Abstract
The accurate timing of organelle biogenesis and the precise regulation of organelle size are crucial for cell physiology. Centriole biogenesis initiates exclusively in S-phase, when a daughter centriole emerges from the side of a pre-existing mother and grows until it reaches its mother’s size. This process is regulated by Polo-like kinase 4 (Plk4), which is recruited to centrioles in oscillatory waves in flies and human cells 1,2. The nature and function of Plk4 oscillations is, however, unknown. Here we discover that Plk4 forms an adaptive oscillator at the base of the growing centriole, whose function is to time and set the duration of centriole biogenesis in Drosophila embryos. We demonstrate that the Plk4 oscillator is free-running of, but is entrained and calibrated by, the core Cdk/Cyclin cell-cycle oscillator, explaining how centrioles can duplicate independently of the cell cycle 3–5. Mathematical modelling and further experiments indicate that the Plk4 oscillator is generated by a time-delayed negative-feedback loop in which Plk4 recruitment to, and dissociation from, the centriole is monitored via changes in the affinity-state of its centriolar receptor, Asterless. We postulate that such organelle-specific autonomous oscillators could regulate the timing and execution of organelle biogenesis more generally.
- Abbreviations and symbols
- A
- Amplitude
- T
- Full width at half maximum
- Ω
- Area under the curve
- CCO
- Cdk-cyclin oscillator
Albert Claude’s landmark paper 6 challenged the idea that cells are a mere bag of enzymes whose contents grow freely in the cytoplasm with no active regulation. We now appreciate the diverse and compact nature of the many organelles in the cytoplasm 7, yet the physical mechanisms that regulate the amount and size of these organelles remain largely unknown.
In order to study the question of organelle size control, we recently established the Drosophila centriole as a model, whose linear structure makes its size easy to define and measure 1. Most cells are born with a single centriole pair that duplicates precisely once during S-phase, when a single daughter centriole grows out from each mother 8. To monitor the dynamics of centriole growth, we used living Drosophila embryos where hundreds of centrioles duplicate in near-synchrony in a common cytoplasm 9. Centriole growth in these embryos is homeostatic: when centrioles grow slowly, they grow for a longer period; when centrioles grow quickly, they grow for a shorter period. As a result, centrioles grow to a consistent size. The centriolar kinase Plk4 is essential for this process and helps to establish the inverse relationship between the rate and period of centriole growth 1. Plk4 can regulate the rate of centriole growth, because it phosphorylates Ana2/STIL to promote centriole assembly 10–12. But, how Plk4 might regulate the period of centriole growth remains unknown.
It has recently been shown that centriolar Plk4 levels oscillate in both fly embryos and human cultured cells, but the significance of these oscillations is unclear 1,2. To investigate the physical nature of these oscillations, we generated flies transgenically expressing Plk4-mNeonGreen (Plk4-NG) under the control of its own promoter in a Plk4 mutant background and monitored Plk4-NG recruitment to centrioles over nuclear cycles 11-13 in living embryos (Figs. 1, A and B, S1 and S2, and Movie S1). Plk4-NG oscillations were maximal in early-mid S-phase and minimal by the start of mitosis, and they showed evidence of adaptive behaviour: the amplitude (A) of the oscillations dampened during successive cycles, while their period—as judged by the full width at half maximum amplitude (T)—lengthened, so that the total amount of Plk4 recruited to centrioles remained approximately the same over each cell cycle (i.e., the area under the oscillation curve, Ω) (Fig. 1C). Moreover, A and T were negatively correlated, explaining how Ω is kept constant through successive nuclear cycles (Fig. S3; Pearson r =-0.5666, P<0.0001).
We next used 3D-Structured Illumination super-resolution Microscopy (3D-SIM) in living embryos to assess the precise location of Plk4 oscillations around the mother centriole. Plk4 is initially recruited to a ring around the mother centriole and then resolves into a single hub that defines the site of daughter centriole assembly 13,14. In our experiments, Plk4-NG was only very briefly detectable in a ring during late-mitosis; at all other stages it appeared as a single-focus (Fig. S4). Thus, in these embryos, centriolar Plk4-NG levels are largely oscillating during S-phase at the single-focus present at the base of the growing daughter centriole.
To test whether the Plk4 oscillations were important for centriole biogenesis we generated flies co-expressing Plk4-NG (in a Plk4 mutant background) and Sas-6-mCherry, which is incorporated into the base of the growing daughter centriole and can be used to monitor centriole growth 1. These flies laid embryos that often failed to hatch (Fig. S5A), but we could simultaneously measure Plk4 oscillations (Fig. S6A) and centriole growth (Fig. S6B) in the embryos that developed normally (Fig. 2A; Movie S2). The centrioles in these embryos were often slightly delayed in initiating growth (Figs. 2A and S5, B and C), allowing us to determine the precise points within the Plk4 oscillation at which daughter centrioles started and stopped growing (Fig. 2A and S6). We found that during each cycle the centriolar levels of Plk4 at which centriole growth was initiated (“start”, Fig. 2A) was not significantly different from the centriolar Plk4 levels at which the growth stopped (“stop”, Fig. 2A). These data suggest that there is a “threshold” level of centriolar Plk4 at each cycle that is required to support centriole growth: above this threshold the centrioles can grow, below this threshold they cannot. Although this threshold was similar at successive nuclear cycles (Fig. 2, A and B), its absolute level may depend upon the relative concentrations of key centriole biogenesis regulators (see, for example, Plk4-NG and asl “half-dose” embryos, below).
If the centriolar Plk4 “threshold” concept is correct, then mother centrioles that failed to recruit the threshold level of Plk4 should not be able to grow a daughter. We observed that the centrioles in several of these embryos separated at the start of S-phase, but did not detectably incorporate Sas-6-mCherry, indicating that daughter centrioles did not grow (Fig. S5, B and C). Importantly, centriolar Plk4 levels still oscillated in these embryos, but the peak amplitude of these oscillations was almost always below the average threshold at which centriole growth was initiated in the other embryos (Fig. S5D). Taken together, these results strongly indicate that the Plk4 oscillations initiate, and then determine the duration of, centriole biogenesis.
We next investigated whether Plk4 oscillations were generated by the core Cdk-Cyclin Oscillator (CCO) that is thought to drive cell cycle progression in most eukaryotic cells. This seemed to be the case, as there was a strong positive correlation between the time at which the Plk4 oscillations peak (oscillation centre, C) and the length of S-phase (Fig. S7; Pearson r =0.8668, P<0.0001). Moreover, if we manipulated S-phase length—by decreasing cytoplasmic Cyclin B levels to elongate S-phase 15 (CycB1/2 embryos), or decreasing cytoplasmic Chk1 (grapes) levels to shorten S-phase 16 (grp1/2 embryos) (Fig. 3, A and B; a total of ~23% difference in S-phase length, denoted with N)—the phase of the Plk4 oscillation shifted accordingly (Fig. 3A and B). Thus, the Plk4 oscillator is entrained by the CCO.
There has been some debate, however, about whether the CCO acts as a “ratchet” that triggers the sequential execution of cell cycle events 17,18, or as a “phase-locker” that entrains the phase of several autonomous oscillators to ensure the timely execution of different cellular events 19–21. In many systems centrioles can continue to duplicate when other aspects of cell cycle progression have been blocked, but it is unclear how they do so 3–5. To test whether centriolar Plk4 levels could continue to oscillate and drive centriole biogenesis if the CCO was perturbed, we injected embryos with doublestranded RNAs (dsRNAs) targeting the three embryonic mitotic cyclins: A, B and B3. This arrests embryos in an interphase-like state with intact nuclei that do not duplicate their DNA or divide, but where centrosomes can continue to duplicate 5. We initially injected embryos in nuclear cycle 7-8 and monitored Plk4-NG behaviour ~30min later. The nuclei in these “late” embryos were arrested in an interphase-like state, but the centrioles continued to duplicate (Movie S3); the timing of these duplication events was more erratic and asynchronous than normal, but a clear Plk4-NG oscillation was associated with each duplication event (Figs. 3C and S8).
We next injected embryos at an earlier stage (nuclear cycles 2-4) and monitored them ~90min later. These “early” embryos contained a small number of nuclei but many more centrioles (Movie S4). Some of these centrioles were “fertile”, and duplicated one or more times in an apparently stochastic fashion (i.e. in a manner that was not synchronised with the other duplicating centrioles), while others were “sterile” and did not duplicate at all (Fig. S9A; Movie S4). We measured Plk4-NG levels at individual centrioles in these embryos. This data was noisy, but fertile centrioles associated with more Plk4-NG than sterile centrioles, and they exhibited clear Plk4-NG oscillations, whereas sterile centrioles did not (Fig. S9, B–D). Moreover, an unbiased computational analysis of all 45 centrioles tracked from 3 different embryos revealed that centriolar Plk4-NG levels could accurately predict whether centrioles were fertile or sterile (Fig. S9, E and F), and that the timing of individual centriolar Plk4-NG oscillations was strongly correlated with duplication events (Fig. S9, F and G), which were statistically stochastic (Fig. 9H). Together, these results indicate that the CCO normally entrains an autonomous centriolar Plk4 oscillator to synchronize individual centriole duplication events to the frequency of nuclear divisions in the early fly embryo. In the absence of a robust CCO, the Plk4 oscillator can continue to drive centriole duplication events, but in a stochastic manner.
It is well documented that centrioles can continue to duplicate when systems are arrested in an interphase- or S-phase-like states, but a mitotic-state is generally considered to be refractory for centriole duplication 22. To test whether Plk4 oscillations could continue when embryos were arrested in mitosis, we injected embryos with the microtubule-depolymerising drug colchicine 23 (Movie S5). Plk4-NG oscillations were detectable in these arrested-embryos (Fig. 3D and S10), but their amplitude was dramatically reduced; these oscillations were all sub-threshold for centriole biogenesis (Fig. S11), and we observed no centriole duplication events. Thus, the Plk4 oscillator appears to be calibrated by the CCO so that it can normally execute centriole biogenesis only at the right stage of the cell cycle.
Finally, we injected the protein synthesis inhibitor cycloheximide 24 (Movie S6). This treatment blocked Plk4 oscillations, and we observed no further centriole duplication (Fig. 3E and S12), indicating that the machinery that generates the Plk4 oscillations in these embryos requires active protein synthesis. Taken together, our observations are consistent with a model where the CCO functions to phase-lock autonomous oscillators that are responsible for different events in the cell cycle 19,20. We propose that Plk4 forms one such autonomous oscillator that governs centriole biogenesis, and that Plk4 oscillations are induced by a self-sustained enzymatic event on the centriole.
How are Plk4 oscillations generated? Our observation that the Plk4 oscillator is adaptive (i.e., the A and T of the oscillations are inversely correlated to ensure a constant Ω; Fig. 1C and S3), suggests that centrioles are somehow measuring (i.e. integrating) the total amount of Plk4 bound to them in each nuclear cycle, which remains constant. Drosophila Asterless (Asl) recruits Plk4 to centrioles and appears to activate it, allowing Plk4 to phosphorylate itself and Asl at multiple sites 25–27. Moreover, human Asl (Cep152) also binds, and is phosphorylated by, Plk4 in vitro 28,29. These findings, along with our observation that the Drosophila Asl is specifically phosphorylated at centrioles in vivo (Fig. S13), led us to propose the following model: At the start of each oscillation cycle, Asl binds Plk4 with high affinity to recruit it to centrioles and activate it (Fig. 4A, step 1; k1 in Fig. 4B). The active Plk4 then phosphorylates itself 30–32 and Asl 27 at multiple sites, providing a mechanism for each Asl receptor to monitor how much Plk4 activity it has experienced over time (Fig. 4A, step 2; k2 in Fig. 4B). The active Plk4 also phosphorylates Ana2/STIL to promote centriole growth 10–12—explaining why a threshold level of centriolar Plk4 is required to promote centriole growth—but this reaction is not important for the generation of Plk4 oscillations per se, and so we do not consider it further here. Once Asl is phosphorylated at multiple sites, we hypothesise that it switches to a state that binds Plk4 with low-affinity. The Asl-bound phosphorylated Plk4 is then released and/or degraded (Fig. 4A, step 3; k3 in Fig. 4B), leaving behind the unbound, but phosphorylated, Asl-receptor that can no longer recruit Plk4 and so can no longer promote centriole growth.
This model is a regulatory network whereby an inhibitor (Plk4) inhibits its own activator (Asl)33. This network maps onto a set of coupled linear ordinary differential equations (see Mathematical Modelling section in Materials and Methods) (Fig. 4B), which we solved analytically. This model fits the discrete Plk4 oscillation data from each S-phase of nuclear cycles 11-13 well (Fig.4C; R2=0.998, 0.999 and 0.998, respectively; Table S1 and Figs. S14 and S15), so we tested two predictions suggested by this model.
First, the model predicts that the main difference between the Plk4 oscillations at nuclear cycles 11-13 is the initial rate at which Plk4 binds to Asl (k1, Fig. 4B and Table S1), which depends on the concentration of cytoplasmic Plk4. This cytoplasmic concentration of Plk4 is predicted to decrease at successive cycles (k1, Table S1). We therefore tested whether simply decreasing the cytoplasmic concentration of Plk4 would make the A and T of cycle 12 more like that of cycle 13, while ensuring a constant Ω. We halved the genetic dose of Plk4-NG in embryos (hereafter Plk4-NG1/2 embryos) and found that this was indeed the case, and that our mathematical model fitted the experimental Plk4-NG1/2 embryo data well (R2=0.996) (Figs. 4D and S16, A and B; see Table S2 and Supplementary Text). Thus, when we perturb the input to the system (cytoplasmic Plk4 levels) and measure the output (total centriole-bound Plk4 over S-phase), the system adapts to changes in the input to produce the same output. Importantly, we previously showed that centrioles in Plk41/2 embryos grow at a slower rate, but for a longer period, and so grow to their normal size 1. This confirms that the ultimate output of the Plk4 oscillator, i.e., centriole biogenesis, also adapts to changes in the input.
Second, the model predicts that the total amount of the Asl-receptor at the centrioles is the same at each cycle (Atot, Table S1). We confirmed that this was the case (Fig. S17), and used Fluorescence Correlation Spectroscopy (FCS) to show that the cytoplasmic concentration of Asl-GFP—that determines the amount of Asl on the centrioles 34—also remained constant at the start of each nuclear cycle (Fig. S18). Interestingly, this suggests that Asl may function as an integrator (i.e., a monitoring factor) whose levels are kept constant to allow the oscillation machinery to sense any changes in the input (e.g. Plk4 concentration) and adapt accordingly 35,36. To test this possibility, we monitored Plk4-NG oscillations in embryos laid by mothers in which we genetically halved the dose of asl (hereafter asl1/2 embryos) (Fig. 4E). This change broke the adaptive nature of the Plk4 oscillator: The A was reduced, but T hardly changed, and so Ω was reduced (Fig. 4E). Again, the mathematical model fitted the experimental data well (R2=0.999) (Figs. 4E and S16, C – E; see Table S3 and Supplementary Text). If centriole size is kept constant by adapting A and T to maintain a constant Ω, then the centriole cartwheel should be too small in asl1/2 embryos (as Ω is now too small), and we confirmed that this was the case (Fig. S19). Thus, the Plk4 oscillator, and so centriole biogenesis, can adapt to lower Plk4 levels, but not to lower Asl levels—supporting the idea that Asl functions as an integrator in this oscillatory network.
Here we show that Plk4 forms a free-running sub-cellular oscillator that times and executes centriole biogenesis in flies. Importantly, our data mining of a recently published diurnal proteome from mouse liver 37 indicates that Plk4 may also form a free-running oscillator in non-dividing mammalian cells that, like the basic cell cycle oscillator 38,39, can be entrained to the circadian clock (Fig. S20). We speculate that the Plk4 oscillator may exemplify a more general mechanism for regulating organelle biogenesis, whereby autonomous oscillations in the levels/activity of key regulatory factors essential for organelle biogenesis could precisely define the timing and duration of the growth process. Such oscillations are likely to be phase-locked to the oscillators of both the cell cycle and the circadian clock to precisely coordinate organelle biogenesis with other cellular events and processes.
Author contributions
This study was conceptualised by M.G.A., M.A.B. and J.W.R. Investigation was done by M.G.A., T.L.S., M.M., L.G., A.W., S.S., P.T.C. and M.B.A. Data were analysed by M.G.A., T.L.S., L.G., F.Y.Z. and M.B.A. Methodology was developed by M.G.A., T.L.S., M.M., F.Y.Z., M.A.B. and J.W.R. Project was administered by M.G.A., M.A.B and J.W.R. Resources were shared/made by M.G.A., M.M., L.G., A.W., S.S. and M.A.B. Software development was carried out by M.G.A., T.L.S., F.Y.Z., and M.A.B. Overall supervision was done by M.G.A. and J.W.R. Validation experiments/analyses were carried out by M.G.A., A.W., S.S. and J.W.R. Finally, M.G.A., M.A.B. and J.W.R. wrote, reviewed and edited the draft.
Competing interests
Authors declare no competing interests for this study.
Data and materials availability
All the equations used for mathematical modelling and regressions are available in Supplementary Materials and in the following web link: <https://github.com/MBoemo/centriole_oscillator_model.git>. The rest of the data is available upon request.
Acknowledgements
We are grateful to Laura Hankins, Fabio Echegaray, Drs. Marjorie Fournier, Christoffer Lagerholm, Pedro Carvalho and Bela Novak for advice and discussion, the Carvalho Lab (Oxford, UK) for sharing cycloheximide, and Dr. Alain Goriely, Alissa Kleinnijenhuis and the members of our laboratory for critically reading the manuscript. Super-resolution microscopy was performed at the Micron Oxford Advanced Bioimaging Unit, funded by a Strategic Award from the Wellcome Trust (107457). The research was funded by a Wellcome Trust Senior Investigator Award (104575; T.L. Steinacker, M. Mofatteh, A. Wainman, S. Saurya, P.T. Conduit, J.W. Raff), an Edward Penley Abraham Scholarship (to M.G. Aydogan and L. Gartenmann), Ludwig Institute for Cancer Research funding (to F.Y. Zhou), and a Biotechnology and Biological Sciences Research Council grant BB/N016858/1 (to M.A. Boemo).