Cancer and Metastasis Reviews

, Volume 28, Issue 1, pp 85–98

Centrosomes and cancer: how cancer cells divide with too many centrosomes


  • Susana A. Godinho
    • Department of Pediatric Oncology, Dana-Farber Cancer InstituteHarvard Medical School
  • Mijung Kwon
    • Department of Pediatric Oncology, Dana-Farber Cancer InstituteHarvard Medical School
    • Department of Pediatric Oncology, Dana-Farber Cancer InstituteHarvard Medical School
    • Department of Pediatric Hematology/OncologyChildren’s Hospital
    • Howard Hughes Medical Institute

DOI: 10.1007/s10555-008-9163-6

Cite this article as:
Godinho, S.A., Kwon, M. & Pellman, D. Cancer Metastasis Rev (2009) 28: 85. doi:10.1007/s10555-008-9163-6


Precise control of centrosome number is crucial for bipolar spindle assembly and accurate transmission of genetic material to daughter cells. Failure to properly control centrosome number results in supernumerary centrosomes, which are frequently found in cancer cells. This presents a paradox: during mitosis, cells with more than two centrosomes are prone to multipolar mitoses and cell death, however, cancer cells possessing extra centrosomes usually divide successfully. One mechanism frequently utilized by cancer cells to escape death caused by multipolar mitoses is the clustering of supernumerary centrosomes into bipolar arrays. An understanding of the molecular mechanisms by which cancer cells can suppress multipolar mitoses is beginning to emerge. Here, we review what’s currently known about centrosome clustering mechanisms and discuss potential strategies to target these mechanisms for the selective killing of cancer cells.



1 The centrosome

The centrosome, the primary microtubule-organizing center (MTOC) in animal cells, regulates the nucleation of microtubules (MTs) as well as their spatial organization [for review see [1]]. In addition to its facilitating role in mitosis, the centrosome influences MT-dependent processes in interphase: cell shape, cell polarity and cell motility.

A single centrosome contains two centrioles, joined by fibers connecting their proximal ends. The centrioles are embedded in an electron-dense protein matrix, known as the pericentriolar material (PCM). Centrioles within a pair are structurally different at their distal ends, the mother centriole contains a set of distal appendages important to anchor MTs, while the daughter centriole does not [2]. Centrioles are also important for the recruitment of the PCM, as centriole loss leads to PCM dispersal and loss of centrosome integrity [37].

Like DNA, centrioles normally duplicate in a semi-conservative fashion once per cell cycle during S-phase, ensuring that each daughter cell inherits one centrosome at the end of cell division. The centrosome cycle can be subdivided into five distinct steps: centrosome segregation (M-phase), centriole disengagement (late M-phase/G1-phase), centriole duplication (S-phase), centrosome maturation and centrosome separation (G2-phase) [reviewed in [1, 8, 9]. Recently, the PCM has been shown to have an important role in centriole duplication by controlling the formation and stabilization of daughter centrioles [10, 11], suggesting that centrioles and PCM have a symbiotic relationship for their assembly and stabilization.

Coordination of the DNA and centrosome replication cycles is essential to avoid centrosome overduplication [1]. Indeed, defective coordination between the DNA and centrosome cycles is commonly observed in many cancers [12, 13]. The mechanisms that control centrosome duplication are only recently starting to be understood [1, 8, 9, 14]. Similar to the DNA cycle, the control of centrosome duplication involves: (1) centrosome-duplication machinery; (2) licensing molecules essential for centriole duplication; (3) activities that integrate licensing with cell cycle controls, maintaining the coordination of both processes [for review see [1, 8, 14]].

2 Centrosomes and cancer

  1. 1.

    Extra centrosomes and Cancer


Centrosome amplification (supernumerary centrosomes) is a common characteristic of most solid and hematological cancers examined to date [12, 13, 1517]. Extra centrosomes in cancers can originate by several mechanisms: cell fusion, centrosome overduplication, de novo synthesis of centrosomes, mitotic slippage and cytokinesis failure [13, 1820]. One way to distinguish whether centrosome amplification is derived from overduplication of centrosomes or from cytokinesis failure is to specifically label mature centrioles [21]. The presence of only one mature centriole among numerous immature centrioles indicates centriole overduplication rather than cell division failure [21]. Although this type of analysis is an important technical advance, it has yet to be applied to primary tumors on a large-scale basis. Thus, the relative frequency that each of these different paths contributes to centrosome amplification in human tumors remains unknown. Whatever the cause leading to extra centrosomes, the percentage of cancer cells containing extra centrosomes within a given tumor is highly variable. Why some cancer cells maintain their extra centrosomes more than others, how they maintain them, and whether extra centrosomes are beneficial or deleterious to cancer cells remain interesting open questions.

The link between supernumerary centrosomes and tumorigenesis dates back over 100 years. Extra centrosomes may contribute to tumorigenesis by promoting genetic instability through multipolar mitoses or by disrupting cell polarity and asymmetric cell division [13, 15, 19, 22]. Genetic instability in human cancers was first hinted at in 1890 by the German pathologist, David Hansemann, who observed the presence of bipolar divisions but with asymmetric segregation of chromosomes in human epithelial tumors [23]. Following Hansemann’s report, in 1902, Theodor Boveri recognized that multipolar divisions in the presence of extra centrosomes could have similar consequences; each daughter cell inherits an incorrect and unequal set of chromosomes [24]. These findings led Boveri to propose a simple but appealing hypothesis that tumors can originate from premalignant precursors and that aneuploidy resulting from multipolar divisions in cells harboring supernumerary centrosomes has a causal role in tumorigenesis [19]. This idea is further supported by a strong correlative relationship between extra centrosomes, aneuploidy and tumor aggressiveness in tumor-derived cell lines, mouse models and human patients [15, 16, 2527]. Deregulation of many genes implicated in human cancers leads to centrosome amplification. For example, centrosome overduplication can result from mutation of the p53 tumor suppressor, a genetic alteration associated with more than half of human tumors [28]. Moreover, centrosome overduplication is associated with aneuploidy in nucleophosmin (NPM) haploinsufficient mice. NPM haploinsufficiency induced tumorigenesis and this was accompanied by unrestricted centrosome duplication and genetic instability [29]. Likewise, many cell division errors can lead to cytokinesis failure which is linked, by a series of recent studies, to genetic changes in tumors [20, 3034]. Fujiwara et al. demonstrated that tetraploid cells containing extra centrosomes could promote genetic instability and tumorigenesis in a mouse mammary model [35]. Several recent studies observed a similar tumor-promoting effect of tetraploidy [18, 36].

These studies provide experimental support for Boveri’s hypothesis, but many important issues remain unresolved. How can we disentangle the effects of a specific mutated gene from the effects of extra centrosomes? In the case of tetraploidy, how do we separate the effects of extra centrosomes from extra chromosomes? An important recent advance comes from Basto et al. [37]. This paper elegantly demonstrated that extra centrosomes per se can initiate tumorigenesis in flies overexpressing polo-like kinase 4, SAK/PLK4, a kinase known to be crucial for centriole duplication [4, 3739]. The surprising aspect of this study is that tumorigenesis is accompanied by only a minor increase in aneuploidy. Instead, in this model, the primary effect of extra centrosomes appears to be the disruption of asymmetric cell division and normal cell differentiation, thereby increasing the proportion of proliferating progenitor cells [37]. This effect is not specific to extra centrosomes, because abnormal asymmetric cell division and tumorigenesis also occurs in cells lacking centrosomes [40]. Thus, these studies illustrate an interesting mechanism by which cell division defects can affect lineage commitment and tumorigenesis. However, even in this model we cannot take aneuploidy completely off the table. The tumor cells initiated by centrosome amplification vary strikingly in cell size, consistent with the possibility that they are aneuploid or polyploid (Basto et al., personal communication). Therefore, although aneuploidy may not be an initiating event in this model, aneuploidy could affect late stages of tumor development. How frequently the impairment of asymmetric cell division occurs as a tumor initiating event in human cancers is an important question that needs to be addressed in future work.
  1. 2.

    Cancer cell division: the extra centrosome paradox


The idea that extra centrosomes lead to multipolar mitoses, aneuploidy, and thus tumor formation is based on the correlation between extra centrosomes, multipolar mitoses and tumorigenesis. However, this does not account for the fact that multipolar mitoses are likely to be detrimental to cancer cells because multipolar mitoses can lead to gross chromosome missegregation and death. A recent study by Weaver et al. highlights the fact that there is no black-and-white answer for the relationship between aneuploidy and tumorigenesis [41]. The degree of aneuploidy and the context in which it occurs can result in different outcomes: moderate levels of genetic instability can induce tumorigenesis, whereas high levels of genetic instability, which might be intolerable to cancer cells, can instead suppress tumorigenesis in vivo [41]. Using mice heterozygous for the mitotic kinesin CENP-E, which is important for chromosome segregation and mitotic checkpoint function, the authors demonstrate that one or few missegregations of whole chromosomes could promote tumorigenesis in some tissues. Interestingly, when this mouse was treated with a chemical mutagen or crossed with a mouse deficient for the tumor suppressor gene ARF which by itself promotes tumor development, suppression of tumorigenesis was observed. Whether or not the level of aneuploidy in those tumors is higher than that of CENP-E heterozygous mice was not examined in this study [41]. Altogether, high levels of aneuploidy expected from a multipolar mitosis might set unfavorable conditions for the survival of cancer cells with supernumerary centrosomes.

How then do cancer cells “cope” with extra centrosomes? For tumors with a small number of extra centrosomes-containing cells, coping mechanisms may not be important. A minor fraction of cells undergoing multipolar mitoses and death might have only minimal effects on the growth of the tumor. However, as the fraction of cells with extra centrosomes within a tumor rises, there might be a stronger requirement to suppress multipolar mitoses.

Several mechanisms exist to limit the detrimental consequences of supernumerary centrosomes and enable a bipolar mitosis. These include centrosome removal, centrosome inactivation, asymmetric segregation of centrosomes followed by clonal expansion and centrosome clustering (Fig. 1). Centrosome elimination is a common event during the production of gametes. In most organisms, the bipolar mitosis during the first zygotic cell division depends on the paternal contribution of centrosomes [42]. Maternal centrosomes are eliminated during oogenesis to prevent multipolar mitosis resulting from supernumerary centrosomes. In C. elegans, the cyclin dependent kinase (cdk) inhibitor Cki-2 is responsible for the removal of maternal centrosomes in the oocyte as evidenced by the persistent presence of maternal centrioles in oocytes deficient for Cki-2 [43]. Thus, the restriction of the levels of the G1/S Cdk, Cdk2/cyclin E, ensures that the first mitosis occurs in a bipolar fashion. Although this result demonstrates a role of Cdk/cyclin E in centrosome stabilization, the precise mechanism by which maternal centrosomes are eliminated/degraded was not defined. In principle, centrosomes could either be degraded or extruded from cells. Dictyostelium cells can eject extra centrosomes by cystoplast formation, a mechanism reminiscent of polar body exclusion in female meiosis (Fig. 1a) [44].
Fig. 1

Mechanisms to suppress multipolar mitoses in cells with extra centrosomes. Multiple distinct mechanisms are employed to allow bipolar mitoses in cells with extra centrosomes. (a) Centrosome extrusion; cells can expel extra centrosomes by the formation of cytoplasts containing centrosomes. (b) Centrosome inactivation; a bipolar mitosis is achieved in the presence of extra centrosomes by silencing MTOC activity of additional centrosomes. Some extra centrosomes are incapable of function as MTOC due to loss of pericentriolar material (PCM). Green circles around a pair of centrioles represent the PCM. (c) Asymmetric segregation of centrosomes and potential clonal expansion; centrosomes can be asymmetrically segregated into two daughter cells. A daughter cell that inherits one centrosome forms a bipolar spindle in a subsequent mitosis. (d) Centrosome clustering; cells can coalesce their extra centrosomes into two groups to form a bipolar spindle

In addition, supernumerary centrosomes can be inactivated, allowing only two centrosomes to function as MTOCs during mitosis (Fig. 1b). In physiologically polyspermic newt eggs, all sperm nuclei incorporated into the egg develop sperm asters, however eventually only one sperm nucleus and associated centrosomes contribute to bipolar mitosis of the zygote [45]. Several studies demonstrated that a gradient of cyclin B1 and γ-tubulin distribution (high in the animal hemisphere while low in the vegetal hemisphere) is responsible for the preferential accumulation of cyclin B1 and γ-tubulin with the centrosome associated with the principal sperm nucleus [46]. This ensures that only the centrosome associated with the zygote nucleus located in the animal hemisphere contributes to the formation of bipolar spindle whereas centrosomes associated with accessory nuclei degenerate [46, 47]. MTOC activity of the centrosomes is regulated not only by the accumulation of PCM but also by their phosphorylation, dependent upon cdk1/cyclin B [48]. Thus, a gradient of cyclin B1 and γ-tubulin in a large oocyte may confer selectivity to centrosomes to be maintained in this study. In addition, given that PCM also have a important role in centrioles duplication by the formation and stabilization of daughter centrioles [6, 10], loss of PCM could account for the mechanism for centrosome degeneration by limiting their duplication potential in this system. Centrosome inactivation also seems to occur in flies overexpressing SAK/PLK4: centrosomes that are scattered along spindles contain significantly less γ-tubulin than those at the spindle poles [37]. Thus, a decrease in PCM levels may account for, at least partially, the inactivation of extra centrosomes. However, the mechanisms leading to selective removal or inactivation of centrosomes still remain a mystery. How frequently the removal or inactivation of extra centrosomes occurs in mammalian cells and what distinguishes the centrosomes to be removed or inactivated from those that form the MTOCs is largely unknown.

Alternatively, extra centrosomes may be segregated asymmetrically during cell division (Fig. 1c). In this case, the daughter cell that inherited one centrosome would proliferate and take over the population if it harbors sufficient genetic material or any mutation to support their survival [49]. Predominantly in cancer cells, supernumerary centrosomes can be clustered into two groups to achieve a bipolar mitosis (Fig. 1d) [13, 5052]. Even in Dictyostelium and SAK overexpressing flies, where centrosome removal or inactivation was observed, centrosome clustering appears to be a dominant mechanism to prevent multipolar mitosis [37, 44].

Centrosome clustering has long been observed in a variety of tumor cells (Fig. 2) [5155]. The classic example demonstrating centrosome clustering is the mouse neuroblastoma N1E-115 cell line in which nearly 100% of cells contain extra centrosomes but still undergo bipolar divisions by clustering extra centriole pairs as judged by centrin staining [51]. Many cancers with a large fraction of cells containing extra centrosomes (>40%), such as the human breast cancer cell MDA-231 and the mouse neuroblastoma NHO2A, also appear to cluster extra centrosomes efficiently [52]. In addition, there is also abundant indirect evidence of centrosome clustering, especially in tissue sections. There are reasons to suspect that immunohistochemical methods using PCM markers, such as γ-tubulin, for analyzing pathological specimens may underestimate the numbers of centrosomes, particularly if they are tightly clustered [12, 13, 56]. Despite technical difficulties, several studies have utilized serial section EM of tissue sections and observed that tumors containing extra centrioles frequently undergo bipolar mitoses in vivo, possibly due to centrosome clustering [26, 56]. More directly, using centriole markers, Basto et al. showed that flies containing cells with extra centrosomes are viable due to robust centrosome clustering [37]. The development of automated techniques to reliably count centrosomes in tissue sections would permit the identification of more classes of cancers that harbor extra centrosomes and would also allow to accurately measure the frequency of the centrosome clustering phenomenon in vivo [57].
Fig. 2

Cancer cells can cluster extra centrosomes to achieve a bipolar mitoses. The cartoon illustrates a normal bipolar spindle with 2 centrosomes (a) and a bipolar spindle with clustered extra centrosomes (b). c Image from the colon cancer cell line CaCo-2 showing the clustering of supernumerary centrosomes. Cells were stained for centrioles (centrin, green), MTs (α-tubulin, red) and DNA (blue)

In principle, centrosome clustering could be a unique adaptation mechanism in cancer with extra centrosomes. Alternatively, and perhaps more intuitively, mechanisms to cluster centrosomes may exist in normal cells, but only cells with extra centrosomes depend on this machinery to survive [13, 58, 59]. Consistent with the latter hypothesis, several non-transformed cell types are known to cluster extra centrosomes in various circumstances: after cytokinesis failure [52, 60, 61], during normal development, or in the setting of stress or disease [58, 59]. Nevertheless, the efficiency of centrosome clustering varies among cancer cell lines [52]; it remains an interesting possibility that under selective pressure for bipolar mitoses, the strength of these pathways is modulated in different tumor types.

3 Mechanisms to facilitate centrosome clustering

Despite its potential implications for cancer biology, the mechanisms mediating centrosome clustering have not been well studied. However, recent studies suggest that multiple redundant mechanisms facilitate the clustering of supernumerary centrosomes. Centrosome clustering is achieved by a combination of spindle-intrinsic MT binding forces and actin-regulating forces at the cell cortex whose distribution is regulated by the interphase adhesion pattern. In addition, the spindle assembly checkpoint (SAC) ensures that cells have enough time to cluster extra centrosomes before anaphase onset.
  1. 1.

    Spindle assembly checkpoint (SAC)


The SAC is a signaling network that delays anaphase onset if kinetochores are improperly attached to the mitotic spindle [reviewed in [62]. Recent studies suggest that the SAC is one important factor in suppressing multipolar mitoses. In addition to extra centrosomes, it has been observed by Lambert in 1913 that tumors frequently display a high mitotic index (the ratio between the number of cells in mitosis and the total number of cells) [63]. Multiple subsequent studies suggest that this increased mitotic index might be explained by a prolonged metaphase [64, 65]. Although similar results have now been observed in numerous studies, the precise nature of the delay was unclear. One complicating issue when assessing mitotic index for a given tumor is defining the appropriate “normal” cell for comparison. Nevertheless, using time-lapse microscopy, direct comparison of SV40 transformed human fibroblasts and their counterpart untransformed cells clearly revealed that the delay in transformed cells occurs in metaphase [66], suggesting that transformation-dependent events might account for the metaphase delay. In SV40 transgene-induced pancreatic cancer model, SV40 transformation generated tetraploid cells with extra centrosomes. Thus, the mitotic delay in tumors could possibly be explained by tetraploidy and/or extra centrosomes that are frequently associated with cancer [26]. Indeed, work from Rieder and colleagues, directly demonstrated that either increased chromosome or centrosome numbers can increase the duration of mitosis in human cells [67].

Initially it was thought that the SAC was not activated by multipolar mitoses because no delay in anaphase onset was observed in spontaneously occurring binucleated PtK1 cells that undergo multipolar mitoses [68]. However, recently several groups revisited this question and observed a SAC-dependent mitotic delay in cells with extra centrosomes [37, 52, 67]. Using time-lapse microscopy, two independent studies showed extra centrosomes cause a Mad2-dependent delay in anaphase onset in mammalian and Drosophila cell lines [52, 67]. Another study reported similar results in vivo in flies: flies with extra centrosomes, generated by overexpressing SAK/PLK4, are viable because of efficient centrosome clustering [37]. However, combining SAK/PLK4 overexpression with a Mad2 mutation results in lethality. This experiment is feasible in flies because the SAC is not essential [69]. The metaphase delay, not specifically the SAC, is required to suppress multipolar mitoses; an artificial mitotic delay imposed by a proteasome inhibitor can bypass the Mad2 requirement for suppressing multipolar mitosis [52]. These results explain previous observations that multipolar anaphases were very rare despite the high frequency of multipolar metaphases in other cell types [60, 70]. The simplest interpretation of these results is that the SAC does not recognize abnormal spindles per se (i.e., it does not really monitor spindle assembly) but rather detects improper kinetochore attachment/tension during a multipolar mitosis.
  1. 2.

    Molecular motors and spindle associated proteins


Centrosome clustering in tumor cells relies to a significant degree on MT-based motors and microtubule-associated proteins (MAPs) that organize the spindle poles. Numerous studies have identified proteins required to form spindle poles. Interestingly, the importance of these factors vary from cell type to cell type; loss of the minus-end directed motor dynein in Drosophila S2 cells inhibits the attachment of centrosomes to the spindles, but does not strikingly increase the frequency of multipolar mitoses [71]. In contrast, dynein is absolutely required for the organization of supernumerary centrosomes in many human cell lines [72]. In cancer cells containing extra centrosomes, elevated levels of NuMA (mitotic apparatus protein), a main structural component of the spindle poles, and accompanying loss of dynein from the spindle lead to centrosome de-clustering. Interestingly, this defect was rescued by the titration of NuMA levels, which restore dynein localization to the spindle, supporting the role of the NuMA/dynein complex in clustering supernumerary centrosomes [72, 73].

Another minus-end directed motor Ncd/HSET, a Kinesin-14 member, also has a critical role in suppressing multipolar mitoses. Ncd is known to function in efficient focusing of poles in acentrosomal female meiotic spindle and of multiple γ-tubulin foci in Drosophila S2 cells [71, 74, 75]. More recently, the critical role of Ncd in centrosome clustering was confirmed by visualization of centrioles using centriolar markers, such as SAS-6 and D-PLP [37, 52]. Most significantly, the requirement of HSET, a human Ncd homologue, for spindle bipolarity was shown to be unique to cells with extra centrosomes. siRNA of HSET specifically induces multipolar mitoses via de-clustering of centrosomes in cancer cells with extra centrosomes, while it did not induce multipolar mitosis in cells with normal centrosome number [52]. Although Ncd/HSET is not required for spindle assembly in normal somatic cells, the force generated by HSET/Ncd is necessary in meiotic divisions where centrosomes are absent [74, 76]. Thus, it is likely that centrosomes impose a strong bundling force on the spindle poles that, in somatic cells, overrides any requirement for HSET/Ncd. The forces holding centrosomes together in cancer cells may have much in common with the forces that bundle the minus ends of MTs in acentrosomal spindles [13, 52]. Thus, it remains an intriguing possibility that other proteins involved in the assembly of female meiotic spindles [77] also have critical roles for increasing the survival of cancer cells with extra centrosomes.
  1. 3.

    The actin cytoskeleton


A critical role for the actin cytoskeleton in suppressing multipolar mitoses in cells with extra centrosomes was recently described [52, 78]. In principle, the actin cytoskeleton could affect the distribution of centrosomes by at least four mechanisms: (1) by controlling cell shape; (2) by controlling the contractility and stiffness of the cell cortex; (3) by influencing the cell adhesion pattern and the organization and integrity of retraction fibers (RFs); and (4) via a direct role on the centrosomes and the spindle. Although studies by Kwon and colleagues support a key role for the interphase adhesion pattern and thus the distribution of RFs (see next section), we cannot exclude the possibility that other roles mediated by the actin cytoskeleton could be important for the clustering of supernumerary centrosomes.

Actin, along with myosin II, is important for centrosome separation by providing cortical forces that drive the separation of newly duplicated centrosomes [78]. One model for how actin-based contractility could influence centrosome organization, proposed by Rosenblatt et al., envisions astral MTs attached to a continuously tensioned cortical actin cytoskeleton. Astral MTs are known to inhibit type II myosin-dependent cortical contractility. These authors reasoned that if astral MTs attach to the cortical actin cytoskeleton and stimulate a local relaxation of myosin II, the induced asymmetric contractile forces would drive centrosome separation. If cells have two centrosomes, the expected lowest energy state would be to have two centrosomes at opposite poles. With more than two centrosomes, the situation might be more complex; contractility would drive centrosomes together or apart, depending on their starting positions. Under this sort of model, cortical rigidity would also be expected to be a prerequisite for exerting forces on astral MTs. Indeed, the contribution of cortical rigidity to spindle formation has been recently shown in multiple studies and this appears to be mediated in large part by moesin activation [79, 80]. Moesin is a member of the ezrin/radixin/moesin (ERM) family. The ERM proteins bind to actin and stiffen the cortex by cross linking the actin cortex to the plasma membrane [81]. Depletion of moesin, the sole ERM family member in Drosophila, impairs cortical rigidity and is important for spindle formation and positioning [79, 80]. Strikingly, in mammalian tissue culture cells, the ERM family member ezrin localizes heterogeneously at the cell cortex where RFs form (see next section). RFs are actin-rich structures that form during mitotic round-up at the sites of strong interphase cell adhesion. The regions where RF form are important to promote interaction of the astral MTs with the cortex, which drive spindle orientation [82]. Thus, actin-dependent cortical stiffness during mitosis might be important for centrosome clustering by facilitating astral MTs interaction with the cortex.

An alternative view of how cortical forces act on astral MTs is that the distribution of cortical forces that act on astral MTs is heterogeneous rather than being continuously tensioned. Under this model, local forces rather than global contractility would be the dominant factor in positioning centrosomes during mitosis. This hypothesis is supported by the heterogeneous distribution of ezrin in mammalian tissue culture cells [82]. It is also strongly supported by the influence of the adhesion pattern on the distribution of RFs (see next section). Irrespective of the precise nature of the cortical forces acting on centrosomes, the importance of the actin cytoskeleton and myosin II are clear; actin-based contractility is not merely required, but enhanced cortical contractility can suppress multipolar mitoses in certain circumstances [52].

The machinery that drives centrosome separation or spindle positioning in cells could also contribute to the organization of multiple centrosomes in extra centrosomal cells. This would suggest a role for the astral MTs, which link the cortex and centrosomes, in the clustering of extra centrosomes. Interactions between astral MTs and the cell cortex are mediated by MT plus-ends proteins, such as MT-associated proteins (MAPs) and plus- and minus-end directed MT motors [for review see [83]). The MAPs associated with the MT plus-ends, such as CLIP-170, are important for the anchoring of MTs to the cortex; MT motors, such as the minus-end directed motor dynein, are though to be responsible for the pulling forces necessary for spindle positioning [8385]. More recently, the actin-based motor protein Myosin-10 (Myo10) was demonstrated to be important for spindle positioning in mammalian cells [86]. The unconventional Myo10 contains a motor domain that binds to actin and a C-terminal MyTH4/FERM domain, which in turn binds to MTs. This uncommon characteristic makes Myo10 an ideal candidate to link astral microtubules to cortical actin to promote spindle orientation. However, how and where astral MTs contact the cortex is not well understood. Several studies suggest the spindle is positioned according to adhesion cues that are translated into cortical domains where astral MTs will preferentially bind [82, 87]. Interestingly, depolymerization of astral MTs with low doses of nocodazole affects not only spindle positioning but also centrosome clustering [52, 82]. Furthermore, consistent with a role for the spindle positioning machinery in centrosome clustering, other molecules involved in spindle positioning seem to cause a multipolar phenotype, namely CLIP-170 (Drosophila CLIP-190 homolog), Phosphoinositide-3 kinase (PI3K), p150glued (dynactin subunit), dynein and Myo10 [52, 72, 8689]. Thus, the spindle positioning machinery that operates in normal cells is involved in the clustering of extra centrosomes.

The role of actin in mitotic spindle function has been highly controversial, as morphologically normal mitotic spindles can assemble in Xenopus egg extracts in vitro when filamentous actin assembly is prevented [90]. However, we cannot exclude a direct role of actin and actin-based contractility in the clustering of extra centrosomes. For example, the actin-filament-binding protein cortactin can associate with the centrosomes early in mitosis and it seems to be important for centrosome separation in G2 [91]. Recently, Bement and colleagues showed that actin filaments can associate with the mitotic spindle and extend from the spindle to the cortex, in Xenopus leavis. In this system, actin filaments seem to be important at least to anchor the spindle to the cortex, a function shared with the unconventional Myo10 [89]. In Xenopus embryos Myo10 localizes to the spindle poles and is also important for spindle pole integrity leading to multipolar spindles, but this function seems to be independent of its association with actin filaments [89]. However, in cultured mammalian and Drosophila cells that does not seem to be the case. Similar to actin disruption, depletion of Myo10 causes multipolar spindles only in cells with extra centrosomes [52]. Several evidences strongly support that Myo10 functions in actin dependent manner [52]. There is no additive effect in inducing multipolar spindles between Myo10-depletion and actin depolymerization. In addition, in an assay to visualize trajectories of centrosomes, centrosome movements towards the cortex which are dependent on cortical actin are strongly attenuated upon loss of Myo10. Therefore in these cells, we favor a model where Myo10 interacts with cortical actin to promote centrosome clustering.
  1. 4.

    Cell ECM-adhesion

It is well established that cortical cues are important to orient the spindle during asymmetric cell division [reviewed in [92]]. Localization of cell fate determinants and polarity proteins determine the position of the spindle during asymmetric cell division. Similarly, it was proposed that extrinsic cues such as ECM-integrin adhesion are also important to position the spindle in mammalian tissue culture cells [82, 86]. When cells round-up during mitosis, RFs, actin-rich structures linked to the sites of adhesion, remain attached to the substrate [93]. Elegant work by Thery et al. using fibronectin (FN) micropatterns showed that RFs formed at the sites of strong interphase adhesion [82]. In addition, they showed that the positioning of RFs might localize the cortical forces necessary for spindle orientation. This hypothesis is in agreement with previous studies in C. elegans demonstrating that cortical cues activate force generators that pull astral MTs under tension and induce spindle pole movement [94]. These findings can be accounted for by a simple mathematical model where the spatial distribution of these cortical cues induces torque on astral MTs, guiding spindle positioning [95]. Interestingly, the position of the RFs is also important for clustering supernumerary centrosomes. In fact, the patterns of adhesion can promote (H) or disrupt (O or Y) centrosome clustering, depending on the distribution of the cortical cues and therefore, the distribution of the corresponding forces that pull astral MTs (Fig. 3) [52]. Furthermore, disruption of cortical markers such as ezrin by inhibiting Src kinase affects both spindle positioning and centrosome clustering [52, 82]. The accumulation of the ERM protein ezrin at RFs might locally enhance cortical rigidity, enabling pulling forces on astral MTs. Clearly an important future direction will be to elucidate the nature of the forces that pull astral MTs, as well as the molecular mechanisms that link cell adhesion to the mitotic spindle.
Fig. 3

Interphase adhesion pattern can influence spindle formation in cells with extra centrosomes. The centrosomes are pulled towards the sites of strong cell adhesion, resulting in the formation of bipolar or multipolar spindles, depending on the pattern of cell matrix adhesion. When cancer cells with extra centrosomes are platted on H-shaped fibronectin (FN) micropatterns (upper panel), two cortical sites are formed in front of each bar of the H (right) facilitating centrosome clustering (left and middle). In contrast, when cells are plated on Y-shaped FN micropatterns (lower panel), three cortical sites formed (right) and supernumerary centrosomes are pulled away from each other, leading to a multipolar spindle formation (left and middle). The human breast cancer MDA-231 cells were plated on fibronectin patterns (H and Y) and stained for MTs (α-tubulin, green), centrosomes (pericentrin, red) and DNA (blue). The FN patterns are visualized in red. See Kwon et al. for details [52]

Similar to actin, adhesion molecules can associate with the centrosomes and directly influence spindle assembly. For example, molecules involved in focal adhesion regulation during interphase, such as HEF1 and PAK1, were shown to associate with centrosomes during mitosis and to regulate Aurora-A activity [96, 97]. Moreover, the integrin-linked kinase (ILK), which functions in integrin-mediated cell adhesion, has also been shown to localize to the centrosomes and to play a role in spindle assembly [98]. Importantly, ILK signaling effectors such as Akt, GSK3 and β-catenin have also been found at the centrosomes and mitotic spindles, suggesting a direct role for ILK in spindle assembly [99101]. These results suggest an intriguing interplay between adhesion complexes and centrosomes. However, whether there is a genuine functional link or whether the findings represent “moonlighting” jobs for these proteins remains poorly understood. Interestingly, a mutation in the cytoplasmic domain of β1 integrin essential for integrin activation inhibits MT nucleation from the centrosomes and also disrupts cytokinesis, most likely due to spindle defects such as multipolar spindles [102]. Surprisingly, the aforementioned phenotypes can be restored by the addition of an integrin activating antibody that bypasses the requirement of the cytoplasmic domain of β1 integrin for integrin activation [102]. Thus, it seems that, at least in this case, the role of the integrins in cell adhesion and centrosome function might be connected.

4 Interplay between the tumor microenvironment, centrosomes and the stability of the genome

Tumor microenvironments are known to play important roles in the later stages of tumorigenesis, such as invasive behavior and motility of malignant cells. Alterations in geometry/shape, adhesion and stiffness of tumor cells or tissues can lead to changes in the gene expression and signal transduction in cancer cells. One example of how cell shape can control metastatic potential is illustrated in a study by Raz et al. Cells grown in conditions promoting a round morphology (plating cells on a nonadhesive substrate) can lead to an increased metastatic potential when injected to nude mice, by comparison with conditions that promote flattened cell morphology [103]. Moreover, cell shape per se can govern whether cells grow or die. This idea was directly tested by Ingber and colleagues using fibronectin (FN) micropatterns [104]. The authors demonstrated that human capillary endothelial cells can switch from apoptosis to growth when cell spreading increases, suggesting a critical role for cell shape in cell growth control [104]. In addition, tissue stiffness, which can be used as a tumor detection method and is a predictor of tumor risk profiles, can alter signal transduction mechanisms, that promote tumor progression [105107].

Interestingly, tumor microenvironments can also affect early stages of tumorigenesis. The impact of the tumor microenvironment on genomic stability is receiving growing attention. Several studies highlight the fact that cell adhesion has an impact on genetic instability through multiple routes. Loss of cell adhesion can promote genetic instability by attenuating key genomic surveillance mechanisms. This idea is supported by the findings that the loss of adhesion leads to a decrease in the expression levels of p53 in human epithelial cells [108], which could attenuate the cell cycle checkpoint controls thus contributing to genomic instability [109]. p53-dependent apoptosis has also been shown to play a role in limiting the accumulation of cells with supernumerary centrosomes and genomic instability [110]. Thus, the loss of cell adhesion and a concomitant decrease in p53 levels could compromise the fidelity of chromosome segregation and bipolar spindle formation. In addition, loss of adhesion could also induce genomic instability by generating genotoxic metabolites such as reactive oxygen species (ROS) [111], which have previously been implicated in genomic instability and tumorigenesis [112]. Targeted degradation of cell–cell adhesion by overexpression of MMP-3 (matrix metalloproteinase-3) can induce genetic instability and tumors [113]. This MMP-3 effect in promoting genomic instability is mediated by the production of ROS [111].

Another important mechanism by which cell adhesion can affect genome stability is by influencing centrosome clustering [52]. Using a genome-wide RNAi screen, Kwon et al. unexpectedly found that proteins involved in cell adhesion are important for centrosome clustering. To directly test the role of cell adhesion in centrosome clustering, the authors manipulated cell adhesion using FN micropatterns, instead of genetically modulating proteins involved in cell adhesion, which might also have direct roles in mitosis [52]. Using this method, the authors demonstrated that the interphase cell adhesion pattern can influence the organization of supernumerary centrosomes and thus determine whether mitosis is bipolar or multipolar in cancer cells. Given the reorganization of tissue architecture during tumorigenesis, testing how the shape of cancer cells in three-dimensional space affects centrosome clustering, genomic instability and cell survival is an interesting topic for future research.

5 Targeting cancer cells with extra centrosomes: a novel therapeutic approach?

Many useful chemotherapy drugs target the mitotic spindle. For example, microtubule poisons, such as taxanes and vinca alkaloids, are extensively used and are effective in a wide variety of cancers [114]. Despite their efficacy, targeting MTs as therapeutic agents affects not only cancer cells, but also normal dividing cells, leading to toxic side effects for example myelosuppresion (suppression of both white and red blood cells), or gastrointestinal toxicity. Moreover, MT poisons cause toxicity in non-dividing cells, such as neurons, leading to neuropathy (peripheral nerve damage). While myelosuppression can, to some degree, be managed, neuropathy resulting from inhibiting MT dynamics is irreversible and untreatable [115, 116]. This has provided a rationale for the development of pure mitotic inhibitors that do not affect MTs [117].

Kinesin is a mechanochemical enzyme that uses ATP to generate force. Many kinesins are essential for accurate cell division and have recently emerged as a promising “druggable” target class [118]. Kinesin-5 (Eg5) inhibitors are some of the first mitotic inhibitors being evaluated in the clinic. Eg5 is an essential kinesin required for bipolar spindle assembly [119]. A cell based-screen developed by Mitchison and colleagues identified a specific Eg5 inhibitor named monastrol [120]. Extensive study of the structure and enzymatic mechanisms of kinesin motors provided the basis for the development of specific chemical inhibitors [121]. Indeed, new Eg5 inhibitors are currently being tested as potential chemotherapeutic agents and while they seem to have little neutoxicity in patients, it is still unknown if they will be efficient in inhibiting tumor progression [122, 123]. Mitotic kinesins might therefore be suitable anti-cancer targets. For similar reasons, mitotic kinases are also attractive chemotherapeutic targets. Recently, inhibitors against the mitotic kinases polo-like kinases (PLKs) and Auroras have been identified and are currently in clinical trials [117, 124, 125]. Whether these new mitotic inhibitors will improve anti-cancer therapies when compared with MT poisons remains unknown. However, Aurora and PLK inhibitors appear to cause no significant neuropathy and the main dose-limiting toxicity observed was neutropaenia (low number of neutrophils), while other haematopoietic lineages seemed to be spared [117].

Although these new mitotic inhibitors are a promising advance, they still have the drawback of blocking essential functions in normal cells. Therefore, the cancer selectivity or therapeutic window of these agents might be narrow. Ultimately, the ideal therapeutic strategy will rely on the identification of unique requirements for cancer cell survival; effective cancer therapeutics must exploit biological differences between tumor cells and the normal cells from which they arose. One striking difference is that many cancers have extra centrosomes and seem to rely on centrosome clustering mechanisms for their survival [52]. The disruption of these mechanisms could selectively kill tumor cells with extra centrosomes. Indeed, the kinesin HSET, which is non-essential in normal cells [76], becomes essential for the survival of cancer cells with extra centrosomes [52]. Hence, HSET might be an appealing chemotherapeutic target.

Interestingly, the levels of HSET expression vary in tumors ( Perhaps in some of those tumors HSET expression is increased as an adaptive response to promote centrosome clustering. Determining if there is a correlation between the levels of HSET expression and the number of extra centrosomes in tumors would be an interesting future direction, and could help define the type of tumors that could be sensitive to HSET inhibition.

The actin cytoskeleton and its interactions with cell–cell and cell–matrix adhesion might similarly be modulated in tumor cells in response to extra centrosomes. Importantly, the strength and organization of interphase adhesion sites strikingly influences whether cells with multiple centrosomes undergo bipolar or multipolar mitoses.

Therefore, other cellular characteristics could influence the extent of the requirement for HSET [52]. For example, N1E-115 neuroblastoma cells are a classic “small round blue-cell” tumor whose cells express low levels of integrin β-1 [126]. The round cell shape and inefficient substrate attachment may therefore contribute to their dramatic requirement for HSET. Likewise, cancer cells with compromised SAC signaling might also be more sensitive to HSET depletion. Current cancer treatments often combine different chemotherapeutic agents in order to improve efficacy and to minimize toxicity. Identifying targets that induce loss of cancer cells viability synergistically with HSET inhibition will be a valuable future direction. For example, PI-3 kinase inhibitor, an important cancer therapeutic target [127], might be synergistic with HSET inhibition. PI-3 kinase appears to affect centrosome clustering through regulation of the actin cytoskeleton whereas HSET affects spindle MTs bundling. Therefore, inhibition of both pathways could synergistically kill cancer cells.

Despite its potential appeal, HSET inhibitors as chemotherapeutic agents need further validation. What are the long term consequences of HSET inhibition in cells? Even though flies develop normally until adulthood in the absence of Ncd/HSET [74], whether or not HSET is essential for mouse development and/or survival is still unknown. Furthermore, it is important to stress that such an inhibitor would be efficient only in a subset of tumors, the ones with high percentage of cells containing supernumerary centrosomes. Thus, identifying a strategy that enables cost effective and accurate quantification of the percentage of extra centrosomes in tumor samples will be a major challenge to address in the future.

Many exciting questions lie ahead in the field. It will be interesting to investigate the dynamics of the centrosome number in cancer cells. In a given tumor, are there separate stable populations of cells with either normal centrosome numbers or centrosome amplification? Alternatively, do tumor cells with normal centrosome number generate progeny with centrosome amplification and vice versa? One fascinating open question is why centrosome amplification correlates with the malignancy of the tumor. Are tumors addicted to extra centrosomes or are extra centrosomes a byproduct of defective cell divisions? Do extra centrosomes bring any advantages to the tumor? Addressing these questions would allow us to understand the role of extra centrosomes in tumorigenesis and to potentially define selective therapeutic strategies.

6 Conclusions

The presence of extra centrosomes is a common feature of many human cancers. Understanding whether extra centrosomes are beneficial or deleterious to cancer cells is an important topic in cancer cell biology. Extra centrosomes can have a causative role in tumorigenesis, at least in flies, however, it is unlikely that this is achieved through multipolar mitoses. Recent studies showed that multipolar mitoses are detrimental for cells and several mechanisms ensure that cells cluster the extra centrosomes to form a bipolar spindle prior to anaphase onset. The fact that extra centrosome-containing cancer cells have unique requirement for some proteins primarily known to function in acentrosomal spindle assembly is intriguing. The fact that cell geometry and the adhesion pattern can influence centrosome clustering demonstrates a new mechanism by which the microenvironment can affect mitosis and the stability of the genome. Centrosome clustering can be a unique requirement for the survival of certain cancer cells. Can we use this knowledge to selectively kill cancer cells? In vitro studies suggest that this may be the case. This could provide an avenue towards the individual tailoring of therapies based on morphological features of cells.

Note added in proof: Two new manuscripts report spindle associated actin filaments in mouse oocytes that is assembled in a Formin-2 dependent manner [128, 129].


We apologize to the authors whose work was not discussed or cited owing to space constraints. We are grateful to N. Ganem, S. Buttery, K. Crasta, N. Chandhok and M. Thery for critical reading of the manuscript. Our work is supported by a Fundação para a Ciência e Tecnologia grant to S. Godinho, by a Susan Komen postdoctoral fellowship to M. Kwon, and by NIH grant GM083299 to D. Pellman. D.P. is an Investigator of Howard Hughes Medical Institute.

Copyright information

© Springer Science+Business Media, LLC 2009