therefore, we investigated whether Aurora A exerts control over LIMK2 level. Aurora A was transiently depleted in MDA cells using two different shRNAs, causing a concomitant decrease in LIMK2 level (Fig. 2A). Wild-type and mutant AA (AA-as7) were stably overexpressed in MDA cells (AA-MDA and AA-as7-MDA cells, respectively), leading to a considerable increase in LIMK2 levels (Fig. 2B), thereby confirming that Aurora A positively regulates LIMK2 levels. Notably, Aurora A overexpression is highly toxic in many cell types and causes cell death. However, it was well tolerated in MDA cells. AA-MDA and AA-as7-MDA cells were viable and showed robust transformation phenotype (Johnson et al., 2011).

Aurora A can regulate its substrates using either its kinase activity or scaffolding function. Aurora A regulates p53 levels by phosphorylation (Mao et al., 2007), but regulates ASAP levels through the scaffolding function (Venoux et al., 2008). To dissect the mechanism by which Aurora A regulates LIMK2, AA-MDA and AA-as7-MDA cells were treated with 1-NM-PP1 for 12 hours and LIMK2 expression was analyzed. 1-NM-PP1 is a highly selective and potent inhibitor of AA-as7, and does not inhibit wild-type Aurora A, even at higher concentrations (Johnson et al., 2011). 1-NM-PP1 treatment caused significant decrease in LIMK2 level in AA-as7-MDA cells, but not in wild-type AA-MDA cells (Fig. 2C). This result demonstrates that Aurora A upregulates LIMK2 levels using its kinase activity, and not by scaffolding interactions.

The involvement of the Aurora A kinase function in positively regulating LIMK2 suggested that Aurora A might inhibit degradation of LIMK2 by phosphorylation. Therefore, the LIMK2 protein degradation profile was examined in MDA and AA-MDA cells using cycloheximide. Because Aurora A has a half-life of ∼2 hours (Honda et al., 2000), 2 hour and 4 hour time-points were selected. Aurora A overexpression reduced LIMK2 degradation (Fig. 2D–F), suggesting that it regulates the level of LIMK2 by inhibiting its degradation. This study revealed that the half-life of LIMK2 was ∼2 hours (Fig. 2F).

LIMK2 degradation could be mediated by ubiquitin or non-ubiquitin pathways. 6x-His–ubiquitin was transfected in MDA and Aurora-A-depleted MDA cells, and LIMK2 ubiquitylation analyzed. Aurora A depletion led to increased LIMK2 ubiquitylation (Fig. 2G), thereby demonstrating that Aurora A stabilizes the level of LIMK2 by inhibiting its ubiquitylation.

After confirming Aurora-A-mediated stability of LIMK2 protein, we sought to determine whether Aurora A regulated LIMK2 kinase activity. Cofilin, a well-characterized substrate of LIMK2, was used to evaluate LIMK2 kinase activity. Aurora A did not phosphorylate cofilin (Fig. 2H, lanes 1 and 4). LIMK2 efficiently phosphorylated cofilin as expected (Fig. 2H, lane 3). Importantly, LIMK2 pre-incubated with Aurora-A–TPX2 showed increased kinase activity (Fig. 2H, lane 2), confirming that Aurora A also boosts the kinase activity of LIMK2.

Aurora A can increase LIMK2 activity either by phosphorylation or by association. Therefore, LIMK2 kinase activity was tested by pre-incubating it with either wild-type or dominant-negative (D274A) Aurora A. Although wild-type Aurora A increased LIMK2 kinase activity significantly (as measured by increased cofilin phosphorylation, (Fig. 2I, compare lanes 1 and 2), dominant-negative Aurora A had no effect on LIMK2 activity (Fig. 2H, lane 3). This result shows that the kinase activity of LIMK2 is predominantly regulated by Aurora-A-mediated phosphorylation, and not by physical association with LIMK2.

Several Aurora A substrates are known to regulate Aurora A activity or expression by a feedback mechanism. Aurora A phosphorylates FAF1 at S289 and S291, which in turn degrades Aurora A (Jang et al., 2008). Similarly, PP1 inhibits Aurora A kinase activity upon phosphorylation by Aurora A (Katayama et al., 2001). Our previous report showed that the Aurora A substrate PHLDA1 negatively regulates Aurora A in a feedback loop (Johnson et al., 2011).

LIMK2 shRNAs were generated and used to reduce LIMK2 levels in cells, which decreased Aurora A level significantly, suggesting the existence of a positive-feedback loop between the two proteins (Fig. 3A). LIMK2-overexpressing stable MDA cells (LIMK2-MDA) were generated and Aurora A levels analyzed. LIMK2 overexpression increased the Aurora A level, confirming that LIMK2 positively regulates Aurora A (Fig. 3B).

Because LIMK2 can also increase Aurora A levels by inhibiting its degradation, Aurora A and LIMK2 levels were analyzed in cycloheximide-treated MDA and LIMK2-MDA cells. LIMK2 overexpression considerably reduced Aurora A degradation (Fig. 3C–E), suggesting that LIMK2 stabilizes Aurora A protein levels, presumably by inhibiting its ubiquitylation. To test this possibility, 6x-His–ubiquitin was transfected in MDA cells and LIMK2-depleted MDA cells and Aurora A ubiquitylation analyzed using antibody against 6x-His. LIMK2 depletion indeed increased ubiquitylated Aurora A (Fig. 3F, lane 2). These results show that Aurora A and LIMK2 positively regulate each other by stabilizing the other’s protein levels.

As LIMK2 is a kinase, we examined whether it phosphorylates Aurora A leading to its stabilization. Recombinant 6x-His-tagged Aurora A showed negligible kinase activity in vitro in the absence of TPX2 (Fig. 3G, lane 1), suggesting that it could be used as a putative substrate for LIMK2. However, when LIMK2 was incubated with Aurora A (without TPX2), Aurora A did not become phosphorylated (Fig. 3G, lane 2). By contrast, LIMK2 efficiently phosphorylated cofilin, which was used as a positive control (lane 3). This result suggested that, unlike Aurora A, which stabilizes LIMK2 by phosphorylation, LIMK2 stabilizes Aurora A by protein–protein interactions.

To confirm this finding, stable MDA cell lines expressing wild-type LIMK2 and kinase-inactive LIMK2 (K339M) were generated. Interestingly, both wild-type and kinase-inactive LIMK2 caused an increase in Aurora A levels, although, wild-type LIMK2 increased it more (Fig. 3H). This result suggests that LIMK2 stabilizes Aurora A using both its scaffolding interactions and kinase function. However, Aurora A is not a direct substrate of LIMK2 (Fig. 3G,H). It is likely that LIMK2 stabilizes Aurora A indirectly by phosphorylating some other targets. Future studies are required to uncover this pathway.

Aurora A regulates LIMK2 by phosphorylation (Fig. 2C). Thus, we set out to identify Aurora-A-mediated phosphorylation sites on LIMK2. Aurora A preferentially phosphorylates R/K/N-R-x-S/T-B, where B denotes any hydrophobic residue except Pro (Ferrari et al., 2005). This preference revealed S283, T494 and T505 as putative Aurora A phosphorylation sites on LIMK2. S283A, T494A and T505A LIMK2 alleles were generated and their phosphorylation analyzed in vitro. Aurora A phosphorylates LIMK2 at all these sites (Fig. 4A).

To investigate whether S283, T494 and T505 residues on LIMK2 are phosphorylated in cells, a gel shift assay was performed by transfecting LIMK2 phosphorylation-resistant single mutants in MDA cells. After 30 hours, these cells were treated with MLN8237 for an additional 12 hours. Inhibition of Aurora A decreased both phosphorylation level and overall protein level of LIMK2 mutants, suggesting that these three sites are phosphorylated by Aurora A in cells (Fig. 4B). This result also suggested that phosphorylation of all three sites (S283, T494 and T505) by Aurora A contributes to LIMK2 stability.

To confirm these findings, HA-tagged LIMK2 double mutant (S283A,T494A) and HA-tagged LIMK2 triple mutant (S283A,T494A,T505A) were generated and transfected in MDA cells. HA-tagged wild-type LIMK2 was used as a control. After 24 hours, Aurora A was inhibited using MLN8237 for an additional 12 hours, and LIMK2 levels analyzed using antibody against the HA tag. Wild-type LIMK2 showed the highest expression levels, followed by the LIMK2 double mutant. The LIMK2 triple mutant showed the least expression (Fig. 4C). Importantly, Aurora A inhibition decreased phosphorylation and protein levels of both wild-type LIMK2 and LIMK2(S283A,T494A) (Fig. 4C, top bands), but the protein and phosphorylation levels of the LIMK2 triple mutant remained unchanged. Together, these findings demonstrate that Aurora A modulates LIMK2 stability by phosphorylating S283, T494 and T505 sites in cells.

Because LIMK2 levels modulate Aurora A levels in a feedback loop, we compared LIMK2 and Aurora A levels in MDA, wild-type LIMK2-MDA, LIMK2(S283A,T494A)-MDA and LIMK2(S283A, T494A,T505A)-MDA cells. Whereas wild-type LIMK2 MDA cells showed robust expression of Aurora A and LIMK2, LIMK2 (S283A,T494A)-MDA cells showed reduced expression, and LIMK2(S283A,T494A,T505A)-MDA cells negligible expression (Fig. 4D, top panel). These findings confirm that Aurora A stabilizes LIMK2 levels by direct phosphorylation at S283, T494 and T505, which in turn promotes Aurora A stabilization in a positive-feedback loop (Fig. 4D, middle panel). This result is consistent with our previous data whereby inhibition of Aurora A kinase activity reduces the LIMK2 protein level, in turn reducing the level of Aurora A (Fig. 2C).

We next analyzed ubiquitylation of wild-type and triple mutant LIMK2 in MDA cells, which showed robust ubiquitylation of the LIMK2 triple mutant, but not of wild-type LIMK2 (Fig. 4E), suggesting that Aurora-A-mediated phosphorylation of LIMK2 stabilizes its protein levels.

Previous studies have shown that LIMK2 phosphorylation at T505 by ROCK and MRCK-α kinases increases its kinase activity (Sumi et al., 2001a; Sumi et al., 2001b). Because we observed T505 phosphorylation by Aurora A, it provides the mechanism by which Aurora A increases LIMK2 kinase activity (Fig. 2H). However, to confirm that the increased activity was only due to T505 phosphorylation and not due to phosphorylation at other sites, we measured Aurora-A-mediated activation of wild-type LIMK2, LIMK2(S283A), LIMK2(T494A) and LIMK2(T505A) in an in vitro kinase assay using cofilin as the substrate. As expected, Aurora A activated wild-type LIMK2, LIMK2(S283A) and LIMK2(T494A), but not LIMK2(T505A) mutant (Fig. 4F). Furthermore, Aurora A activated the LIMK2(S283A,T494A) double mutant, but not the LIMK2(S283A,T494A,T505A) triple mutant, thereby demonstrating that Aurora A activates LIMK2 by phosphorylating T505 (Fig. 4G).

Previously, PKC has been shown to phosphorylate LIMK2 at S283 and T494 sites, which reduces its nuclear import (Goyal et al., 2006). We also observed nuclear localization of LIMK2 upon Aurora A inhibition (Fig. 1D). These results were further confirmed in Aurora-A-ablated MDA cells, which showed nuclear localization of LIMK2 (Fig. 4H, middle panel), suggesting that similar to PKC (Fig. 4H, third panel), Aurora-A-mediated LIMK2 phosphorylation also localizes it in the cytoplasm.

Aurora A and LIMK2 showed strong association (Fig. 1B,C). Thus, we examined the LIMK2 domain responsible for the Aurora A association. As shown in Fig. 4I, LIMK2 contains two LIM domains (12–129), a PDZ domain (152–236), an S/P domain (240–328) and a kinase domain (331–601) (Scott and Olson, 2007). Therefore, different HA-tagged truncated LIMK2 mutants were generated containing two LIM domains (1–145), LIM and PDZ domains (1–298) or predominantly kinase domain (298–638) (Fig. 4I). These mutants were transiently expressed in MDA cells, isolated and analyzed for Aurora A binding. Our results show that LIMK2 binds Aurora A through its LIM domains (1–145) (Fig. 4J, lane 2).

We next investigated whether the association of LIM domains of LIMK2 is sufficient for Aurora A stabilization. Aurora A levels were examined in MDA cells, LIMK2-MDA cells and LIMK2(1–145) MDA cells. Although expression of full-length LIMK2 robustly increased Aurora A levels, LIMK2(1–145) expression partially increased it (Fig. 4K). This result suggests that although the LIMK2–Aurora A association increases Aurora A levels, LIMK2 kinase activity is also an essential contributor to Aurora A stability (Fig. 4K). This result is consistent with our previous finding that showed LIMK2 stabilizes Aurora A using both its scaffolding interactions and kinase function (Fig. 3H).

To investigate whether LIMK2 plays a role in mitosis similar to Aurora A, G1–S-arrested MDA cells were released for varying periods, and LIMK2 and Aurora A levels analyzed. LIMK2 was expressed throughout the cell cycle, whereas Aurora A expression markedly increased during mitosis (Fig. 5A), suggesting that LIMK2 expression is not regulated by the cell cycle. However, a positive correlation was observed between Aurora A and LIMK2 levels, and increased Aurora A levels were closely linked with increased LIMK2 levels (Fig. 5A). We also examined the levels of LIMK2 and Aurora A in HCT116 cells, which show tight cell cycle regulation. The expression of LIMK2 was strongly correlated with Aurora A levels, and markedly increased after 4 hours and 8 hours (Fig. 5B). Taken together, our data reveal that Aurora A regulates a significant population of LIMK2 in cancer cells.

FACS analysis was conducted to examine a potential role of LIMK2 in mitosis using unsynchronized MDA cells, LIMK2-MDA stable cells (Fig. 3H) and LIMK2-ablated MDA stable cells (Fig. 5C). Whereas MDA cells showed no aneuploidy, LIMK2-MDA and LIMK2-ablated cells showed some aneuploidy (Fig. 5D–F).

To probe the mechanism further, histone H3 (Ser10) phosphorylation (a mitotic marker) was analyzed in synchronized MDA, LIMK2-ablated and LIMK2-MDA cells. Both MDA and LIMK2-MDA cells showed similar percentages of phospho-histone-H3-positive cells at different times upon release from thymidine block; however, LIMK2-ablated MDA cells showed slightly fewer cells positive for phosphorylated histone H3 (Fig. 5G). Importantly, Aurora A levels were reduced (but not depleted) in LIMK2-ablated stable MDA cells (Fig. 5C), suggesting that reduced Aurora A levels might be sufficient to carry out normal mitotic functions in MDA cells. Similarly, LIMK2-overexpressing MDA cells show increased Aurora A levels (Fig. 3B), which were well tolerated in MDA cells, as observed with AA-MDA and AA-as7-MDA cells.

Because Aurora A plays a vital role in mitotic spindle formation, we further examined the consequences of LIMK2 overexpression or ablation in mitotic spindle assembly. MDA, LIMK2-MDA, LIMK2-ablated MDA and LIMK2-ablated AA-MDA cells arrested at G1–S were released for varying periods, and mitotic spindle was analyzed. LIMK2 overexpression did not affect mitotic spindle assembly in the majority of the cells (∼95%) (Fig. 5H). By contrast, ∼40% of LIMK2-ablated cells showed defect in the mitotic spindle (Fig. 5H). The remaining 60% of cells were normal. Because LIMK2-ablated MDA cells are comprised of a pooled population, varying LIMK2 levels are expected in different populations. We speculate that the cells with severely reduced LIMK2 levels show defective mitotic spindle and multiple nuclei, which is consistent with Aurora A ablation. This result is also consistent with FACS data, which revealed relatively more aneuploidy in LIMK2-ablated cells than in LIMK2-MDA cells. These results suggest that LIMK2 is not directly involved in mitosis; however, LIMK2 indirectly affects mitosis by regulating Aurora A levels. LIMK2-ablated AA-MDA cells showed normal mitotic spindles in the vast majority of cells, which was expected because these cells contain relatively higher Aurora A levels compared with LIMK2-ablated MDA cells (Fig. 5H, bottom panel).

LIMK2 promotes metastasis in pancreatic cancer and fibrosarcoma, but has not been analyzed in breast malignancy. Aurora A overexpression increased cell proliferation in MDA cells (Fig. 6A). LIMK2 depletion from both MDA and AA-MDA cells reduced cell growth significantly (Fig. 6A). Previous studies have shown that LIMK2 promotes metastasis in fibrosarcoma by increasing anchorage-independent growth and cell motility, but not cell proliferation (Suyama et al., 2004). This is the first study that shows a positive role of LIMK2 in cell proliferation.

The effect of LIMK2 was further evaluated in MDA and AA-MDA cells under anchorage-independent conditions. Both MDA and AA-MDA cells revealed dramatic loss in colony-forming ability upon LIMK2 depletion (Fig. 6B). These findings show that LIMK2 depletion inhibits cell proliferation both under attached and anchorage-independent conditions in breast cancer cells. More importantly, the striking increase in colony-forming ability of MDA cells upon Aurora A overexpression (Fig. 6B, compare columns 2 and 4) is almost completely lost upon LIMK2 depletion (columns 3 and 5), suggesting that LIMK2 upregulation is one of the key mechanisms by which Aurora A promotes cellular transformation.

We next determined the contribution of LIMK2 to promoting chemotaxis using serum-starved MDA, AA-MDA, LIMK2-depleted MDA and LIMK2-depleted AA-MDA cells. Although AA-MDA cells were highly motile, LIMK2 depletion reduced cell motility considerably (Fig. 6C). As chemotaxis plays a key role in cancer metastasis, these results underscore a crucial oncogenic role of LIMK2 in breast malignancy.

Our cellular data strongly suggested that Aurora-A-mediated oncogenic phenotypes can be reversed by LIMK2 ablation in vivo (Fig. 6A–C). Therefore, we initially determined the contribution of Aurora A to inducing tumorigenesis in a mouse xenograft model. MDA and AA-MDA cells were subcutaneously inoculated on the right and left shoulders, respectively, in athymic nude mice, and tumors were measured every 2 days for a period of 5 weeks. Aurora A overexpression caused a robust increase in tumorigenesis in vivo (Fig. 6D–F).

We investigated whether LIMK2 ablation in MDA cells reverses tumorigenesis in mouse xenograft models. Athymic nude mice were subcutaneously inoculated with AA-MDA cells and LIMK2-depleted AA-MDA cells on the right and left shoulders, respectively. The tumors became measurable 12–14 days after implantation, and were measured every 2 days. Mice were observed over a period of 5 weeks and no signs of toxicity were observed. As shown in Fig. 6G–I, LIMK2 ablation prevented tumor formation in nude mice. These results confirm that LIMK2 is a key oncogenic effector of Aurora A in breast malignancy. Because Aurora A is an essential kinase, inhibition or ablation of LIMK2 appears to be a viable alternative for preventing or treating breast cancer.

Aurora A is overexpressed in many cancers. Because LIMK2 was found to be a key oncogenic regulator and effector of Aurora A in breast cancer cells, we investigated whether a similar mechanism exists in other types of cancers. Aurora A has been shown to be overexpressed in 96% of high-grade prostate intraepithelial neoplasia (PIN) and 98% of prostate cancer lesions (Lee et al., 2006). Aurora A ablation in PC3 cells suppresses cell growth, induces apoptosis, attenuates cell migration and inhibits the growth of human prostate cancer xenografts in nude mice (Qu et al., 2008). Thus, we chose prostate cancer cells to examine whether a LIMK2 and Aurora A feedback loop exists in these cells. LIMK2 has not been previously analyzed in prostate cell lines or tissues.

LIMK2 and Aurora A levels were examined in PC3 cells, which are highly metastatic and express high levels of Aurora A (Fig. 7A). Most importantly, Aurora A ablation in PC3 cells depleted LIMK2 levels, confirming that Aurora A positively regulates LIMK2 levels in these cells (Fig. 7A). Furthermore, similar to the situation in breast cancer cells, LIMK2 positively regulates Aurora A in prostate cancer (Fig. 7B), because LIMK2 ablation depleted Aurora A. These findings demonstrate that LIMK2 is also an important effector of Aurora-A-mediated oncogenesis in prostate cancer cells.

Aurora A and LIMK2 levels were further analyzed in pancreatic cancer (PaCa2) and colorectal cancer cells (HCT116). Both Aurora A and LIMK2 were highly expressed in these cells (Fig. 7C). Aurora A ablation in PaCa2 and HCT116 cells depleted LIMK2 (Fig. 7D). Similarly, LIMK2 ablation depleted Aurora A, suggesting that a feedback loop involving Aurora A and LIMK2 is a common mechanism in Aurora-A-induced malignancy (Fig. 7E).

Over a dozen Aurora A-targeted drugs are in clinical trials. Owing to a lack of oncogenic Aurora A targets, Aurora A autophosphorylation is used as a pharmacodynamic biomarker. Because we observed a short half-life of LIMK2 (∼2 hours) and rapid depletion of LIMK2 upon Aurora A ablation and inhibition (Fig. 2A,C), we examined whether Aurora A inhibition using a pharmacological inhibitor affects the LIMK2 level. Aurora A was inhibited using MLN8237 and LIMK2 levels were analyzed. As expected, Aurora A inhibition depleted LIMK2 levels, suggesting that LIMK2 levels could be used as a pharmacodynamic biomarker for investigational drugs that target Aurora A (Fig. 7F).

The positive role of LIMK2 in Aurora-A-mediated oncogenic pathways suggested that LIMK2 downregulation might work synergistically with Aurora A inhibition in promoting cell death. Aurora A was inhibited using MLN8237 in MDA and LIMK2-ablated-MDA cells. Although ∼25% loss in cell viability was observed in MDA cells, >50% loss was observed in LIMK2-ablated MDA cells (Fig. 7G). These findings suggest that LIMK2 inhibition or ablation might be an alternative approach for modulating Aurora-A-mediated breast oncogenesis.

Aurora A kinase is overexpressed in cancers of many origins, which include both solid tumors and hematological malignancies. Over a dozen Aurora A inhibitors are in advanced clinical trials. Although Aurora A inhibition has shown high efficacy in clinical trials, it is also associated with significant side effects, including neutropenia, somnolence, asthenia and transaminitis (Carol et al., 2011), presumably because it is expressed in all dividing cells. In normal cells, Aurora A is essential for centrosome duplication and separation, microtubule–kinetochore attachment, spindle checkpoint formation and cytokinesis during mitosis. Aurora-A-null mice die at the blastocyst stage (Cowley et al., 2009). These findings suggest that selective inhibition of cancer-specific targets of Aurora A should reduce the toxicity associated with systemic Aurora A inhibition in cancer. However, the underlying molecular mechanisms of Aurora-A-mediated malignancy remain elusive, primarily because of the lack of known cancer-specific targets of Aurora A.

The goal of the present study was to identify and validate oncogenic effectors of Aurora A malignancy using a chemical genetic approach in highly malignant breast cancer cells. The chemical genetic approach for the identification of direct substrates of kinases is highly versatile and has been applied to over 40 kinases to date. This technique revealed several new substrates of Aurora A, including LIMK2 and PHLDA1, in breast cancer cells (Johnson et al., 2011).

In normal cells, LIMK2 promotes cell cycle progression by regulating actin dynamics. LIMK2 directly phosphorylates cofilin, which inhibits its actin depolymerization activity (Sumi et al., 1999). Previous studies have shown that LIMK2 promotes metastasis in fibrosarcoma and pancreatic cancer cells (Suyama et al., 2004; Vlecken et al., 2009); however, the mechanism remains unknown. Our studies reveal that LIMK2 is highly expressed in both breast and prostate cancer cells in an Aurora-A-dependent manner. LIMK2 and Aurora A have never been associated before.

We show that Aurora A and LIMK2 are involved in a positive synergistic feedback loop, where each potentiates the other’s protein level: inhibition or ablation of one protein depletes the other. Aurora A also regulates LIMK2 kinase activity and subcellular localization. Aurora A phosphorylates LIMK2 at S283, T494 and T505. Aurora-A-mediated LIMK2 phosphorylation at T505 increases its kinase activity, whereas phosphorylation at all three sites contributes to protein stability. Sequence analysis revealed that LIMK2 contains a KEN box motif (272–274), which is one of the mechanisms by which proteins are targeted for proteasomal destruction by the APC/C complex. Our data indeed show that Aurora A stabilizes LIMK2 levels by inhibiting its ubiquitylation.

Aurora A also promotes cytoplasmic localization of LIMK2, presumably by phosphorylating S283 and T494. Previously, LIMK2 has been shown to be phosphorylated by ROCK at T505 and by PKC at S283 and T494. PKC-mediated LIMK2 phosphorylation reduces its nuclear retention.

Our results show that LIMK2 stabilizes Aurora A levels exploiting both its protein–protein interactions and kinase activity. Domain-mapping experiments demonstrated that Aurora A binds LIMK2 by binding to its LIM domains, which contributes to Aurora A stabilization to some extent. Although LIMK2 did not appear to directly phosphorylate Aurora A, it might phosphorylate additional substrates leading to stabilization of Aurora A. This result is very important because it suggests that LIMK2 inhibitors are likely to be effective in abrogating Aurora A malignancy.

Most importantly, LIMK2 depletion in Aurora-A-transformed MDA cells completely inhibits tumor formation in nude nice, thereby revealing LIMK2 as a key oncogenic effector of Aurora A in breast malignancy. Interestingly, Limk2-null mice do not exhibit embryonic lethality or phenotypic abnormalities in postnatal growth and development, except for spermatogenesis in the testis (Takahashi et al., 2002), suggesting that it is not an essential cell cycle gene. Therefore, systemic inhibition using LIMK2 inhibitors might have fewer side effects than inhibition of Aurora A. Furthermore, LIMK2 depletion acts synergistically with Aurora A inhibition in promoting cell death. These findings suggest that a combination regimen comprising of sub-threshold levels of LIMK2 and Aurora A inhibitors might be a viable therapeutic approach for inhibiting breast tumorigenesis and perhaps other Aurora-A-mediated malignancies.

Several Aurora A drugs are in clinical trials. Because most known Aurora A substrates are mitotic targets, Aurora A autophosphorylation is used as a pharmacodynamic biomarker. Our data show that LIMK2 is highly expressed in breast cancer and prostate cancer due to Aurora A kinase activity. Aurora A inhibition rapidly degrades LIMK2 (half-life ∼2 hours), suggesting that it can be used as a pharmacodynamic biomarker for Aurora-A-targeted drugs. Sensitive pharmacodynamic biomarkers are essential for determining appropriate drug doses in clinical trials, thus preventing unnecessary toxicity.

Another interesting consequence of the feedback loop between Aurora A and LIMK2 was the observation that Aurora A inhibition using 1-NM-PP1 reduces Aurora A levels significantly (Fig. 2C). Because 1-NM-PP1 binds to the Aurora A active site, it is expected to have no effect on Aurora A stability. Several Aurora A inhibitors have been developed, but none of the studies have reported downregulation of Aurora A levels upon inhibition. Our data suggest that Aurora A downregulation upon inhibition is due to the loss of LIMK2-mediated suppression of Aurora A ubiquitylation. Interestingly, when MLN8237 was used to inhibit Aurora A in MDA cells, it also caused reduced LIMK2 and Aurora A levels. These results highlight the relevance of the mechanism identified in this study.

In conclusion, Aurora A upregulation is a reliable predictor of poor prognosis in breast cancer patients. LIMK2 has not previously been analyzed in breast cancer. Our data show that LIMK2 is a strong activator of cell motility, proliferation and transformation in breast cancer cells and tumorigenesis in vivo, suggesting that LIMK2 inhibition or ablation might be an alternative approach to modulate Aurora A deregulation in breast cancer. Most importantly, LIMK2 and Aurora A are engaged in a positive-feedback loop in several other cancer cells that were analyzed in this study, suggesting that it might be a common mechanism in Aurora-A-mediated malignancy. Because Aurora A is overexpressed in various types of cancer, analysis of LIMK2 and Aurora A levels could supplement standard staging information in primary biopsy samples. Results from these studies have the potential to facilitate the development of combination therapies using drugs targeted at both Aurora A and LIMK2.

Antibodies against Aurora A (H-130), α-tubulin (B-7), actin (C-2), histone H3 (Ser10)-R and LIMK2 (H-78) were purchased from Santa Cruz Biotech (Santa Cruz, CA). Matrigel was obtained from BD Biosciences (Bedford, MA).

HCT116, MDA-MB-231, HEK-293T, PC3 and PaCa2 cells were purchased from ATCC (Manassas, VA). All cell lines except for HEK293 cells were cultured in RPMI with 10% FBS supplemented with 2 mM glutamine and antibiotics (penicillin-G, streptomycin). HEK293T cells were cultured in DMEM with 10% FBS supplemented with 2 mM glutamine and antibiotics (penicillin-G-streptomycin).

HA-tagged LIMK2 was cloned into VIP3 mammalian vector and pTAT-HA vector at BamHI and XhoI sites. HA-tagged LIMK2 mutants were generated using overlapping PCR.

Aurora A and TPX2 were prepared using the baculovirus Bac-to-Bac expression system according to the manufacturer’s instructions (Invitrogen). Protein concentration was determined using Bradford assay, and the protein purity was assessed by western blotting using 6x-His antibody. 6x-His–LIMK2 was expressed in E. coli and purified as described before (Sun et al., 2008a; Sun et al., 2008b).

For generating stable cell lines, Aurora A and LIMK2 plasmids were transiently transfected using calcium phosphate into Phoenix cells. The retroviruses were harvested and used to infect MDA cells as reported previously (Shah and Shokat, 2002).

For in vitro labeling, 2 μg of 6x-His-tagged recombinant protein (such as LIMK2 or cofilin) was added into kinase buffer along with Aurora-A–TPX2 complex and 0.5 μCi of [γ-³²P]ATP. Reactions were terminated by adding SDS sample buffer, separated by SDS-PAGE gel and then transferred to PVDF membrane and exposed to Biomax MS film.

LIMK2 shRNAs were cloned into pLKO.1 TRC vector, which was a gift from David Root (Moffat et al., 2006). LIMK2 shRNA were designed as follows: (1) LIMK2-shRNA1 forward oligo, 5′-CCG GCC AAC TGG TAC TAT GAG AAC TCG AGT TCT CAT AGT ACC AGT TGG TTT TTG-3′ and reverse oligo, 5′-AAT TCA AAA ACC AAC TGG TAC TAT GAG AAC TCG AGT TCT CAT AGT ACC AGT TGG-3′; (2) LIMK2-shRNA2 forward oligo, 5′-CCG GGC TAT TCA CAG CAG ATC TTC TCG AGA AGA TCT GCT GTG AAT AGC TTT TTG-3′ and reverse oligo, 5′-AAT TCA AAA AGC TAT TCA CAG CAG ATC TTC TCG AGA AGA TCT GCT GTG AAT AGC-3′; (3) LIMK2-shRNA3 forward oligo, 5′-CCG GCC TGC TGA CAG AGT ACA TTC TCG AGA ATG TAC TCT GTC AGC AGG TTT TTG-3′ and reverse oligo, 5′-AAT TCA AAA ACC TGC TGA CAG AGT ACA TTC TCG AGA ATG TAC TCT GTC AGC AGG-3′. Aurora A shRNA were generated in our previous study (Johnson et al., 2011). Control shRNA (scrambled shRNA) and LIMK2 shRNA Aurora A were transfected into MDA cells using Lipofectamine (Invitrogen) following the manufacturer’s instructions. Alternatively, AA shRNA and LIMK2 shRNA lentiviruses were generated and used for infecting MDA cells. LIMK2-ablated MDA and LIMK2-ablated AA-MDA stable cells were generated following puromycin selection.

AA-MDA and AA-as7-MDA cells were treated with either DMSO or 1-NM-PP1 (250 nM) for 12 hours, followed by cell lysis. For the MLN8237 experiment, MDA cells were treated either with 1 μM MLN8237 for 12 hours or with 0.5 μM MLN8237 for 16 hours.

MDA, AA-MDA, LIMK2-ablated MDA and LIMK2-ablated AA-MDA cells were plated in RPMI (10³, 10⁴ and 10⁵ cells per dish in triplicate), 0.3% agar and 10% calf serum in six-well plates as reported previously (Shah and Shokat, 2002). Transformed colonies were counted after 3 weeks.

MDA, LIMK2-MDA or LIMK2-ablated MDA cells were treated with 2.5 mM thymidine for 16 hours, released for 8 hours, and then treated with thymidine for an additional 16 hours. After two washes with phosphate-buffered saline (PBS), cells were cultured for different times as indicated in the experiment and harvested.

Cells were harvested and lysed in modified RIPA buffer, supplemented with protease inhibitors. Equal amounts of cell extracts were then used for western blotting.

MDA cells were co-transfected with LIMK2 or Aurora A shRNA along with 6x-His–ubiquitin. After 36 hours, MG132 (Sigma) was added at 10 μM final concentration for additional 12 hours. Cells were then harvested, and ubiquitylated proteins were isolated using Ni-NTA beads. The proteins were separated by SDS-PAGE and analyzed using antibodies against Aurora A and LIMK2.

MDA, AA-MDA, LIMK2-ablated MDA and LIMK2-ablated AA-MDA cells were serum starved in serum-free RPMI for 12 hours and isolated by limited trypsin digestion. Cell migration was determined in Boyden chambers as reported previously (Shah and Vincent, 2005). The assays were performed in triplicate, four times. To allow for comparison between multiple assays, the data were normalized, and expressed as a percentage of the number of cells present on the membrane.

Cells were seeded in 96-well plates at 1500 cells per 100 μl per well and cultured for 6, 12, 24, 36, 48, 60 and 72 hours. MTT assay was conducted as reported previously (Sun et al., 2008a; Sun et al., 2008b; Sun et al., 2009). Experiments were repeated three times in quadruplicate wells to ensure the reproducibility of results.

MDA, LIMK2-MDA, LIMK2-ablated MDA or LIMK2-ablated AA-MDA cells were grown on poly-L-lysine coated coverslips for 24 hours, fixed with 4% formaldehyde in PBS for 15 minutes at room temperature, and then washed three times with PBS. The cells were permeabilized with 0.2% Triton X-100 in PBS for 5 minutes, washed twice with PBS, and blocked in 5% BSA in PBS for 2 hours at 25°C. Cells were labeled with antibodies (Aurora A, α-tubulin, phosphorylated histone H3 or LIMK2) for 3 hours in 1% BSA in PBS, followed by incubation with fluorescein-isothiocyanate- or Texas-Red-conjugated secondary antibody. Cells were counterstained with prolong antifade and visualized using a Nikon TE2000 inverted confocal microscope (Nikon, Tokyo, Japan) with a Radiance 2100MP Rainbow Laser (Bio-Rad Laboratories).

All the animal experiments were done in accordance with institutional guidelines of the Purdue Animal Care and Use Committee. Female athymic nude mice 4–5 weeks of age were obtained from Harlan Laboratories (Indianapolis, IN). 5- to 6-week-old nude mice weighing 18–22 g were anesthetized and inoculated with MDA (5×10⁶/mouse), AA-MDA cells (5×10⁶/mouse) and LIMK2-ablated AA-MDA cells (5×10⁶/mouse). AA-MDA cells were implanted on the right shoulders and LIMK2-ablated AA-MDA cells were implanted on the left shoulders of the same four mice. To compare the tumorigenic potential of MDA and AA-MDA cells, MDA cells were planted on the right shoulder, and AA-MDA cells were planted on the left shoulder of athymic nude mice in a different set of experiments. Before implantation, cells were harvested by trypsinization, washed twice with PBS and resuspended in 200 μl RPMI and matrigel in 1:1 ratio. Tumor growth was monitored and measured every 2 days in two perpendicular directions using a caliper (body weights were monitored on the same schedule), and the volumes of the tumors were calculated as 0.5×L×W², where L is the longest axis and W is the axis perpendicular to L in millimeters. All mice were sacrificed 34 days following inoculation, tumors were dissected and their sizes were compared. Mice bearing tumors did not display any weight loss compared with control mice at the time of sacrifice.

Bar graphs results are plotted as the average ± s.e.m. Significance was evaluated using Student’s t-test analysis and is displayed as follows: ^*P<0.05, ^**P<0.01, ^***P<0.001.

We thank David Root for providing pLKO.1 TRC vector, Eric Nigg (Max Planck Institute of Biochemistry, Germany) for Aurora A plasmid, Ora Bernard (St Vincent’s Institute, Victoria, Australia) for LIMK2 plasmid and the Purdue Bioscience Imaging Facility for confocal microscopy.