Models, Maps, and the Mitotic Checkpoint

02.11.2016, Lukas Hutter

For **cell division to succeed**, cells generally have to grow, replicate their DNA, and finally divide their DNA and cellular constituents between the nascent daughter cells. **The final stage** in this process is called **mitosis**, when cells divide their nuclei and genetic material. Higher eukaryotes undergo **open mitosis**, characterised by a **breakdown of the nuclear envelope**, **exposing the condesed chromatin to the mitotic spindle**. **Microtubules emanating** from the poles of the mitotic spindle invade the centre of the cell and **bind to kinetochores** - large protein compelxes that assemble in the contromeric region of each chromatid. **Once correspodning sisters attach to opposite poles** of the mitotic spindle, they are said to have become **bi-oriented**. This is the hallmark of metaphase, and marks a turning point: When all pairs of sister chromatids have become bi-oriented, they are ready to become segregated towards opposing poles of the mitotic spindle in anaphase. I would like to **stress the crucial problem** in this context: **The identity of chromatids and their "belongingness" is topologically encoded.** The **ring-shaped cohesin complex** holds sister chromatids together **from the the point of DNA replication to the onset of anaphase**, when they are safely attached to opposite poles of the mitotic spindle and **cohesin is cleaved** by the specific protease called separase. **It is absolutely pivotal** that they become correctly attached before the cohesin is cleaved, as **premature cleavage** of cohesin would lead to **loss of the information of where they belong**. - And correct division is likely to fail. **How do cells achieve this impressive feat?** They have evolved a **control mechanism** called the **mitotic checkpoint** that **combines** the activity of a **mechanism correcting faulty attachments** with a **mechanism that delays anaphase**.

The **error correction mechanism makes sure that kinetochores are only connected to one of the two poles, and ensures that corresponding sisters are connected to opposite poles.** To ultimately obtain these so-called amphitelic attachments **Error Correction must integrate the attachment state of the kinetochore on each of the two sisters**. This is achieved through a **tension-sensitive mechanism**: **Aurora B kinase**, which is part of the **chromosomal passenger complex**, phosphorylates the binding interface between kinetochores and microtubules. It thus **abolishes the electrostatic interaction** between kinetochore and microtubule and destabilises the attachment. **Only correct, amphitelic attachment configuration will experience equal drag** from the two poles of the mitotic spindle. This results in a sufficiently strong **deformation of the kinetochore**, that the phosphorylation targets are **displaced out of reach of the sharp aurora b activity gradient**. This provides a mechanism that biases the attachment process towards the formation of correct attachment.

**To be effective** the checkpoint requires a **mechanism to signal incomplete biorientation** to the molecular machinery that drives anaphase, and delays anaphase onset. This is achieved via the checkpoint effector mechanism. **Anaphase is initiated** by the **activation of** the E3 ubiquitin ligase anaphase promoting complex/cyclosome **(APC/C) by its co-activator Cdc20**. This confers it with **susbtrate specificity** towards it's early anaphase substrates **cyclin B and securin** and allows it to **target them for proteasomal degradation**. **Securin** acts as chaperone and stoichiometric inhibitor of separase, which promotes the segragation of sister chromatids as specific protease to cohesin. **Cyclin B** is the crucial activator of the mitotic cyclin-dependent kinase, CDK1, whose activity high in mitosis and crucial in maintaining the mitotic state.

**To delay the singular event of APC/C:Cdc20 activation** that drives the cell into anaphase in order to give the error correction time to promote amphitelic attachments, cells produce a **soluble inhibitor** called mitotic checkpoint complex (MCC). If **unattached kinetochores** are present, the checkpoint is active. They **serve as catalytic platforms** in the assembly of MCC, which - once assembled - **binds very tightly to APC/C**, and acts as a stoichiometric inhibitor. In **prometaphase** the **rate of production of MCC** by unattached kinetochores is sufficiently high that a single unattached kinetochore is sufficient to delay the onset of anaphase. Once all chromosomes have become attached, the checkpoint is **satisfied, the assembly of MCC from its components stops**, and a set of **processes directing the disassembly of MCC remove the APC/C-inhibitor** in metaphase, and lead to timely activation of APC/C. Noteably, the checkpoint retains the **capacity to become re-engaged** when it is satisfied. Only after the onset of anaphase does it become irreversibly inactivated. **The literature on this topic is rich in detail and vast in scope**, my account of the molecular detail was merely superficial. Yet our understanding of **how these parts come together to elicit control** is far complete. This is where my work as a modeller comes in: The combination of data-driven modelling and model-driven experiments is uniquely positioned address problems at systems scale...

**This diagram depicts** a simple network that describes how components of the checkpoint interact: The **top** panel shows the **error correction** module, where aurora B destabilises attachments in a tension dependent manner. The **bottom** half depicts the **checkpoint effector** module, where unattached kinetochores produce MCC that inhibits APC and prevents it from degrading cyclin B. **Note that cohesin plays an important role** in error correction, as it is required for the formation of amphitelic attachments, and the coupling of tension between two sisters. As such it is **well-posed as an entry point for perturbing** the regulatory system and studying the interactions of its components over time. **What can we learn about the checkpoint, if we challenge cells with precociously cleaved cohesin?** **To address this question**, my collaborators -Raquel Oliveira and Mihailo Mirkovic at IGC in Lisbon- could draw on a **neat experimental system** in **Drosophila neuroblasts**, that allowed the replacement of endogenous cohesin, with a mutant cohesin that was cleavable by a tev-proteas, which could be expressed from a temperature sensitive promotor prior to mitosis.

### They first looked at timing of mitosis: **Control cells** take about 10-15 minutes to complete mitosis, whereas cells treated with the microtubule poison **colchicine**, that depolymerises microtubules take more than five hours. **If premature cohesin cleavage activated the checkpoint fully**, one would expect to oserve a similar duration, **however strikingly**, they observed that in this case cells exit mitosis within 40-60 min. Interestingly, when additionally treated with colchicine, a similar duration of mitosis to pure colchicine treatment was oserved. **To complement these data**, my collaborators performed an extensive **charcterisation of the attachment state of kinetochores** in these cells: When **sampled randomly in mitosis**, the **majority ofkinetochores in control cells display end-on** attached kinetochores, whereas in **colchicine, all kinetochores are unattached**. Interestingly, in cells subjected to **premature cohesin cleavage, still roughly 70% of all kinetochores display end-on** attachments. **These data suggest that error correction is not working properly**, as the **frequency of end-on attachments is significantly higher than expected**. This **finding was further corroborated by kinetochore tracking experiments** that showed that the error correction was **losing its capacity to destabilise erroneous attachments** in cells subjected to premature cohesin cleavage.

**For the short duration of mitosis to be caused by gradual stabilisation of attachments due to the error correction losing grit, we would expect to observe an effect at the level of checkpoint effector signalling: gradual activation of APC/C and declining rates of MCC-assembly** Thus my collaborators next performed **live-cell imaging** of the APC/C-substrate cyclin B. Cyclin B levels are relatively **stable in colchicine** treatment, and **rapidly drop in switch-like fashion** in cells undergoing unperturbed mitosis. **Upon premature cohesin cleavage however**, they become degraded with a **characteristic biphasic pattern**, consistent with a partial activity of APC/C due to incomplete inhibition by the checkpoint. These findings were **further supported by analysis of the localisation of the MCC-component Mad2 to kinetochores, which is a necessary condition for checkpoint signalling**. Upon premature cohesin cleavage, the localisation of Mad2 to kinetochores **decreased gradually**, indicating a decreasing rate of MCC-assembly, and ultimately **disappear completely**.

**Can this gradual collapse of mitotic checkpoint signalling upon premature cohesin cleavage be explained by a purely linear model, where checkpoint effector signalling is downstream of error correction and Cdk1 activity is downstream of MCC-production?** To test this hypothesis, I converted the influence diagram shown on the left into a mathematical model that allowed me to simulate the activities of components of the regulatory system over time in a manner that accounts for the inherent stochasticity of the system. The **left column** depicts simulations for a **control** cell, and is shown to **demonstrate the model produces consistent results with the experimental data**. The simulations shown in the **right column** correspond to cells subjected to **premature cohesin cleavage**: Here the **simulations highlight two problems with the linear model**: It predicts that error **correction stays active and continues to destabilise attachments**. Moreover, the model predicts a **degradation pattern for cyclin B that appears exponential and converges towards a non-zero steady-state** of cyclin B Thus is **very unlike the biphasic degradation pattern observed in the live-cell imaging** of cyclin B! **Studies in higher eukaryotes have shown that the activity of cyclin B:Cdk1 is required to sustain the delay** elicited by the checkpoint effector module. - **Can we capture the experimental data by accounting for this interaction?**

The **feedback mechanism within the checkpoint effector module was the focus of a second line of work I conducted with my collaborators Amalie Dick and Daniel Gerlich at Imba, Vienna.** In our study we focussed on **live-cell imaging of human RPE1 cells expressing endogenously tagged cyclin B-Venus and their response to inhibition of CDK1**. To **isolate the checkpoint effector module from error correction**, cells were treated with **nocodazole**, which depolymerises microtubules and arrests them in prometaphase. To **probe for the role of a CDK1-mediated feedback mechanism in the checkpoint effector** module, my collaborators then added **different doses of the CDK1-inhibitor RO-3306 and quantified levels of cyclin B-Venus over time**. As in the previous study, cyclin B levels remain relatively **stable over many hours when treated with microtubule poison** alone. However, upon addition of **10uM RO-3306 or higher**, cyclin B became **rapidly degraded in sigmoidal fashion** within 25 minutes. The addition of a **lower dose**, such as 3uM RO-3306, elicited a **striking response**: It **neither** led to an **abrupt switch** like degradation, **nor to a biphasic degradation pattern**, but cyclin B levels remained **stable for about an hour until rapid degradation of cyclin B set in**.

To **understand the unexpected response to CDK1 inhibition better**, I **adapted the model presented in the previous part** of the talk, however I **simplified it structurally** - disregarding the **error correction** module, which is **irrelevant in nocodazole treated cells**. I initially **conditioned the model to capture the degradation of cyclin B during normal mitotic progression, and in a nocodazole arrest**, and **then simulated treatment with a CDK1-inhibitor** by **first** simulating treatment with **nocodazole alone** for some time, **then** reducing the parameter representing the **pool of total available CDK1** in the system. Using this model, I was able to **qualitatively reproduce the cyclin B-timecourses observed upon addition of 10 and 3 uM RO-3306**. To get a more **global perspective** at the response of the system to different extents of CDK1-inhibition, I **created a bifurcation diagram** which shows how the **cyclin B steady states change as a function of CDK-inhibition**. Here, the **solid lines** depict stable steady states, that form **valleys attraction**, whereas the **dashed line** depicts unstable steady states, **akin to a mountain ridge**. The model exhibits **bistability for CDK1 inhibitions of less than 50%**, where the upper steady state corresponds to a checkpoint effector-**"ON" state**, and the lower steady state correspond to a checkpoint effector-**"OFF" state**. As the **CDK1-inhibition is increased beyond 50%**, the domain, where ON and OFF state co-exist **disappears and only the "OFF"-state is stable**. In this framework, **both doses of CDK-inhibitor are best explained by inhibitions of CDK1 beyond 50%** residual activity. In both cases, the checkpoint **"ON" steady state is not stable**, however for the **lower dose**, CDK1 activity is still high enough that it can** delay, yet not block activation of APC/C via MCC**.

**If you have head about the phenomenon of mitotic slippage this may sound very puzzling to you.** What you see **here is the timecourse of cyclin B in a nocodazole-treated cell, which was recorded over around 30 hours**. **Even if no Cdk1-inhibition was applied**, the cell eventually **exits mitosis**. - **So how can we consider the ON state stable**? The answer is that **the bfurcation diagram I have shown you before** was merely a **snapshot**, where I **froze the system at a particular point** in time. To **recapitulate mitotic slippage**, the model incorporates a **mechanism exterior to checkpoint effector mechanism**, which **very slowly reduces** the cyclin B steady state. The **mountain range** that separates two valleys corresponding to checkpoint effector on and off is really **more like a glacier: it is melting away over time!**

**Where does this analysis leave us in terms of our understanding of the mitotic checkpoint?** The **transient bistability **as a function of maximal CDK-activity suggested by this analysis **translates to bistability **of the system in terms of **another factor that influences MCC-production** unattached kinetochores. This is an **incredibly powerful framework** to **reason about system-level properties** of the mitotic checkpoint: it allows us to **define states in terms of basins of attraction (valleys)**, and explains system-level properties of the checkpoint such as its reported **sensitivity** towards single unattached kinetochores, and its **irreversible inactivation in anaphase**. Whereas we have **not conclusively shown** that a bistable switch underpins mitotic checkpoint control, **we observed fingerprints of bistability**, and are **now able to formulate a wide range of predictions that are yet to be tested experimentally.

Acknowledgements

Bela Novak
Stefan Heldt
Scott Rata
Michael Hopkins
Vinod PK
Tongli Zhang
Raquel Oliveira
Mihailo Mirkovic
Daniel Gerlich
Amalie Dick

Francis Barr
Michael Cundell
Ricardo Nunes Bastos

M Cundell et al. 2016 JCB

EPSRC Systems Biology DTC, Boehinger Ingelheim Fonds, iv Kärnten