Anti-mitotic cancer drugs include classic microtubule-targeting drugs, such as taxanes and vinca alkaloids, and the newer spindle-targeting drugs, such as inhibitors of the motor protein; Kinesin-5 (aka KSP, Eg5, KIF11); and Aurora-A, Aurora-B and Polo-like kinases. Microtubule-targeting drugs are among the first line of chemotherapies for a wide spectrum of cancers, but patient responses vary greatly. We still lack understanding of how these drugs achieve a favorable therapeutic index, and why individual patient responses vary. Spindle-targeting drugs have so far shown disappointing results in the clinic, but it is possible that certain patients could benefit if we understand their mechanism of action better. Pre-clinical data from both cell culture and mouse tumor models showed that the cell death response is the most variable point of the drug action. Hence, in this review we focus on current mechanistic understanding of the cell death response to anti-mitotics. We first draw on extensive results from cell culture studies, and then cross-examine them with the more limited data from animal tumor models and the clinic. We end by discussing how cell type variation in cell death response might be harnessed to improve anti-mitotic chemotherapy by better patient stratification, new drug combinations and identification of novel targets for drug development.
Despite the development of new targeted anti-cancer drugs and immunotherapeutics, classic chemotherapeutics, including DNA damaging drugs, anti-metabolites and anti-mitotic drugs, remain the primary treatment modality for cancer patients. Anti-mitotic drugs currently used in the clinic are mostly taxanes (paclitaxel and its derivatives) and vinca alkaloids (vinblastine, vincristine and their derivatives) (Jordan & Wilson 2004, Perez 2009, Mukhtar et al. 2014). These classic anti-mitotics are called ‘microtubule poisons’, as they bind to microtubules and block their polymerization dynamics. In the absence of microtubule dynamics, mitotic spindle cannot properly form, resulting in prolonged mitotic arrest and subsequent growth arrest and cell death (Stanton et al. 2011). Taxanes and vincas are widely used, mostly in combination therapies with other classes of chemotherapeutics; and taxanes in particular have shown a significant positive effect on solid tumors. However, these drugs suffer significant limitations: they are ineffective for many types of cancer; sensitive cancers tend to acquire resistance and they cause serious toxicity, particularly to dividing cells in the bone marrow and gut, and also to neurons (Rowinsky 1997, Crown et al. 2004, Jordan & Wilson 2004, Canta et al. 2009, Rebucci & Michiels 2013).
In a large effort to improve anti-mitotic chemotherapy, inhibitors of more mitotic spindle-specific proteins, such as motor proteins (e.g., Kinesin-5) and several kinases (e.g., Aurora and Polo-like kinases), were developed (Jackson et al. 2007, Marzo & Naval 2013). We will refer to these as spindle-targeting drugs. Unlike microtubules, which play diverse roles not only in cell division but also in cell dynamics during interphase, the spindle proteins targeted by these drugs only, or mainly, function in mitosis. Hence, targeted inhibition of these proteins is believed to mainly affect, and selectively kill, dividing cells, but not the majority of non-proliferating tissues in our body. At least conceptually, this should render a better efficacy-to-toxicity ratio (‘therapeutic index’) compared to classic microtubule inhibitors. For instance, the significant neurotoxicity of taxanes and vincas, presumably due to their inhibition of microtubule dynamics in non-dividing neurons, should be eliminated with spindle-targeting anti-mitotic drugs. However, clinical trials of the spindle-targeting inhibitors have so far shown disappointing results, fairing no better, and often less effective than the classic microtubule-targeting drugs (Katayama & Sen 2010, Purcell et al. 2010, Komlodi-Pasztor et al. 2011, McInnes & Wyatt 2011, Bavetsias & Linardopoulos 2015).
This failure divided the anti-mitotic field into two camps: some now argue mitosis is not a useful target for cancer chemotherapy, due to slow proliferation of tumor cells compared to normal proliferating compartments; and by extension, that microtubule poisons must promote tumor regression by killing interphase cancer cells (Komlodi-Pasztor et al. 2011). Others argue that mitosis is still a useful target, but the details matter, for example, it may be more effective to promote chromosome mis-segregation without mitotic arrest than to strongly block mitotic progression (Weaver 2014, Zasadil et al. 2014). We see merit in both arguments, but will not rehash them here. Rather, we will focus on how cells that try to divide in the presence of anti-mitotic drugs make decisions, and in particular on differences between cell types in these decisions and how those might affect therapy.
The therapeutic index of a cytotoxic cancer drug depends on variation between cell types in cell death responses. For the drug to succeed, cancer cells of some particular genotype and phenotype must die at a level of drug exposure (combining dose and duration) that does not kill normal cells. In this review, we will explore this kind of variability specifically for cell passing through an abnormal division event in drug. We will start by summarizing recent results on cellular mechanism by which anti-mitotic drugs induce cell death, and then examining the points of variation in mechanism of action between classic anti-microtubule drugs and the spindle-targeting inhibitors. By further comparing the cell culture results with those from mouse tumor models, we would discuss potential therapeutic and toxic mechanisms of anti-mitotic drugs seen in the clinic.
In our discussion of cell type variability, we will focus on decisions made at saturating drug concentrations, and not on variation in the minimal drug concentration needed to trigger mitotic arrest. Both types of variability are likely to be clinically important, but most reviews of drug sensitivity focus only on variation in IC50 values, i.e., on variation in the concentration-dependence of response, and assume similar behavior once drug concentration is saturated, which is clearly not the case, as we would discuss below. A discussion of factors that alter IC50 values, which includes drug efflux pump activity, lies outside the scope of this review.
Overall response dynamics to anti-mitotic drugs in cultured cancer cells
Single-cell live imaging revolutionized the analysis of anti-mitotic drug responses by avoiding the need for synchronization protocols, and revealed considerable cell-to-cell variation even in clonal populations (Gascoigne & Taylor 2008, Orth et al. 2008, Shi et al. 2008, 2011, Zhu et al. 2014). Microtubule-targeting drugs, like paclitaxel, and spindle-targeting inhibitors of Kinesin-5 and Polo-like kinase 1 (Plk1) induced a similar sequence of cellular alterations, as summarized in Fig. 1 (Lénárt et al. 2007, Shi et al. 2008). Despite difference in the primary targets of these anti-mitotic drugs, they all disrupt formation of the mitotic spindle, which subsequently leads to activation of the spindle assembly checkpoint (SAC) that arrests the cells in mitosis for prolonged duration (~10–20 h). Depending on the cell type, cells may die during the prolonged mitotic arrest or slip out of the arrest (i.e., mitotic slippage) into an abnormal G1 state without division. From the 4N G1 state, cells may die, arrest or progress again through cell cycle. Different from the other spindle-targeting inhibitors, Aurora-A kinase inhibitors only induced a transient mitotic arrest (Kaestner et al. 2009); and inhibitors of Aurora-B kinase inhibit the SAC itself so drug-treated cells do not arrest in mitosis, despite spindle defects (Hauf et al. 2003). Because Aurora-B kinase activity is required for cytokinesis, these cells subsequently become polyploid and undergo apoptosis for unclear mechanism (Wilkinson et al. 2007).
Figure 1 illustrates how cells treated with anti-mitotic drugs choose between different final outcomes, which are determined by alternative cell behaviors at two major decision points. The first decision point occurs during mitotic arrest, between cell death during the arrest and slippage out of the arrest to an abnormal G1 state without division. The second decision point occurs after mitotic slippage, between cell death, cell cycle arrest and progression through cell cycle in the abnormal G1 state.
The question of how cells make choices between alternative cell fates has been much studied across biology, most recently using single-cell analysis. Some decisions, such as a gene turning on or off, depend on two states of the same biochemical circuit. For choosing different fates during abnormal cell division, a different model was proposed, in which alternative decisions are due to competing actions of alternative pathways from different, non-interacting molecular circuits. The fastest process to execute wins; and the biology is such that when one pathway executes, the competing alternative is immediately blocked. For example, cells that die cannot exit mitosis, and cells that exit mitosis may turn off some timer that will eventually trigger apoptosis in mitosis. In this type of model, alternative decisions arise from variable rates of the competing processes. The choice between cell death and mitotic exit during prolonged mitotic arrest depends on two simultaneously active pathways that function independently, one promoting cell death via intrinsic apoptosis and the other promoting gradual proteolysis of cyclin B1 that leads to mitotic slippage to G1, overcoming a still active spindle assembly checkpoint (SAC) (Brito & Rieder 2006, Huang et al. 2010). Cell types with a faster rate of cell death induction than cyclin B1 degradation tend to reach the threshold of mitochondrial outer membrane permeabilization (MOMP) before they slip out of mitosis, and thus exhibiting mostly death in mitotic arrest in response to saturating dosage of anti-mitotic drugs. Similarly, cells with a slow rate of cell death activation, or fast cyclin B1 degradation, mostly slip out of mitotic arrest to the 4N G1 state. Large cell-to-cell variation in these rates causes variable decision-making, so individual cells may make different choices even in clonal populations. The second major point of cell fate decision occurs in the abnormal 4N G1 state after mitotic slippage, and it is most likely determined by the status of DNA damage response, which we would discuss below in more detail.
Cell death response to anti-mitotic drugs in cultured cancer cells
While anti-mitotic drugs at sufficiently high concentration can induce mitotic arrest in all proliferating cells, kinetics and the extent of cell death induction during or after the arrest are highly variable across different cancer cell types, as observed by the studies of both syngeneic mouse tumors (Milross et al. 1996) and cultured human cell lines (Shi et al. 2008). Therefore, the cell death pathways shown in Fig. 1 are arguably the most important with respect to understanding tumor response to anti-mitotic drugs, and drug toxicity. Many studies have shown that paclitaxel and other microtubule-targeting drugs promote apoptosis through the mitochondrial or intrinsic pathway (Woods et al. 1995, Wang et al. 1999, Park et al. 2004, Shi et al. 2008). The hallmark of this pathway is mitochondrial outer membrane permeabilization (MOMP), which leads to release of cytochrome C and other pro-apoptotic proteins into the cytoplasm, subsequently activating caspases that execute cell death (Fig. 2). The spindle-targeting Kinesin-5 inhibitor and Plk1 inhibitor also activates this pathway similarly (Tao et al. 2005, Shi et al. 2008, Choi et al. 2015). Cell death induced by anti-mitotics can occur both during mitotic arrest, and after mitotic slippage, depending on the cell type and drug. Although some studies concluded that anti-mitotics promote cell death by non-apoptotic pathways (Broker et al. 2004, 2005), in our hands, cell death occurring at both cell cycle states are mediated by mitochondrial apoptosis. However, the upstream regulation and specific triggers of MOMP differs, as detailed below (Shi et al. 2008, 2011, Zhu et al. 2014).
Apoptotic response in mitotic arrest
Time-lapse microscopy data revealed long and variable delays from cells entering anti-mitotics-induced mitotic arrest to induction of MOMP and subsequent cell death in the arrest, typically 10–20 h. A key pro-apoptotic signal that slowly accumulates during mitotic arrest and eventually triggers MOMP was identified to be the depletion of an anti-apoptotic protein, Mcl-1 (Harley et al. 2010, Tunquist et al. 2010, Shi et al. 2011, Wertz et al. 2011) (Fig. 2). Transcription is silenced during mitosis due to chromosome condensation, and translation is attenuated due to phosphorylation of elongation factors (eEF1 and eEF2) (Gottesfeld & Forbes 1997, Blagosklonny 2007, Sivan & Elroy-Stein 2008). Therefore, proteins with short half-life are significantly depleted during anti-mitotics-induced prolonged mitotic arrest. Uniquely among the Bcl-2 family proteins, Mcl-1 turns over rapidly through ubiquitination by E3 ligases, such as MULE/HUWE1 (Zhong et al. 2005), with a typical half-life of < 1 h (Nijhawan et al. 2003, Adams & Cooper 2007). Rescuing Mcl-1 by knockdown of MULE/HUWE1 significantly attenuates cell death in mitotic arrest, confirming the loss of Mcl-1 is a key pro-apoptotic signal during mitotic arrest (Shi et al. 2011).
However, the loss of Mcl-1 alone cannot explain the variable cell death response during mitotic arrest between different cancer cell lines. Both cell lines that are sensitive and resistant to mitotic death showed similar kinetics and extent of Mcl-1 depletion (Shi et al. 2011). Further experiments showed that another anti-apoptotic protein, Bcl-xL, acts with Mcl-1, as the major negative regulators of apoptosis during mitotic arrest (Shi et al. 2011, Topham et al. 2015). For cell lines that express low basal level of Bcl-xL, such as HeLa, the loss of Mcl-1 alone due to mitotic arrest is sufficient to release adequate amount of Bax/Bak to form pores on the mitochondria membrane and trigger MOMP. In cancer cell lines with high levels of Bcl-xL, e.g., U-2 OS, OVCAR-5 and A549, the loss of Mcl-1 alone is insufficient to release Bax/Bak to a level that can cross the MOMP threshold, due to significant Bax/Bak sequestration by Bcl-xL. These results suggest that inhibiting Bcl-xL is potentially an effective combinatorial strategy with anti-mitotic drugs to enhance and accelerate cell death even for cancer cells that are resistant to anti-mitotics alone. This indeed was confirmed in both cell culture and mouse tumor model that combined anti-mitotics, such as paclitaxel and Kinesin-5 inhibitor, with a Bcl-xL inhibitor, Navitoclax (Shi et al. 2011, Tan et al. 2011, Bennett et al. 2016).
Although the mechanism of apoptosis regulation during anti-mitotics-induced mitotic arrest was partly elucidated by studies as discussed above, it was not solved. Combined inhibition of Mcl-1 and Bcl-xL induced significant cell death only in mitosis, but not in interphase (Shi et al. 2011), indicating additional, mitosis-specific pro-apoptotic signal(s) that contributes to MOMP induction in prolonged mitotic arrest. Given that mitotic arrest is a hyper-phosphorylated state resulting from, for example, prolonged activation of cyclin-dependent kinase-1 (Cdk1) (Castedo et al. 2002, Enserink & Kolodner 2010), we suspect the additional pro-apoptotic signal could be triggered by the activation of kinase signaling pathways. Phospho-proteome analysis of mitotic arrest cells may help to shed light on additional proteins/pathways involved in mitotic death control.
As the pro-apoptotic signals that trigger death during mitotic arrest are mitosis specific, the process of slippage, which moves the cells out of mitotic arrest, protects cells against the initiation of mitotic death. This led to the proposal that inhibiting mitotic slippage is a better strategy to induce cell death than anti-mitotic drugs, as it renders a nearly permanent mitotic arrest that allows the mitosis-specific pro-apoptotic signals to accumulate indefinitely. Compared to anti-mitotic drugs that typically activate mitotic arrest of 10–20 h before cells exit the arrest, inhibiting the mitotic slippage pathway, e.g., by Cdc20 knockdown, induced significantly longer arrest, about 30–40 h (Huang et al. 2009). This long arrest efficiently activates death in even the slippage-prone cancer cell lines that exhibit little mitotic death in response to anti-mitotic drugs, e.g., MCF7 and A549 (Huang et al. 2009). Similar effect of Cdc20 knockdown in attenuating cancer growth was also observed in mouse tumor model (Manchado et al. 2010). The type of cell death activated in this ultra-long arrest is again apoptosis mediated by the mitochondrial, intrinsic apoptotic pathway. Considering the measured half-life of Bcl-xL is about 10 h (Kutuk et al. 2010), mitotic arrest of 30–40 h should have enabled Bcl-xL to deplete to a significantly low level, thus triggering apoptosis in cells with even high level of Bcl-xL. Therefore, we think similar cell death mechanism is likely to operate in mitotic arrest, irrespective of whether it arises from abnormal spindle assembly due to anti-mitotic drug treatment or inhibition of mitotic slippage.
Apoptotic response after mitotic slippage
Anti-mitotic drugs, particularly microtubule inhibitors, also trigger interphase cell death after slippage out of mitotic arrest into an abnormal, 4N G1-like state. For some cancer cell lines, e.g., U-2 OS, apoptosis after mitotic slippage is the major cell death response to anti-mitotic drug treatment and highly effective (~90% death after slippage) (Shi et al. 2008). The key pro-apoptotic signal that activates cell death after slippage was identified to be DNA damage (Lanni & Tyler 1998, Quignon et al. 2007, Lai et al. 2011, Orth et al. 2012, Hayashi & Karlseder 2013, Zhu et al. 2014, Hain et al. 2016) (Fig. 3). Several studies have shown that caspase activities, which were partially activated during prolonged mitotic arrest, engendered DNA fragmentation in the post-slippage cells and thus induce p53-mediated DNA damage response (Quignon et al. 2007, Lai et al. 2011, Orth et al. 2012, Hain et al. 2016). Our own results confirmed the involvement of DNA damage in post-slippage cell death and further illustrated that the degree of DNA damage induced by microtubule-targeting drugs, e.g., paclitaxel, is substantially higher than the spindle-targeting Kinesin-5 inhibitor (Zhu et al. 2014). This variation in DNA damage level well explains the much higher cytotoxicity of paclitaxel than Kinesin-5 inhibitor, in particular to the post-slippage population of cells. The critical role of DNA damage in triggering post-slippage apoptosis was also confirmed by experiments showing that the attenuation of DNA damage or knockdown of p53 significantly reduced cell death after mitotic slippage in response to paclitaxel (Zhu et al. 2014).
Although the involvement of DNA damage in post-slippage apoptosis is clear, ambiguity remains regarding the triggers of this DNA damage. Results from our lab showed that the combined treatment of paclitaxel and a pan-caspase inhibitor, zVAD-FMK, only reduced the level of DNA damage in post-slippage U-2 OS cells by 40%. And MOMP was still triggered in nearly 70% of these post-slippage cells in the absence of functioning caspases. The data suggest induction of DNA damage after mitotic slippage cannot be attributed to caspase activity alone. One distinct feature of the post-slippage cells rendered by paclitaxel treatment is that they consist of multiple small nuclei (Fig. 4). Work by Crasta et al. showed defective and asynchronous replication of small micronuclei results in DNA fragmentation (Crasta et al. 2012). To test whether this mechanism contributes to the DNA damage seen in the paclitaxel-induced multinucleated cells after mitotic slippage, we inhibited DNA replication in the post-slippage cells using thymidine or mimosine. However, we did not observe significant reduction in DNA damage with either treatment. In addition, bromodeoxyuridine (BrdU) staining indicated no chromosome replication in the post-slippage cells, indicating cells arresting in G1. Hence, defective replication of the multiple small nuclei is unlikely a major contributor to DNA damage after mitotic slippage. Further investigation is clearly needed to establish the potential links between multinucleation and DNA damage, which appear to determine apoptotic response to anti-mitotic drugs after mitotic slippage, at least in cell culture.
p53 monitors the duration of mitosis independent of DNA damage
p53 is the most commonly mutated gene in cancer, and is thought to act as a ‘guardian of the genome’ by triggering apoptosis or senescence in cells where the genome is compromised. Recently, an exciting new circuit was discovered, in which p53 monitors the duration of mitosis independent of DNA damage and triggers senescence in cells that take too long to progress through mitosis. In normal RPE cells, any lengthening of mitosis beyond ~90 min was shown to cause G1 arrest in the following cell cycle (Uetake & Sluder 2010). In 2016 three independent groups used CRISPR-based loss-of-function genetic screens to uncover the underlying circuitry (Fong et al. 2016, Lambrus et al. 2016, Meitinger et al. 2016). These screens revealed four key genes required to halt cells in G1 after an abnormally long mitosis, including TP53 (p53), CDKN1A (p21), TP53BP1 (53BP1) and USP28. p21 is a known transcriptional target of p53 that halts cell cycle progression in G1 by inhibiting CDKs. 53BP1 and USP28 somehow measure mitotic duration and activate p53 upon mitotic exit. The mechanism of the timing measurement is not yet known, but it likely involves transient localization of 53BP1 to kinetochores, which is linked to stabilization of p53 by USP28 removing ubiquitin. This pathway is lost in p53-null cancer cells, where it likely contributes to development of aneuploidy. Whether this pathway helps to activate mitosis in p53-positive cancer cells, and more generally its role in response to anti-mitotic drugs, is currently unknown.
Cell death responses in tumors: mechanistic data from animal model and the clinic
Microtubule-targeting anti-mitotic drugs have been extensively studied in mouse tumor models, although most early investigations only acquired mechanistically uninformative data of tumor mass vs. dose/time. The Milas group performed a more informative series of studies, where they treated mice carrying a diverse panel of syngeneic mouse tumors with a single dose of taxane (either paclitaxel or docetaxel), killed animals at multiple time points for quantitative histology and related the histological data to overall tumor responses across a panel of tumors (Milross et al. 1996, Mason et al. 1997, Schimming et al. 1999). The kinetics of drug response that they observed were consistent with the sequence of events summarized in Fig. 1, with the peaks of mitotic arrest typically occurring 3–12 h after drug injection in most tumors, and the peaks of cell death at 6–36 h in drug-sensitive tumors (Milross et al. 1996). Comparing between tumor lines, peak mitotic arrest levels following a single dose of paclitaxel varied to a limited extent between tumors, while peak apoptosis levels varied greatly. Only the apoptosis response correlated with the ‘clinical response’, measured as delay in growth of tumor volume (Fig. 4A and B). These data showed that efficient triggering of apoptosis is required for effective treatment of the tumor, and that different tumors are highly variable in this response. Remarkably, the tendency to die from mitotic arrest as well as the bulk growth delay correlated well with the rate of spontaneous apoptosis in untreated animals across the tumor panel, suggesting that the more treatable tumors were generally more apoptosis prone. Both conclusions are consistent with our cell line study (Shi et al. 2008) (Fig. 4C and D), pointing to variation in apoptosis sensitivity, rather than mitotic arrest, as the determining factor in anti-mitotic drug response. In a broader sense, the conclusions from the Milas mouse tumor studies match the concept advanced by Letai and coworkers that the response of tumors to chemotherapy depends on the extent to which their mitochondria are ‘primed’ to undergo MOMP and initiate apoptosis (Del Gaizo Moore & Letai 2013). Higher priming is expected to promote spontaneous apoptosis as well as drug-induced apoptosis. In support of this model, Letai and coworkers showed that the degree of priming predicts survival across a broad range of human cancers, including multiple myeloma, acute myelogenous and lymphoblastic leukemia, and ovarian cancer (Ni Chonghaile et al. 2011).
Milas and coworkers went on to compare the effects of two taxanes, paclitaxel and docetaxel, using a dose of docetaxel that promoted stronger average tumor responses than the respective dose of paclitaxel (Schimming et al. 1999). At these doses, docetaxel caused mostly death in mitosis, while paclitaxel caused mostly death after slippage (Fig. 5). This analysis is one of the few detailed head-to-head comparisons of these widely used taxanes in a tumor model; their effects, as scored by histopathology, were surprisingly different. Interestingly, death in mitosis released more cellular debris, including condensed chromosomes, and was histologically classified as necrosis rather than apoptosis. This was different from the results from cell culture studies that suggest both death in mitosis and after slippage are caused by mitochondrial apoptosis. Considering Mcl-1 and Bcl-xL were also found to be the major predictors of tumor response to docetaxel in mouse xenograft (Tan et al. 2011), similar to our data from cell culture, we think tumor death in mitosis in vivo is also mediated by apoptosis. We suspect the distinct histological classification of the two types of cell death by Milas and coworkers does not correspond to different molecular mechanisms of death. Rather, we interpret these data as suggesting that the downstream aspects of the mitochondrial apoptotic program (for example, the tendency of apoptotic cells to elicit their own phagocytosis) differ depending on whether the death program is executed in mitosis or interphase (i.e., after mitotic slippage). Milas and coworkers observed massive recruitment of mononuclear leukocytes in response to docetaxel, but not paclitaxel, in some of their treated tumors, which they interpreted as a response to release of cell debris following tumor cell lysis in mitosis. They went on to argue that this leukocyte recruitment amplified the therapeutic effect of the drug, because the leukocytes killed non-dividing tumor cells in the treated tumor (Mason et al. 2001). Moreover, the extent of leukocyte recruitment also correlates with the extent of tumor shrinkage following taxane treatment in human tumors (Demaria et al. 2001), though it is not clear how much this is a cause vs a consequence of cell death. The idea that when a dividing tumor cell dies, it recruits leukocytes to help kill its non-dividing neighbors is an example of bystander effect. Bystander effects have long been recognized in ionizing radiation responses (Little 2006). If they occur with anti-mitotic drugs, they might amplify drug responses in tumors with a low mitotic index, for example, due to slow proliferation. Paclitaxel-treated cells have been shown to damage untreated neighbors in culture by emitting reactive oxygen species (ROS) (Alexandre et al. 2007), a second bystander mechanism that might be important in tumors. Intra-vital imaging of tumor response in mouse xenograft had shown paclitaxel treatment did not induce mitotic arrest in most tumor cells, nonetheless the tumor still significantly regressed (Orth et al. 2011). This suggests killing by bystander effect could be an important mechanism by which taxanes exert cytotoxicity in vivo. Further investigation of bystander effects induced by anti-mitotic drugs is clearly warranted.
Then, what do we know about the sequence of events that follow tumor cell mitosis in the presence of an anti-mitotic drug in the clinic? Sampling human tumors or tissues at multiple time points after drug treatment is difficult. Symmans and coworkers were able to take fine needle biopsies at 24-h intervals from a panel of breast cancer patients treated with paclitaxel prior to surgery (Symmans et al. 2000). Their data were consistent with the response kinetics shown in Fig. 1, and also with conclusion from Milas and coworkers that clinical response depends on apoptosis, not just mitotic arrest. They found large variation between patients in mitotic index as well as in apoptotic index. As with the mouse tumor data by Milas and coworkers, apoptotic index correlated with overall response much better than mitotic index, again reflecting the importance of apoptosis sensitivity in overall tumor response. Peak mitotic index values in drug reported by Symmans and coworkers were quite low (< 6%), presumably reflecting slow proliferation in human solid tumors. An unanswered question is what fractions of the cells that went on to die in response to drug did so after experiencing drug-induced mitotic arrest or chromosome mis-segregation, vs dying independent of mitosis via poorly characterized pathways that convert microtubule stabilization to cell death in interphase.
A different approach to obtain histological data from human tumors responding to drug is to explant human tumor samples directly into nude mice (i.e., patient-derived xenograft, PDX) and treat them with anti-mitotic drug in that context. This may preserve proliferation rates and drug responses of primary tumors more reliably than cell culture. Perez-Soler and coworkers used this approach to investigate paclitaxel responses in non-small cell lung cancers (Perez-Soler et al. 2000). They found no correlation between drug response and proliferation rate or the increase in mitotic arrest after drug treatment, again suggesting that the critical factor is efficient triggering of cell death. A striking aspect of this study is the slow proliferation rate of the directly explanted human tumors. Efficient drug-induced shrinkage in tumors with low proliferation rate, and therefore low mitotic index following drug treatment, provide credence to the concept of bystander killing discussed above and/or other unknown mechanism of killing independent of mitotic arrest, which were reviewed previously (Mitchison 2012).
Clinical responses to microtubule-targeting vs spindle-targeting anti-mitotic drugs
Clinical trials of the spindle-targeting drugs provide an important test of whether perturbation of mitosis alone can be an effective therapeutic strategy to treat cancer. Results from phase I, and limited phase II, trials so far showed rather disappointing results. At the neutropenia-determined exposure limit, the spindle-targeting drugs showed little anti-cancer effect in shrinking tumor mass, though the drugs may stabilize tumor growth in some patients (Jackson et al. 2007, Katayama & Sen 2010, Purcell et al. 2010, Komlodi-Pasztor et al. 2011, McInnes & Wyatt 2011, Bavetsias & Linardopoulos 2015). It is of course problematic to evaluate the anti-cancer efficacy of these new drugs on the basis of tumor response data from heavily pre-treated patients. Nonetheless, we probably can conclude from a lack of tumor regression reports that none of the spindle-targeting drugs tested so far could be much more effective than taxanes for treating solid tumors. Toxicity data are more reliable, and an interesting surprise has emerged in that end. At doses that are effective for treating cancer, taxanes and vincas damage all proliferating tissues, including bone marrow, gut and hair follicles, as well as non-dividing neurons. The dose-limiting toxicity is spread between these effects, and varies for particular compounds, with most taxanes and vincas causing some gut toxicity and hair loss. The spindle-targeting drugs tested so far are, in contrast, dose limited by only one toxicity, neutropenia. At this dose limit, patients suffer no hair loss, and little, if any, gut toxicity (Jackson et al. 2007). Remarkably, the spindle-targeting drugs are similar to each other in this respect and thus different from all the microtubule-targeting drugs, despite their different targets and pharmacokinetics. Lack of neurotoxicity of the spindle-targeting drugs was expected, but their apparent selectivity for bone marrow damage, with little evidence of gut damage or hair loss at the dose limit, is a surprise that was not predicted by studies in cell culture or mice.
Neutrophils are among the shortest-lived cells in our body, and their level depends on continuous proliferation of promyelocytes in the bone marrow. Neutropenia is diagnostic of cytotoxic effects that are specific for dividing cells, and this dose-limiting toxicity strongly suggests that all the candidate spindle-targeting drugs are doing exactly what they were designed to do, that is, to kill dividing cells without neurotoxicity. The lack of gut toxicity and hair loss at the neutropenia dose limit implies an altered selectivity profile among dividing cells for the spindle-targeting agents compared to taxanes and vincas, which is currently difficult to explain. This altered selectivity profile seems to point in the wrong direction for treatment of epithelial cancers, but might be an advantage for hematological cancers. Why might mitosis-specific inhibitors show a different selectivity profile compared to microtubule-targeting drugs? One possibility is that effects on non-dividing cells in tumors are important; these occur only with the microtubule-targeting drugs. For example, vincas and taxanes have potent effects on the cytoskeleton of interphase endothelial cells that may contribute to their effects on solid tumors (Nihei et al. 1999, Hotchkiss et al. 2002). Alternatively, the anti-mitotic drug classes may differ in the responses that they elicit from dividing cells. Consistent with the latter possibility, we observed that while a Kinesin-5 inhibitor and paclitaxel are equal in their ability to kill cells in mitotic arrest, taxanes are significantly more able to trigger apoptosis after mitotic slippage to the G1 state, due to induction of DNA damage (Shi et al. 2008, Zhu et al. 2014). Why paclitaxel triggers more DNA damage after slippage than a Kinesin-5 inhibitor is not known. An interesting hypothesis based on recent data is that micronucleation is important. When cells slip out of mitosis in paclitaxel, their chromosomes are spread far apart, and as a result, many micronuclei form. When they slip in a Kinesin-5 inhibitor, the chromosomes tend to be more clumped, and often a single, large nucleus forms (Zhu et al. 2014). DNA in micronuclei tends to undergo extensive fragmentation for reasons that are not fully understood (Crasta et al. 2012). The tendency to drive cells into a highly multinucleated state after slippage might account for part of the extra efficacy of taxanes compared to spindle-targeting drugs.
Implications for improving anti-mitotic therapy
It is not clear whether major improvements in anti-mitotic therapy are possible; if they are, progress is likely to come either from new methods for identifying sub-populations of patients that respond well to current drugs, developing improved combination therapies, or from developing new anti-mitotic drugs with novel targets that give rise to more cancer-selective actions. In order to answer the basic question of whether improvement is possible, and to rationally pursue these avenues toward improvement, we definitely need to further improve our understanding of how anti-mitotic drugs work in the complex environment of tumor in vivo.
For current therapies in the clinic involving taxanes and vincas, a pressing need is to identify patients most likely to respond well or poorly by molecular diagnosis, so as to avoid unnecessary treatment with the highly toxic drugs. If we could predict highly responsive patients, and avoid treating non-responders, these drugs would become ‘precision medicines’, with a much higher therapeutic index averaged across the treated population. The identification of Mcl-1 and Bcl-xL as the major negative regulators of apoptotic response to anti-mitotic drugs, at least in cell culture and mouse tumor models, suggests that basal expression of Mcl-1 and Bcl-xL, in particular Bcl-xL, may be employed as diagnostic predictors of tumor response. However, we already know additional mechanisms are likely involved in mediating the anti-cancer effect of anti-mitotics in tumor, e.g., bystander effects and interphase toxicity, so we still need more mechanistic data to curate a comprehensive set of molecular biomarkers. We advocate grounding the search of predictive biomarkers on phenotypic and mechanistic data, rather than the more popular approaches using only genomic data (profiles of mutation, mRNA and miRNA). This popular approach allows for easy measurement, but suffers from the large amount of uninformative data that is inevitable with broad profiling measurements, which may swamp the useful data in noise. We believe it would be very helpful if we first have more mechanistic data on what makes tumors respond well or poorly, and then use that understanding to filter the genomic data, or even replace it with a smaller number of mechanism-relevant measurements at the protein level, akin to measuring receptor status for hormonal drugs. We argue that finding such predictive measurements will require not only elucidating the pathways that connect spindle damage and alteration of tumor microenvironment to induction of mitochondrial apoptosis and tumor regression, but also understanding how these pathways vary and are differentially activated in distinct cancer types so as to exploit these differential points as diagnostic markers and also potential new targets for drug combination.
One encouraging example of a phenotypic measurement that predicts tumor responses is Letai’s mitochondrial priming assay (Ni Chonghaile et al. 2011). Measurement of priming in untreated cells predicts general chemo-sensitivity, although not the specific drugs that a given patient will best respond to. Recently, this approach was extended to measurement of changes in mitochondrial priming in response to ex-vivo drug treatment (Montero et al. 2015). This can predict specific drug responses, but with the usual caveat that the drug exposure occurs under standard tissue culture conditions, which poorly mimic the complex tumor environment.
Since taxanes are only partially effective on a variety of solid tumors, it requires additional strategies, such as new drug combinations, to increase its efficacy and combat acquired resistance. One potentially viable combination is taxanes plus a Bcl-xL inhibitor, e.g., Navitoclax (previously known as ABT-263), as both cell culture and mouse tumor data illustrated that this combination can significantly enhance and accelerate cell death, even for cancer types that are resistant to taxanes alone. A phase I trial of Navitoclax plus paclitaxel showed both hematological and non-hematological toxicity, including neutropenia, thrombocytopenia, small intestinal obstruction and pulmonary embolism (Vlahovic et al. 2014). It remains to be seen whether this combination indeed can improve patient response. A major challenge to clinical development of Navitoclax is that Bcl-xL inhibition tends to kill platelets, whose lifespan is naturally limited by apoptosis (Zhang et al. 2007). This is a classic case of on-target toxicity in an off-target cell type. Given the huge potential of Bcl-xL inhibition to sensitize tumors to chemotherapy, we wonder if it might be possible to generate drugs that specifically protect platelets as part of a two-drug strategy. Combining taxanes with other reagents that directly enhance apoptosis sensitivity of tumor cells, such as TRAIL, may also synergize with anti-mitotic drugs. In addition, a new modality of anti-cancer therapy is immuno-oncology drugs, such as checkpoint inhibitors (e.g., PD-1/PD-L1 and CTLA-4 neutralizing antibodies), and cancer vaccines (Hodge et al. 2012, Melero et al. 2014, Shin & Ribas 2015, Khalil et al. 2016). Combination of taxane with either checkpoint inhibitor or cancer vaccine has shown notable improvement in progression-free survival in certain solid tumors than taxanes treatment alone (Lynch et al. 2012, Slovin 2012, Reck et al. 2013). Given tumor response to anti-mitotic drugs clearly involves leukocytes and changes in the tumor microenvironments, and cell death induced by anti-mitotics is strongly inflammatory, stimulating anti-tumor immunity is certainly a promising combinatorial strategy to improve anti-mitotic therapy. However, it is still early stage of this line of research; we need more mechanistic and clinical data to determine whether an improved efficacy-to-toxicity ratio can indeed be achieved by such combination; and if so, what patient population would benefit and what would be the optimal dose, schedule and sequence (Antonia et al. 2014).
Compared to better patient stratification and new drug combinations, developing new anti-mitotic drugs of novel targets seem to be less promising in terms of therapeutic development. The similar clinical profiles, so far, of Kinesin-5, Aurora and Plk-1 inhibitors suggest that targeting another highly conserved mitotic spindle protein is unlikely to yield a more selective drug. However, this does not mean that we should ignore the mitotic spindle and cleavage apparatus in the search for better targets. It is possible that certain oncogenic pathways generate specific, druggable vulnerabilities in mitosis, although none has yet been found. Recent work suggested that MELK kinase activity is required for successful mitosis in certain breast cancer lines, but not others (Wang et al. 2014, 2016). Whether MELK is a useful target remains to be seen, but this work is encouraging in the context that it is at least theoretically possible to perturb mitosis selectively in some cancer genotypes. A protein required for spindle assembly in epithelial lineages, but not hematopoietic lineages, might make an excellent target for a novel anti-mitotic that lacks bone marrow toxicity. Although no such protein is yet known, the confluence of RNAi data and informatics might reveal one. The systems that position the spindle may be more complex in epithelial cells compared to hematopoietic cells, arguing such proteins may exist. We also believe that further research into the pathways that connect spindle damage to induction of apoptosis will reveal targets, whose inhibition could more efficiently trigger apoptosis downstream of spindle damage. Given the complexity and variation in apoptosis regulating pathway as well as its secondary effects on immune response, it is reasonable to hope that aspects of this pathway may differ between normal cell types, and change in certain cancers. So it may be possible to develop small molecules that exploit this variation to selectively block apoptosis downstream of spindle damage in bone marrow cells to reduce toxicity, or selectively activate it in particular cancers to increase efficacy.
Declaration of interest
The authors declare that there is no conflict of interest that could be perceived as prejudicing the impartiality of this review.
The authors’ work on anti-mitotic drugs was supported by the Hong Kong Research Grant Council (#261310) to J Shi and the National Cancer Institute (CA139980 & CA078048-09) to T J Mitchison.
This paper forms part of a special section on 50 Years of Tubulin. The Guest Editors for this section were Karen Crasta and Ritu Aneja.
The authors thank members of their research groups, whose work and efforts have helped shed light on their thoughts on anti-mitotic drug response.
ChoiMKimWCheonMGLeeCWKimJE 2015 Polo-like kinase 1 inhibitor BI2536 causes mitotic catastrophe following activation of the spindle assembly checkpoint in non-small cell lung cancer cells. Cancer Letters 357 591–601. (doi:10.1016/j.canlet.2014.12.023)
HaufSColeRWLaTerraSZimmerCSchnappGWalterRHeckelAvan MeelJRiederCLPetersJM 2003 The small molecule Hesperadin reveals a role for Aurora B in correcting kinetochore-microtubule attachment and in maintaining the spindle assembly checkpoint. Journal of Cell Biology 161 281–294. (doi:10.1083/jcb.200208092)
HotchkissKAAshtonAWMahmoodRRussellRGSparanoJASchwartzEL 2002 Inhibition of endothelial cell function in vitro and angiogenesis in vivo by docetaxel (Taxotere): association with impaired repositioning of the microtubule organizing center. Molecular Cancer Therapeutics 1 1191–1200.
LynchTJBondarenkoILuftASerwatowskiPBarlesiFChackoRSebastianMNealJLuHCuillerotJM 2012 Ipilimumab in combination with paclitaxel and carboplatin as first-line treatment in stage IIIB/IV non-small-cell lung cancer: results from a randomized, double-blind, multicenter phase II study. Journal of Clinical Oncology 30 2046–2054. (doi:10.1200/JCO.2011.38.4032)
ManchadoEGuillamotMde CárcerGEgurenMTrickeyMGarcía-HigueraIMorenoSYamanoHCañameroMMalumbresM 2010 Targeting mitotic exit leads to tumor regression in vivo: modulation by Cdk1, Mastl, and the PP2A/B55α,δ phosphatase. Cancer Cell 18 641–654. (doi:10.1016/j.ccr.2010.10.028)
OrthJDTangYShiJLoyCTAmendtCWilmCZenkeFTMitchisonTJ 2008 Quantitative live imaging of cancer and normal cells treated with Kinesin-5 inhibitors indicates significant differences in phenotypic responses and cell fate. Molecular Cancer Therapeutics 7 3480–3489. (doi:10.1158/1535-7163.MCT-08-0684)
ReckMBondarenkoILuftASerwatowskiPBarlesiFChackoRSebastianMLuHCuillerotJMLynchTJ 2013 Ipilimumab in combination with paclitaxel and carboplatin as first-line therapy in extensive-disease-small-cell lung cancer: results from a randomized, double-blind, multicenter phase 2 trial. Annals of Oncology 24 75–83. (doi:10.1093/annonc/mds213)
SymmansWFVolmMDShapiroRLPerkinsABKimAYDemariaSYeeHTMcMullenHOratzRKleinP 2000 Paclitaxel-induced apoptosis and mitotic arrest assessed by serial fine-needle aspiration: implications for early prediction of breast cancer response to neoadjuvant treatment. Clinical Cancer Research 6 4610–4617.
TaoWSouthVJZhangYDavideJPFarrellLKohlNESepp-LorenzinoLLobellRB 2005 Induction of apoptosis by an inhibitor of the mitotic kinesin KSP requires both activation of the spindle assembly checkpoint and mitotic slippage. Cancer Cell 8 49–59. (doi:10.1016/j.ccr.2005.06.003)
TunquistBJWoessnerRDWalkerDH 2010 Mcl-1 stability determines mitotic cell fate of human multiple myeloma tumor cells treated with the kinesin spindle protein inhibitor ARRY-520. Molecular Cancer Therapeutics 9 2046–2056. (doi:10.1158/1535-7163.MCT-10-0033)
VlahovicGKarantzaVWangDCosgroveDRudersdorfNYangJXiongHBusmanTMabryM 2014 A phase I safety and pharmacokinetic study of ABT-263 in combination with carboplatin/paclitaxel in the treatment of patients with solid tumors. Investigational New Drugs 32 976–984. (doi:10.1007/s10637-014-0116-3)
WilkinsonRWOdedraRHeatonSPWedgeSRKeenNJCrafterCFosterJRBradyMCBigleyABrownE 2007 AZD1152, a selective inhibitor of Aurora B kinase, inhibits human tumor xenograft growth by inducing apoptosis. Clinical Cancer Research 13 3682–3688. (doi:10.1158/1078-0432.CCR-06-2979)
ZasadilLMAndersenKAYeumDRocqueGBWilkeLGTevaarwerkAJRainesRTBurkardMEWeaverBA 2014 Cytotoxicity of paclitaxel in breast cancer is due to chromosome missegregation on multipolar spindles. Science Translational Medicine 6 229ra43. (doi:10.1126/scitranslmed.3007965)