Stable centrosomal roots disentangle to allow interphase centriole independence

The centrosome is a non–membrane-bound cellular compartment consisting of 2 centrioles surrounded by a protein coat termed the pericentriolar material (PCM). Centrioles generally remain physically associated together (a phenomenon called centrosome cohesion), yet how this occurs in the absence of a bounding lipid membrane is unclear. One model posits that pericentriolar fibres formed from rootletin protein directly link centrioles, yet little is known about the structure, biophysical properties, or assembly kinetics of such fibres. Here, I combine live-cell imaging of endogenously tagged rootletin with cell fusion and find previously unrecognised plasticity in centrosome cohesion. Rootletin forms large, diffusionally stable bifurcating fibres, which amass slowly on mature centrioles over many hours from anaphase. Nascent centrioles (procentrioles), in contrast, do not form roots and must be licensed to do so through polo-like kinase 1 (PLK1) activity. Transient separation of roots accompanies centriolar repositioning during the interphase, suggesting that centrioles organize as independent units, each containing discrete roots. Indeed, forced induction of duplicate centriole pairs allows independent reshuffling of individual centrioles between the pairs. Therefore collectively, these findings suggest that progressively nucleated polymers mediate the dynamic association of centrioles as either 1 or 2 interphase centrosomes, with implications for the understanding of how non–membrane-bound organelles self-organise.


Introduction
The centrosome is a major microtubule organising centre, with critical roles in cell migration, cell division and cilia function. Mammalian interphase cells in G1 are generally thought to have one centrosome, consisting of two microtubule based structures, called centrioles. Centriole pairs are proteomically and functionally distinct [1], yet apparently remain physically associated, a phenomenon called centrosome cohesion [2][3][4][5][6][7]. Experimental changes to intercentriolar distance during interphase result in defects in cell migration, ciliary function, and mitosis [8][9][10][11][12], underscoring the functional importance of centrosome cohesion. How two centrioles co-ordinately assemble into a single centrosome yet maintain distinct functions is largely unexplored.
Centrioles are not bounded by a lipid membrane but instead by two distinct structures, termed the pericentriolar material (PCM) and pericentriolar fibres [2,13]. Current models of PCM assembly emphasise high dynamics of constituent proteins, potentially as a liquid-like, toroidal structure [14,15]. In contrast, comparatively little is known about either the structure or assembly of pericentriolar fibres. Rootletin / ciliary rootlet coiled coil protein (gene symbol CROCC) localises to pericentriolar filaments, and rootletin knockout or knockdown results in both loss of filaments and centrosome cohesion [2,8,16,17]. One model posits that rootletin pericentriolar fibres directly connect centriole pairs to keep them spatially restricted [2,5,16,18].
Consistent with this proposal, rootletin is not found on mitotic centrosomes [5, [18][19][20]. The kinetics of pericentriolar fibre dissolution during mitosis, when they reform, and the principles governing their replication are poorly understood, however.
To address these questions, this study uses high resolution imaging, genome editing and cell fusion to obtain unprecedented spatio-temporal information about the morphology, dynamics and assembly properties of rootletin fibres, referred to as roots. Roots are bifurcating adhesive 5 formation influences root structure, interphase centrosomes were classified according to the presence of either two or four GFP-Centrin1 marked centrioles, corresponding to either unreplicated centrioles or diplosomes (mature centriole + nascent procentriole) respectively.
No difference in rootletin intensity or size was detected (Fig 1D), suggesting that procentriole growth does not influence root structure.
Procentrioles mature into centrioles during mitosis, dependent on PLK1 activity, becoming replication competent after physically moving away from a centriole (a process termed disengagement) [24]. Therefore, the effects of PLK1 kinase inhibition on root formation were investigated, using the PLK1 kinase inhibitor BI2536 (Fig 1E). Cells arrested in mitosis through PLK1 blockade contained monopolar spindles [25], which were devoid of roots ( Fig   1F; BI2536), consistent with previous work suggesting that mitotic cells do not have roots [5,16,[18][19][20]. Since the inhibition of PLK1 results in cell cycle arrest, mitotic exit was forced into an ensuing interphase without cell division, by addition of the CDK1 inhibitor RO-3306 [26], to understand subsequent effects on root structure in interphase (Fig 1E). Control cells were also arrested in mitosis, but instead using the Eg5 kinesin motor inhibitor STLC followed by RO-3306. Control cells forced into interphase in this manner reformed roots despite unsuccessful mitotic genome segregation (Fig 1F; right hand panel entitled RO-3306).
Interestingly however, forced mitotic exit after PLK1 blockade resulted in partial root reformation relative to STLC control (Fig 1G). These results suggest that centrioles are capable of root reformation in G1 regardless of PLK1 activity in the previous mitosis. In contrast, procentrioles must be modified by PLK1 dependent processes before they are competent to form roots in the next cell cycle. Furthermore, since PLK1 promotes centrosomal PCM expansion during mitosis [14], mitotic centrosomes disassemble roots even in the absence of centrosome maturation. Taken together, these results demonstrate that roots are large bifurcating fibres, that are found commonly in a range of cell types on mature PLK1-modified 6 centrioles during the interphase.

Diffusionally stable roots are progressively formed from anaphase
The dynamics and biophysical properties of eGFP tagged rootletin were investigated in living cells, by utilising both cDNA transgene overexpression and tagging of endogenous alleles.
Consistent with previous work [2,16], over expression of eGFP-rootletin resulted in fibres and bifurcating fork structures which were longer than endogenous rootletin (e.g. compare Fig S2A   with Fig 1). Time-lapse imaging of eGFP-rootletin fibre formation following transfection showed that eGFP-rootletin first appeared focally in a single location, prior to the emergence of a larger network over many hours (Fig 2A). Fibres increased in size not only by extension in length outwards from a single point, but additionally by the coalescence together of multiple fibres to form larger aggregates, frequently through end-on fusions (compare arrows in Fig 2A; Video S1).
Cell cycle dependent changes in the centrosomal intensity of meGFP tagged rootletin were followed by 3D confocal time-lapse imaging. Since overexpressed rootletin fibres were larger than endogenous antibody stained roots, and because overexpression can influence quantitative measures of protein function in vivo [27], CRISPR Cas9 was used to insert an in-frame fusion of meGFP into the endogenous rootletin (CROCC) locus and therefore study rootletin behaviour with live cell microscopy at endogenous levels for the first time (Fig S3).
Homozygous tagging in the diploid breast cancer cell line Cal51 resulted in fluorescent signal closely resembling antibody staining (Fig S3E). Rootletin-meGFP was barely detectable at the centrosome during mitosis, consistent with immunofluorescent staining (Fig 1), and consequently, stable co-expression of a NEDD1-mRuby3 fluorescent fusion marking the PCM was used to track centrosomes throughout the cell cycle and independently of rootletin levels, in a spectrally distinct fluorescent channel (Fig 2B; Video S2). Additionally, fluorescently 7 labelled chromatin was imaged, to reveal mitotic sub stage.
Rootletin began to be released from the centrosome more than two hours prior to anaphase ( Fig   2C). By anaphase, centrosomal rootletin could not be detected above cytoplasmic levels, suggesting disassembly of all centrosomal roots. Rootletin centrosomal levels increased from anaphase, but unexpectedly, continued to increase at a slow rate for approximately nine hours and thus significantly into G1 phase. Staging of rootletin intensity relative to centrosome separation revealed that its release from the centrosome began prior to centrosome separation and continued after it, with low levels of rootletin still present during centrosome separation which could be ripped apart during poleward centrosome migration (Fig 2D). These results suggest that the removal of rootletin from centrosomes begins early in mitosis or in late G2 phase, prior to both chromatin condensation and centrosome separation, and then continues during these processes. Rootletin assembly at the centrosome begins from anaphase and occurs slowly for approximately nine hours into G1 phase. Some centrosomal PCM components show dynamic exchange of subunits on the seconds timescale -a property which has been suggested to be important for centrosome assembly [13].
Fluorescence recovery after photobleaching (FRAP) was therefore used to ask whether rootletin forms steady state polymers. FRAP of extended eGFP-rootletin fibres showed almost no movement of eGFP-rootletin over a time period of ten minutes however, even after a relatively rapid bleach (Fig S2B; ~1 sec). Lack of recovery was not due to image bleaching or fibre movement out of the field of view, since adjacent unbleached ends of the fibre remained unchanged. To investigate very slow dynamic exchange of endogenous centrosomal rootletin-meGFP, on the hours timescale, cells were arrested at G1/S of the cell cycle using thymidine, to circumvent the effects of cell cycle progression on root morphology (see Fig 2C).
Recovery of rootletin-meGFP fluorescence on this long timescale was limited to ~30%. Therefore, eGFP-rootletin fibres are predominantly diffusionally stable structures which are slowly and progressively assembled over hours following anaphase.

Roots disentangle during transient centriole splitting in interphase
How a single interphase cell co-ordinately organises two disengaged centrioles is unclear. The prevalence of centrosomal cohesion was systematically documented in a range of human tissue culture cell types by automated fluorescence imaging and analysis (Fig 3A). Quantification of the percentage of cells with split centrosomes -defined as two PCM foci >1.5 µm apartshowed it was low at ~10%, dependent on cell type (see Materials and methods for a definition of split centrosome). Thus, in most cell types centrioles remain cohered in close proximity during interphase, consistent with previous work [6,[28][29][30][31][32]. One possibility is that the minority of cells with split centrioles keep them stably separated over time, perhaps due to a permanent failure of centrosome cohesion. Interestingly however, single cell 3D confocal live imaging of centriole pairs marked by GFP-Centrin1 showed transient splitting. Hence, a single centriole pair would split into two and then rejoin, often repeatedly (Fig 3B; Video S3).
Transient centriole splitting was manifest in live cell imaging of several different cell types including Cal51, HeLa, RPE and U2OS cells (Fig 3B-3E; Video S4 and S5).
In agreement with a published report [29], HeLa Kyoto cells had high levels of centrosome separation, with ~50% of cells showing split centrioles in a fixed asynchronous population ( Fig   3A). Since HeLa cells have low levels of rootletin expression and short roots (Fig 1A), and previous work has shown that rootletin knockdown results in the loss of centrosome cohesion [2,33], the effect of increasing root length on centrosome position was investigated in HeLa cells, to ask whether rootletin over expression is sufficient to increase centrosome cohesion.
An increase in rootletin fibre length significantly increased centrosome cohesion in interphase HeLa cells, as measured by automated imaging and analysis of immunofluorescently stained samples (Fig 3F, p<0.001, Fischer's exact test). Together these results show that although centrioles generally remain cohered into a single focal location, they are able to transiently split apart in interphase in a manner that is antagonised by eGFP-rootletin over expression.
How might rootletin fibres respond to transient centriole splitting? Two opposing models for root behaviour after centriole splitting were postulated (Fig 3G). The first is maintenance of a stable root contact between centrioles as they move apart, for example due to stretching. The second is loss of physical connection and disentanglement ("Stable contact" versus "Disentangle" respectively). Surprisingly, and in contrast to cohered centrosomes, split centrioles were not linked by rootletin fibres (Fig 3H). Instead, a transition from linked to unlinked roots occurred at distance greater than ~1.5µm between centrioles (Fig 3I), thus supporting the disentanglement model.
Simultaneous two-colour Airyscan microscopy of root disentanglement in living cells revealed that roots occupy markedly heterogenous orientations which change in response to in vivo centriole movement (Fig 3J; Video S6). The centrosome distal ends of roots have the capacity to pivot relative to centrosome proximal ends, suggesting a common more stable attachment point at the proximal end. Pivoting of centrosome distal tips was not just observed in centrosomes with split centrioles but also in cohered centrosomes, with roots maintained stably at the centriole-centriole interface (Fig 3K; Video S7). As centrosomes remerged after a split, roots did not necessarily join, but could alternatively contact the PCM of the opposing centriole.
Together these observations indicate that although roots can be maintained stably at the interface between centrioles, their orientation is heterogeneous, and notably, in response to centriole movement they disentangle rather than maintaining a continuous direct linkage.

Centriole independence during interphase
Since disengaged centrioles can transiently split (Fig 3), the comparative structure of roots and PCM on centrioles was investigated in detail during splitting. Root area was approximately halved in split versus cohered centrioles (Fig 4A), suggestive of equal partitioning of two independent roots. Due to their mode of replication, mature centrioles are of differing age, with the oldest centriole marked by appendage proteins including CEP164 [34]. Discrimination of centriole age using CEP164 immuno-staining showed that roots are nucleated symmetrically on both centrioles when spatially separated (Fig 4B). A similar comparison of PCM structure with the PCM resident pericentrin showed similarly that centrioles individually nucleate PCM when split (Fig 4C), something also evident in live cell imaging of NEDD1-mRuby3 (Fig 3J-K), and previous work [24,30,32]. These observations show that centrioles independently maintain roots and PCM during centrosome splitting in interphase.
Given the dynamic nature of centrosome cohesion (Fig 3) and root disentanglement, it was of interest to investigate whether mature centriole pairs would be maintained in cells with four mature centrioles (Fig 4D). Interestingly, centriole position in cells forced into interphase after a failed mitosis by STLC treatment (Fig 1E), showed all possible centrosome cohesion configurations. Most commonly, all four centrioles grouped together as one (Fig 4E; ~40%), but notably other spatial arrangements were equally as likely as two pairs. Thus, centrioles are not maintained separately in stable pairs but will cohere together, even in a grouping such as one single centriole and three cohered separately (labelled as "3-1" on Fig 4E). This was examined further using polyethylene glycol mediated cell fusion of two different cell types, one expressing rootletin-meGFP and the other not (see Materials and methods for details).
Fused cells contained two nuclei and four mature centrioles -one mature centriole pair from each different cell line (Fig 4F). As per after mitotic failure (Fig 4E), after cell fusion centriole pairs were not exclusively maintained, despite their origin in different cells (Fig 4G). Instead, 11 fluorescent roots frequently embraced all four centrioles. Therefore, by two independent methods, mature centriole pairs are not stably maintained in cells with four centrioles.

Discussion
Cells must carefully regulate centrosome number and position, coordinating two centrioles which are capable of distinct functions [23,30,35]. The data here provoke an interesting hypothesis; that interphase cells always have two centrosomes which are generally held together by stable fibres which reach outward into the cytoplasm. Three key pieces of evidence support this model. Firstly, both mature interphase centrioles in a pair independently nucleate roots, as well as PCM. Secondly, these units -consisting of a centriole -root -PCM, have the capability to transiently spatially separate during interphase, a phenomenon accompanied by root disentanglement. Thirdly, cells engineered with two centriole pairs do not maintain them separately, but instead dynamically make new groupings. Thus, there is remarkable plasticity in the maintenance of centrosome cohesion, with individual centrioles able to rearrange between pairs, through dynamic splitting of roots. These conclusions are consistent with previous observations of split centrioles [6,[28][29][30][31][32]. Centriole independence could aid plasticity of centrosome function such that the two centrioles can either act as one or separately. Thus, the data presented here explains previous observations that centrioles may have either different or coordinated functions [23,30,35].
How non-membrane bound organelles regulate their position, size and number within the cellular interior is a major open question. Recent work has postulated that organelles such as centrosomes phase separate as liquid-like compartments [36]. This model is characterised by high internal turnover of components parts, spherical shape, and the ability of multiple 12 organelles to fuse [15]. In contrast, roots are diffusionally stable, remain separate through multiple cycles of merging and splitting, and are not spherical, instead potentially engendering polarity to the centrosome as a branched organelle. Hence, roots have surprisingly different organisational principles to the PCM. Further work will be needed to understand whether this has implications for how centrosomal position is regulated; a testable prediction from this work is that the two PCM clouds could merge with liquid like properties as centrosomes cohere.
In conclusion, root mediated splitting of two centrosomes might allow plasticity of cytoskeletal function, thus explaining how two non-membrane bound organelles co-ordinately function in either one or two locations during the interphase [28,32,37]. It is tempting to suggest that progressively nucleated, diffusionally stable polymers might be used generally to maintain organelle subcellular position and number [38].

Antibody validation
Multiple lines of evidence indicate that a commercially available anti-rootletin antibody (Novus Biologicals NBP1-80820) specifically recognises the product of the CROCC gene. siRNA depletion of rootletin (CROCC) using RNA interference removed signal by both immunofluorescence and western blot in multiple cell types (Figs S1A, S1B and S1D). An antibody independent method, GFP tagging, showed significant overlap to protein abundances measured by immunofluorescence, both in time and space (this is apparent throughout Figs 1-3). For example, rootletin signal was virtually undetectable in metaphase by either antibody or GFP tagging. Finally, anti-rootletin antibody also stained GFP-rootletin when overexpressed as a transgene (Fig S1C). 13 Immunofluorescence Cells were fixed in 4% paraformaldehyde or ice cold 100% methanol for ten minutes, CTGGAAGGCCTGTCACTGTC.
Mitotic exit was then forced with RO-3306 (10µM) for six hours. Cells were confirmed as being in interphase during interphase.

Image analysis
Images are presented as maximum intensity projections from 3D data unless otherwise stated.
Image brightness and contrast settings were changed linearly and consistently between samples for display purposes of representative images, but not for quantitation. as the definition of split centrioles since this distance was the threshold above which roots 16 rarely linked centrioles in high resolution imaging (Fig 4D).
For segmentation of roots, various thresholding strategies were used in CellProfiler, including propagation outwards from a GFP-centrin1 seed region, or direct thresholding. Spacing of PCM staining was measured by adaptive thresholding followed by calculation of 2D Euclidean distance between centroids. In this case roots were segmented using propagation from PCM and then manually classified as linked if one pixel overlap occurred between a root from each PCM.

Live cell time-lapse imaging and fluorescence recovery after photobleaching
Cells were imaged without phenol red in either L15 CO2-indepdendent medium, or in Fluorbrite Imaging medium with 5% CO2 at 37 o C, in Ibidi u-slide 8-well dishes. Imaging was with a Carl Zeiss 880 Airyscan, either in Airyscan or standard confocal mode, using either a 63x NA 1.4 or 100x NA 1.4 oil immersion lens. FRAP was performed essentially as described in [14], by bleaching using a 488 argon laser at 100% for the minimum time required to cause ~50% fluorescence loss (but keeping the bleach time the same duration in all samples).
Corrections were made for non-specific image bleaching by using time courses taken with identical settings but in unbleached cells. Hence, lack of recovery was not simply due to nonspecific bleaching. In long FRAP experiments (15 hours total recovery period), significant movement occurred, both of the centrosome and in some cases the whole cell. To account for this, centrosome position was tracked by using a spectrally distinct centrosomal marker, NEDD1-mRuby3.
Prior to fusion, rootletin-meGFP cells were stained with CellTrace Violet dye (Thermo Fisher four centrioles marked by NEDD1-mRuby3 fluorescence, due to turnover of this marker at the centrosome from the cytoplasmic pool. Thus, upon fusion of two cells, NEDD1-mRuby3 was able to bind centrosomes through recruitment from the cytoplasmic pool. Since rootletin-meGFP shows very slow diffusional exchange (Fig 2), this process did not occur to the same extent.

Supporting information
Video S1. Growth of cDNA eGFP-rootletin fibres in a single Cal51 cell after transfection. Each frame is taken at a six-minute interval and shows a maximum intensity z-projection from a 3D confocal stack. confocal Airyscan time-lapse imaging of rootletin-meGFP (green; roots) and NEDD1-mRuby3 (red; PCM). Each frame is taken at a 10-minute interval and shows a maximum intensity zprojection.

19
Video S7. Root behaviour in a stably cohered centrosome, visualised by 3D confocal Airyscan time-lapse imaging of rootletin-meGFP (green; roots) and NEDD1-mRuby3 (red; PCM). Each frame is taken at a 10-minute interval and shows a maximum intensity z-projection.