Constitutive regulation of mitochondrial morphology by Aurora A kinase depends on a predicted cryptic targeting sequence at the N-terminus

Aurora A kinase (AURKA) is a major regulator of mitosis and an important driver of cancer progression. The roles of AURKA outside of mitosis, and how these might contribute to cancer progression, are not well understood. Here, we show that a fraction of cytoplasmic AURKA is associated with mitochondria, co-fractionating in cell extracts and interacting with mitochondrial proteins by reciprocal co-immunoprecipitation. We have also found that the dynamics of the mitochondrial network are sensitive to AURKA inhibition, depletion or overexpression. This can account for the different mitochondrial morphologies observed in RPE-1 and U2OS cell lines, which show very different levels of expression of AURKA. We identify the mitochondrial fraction of AURKA as influencing mitochondrial morphology, because an N-terminally truncated version of the kinase that does not localize to mitochondria does not affect the mitochondrial network. We identify a cryptic mitochondrial targeting sequence in the AURKA N-terminus and discuss how alternative conformations of the protein may influence its cytoplasmic fate.


Introduction
Aurora A kinase (AURKA) was discovered as a mitotic kinase with a key role in bipolar spindle formation [1]. Its prominent localization to the microtubule spindle in mitosis is mediated through interaction of its kinase domain with the microtubule-associated protein TPX2 [2], an interaction that also activates the kinase activity of AURKA through stabilization of the active conformation of the T-loop [3]. Further studies have revealed the existence of alternative pathways to activation of AURKA [4][5][6] and an extensive array of cellular functions regulated by AURKA, in interphase as well as mitosis (reviewed in [7]). One such novel function that has been described for AURKA is the promotion of mitochondrial fission in preparation for mitosis [8].
Mitochondria form a highly dynamic network of interconnected tubules that undergo constant cycles of fission and fusion. These cycles are an essential feature of cellular homeostasis, thought to maintain a healthy mitochondrial population by allowing segregation of damaged mitochondria into the mitophagy pathway [9,10]. It is also proposed that fission/fusion cycles regulate the metabolic output of the cell, because a more interconnected network provides more efficient oxidative metabolism, and that mitochondrial morphology responds to metabolic cues. Mitochondrial morphology depends on the activity of GTPases dynamin-related protein 1 (Drp1) for fission and mitofusins (MFN1, MFN2) for fusion of the outer mitochondrial membrane (OMM) [11,12].
Mitochondrial fission also occurs prior to cell division in mammalian cells. Fragmentation of organelle networks (mitochondria, Golgi, ER) is a common feature of cell division, thought to improve the probability that organelles segregate equally between daughter cells [13,14] and to enable regulated segregation of aged mitochondria during stem cell-like divisions [15]. Mitochondrial fragmentation in mitosis may also be required for efficient microtubuledependent transport of mitochondria towards the cell periphery and away from the cell division plane [16]. Indeed, blocking mitochondrial fission or transport gives rise to cytokinesis failure and aneuploidy [11,17,18], an effect that can be rescued by disrupting mitochondrial fusion [19]. Previous studies have described a pathway leading to increased mitochondrial fission at mitotic entry via Cdk1-dependent activation of Drp1 at the OMM [13]. The mitotic kinase AURKA has been proposed to regulate these events through phosphorylation of RalA to promote RalA-dependent OMM recruitment of both Drp1 and RalBP1-cyclinB-Cdk1 complex [8].
Fragmented mitochondrial networks are a characteristic of cancer cells thought to contribute to the metabolic changes that accompany tumorigenesis, and have been shown to contribute to tumour progression and metastasis [20][21][22]. Overexpressed AURKA (located on the 20q amplicon) is also strongly associated with cancer [23,24]. Given a large number of mitochondrial hits we identified in a search for AURKA interactors, we decided to investigate further the role of AURKA in influencing the fragmentation state of the mitochondrial network in human cell lines. We report here that the mitochondrial network is sensitive to AURKA activity at all phases of the cell cycle, and that this sensitivity contributes to divergent mitochondrial morphology between two cell lines (one transformed and the other nontransformed) expressing different levels of AURKA. Furthermore, we identify an interphase subpopulation of AURKA associated with mitochondria, both by immunofluorescence and fractionation assays. This association depends on the AURKA N-terminal region, which contains a cryptic mitochondrial targeting sequence.

Results and discussion
In using proteomic approaches to identify co-purifying proteins in AURKA-GFP pulldowns from human U2OS cells, we identified large numbers of mitochondrial proteins (electronic supplementary material, figure S1A). We confirmed these potential interactions by reciprocal pulldowns of several GFP-tagged mitochondrial markers, including the OMM translocase complex components TOMM20 and TOMM70 and the inner mitochondrial membrane component Prohibitin (PHB), which identified endogenous AURKA as a partner protein in cell extracts (electronic supplementary material, figure S1B). Repeated experiments suggested that the quantity of AURKA in pulldowns of mitochondrial proteins was at least 10-fold less than that found in pulldowns of TPX2, a major interactor of AURKA in mitotic cells [2].
Given the previous study suggesting that mitochondrial fission was controlled by a cytoplasmic pool of AURKA [8], we investigated the relevance of our own finding-of the association of AURKA with mitochondrial proteins-to mitochondrial morphology. First, we tested whether depletion of endogenous AURKA, using siRNA-mediated knockdown (AURKA-i), would affect mitochondrial morphology in the immortalized retinal pigment epithelial line hTERT-RPE-1 (RPE-1). We found that AURKA-i resulted in markedly reduced fragmentation of the mitochondrial network, observable as a more interconnected network and quantifiable as a change in length distribution of individual mitochondria (figure 1a; electronic supplementary material, figure S2A, B). To confirm the specificity of this effect, we treated RPE-1 cells, expressing an endogenous mRuby-PCNA marker to distinguish cell cycle phases [25], with the small-molecule-specific inhibitor of AURKA, MLN8237. We found that the length of mitochondria varied with cell cycle phase as expected [13,26], being shortest in G1 phase (figure 1b), but that AURKA inhibition led to elongation of mitochondria in all cell cycle phases (figure 1c). Therefore, we conclude that AURKA can regulate mitochondrial organization throughout the cell cycle. To determine whether this response of mitochondria to AURKA is a conserved role of the kinase, we treated cultured Drosophila melanogaster D.mel-2 cells with MLN8237 (figure 1d). The increase in mitochondrial connections and length following treatment suggest that this role for AURKA is conserved in metazoans.
We noticed that cancer cell lines used in our laboratory (e.g. U2OS) had mitochondrial networks in a more fragmented state than non-transformed RPE-1 cells (figure 2a; see also electronic supplementary material, figure S2). Given our finding that AURKA influences mitochondrial morphology, and the well-documented overexpression of AURKA in cancer cells, we tested whether AURKA expression levels might contribute to differences in mitochondrial morphology observed between RPE-1 and U2OS cells. We determined relative expression levels in extracts prepared from known numbers of cells in asynchronous populations, normalized against a panel of cellular proteins (figure 2b). This revealed that U2OS cells contain twofold higher levels of AURKA protein normalized against cell number, and fourfold to sixfold more when normalized against the levels of tubulin, actin or ATP5A1. Therefore, higher AURKA levels correlate with a more fragmented mitochondrial network. Next, we tested whether manipulating AURKA levels would reproduce observed patterns of mitochondrial organization. We used partial siRNA-mediated knockdown to achieve sixfold reduction of AURKA levels in U2OS cells, and found a corresponding lengthening of mitochondria under these conditions (figure 2c,d). Connectivity of the network was also greater when AURKA levels were reduced. Conversely, we found that tetracycline-induced overexpression of a stable AURKA-Venus transgene in RPE-1 cells resulted in a small but significant decrease in the length of mitochondria (figure 2e,f ). In further experiments, we found that transient overexpression of AURKA-Venus in another untransformed line, breast epithelial MCF10A cells, also led to increased fragmentation of the mitochondrial network in cells overexpressing AURKA (electronic supplementary material, figure S2C). Finally, MLN8237 treatment of HCC1143 cells, a breast cancer line characterized by a highly fragmented mitochondrial network, caused an increase in mitochondria length (electronic supplementary material, figure S2D). Therefore we conclude that the different fragmentation state of the mitochondrial network in different cell lines is influenced by AURKA expression. rsob.royalsocietypublishing.org Open Biol. 8: 170272  rsob.royalsocietypublishing.org Open Biol. 8: 170272  Given the prevalence of mitochondrial hits in our AURKA interactome, and the functional relationship between AURKA and mitochondrial organization, we hypothesized that part of the cytoplasmic pool of AURKA might be associated with the mitochondrial network. In support of this, we found that both endogenous and exogenous AURKA co-purify with mitochondria in extracts fractionated by centrifugation (figure 3a). We also examined fixed cells for co-localization of endogenous AURKA with a mitochondrial marker, TOMM20. Immunostaining with two different antibodies against AURKA revealed punctate cytoplasmic staining in which the AURKA 'dots' were frequently apposed to mitochondria, often close to points of mitochondrial constriction (figure 3b). We then sought to confirm that these dots were AURKA-specific using siRNA to deplete endogenous AURKA. This led to the reduction of the punctate staining associated with mitochondria in accord with the downregulation of AURKA by AURKA-i treatment (electronic supplementary material, figure S3A, B). We sought to confirm our finding in live cells, by examining the cytoplasmic localization of AURKA-Venus without fixation. When AURKA-Venus was expressed exogenously from a transgene (as in figure 2e, also electronic supplementary material, figures S2C, S4A), levels of cytoplasmic AURKA were too high to distinguish co-localization patterns in the high cytoplasmic background. To circumvent this issue, we targeted one allele of endogenous AURKA at the C-terminus with mVenus, taking advantage of rAAVmediated homologous recombination (electronic supplementary material, figure S3C). Endogenous mVenus-tagged AURKA was expressed to the same level as untagged endogenous AURKA, was comparably enriched in cells arrested in mitosis (electronic supplementary material, figure S3D, E) and faithfully recapitulated the known localization of untagged AURKA in the cell cycle (electronic supplementary material,  rsob.royalsocietypublishing.org Open Biol. 8: 170272 figure S3F). When we examined endogenous AURKA-mVenus, we could detect a similar pattern of co-localization of cytoplasmic dots of Venus fluorescence with constrictions in mitochondria in the living cells as we had seen when staining fixed cells for endogenous AURKA (figure 3c). Finally, we carried out super-resolution microscopy on an OMX system and used three-dimensional reconstruction of Z-stacks to confirm that punctate GFP-labelled structures in fixed RPE-1-AURKA-Venus cells frequently coincided with constrictions in the mitochondrial network ( figure 3d). Therefore, we concluded that a small fraction of cytoplasmic AURKA resides at mitochondria. The interaction we identified between translocase of the outer membrane (TOM) complex components and AURKA raised the possibility that AURKA might be targeted to mitochondria through a mitochondrial targeting sequence (MTS) [27,28]. These are characterized by an amphipathic helix at the N-terminus. The known structure of AURKA [29] excludes the N-terminal region, which is generally considered to be unstructured. We tested if AURKA lacking its N-terminal 31 amino acids (AURKAD31) would still localize in the mitochondrial fraction in cell fractionation experiments, and we found that it did not ( figure 4a,b). Moreover, AURKAD31 overexpression did not cause fragmentation of mitochondria (figure 4c). We concluded that the N-terminal region of AURKA contains sequences necessary for its localization with mitochondria and that this localization is required for mitochondria to fragment in response to the kinase's activity. To further test this idea, we fused an MTS onto the N-terminus of AURKAD31. As predicted, expression of this gene fusion restored fragmentation of mitochondria (electronic supplementary material, figure S4A, B).
These findings led us to ask whether the natural N-terminal domain of AURKA had any of the characteristics of MTSs. We found that AURKA's N-terminal region is indeed polybasic and has the periodic hydrophobicity consistent with that of an amphipathic helix (figure 4d). This predicted amphipathic helix is a phylogenetically conserved secondary feature, even where the primary amino acid (AA) sequence diverges (electronic supplementary material, figure  S4C). When we analysed the AURKA sequence using the bioinformatic algorithms, TargetP and MitoProt [31,32], we identified a 'cryptic' targeting sequence from AA 7-55. That is, whereas the full-length AURKA sequence was not predicted to localize to mitochondria, in silico removal of the first 6 AAs from its N-terminus generated a predicted mitochondrial localization signal (probability in MitoProt, p . 0.95; figure 4e). Therefore, we tested the prediction that N-terminally clipped versions of AURKA, AURKAD6 and AURKAD8, should localize more strongly to the mitochondrial network than the full-length protein. Comparing the localization of AURKA-Venus with AURKAD6/D8-Venus in fixed cells, we found that in both RPE-1 and U2OS cells, the clipped versions showed a more prominent co-localization with mitochondria, consistent with the presence of cryptic mitochondrial targeting information in the AURKA N-terminus (figure 4g; electronic supplementary material, figure S5). Moreover, mitochondrial fragmentation was also enhanced in the presence of AURKAD6/D8-Venus, supporting the idea that mitochondrial targeting of AURKA promotes mitochondrial fission (figure 4f).
A large number of proteins are described to interact with AURKA [33]. Some of these interactions may depend upon alternative conformations of the N-terminus. Thus, for example, the so-called A-box region can mediate interactions with either the APC/C cofactor Cdh1 or with Calmodulin, in a manner that may be regulated by phosphorylation on Ser51 [5,34]. We suggest that factors affecting the folding of the N-terminal region of AURKA could determine the localization and fate of the kinase by directing alternative interactions. One form of AURKA in this conformational space would display a functional MTS to mediate its mitochondrial targeting. This conformation would be favoured by AURKAD6.
Is AURKA on the surface of mitochondria, as predicted by its known role in promoting Drp1-mediated fission, or on the inside, as predicted by its MTS? Studies of mitochondrial regulation by mitotic cyclin-dependent kinase (Cdk) activity point to roles for cyclinB1 -Cdk1 both at the mitochondrial surface, in regulating Drp1 activity [8], and-in a different study-inside the mitochondrial matrix, in regulating the respiratory chain via Complex I phosphorylation [35]. The targeting sequence on AURKA suggests that it would be translocated into the mitochondria via the TOM complex. Indeed, we have found direct interaction with the TOMM20 and TOMM70 members of this complex and also identified ATP5A subunits and other matrix components in our proteomic survey of AURKA interactors (electronic supplementary material, figure S1A).
The 'MAGIC' pathway, recently described by the Rong Li laboratory, diverts aggregation-prone proteins into mitochondria by an unknown mechanism that results in their ubiquitinindependent clearance by mitochondrial proteases [36]. It is possible that AURKA is an aggregation-prone protein, given the unstructured nature of its extended N-terminus. Exogenously expressed protein would be more likely to be processed via this pathway than endogenous proteins, because it would be more likely to be expressed in the absence of the correct binding partners, and we certainly cannot exclude that this pathway is responsible for co-localization of AURKA with mitochondria. However, both endogenous and exogenous AURKA localize in the same way, presumably sharing the pathway that contributes to the dynamics of the mitochondrial network.
Mitochondrial fragmentation is accompanied by increased glycolysis and decreased oxidative phosphorylation, metabolic changes that are thought to play a critical role in cell fate decisions (discussed in [37]). A switch to high glycolytic flux and low oxidative metabolism is a condition for reprogramming of iPS cells, with the state of high pluripotency being characterized by a fragmented mitochondrial network. It has been reported that AURKA is upregulated in reprogramming of iPS cells, although-curiously-inhibition of AURKA appears to enhance the process [38]. Our finding that the mitochondrial network is sensitive to AURKA levels in different cell types highlights the importance of elucidating the non-mitotic roles of AURKA in order to fully understand its contributions to cell proliferation and to cancer.
RPE-1-mRuby-PCNA cells were previously described [25]. RPE-1-AURKA-Venus (Flp-In) lines were derived as polyclonal populations by pooling cells transfected with pcDNA5-FRT/TO-AURKA-Venus and Flp-recombinase ( pOG44) after 12 days of selection in 500 mg ml 21 geneticin. Expression of AURKA-Venus is achieved by supplementing cell culture medium with 1 mg ml 21 tetracycline (tet; Calbiochem, San Diego, CA, USA). U2OS-AURKA-Venus cells were obtained by clonal selection with 170 mg ml 21 hygromycin B (Calbiochem) after transfection of U2OS tet-OFF cells with pTRE-AURKA-Venus : pCMVhygro in a ratio of 10 : 1. Expression of AURKA-Venus in these cells is achieved with extensive washing in PBS and switching to tet-free cell culture medium. To generate the AURKA-mVenus knock-in, RPE-1 FRT/TO cells were infected with recombinant adeno-associated virus particles harbouring mVenus cDNA flanked by approximately 1500 bp homologous to the AURKA locus as previously described [42]. To identify positive integrands, cells were treated with 10 mM DMA 12 h before single-cell sorting by flow cytometry. mVenus-positive cells were verified by fluorescence microscopy and immunoblot analysis (electronic supplementary material, figure S3C-F). D.mel-2 cells were cultured at 258C in Express Five w serum-free medium supplemented with antibiotics and 2 mM L-glutamine. siRNA duplex targeting human AURKA and control siRNA duplex against GL2 have been previously described [43] (Sigma-Aldrich).
Transfections were carried out using the Invitrogen Neon w system to electroporate cells with siRNA or plasmids, according to the manufacturer's instructions.
For mitochondrial imaging in living cells, cells were incubated in 100 nM MitoTracker TM Red CMXRos (Thermo Fisher Scientific) or Mito-ID w Green (Enzo) diluted in filming medium (see below) for 15 min. MitoTracker TM /Mito-ID w were replaced with fresh filming medium prior to time lapse.
Cells were treated with 100 nM MLN8237 during timelapse experiments, by addition of drug in a volume not less than 1/100 of the existing dish volume and gentle mixing of the medium on the dish.

Cell extracts, fractionation and immunoblotting
Whole-cell extracts were prepared by scraping cells directly into 2Â SDS Sample Buffer with 10 mM DTT. Samples were syringed to shear DNA or sonicated and boiled at 958C for 5 min prior to SDS-PAGE on 4 -12% precast gradient gels (Invitrogen). Transfer onto Immobilon-P or Immobilon-FL membranes was carried out using the XCell IITM Blot Module according to the manufacturer's instructions. Membranes were blocked in PBS, 0.1% Tween-20, 5% dried milk, and processed for immunoblotting with the primary antibodies indicated in the figures. Secondary antibodies used were HRP-conjugated, or IRDye w 680RD-or 800CW-conjugated for quantitative fluorescence measurements on an Odyssey w Fc Dual-Mode Imaging System (LI-COR Biosciences). For cell fractionation experiments, cells were harvested and treated at 48C with 40 mg ml 21 digitonin in 10 mM Tris-HCl ( pH 7.4), 100 mM NaCl, 25 mM MgCl 2 , 1 mM Na 3 VO 4 , 1 mM NaF and EDTA-free protease inhibitor cocktail (Roche). Cells were disrupted by 10 passages through a 25-gauge needle (whole-cell extract). Cell nuclei were pelleted by centrifugation at 3000 r.p.m. for 10 min in a microfuge at 48C. The resulting supernatant was re-centrifuged at 13 000 r.p.m. for 15 min to yield mitochondrial ( pellet) and cytoplasmic (supernatant) fractions.

Immunofluorescence analysis
Cells were grown on 13 mm glass coverslips, synchronized and fixed using either 100% MeOH at -208C or 4% PFA treatment at ambient temperature. Mitochondria were detected using anti-TOMM20 (Santa Cruz), AURKA using anti-AURKA (rabbit antibody Abcam ab1287 or mouse antibody BD Transduction Laboratories 610 939) and AURKA-Venus using anti-GFP (Abcam ab290). Secondary antibodies used were Alexa w Fluor 488-anti-rabbit and Alexa w Fluor 568-anti-mouse (Thermo Fisher Scientific). Cells were stained according to standard protocols in PBS buffer containing 0.1% Triton X100 and 3% BSA and mounted in ProLong w Gold antifade (Thermo Fisher Scientific). Images were captured using an Axiovert 200 M fluorescence microscope (Carl Zeiss Inc.) with a 100Â NA 1.4 oil objective and Coolsnap HQ2 camera (Photometrics) controlled by the METAMORPH w software. Images were deconvolved using 10 iterations of the AUTOQuant X2 (Media Cybernetics) blind deconvolution algorithm and presented as maximumintensity projections of 10 Â 0.1 mm stacks using IMAGEJ (Molecular Devices LLC). Super-resolution images were acquired on a Deltavision Optical Microscope eXperimental (OMX) 3D-SIM System V3 and analysed using SOFTWORX (Applied Precision) and IMAGEJ Volume Viewer plugin. Intensity level, contrast and brightness of images were adjusted using Adobe PHOTOSHOP where indicated.

Live cell microscopy
Cells were seeded onto 8-well plastic-bottom slides (Ibidi GmbH, Martinsried, Germany) at a density of 8 Â 10 4 cm 22 . The imaging medium was Leibowitz L-15 (Thermo Fisher Scientific) supplemented with FBS and antibiotics as rsob.royalsocietypublishing.org Open Biol. 8: 170272 described above. Expression of AURKA-Venus was induced 24 h prior to imaging, which was carried out on an Olympus CellR widefield imaging platform comprising an Olympus IX81 motorized inverted microscope, Orca CCD camera (Hamamatsu Photonics, Japan), motorized stage (Prior Scientific, Cambridge, UK) and 378C incubation chamber (Solent Scientific, Segensworth, UK) fitted with appropriate filter sets and a 60Â NA 1.42 oil objective. Images were acquired using the Olympus CELLR software as 1 mm stacks and exported as 12 bit TIFF stacks for display as maximum-intensity projections in IMAGEJ.

Mitochondrial morphology analysis
MitoTracker TM images were analysed using the MICROP software [44] to derive measurements of mitochondrial length. All analyses and statistical significance tests were verified through a separate manual analysis of the same images using the lengths of 30 mitochondria per cell, selected by beginning at the 12 o'clock position and moving clockwise around the nucleus. MICROP results were plotted in R as kernel density estimations to derive the presented probability density curves (see electronic supplementary material, figure S2A). Statistical significance of probability density distributions were tested using a two-sample Kolmogorov-Smirnov (K-S) test on the raw measurements as indicated in figure legends and illustrated in the electronic supplementary material, figure S2B. All results reported in this study were also significant according to the Mann-Whitney U-test, p , 0.01.
Data accessibility. The datasets supporting this article have been uploaded as part of the electronic supplementary material.