Primary cilium loss in mammalian cells occurs predominantly by whole-cilium shedding

The primary cilium is a central signaling hub in cell proliferation and differentiation and is built and disassembled every cell cycle in many animal cells. Disassembly is critically important, as misregulation or delay of cilia loss leads to cell cycle defects. The physical means by which cilia are lost are poorly understood but are thought to involve resorption of ciliary components into the cell body. To investigate cilium loss in mammalian cells, we used live-cell imaging to comprehensively characterize individual events. The predominant mode of cilium loss was rapid deciliation, in which the membrane and axoneme of the cilium was shed from the cell. Gradual resorption was also observed, as well as events in which a period of gradual resorption was followed by rapid deciliation. Deciliation resulted in intact shed cilia that could be recovered from culture medium and contained both membrane and axoneme proteins. We modulated levels of katanin and intracellular calcium, two putative regulators of deciliation, and found that excess katanin promotes cilia loss by deciliation, independently of calcium. Together, these results suggest that mammalian ciliary loss involves a tunable decision between deciliation and resorption.


Ciliary structures in cells undergoing serum-induced cilium loss
To assess the physical processes underlying ciliary loss, we observed ciliary morphology in a population of mammalian cells entering the cell cycle. We manipulated serum level in the culture medium to synchronize ciliary behavior in inner medullary collecting duct 3 (IMCD3) cells expressing somatostatin receptor 3::green fluorescent protein (SSTR3::GFP) [24, 54,55], a fluorescent plasma membrane marker that is enriched in the ciliary membrane ( Fig 1A and  S1A Fig). The majority of serum-starved cells (60 +/− 9.09%) were ciliated. Subsequent serum stimulation for 6 hrs resulted in a decrease in the fraction of cells with a cilium (30 +/− 0.4%) to levels comparable to asynchronously cycling cells. Serum-induced cilium loss requires the function of histone deacetylase 6 (HDAC6), a deacetylase of tubulin and cortactin [24,34]. Cells treated with 2 μM tubacin, an HDAC6 inhibitor, failed to undergo serum-induced ciliary loss ( Fig 1B) [24, 56,57]. Mitotic cells accumulated following serum stimulation, but this accumulation was inhibited or delayed in tubacin-treated cells (S1B Fig), consistent with a requirement for ciliary loss prior to mitotic entry [10,11,[58][59][60]. Thus, serum stimulation induces semisynchronous, HDAC6-mediated, cell cycle-linked ciliary loss in these cells.

Rapid deciliation is the predominant mode of ciliary loss
These results suggested that ciliary-loss behaviors within a single-cell population might be variable, so we used live-cell imaging to observe individual cilia-loss events. We generated IMCD3-SSTR3::GFP cells that stably coexpress a centrosome marker (mCherry-PACT) to identify the position of the basal body. Live cells were imaged immediately following serum stimulation, with full confocal stacks acquired continuously for 6-12 hrs at 90-sec intervals (see Materials and methods). Apically-facing cilia were selected for analysis to avoid artifacts caused by deformations of basally-facing cilia by interactions with the culture substrate. Control serum-starved cells subjected to this imaging regime retained their cilia over 6 hrs (S2A  Serum stimulation of IMCD3 cells reveals noncanonical ciliary structures. IMCD3-SSTR3::GFP cells were serum starved for 24 hrs, followed by stimulation with 10% serum in order to synchronize ciliary loss. A, C-E) Cells were fixed at indicated time points after serum addition and immunostained for PCNT to mark the basal body (white) and acTub to mark the axoneme (red). Nuclei and cell boundaries are outlined in yellow. A) Morphology of a normal, intact cilium in a starved cell. B) The population of ciliated cells quantified over a serum-stimulation time course. Async and serum-Stv controls were included at 0 hrs and 6 hrs. Cells treated in parallel with 2 μM tubacin. C-E) Noncanonical ciliary structures identified in serum-stimulated cell populations. C) Discontinuous acTub staining, in this case accompanied by narrowing of the membrane. D) A ciliary stub, marked by punctate acTub and SSTR3 fluorescence. E) Full-length axoneme marked by acTub, lacking corresponding SSTR3::GFP signal. F-G) Stacked plots of cilia morphologies observed during serum stimulation in (F) DMSO and (G) tubacin-treated cells. Quantifications are based on means of three independent experiments with 150-200 cells analyzed per condition per replicate. Error bars = SEM. Source data can be found in supporting data file S1 Data. acTub, acetylated tubulin; Async, asynchronous; IMCD3; PCNT, pericentrin; SSTR3::GFP, somatostatin receptor 3::green fluorescent protein; Stv, starved.
Image sequences of cells undergoing cilia loss revealed a striking range of dynamic behaviors (Fig 2A, S1 Movie, S2 Movie, S3 Movie and S4 Movie). We grouped these behaviors into three categories: Gradual, cilium length reduction over at least two consecutive time points resulting in terminal cilium loss (e.g. , Fig 2A.2; 0.02 μm/min; T start 6 ¼ T final-1 ; see S2 Fig); Instant, a single discrete cilium-loss event within a single-imaging frame, i.e., 30-90 sec (e.g. ,  Fig 2A.3; �4.72 μm/min L final-1 > 1.5 μm); and Combined, a period of Gradual disassembly directly followed by Instant loss (e.g. , Fig 2A.4; Gradual phase [0.03 μm/min], preceding rapid loss within 46 seconds, �5.69 μm/min). To reduce bias in our cumulative analysis, we developed an algorithm to normalize ciliary length fluctuations to controls, identify an event start point, and assign each event to one of the three categories described above (Fig 2B and S2 Fig,  see Materials and methods for full description of algorithm strategy). Length curves were normalized by time (to 1,000 arbitrary units) and ciliary length (to the maximum length of each cilium) (Fig 2C-2E). The Gradual averaged curve shows early ciliary shortening with event start points distributed along the curve, followed by a period of consistent shortening in the last approximately 150 normalized time units as the slope of the curve increases ( Fig 2C). The Instant averaged curve appears nearly flat until the last point; start points were nearly all clustered in the last approximately 10 normalized time units (Fig 2D). The Combined averaged curve features a period of slight slope characteristic of Gradual dynamics followed by an Instant cilia-loss event, with start points distributed along the curve (Fig 2E).
Cilia-loss events occurred throughout the entire 12-hr imaging window; hourly frequency of event start points and end points are shown in S2D-S2G Fig. These results demonstrate asynchronicity and further heterogeneity of cilia-loss behaviors. However, end points (completion of cilia loss) were most frequent between 1-3 hrs post serum addition, consistent with the enrichment of noncanonical ciliary structures observed in fixed cells (Fig 1F), further supporting that these structures are intermediates of cilia-loss events. The nature of individual loss events did not change consistently over time, suggesting that factors other than time after serum stimulation influence cilia-loss mechanism.
Ciliary-loss rates spanned several orders of magnitude (10 −3 −10 1 μm/min), with the slowest Gradual events occurring over several hours, and the fastest Instant events (10 1 μm/min) occurring in less than 90 seconds (Fig 2F). Within the Combined category, the rate of the first Gradual step (0.08 μm/min ± 0.26) and second Instant step (3.41 μm/min ± 2.10) were not significantly different from those of the Gradual-only (0.08 μm/min ± 0.11) and Instant-only (3.88 μm/min ± 1.80) rates, respectively ( Fig 2F). Further, the majority of ciliary length (72.6 ± 4.5%) was lost during the Instant stage. Thus, Combined ciliary loss likely represents both Gradual and Instant mechanisms occurring within one loss event, rather than an independent third behavior with biphasic dynamics.
Strikingly, we found that the Instant and Combined groups together comprised 84% (n = 69) of the observed events ( Fig 2G). Thus, events with Instant dynamics, which occur within seconds, are the predominant behavior for terminal ciliary loss in our experimental conditions. Rates of Gradual loss were heterogeneous but approximately consistent with previously reported rates of resorption [36, 43,44], while Instant dynamics were significantly more rapid than would be expected for resorption. To test whether Instant loss dynamics might be consistent with what has previously been described as deciliation, we examined the dynamics of deciliation induced by dibucaine, which likely acts by raising intracellular calcium [47,49,50,61]. Dibucaine-induced deciliation had qualitative features and a dynamic profile similar to serum-induced Instant loss (S3 Fig and Fig 2A.2). Therefore, we propose that deciliation is the physical mechanism underlying Instant cilia loss.
In a rare instance, we observed direct shedding of the entire visible cilium from the surface of a serum-stimulated cell (Fig 3A and S5 Movie). The shed cilium in the z-stack shown in Fig  3A has a fragmented appearance. Due to the rapid timescale of the shedding event, the shed cilium can travel a considerable distance (>1 μm) in the time between individual z-slices (roughly 2.6 seconds) within the stack. As a result, the shift in apparent location of the same object in subsequent slices causes the final imaged object to appear distorted-artificially elongated or with the appearance of separated fragments. We note that the segments in such sequences were visualized as a comigrating cluster, rather than dispersing independently, consistent with the interpretation that they are parts of one structure imaged at slightly different times. Therefore, we interpret the fragmented appearance of the cilium in Fig 3A as the result of motion of the imaged object during acquisition, rather than a biological change in morphology, and suggest that these images represent a deciliation event.
The rapid nature of the deciliation event shown in Fig 3A (<54 seconds) and the diffusion of the shed ciliary remnant(s) away from the site of origin is consistent with the Instant loss dynamics found in 84% of serum-induced disassembly events ( Fig 2G) and in dibucainetreated cells (S3 Fig). In addition, cilia shed under our culture conditions from cells expressing mCherry-α-tubulin contained tubulin in shed ciliary fragments (imaged at 30-second intervals, Fig 3B) suggesting that the axoneme is shed together with the ciliary membrane. These results further support the hypothesis that Instant cilium loss represents ciliary disassembly via deciliation.

Recovery of whole cilia from culture media demonstrates that deciliation occurs during ciliary disassembly
If deciliation is a major mode of cilium disassembly, we would expect to be able to recover whole cilia from culture medium. Such recovered material should have lengths consistent with whole cilia (�1.5 μm, see S2 Fig) and contain membrane-and axoneme-specific components. We developed two methods to enrich for ciliary fragments spontaneously released at low concentration from serum-stimulated cells (Fig 3C-3G and S4 Fig). First, we used immune-capture of fluorescently-labeled cilia to directly visualize unperturbed ciliary fragments. Culture medium from serum-stimulated IMCD3::SSTR3-GFP cells transiently expressing mCherry-αtubulin was incubated on imaging dishes coated with an antibody against the extracellular domain of SSTR3 (S4A Fig). Samples were imaged, without fixation, by fluorescence; we could reproducibly identify cilia marked by both SSTR3-GFP and mCherry-α-tubulin captured in this manner (Fig 3C). We interpret the jagged appearance of the cilia as due to thermal motion of the sample during stack acquisition of material that is not firmly fixed to the substrate. Compared to control serum-starved medium, serum-stimulated medium yielded a 3.5-fold increase in captured cilia. In addition, pretreatment with tubacin decreased the number of captured cilia to control levels ( Fig 3D). Finally, following this procedure in unlabeled IMCD3 cells, we observed objects with dimensions consistent with intact shed cilia after fixation and imaging by scanning electron microscopy, demonstrating that deciliation is not a result of SSTR3::GFP expression ( Fig 3E).
The cilia immune-capture method was limited by low concentration and further sample loss due to the instability of antibody-bound cilia, which hindered subsequent compositional analysis by immunofluorescence or biochemistry. Therefore, we used a complementary method to increase yield, in which culture medium was subjected to a series of filtration and centrifugation steps to concentrate ciliary material approximately 500-fold (Fig 3F-3G and  S4B Fig). As a positive control, serum-starved cells were artificially deciliated with high-calcium buffer [62] (see Materials and methods). Immunoblotting against intraflagellar transport 88 (IFT88), α-tubulin, and acetylated tubulin further confirmed enrichment of ciliary proteins in the concentrated medium (Fig 3F), and lack of DAPI-stained material indicated that isolated samples were free of large cellular debris (S4C Fig). Immunostaining for axoneme markers (IFT88, α-tubulin, and acetylated tubulin) demonstrated increased abundance of ciliary structures, many with the dimensions (1.5-7 μm) expected of whole cilia (Fig 3G).
In Chlamydomonas, flagellar severing can occur at two distinct sites of flagellar autotomy (SOFA): proximal and distal to the transition zone [63]. To determine the site of deciliation with respect to the transition zone in IMCD3 cells, we immunostained isolated ciliary material with an antibody against the transition zone marker RPGRIP1L. Distinct RPGRIP1L puncta asymmetrically colocalized with one end of elongated acetylated tubulin signal, indicating the presence of transition zone components in many isolated cilia ( Fig 3H). Thus, deciliation can occur from the proximal transition zone, consistent with the frequent observation of complete removal of the ciliary membrane in Instant and Combined cilia loss (Fig 2A.3, 2A.4 and 2G). The fact that RPGRIP1L is not present in all acetylated tubulin-positive structures further suggests that deciliation might also occur at intermediate locations along the cilium, as observed in Fig 3B. Together, these results indicate that whole and partial cilia, including membrane and axoneme components, are shed during ciliary loss. In summary, via two independent ciliary-isolation methods, we show that deciliation occurs during serum-induced cilium loss and is likely to be the phenomenon underlying Instant loss dynamics.

p60 katanin overexpression promotes Instant ciliary loss
We next sought to gain insight into the regulation of deciliation. A central requirement for complete deciliation is the disruption of axonemal microtubules near the ciliary base [44,51,63,64]. Although this process has been observed in several organisms [47,49,50,65], the underlying machinery and mechanism of deciliation have been difficult to elucidate. The microtubule-severing enzyme katanin has been proposed as a candidate for this function in Chlamydomonas and Tetrahymena [46,63,66,67]. Katanin has been demonstrated to localize to both proximal and distal regions of the transition zone in Chlamydomonas flagella [67], but pf19 mutants, which are defective in the p60 catalytic subunit of katanin, do not have a deflagellation defect [68]. In mammalian cells, katanin has a general regulatory relationship with primary cilia [51,69]. Katanin is required for microtubule rearrangement in mitotic spindle formation [69][70][71][72], but its role in primary cilia loss is poorly understood.
In addition, in approximately 30% of ciliated cells, p60 localized near the base of the primary cilium ( Fig 4A). Cytoplasmic acetylated tubulin intensity was reduced in serum-starved and serum-stimulated tRFP-p60 cells compared to the control by immunofluorescence and western blot (S5C-S5F Fig), consistent with increased severing and destabilization of microtubules caused by activity of katanin overexpression [65,73,76].
Next, we determined whether tRFP-p60 katanin expression influenced overall assembly and loss of cilia. Total levels of ciliation in response to serum starvation ( Fig 4B) and stimulation (S5C Fig) were unaffected; however, ciliary length was significantly reduced in tRFP-p60 cells ( Fig 4C). Furthermore, inhibition of ciliary loss by tubacin and cytochalasin D was unaffected in tRFP-p60 cells (S5G Fig). Therefore, tRFP-p60 katanin expression influences ciliary structure but is not sufficient to induce ciliary disassembly in serum-starved cells, and any effect of p60 overexpression likely occurs downstream of HDAC6 activity.
We next asked whether tRFP-p60 katanin expression affects ciliary loss dynamics in serumstimulated cells. In cells expressing only tRFP, terminal cilium loss by Instant dynamics comprised 86.7% of disassembly events (n = 61, 33.3% Instant and 53.3% Combined), consistent with our analysis in parental cell lines in Fig 2G. However, in tRFP-p60-overexpressing cells, Gradual dynamics were virtually eliminated, while the frequency of Instant loss increased, resulting in 98% of events featuring Instant terminal cilium loss (n = 50, 50% Instant and 48% Combined) ( Fig 4D). Thus, tRFP-p60 expression shifts the distribution of disassembly behaviors toward Instant ciliary loss dynamics. These results suggest that increased katanin activity is capable of modulating ciliary disassembly behavior by promoting deciliation with Instant dynamics and that the distribution of heterogeneous disassembly behaviors is tunable by mechanistic regulators.
Conversely, we asked what effects depletion of [Ca2+] i would have on ciliary disassembly in tRFP-and tRFP-p60-expressing cells. BAPTA-AM, a cell-permeable Ca 2+ chelator [86], did not affect ciliary length in starved cells but inhibited ciliary loss in serum-stimulated tRFP cells (Fig 4E and 4F), consistent with published work [23]. However, cilium loss was not impaired in serum-stimulated tRFP-p60 cells (Fig 4E), and ciliary length in starved cells was reduced compared to DMSO-treated tRFP-p60 cells (Fig 4F). Therefore, tRFP-p60 katanin expression can overcome the requirement for [Ca 2+ ] i in ciliary loss. We summarize these results in a schematic in Fig 4G. Overall, we found that overexpression of tRFP-p60 eliminated or reversed the effects of both increasing [Ca 2+ ] i with thapsigargin (S6C and S6D Fig) and reducing [Ca 2+ ] i with BAPTA-AM (Fig 4E and 4F). Taken together, these results indicate that [Ca 2+ ] i and katanin do not act cooperatively to promote ciliary loss. Rather, the negative regulation of cilia by [Ca 2+ ] i appears to be mitigated or reversed in the presence of excess p60 katanin.

Discussion
We characterized serum-induced ciliary-loss dynamics and behaviors of primary cilia, a process critical to tissue homeostasis and development in vertebrates. Here, we present a model for ciliary loss based on these results and discuss implications of that model. BAPTA-AM data are compared to DMSO control for each cell line for statistical analysis by Mann-Whitney U test. Source data can be found in supporting data file S4 Data. G) Summary of results from Fig 4E-4F and S6 Fig. Orange bar: tRFP-p60 cells have no defect in overall ciliary disassembly but undergo Instant deciliation more frequently. We make two assumptions to approximate the relative differences in p60 and [Ca 2+ ] i levels between our experimental manipulations: 1) tRFP-p60 cells have "excess" p60 compared to "normal" levels in tRFP control cells, and 2) thapsigargin may produce a [Ca 2+ ] i that is physiological but higher than the baseline state (labeled "excess"), whereas dibucaine and ionomycin induce "high excess" levels of [Ca 2+ ] i . [Ca2+]i, intracellular calcium concentration; IMCD3, inner medullary collecting duct 3 cell line; SSTR3::GFP, somatostatin receptor 3::green fluorescent protein; tRFP, turbo red fluorescent protein. https://doi.org/10.1371/journal.pbio.3000381.g004 Primary cilium loss in mammalian cells occurs predominantly by whole-cilium shedding Our results suggest that there are at least two mechanisms of ciliary loss in mammalian cells: resorption, in which the axoneme is depolymerized and ciliary contents are incorporated into the cell, and deciliation, in which the axoneme is excised near its base. We have incorporated these in a model (Fig 5) in which ciliary loss requires at least two major decision points: decision 1, when to remove the cilium and decision 2, when to invoke Instant deciliation. Decision 1 is controlled by Aurora A, HDAC6, and other elements [24,34,54,87,88]. Decision 2, in IMCD3 cells, results in deciliation immediately (Instant), after a delay (Combined), or never (Gradual). We assume that resorption of the cilium, as the initial behavior in Combined dynamics, is the default mode of cilium loss and that rapid deciliation, when invoked, overrides the slower-acting resorption. Thus, cilium loss is a tunable decision in at least two ways: when to remove the cilium and the mechanism by which that removal is accomplished, both of which are likely to differ in different cell types and different contexts.
In most cases of Instant cilium loss, the entire visible cilium was lost within seconds; however, in some cases, detachment of the distal portion of the cilium could be observed (Fig 3A). In addition, we observed the presence of the transition zone protein RPGRIP1L in some cilia isolated from cell culture medium. This suggests that ciliary detachment can occur at multiple sites along the cilium. In Chlamydomonas, there are two known SOFAs [63], directly above and below the transition zone, and katanin has been shown to localize to both sites via immunogold electron microscopy [67]. Flagellar severing at one site or the other seems to be associated with functional context-distal severing is associated with stress response, and proximal severing occurs during mitotic flagellar disassembly [63]. Whether these deciliation sites translate to a functional difference in the mammalian primary cilium is not understood. For instance, it could be that distal severing, which leaves the transition zone intact, may allow for rapid regeneration of the cilium, while proximal severing may serve as a "terminal deciliation" in a cell cycle context in which regrowth is suppressed; indeed, this could be a third decision that a cell undertakes during cilia loss. Primary cilium loss in mammalian cells occurs predominantly by whole-cilium shedding To better understand how deciliation is regulated in in IMCD3 cells, we manipulated p60 katanin and intracellular calcium levels and found that the activity of overexpressed p60 katanin biases cilium-loss events nearly exclusively toward deciliation, likely representing an intervention at Decision 2. Whether and how katanin might sever axoneme microtubules in deciliation, or whether it is indeed required for deciliation [68], remain critical outstanding questions. Katanin may also indirectly influence cilia, perhaps by severing centrosome-associated microtubules and interfering with ciliary protein trafficking (consistent with the observed diffuse staining pattern at the cilium base, Fig 4A) or by modulating the available pool of cytoplasmic tubulin [89,90]. These hypotheses could help explain why tRFP-p60 expression negatively affects cilia length but not formation in untreated cells.
Calcium is necessary for ciliary loss [23] and is sufficient to drive deciliation [47,50,79,91], thus acting at both Decisions 1 and 2. Although the role of calcium is likely multifaceted, it seems that calcium and increased p60 katanin activity may function independently and also negatively interact at the Decision 2 nexus (Fig 4E-4F, S6C and S6D Fig). Despite the requirement for Ca 2+ for disassembly (Decision 1) and deciliation (Decision 2) in normal conditions, tRFP-p60 cells were able to undergo ciliary loss after [Ca 2+ ] i chelation, likely via deciliation. This suggests the existence of a Ca 2+ -independent deciliation pathway, as has been suggested previously [50]. Alternatively, overexpressed p60 may promote cilium loss via resorption in the absence of [Ca 2+ ] i . This relationship may be consistent with the finding that calcium binding inactivates p60 severing activity in vitro [92]. We speculate that the activities of katanin and [Ca 2+ ] i in modulating ciliary behavior may depend on their relative levels in the cell ( Fig  4G).
What is the significance of the multiple mechanisms of ciliary loss coexisting in the same cells? In Chlamydomonas, the anterograde IFT kinesin mutant fla-10, which normally undergoes passive flagellar shortening due to IFT imbalance [43], switches to deflagellation in the presence of Ca 2+ at the restrictive temperature. Conversely, deflagellation-incompetent mutants (fa1, fa2, adf) undergo flagellar resorption in response to acid shock, which normally induces deflagellation [93]. This work is directly in line with our two-decision model, suggesting that once an internal or external cue to disassemble the cilium/flagellum is detected (Decision 1), the cell will "find a way" to remove its cilium-if the preferred or default mode of disassembly (Decision 2) is unavailable or unfavorable, an alternative mechanism can be used. The recent development of an engineered inhibitable KIF3A/3B recapitulated the fla-10 phenotype of in NIH3T3 cells-inhibition of motor activity induced cilium loss. Intriguingly, the consistent reduction in cilia number over 8 hrs was accompanied by only a minor reduction in cilium length, suggesting that cilia were lost by rapid deciliation [94]. Future studies could take advantage of this tool to examine the dynamic behaviors of cilia during this process, as well whether modulation of p60 katanin, intracellular calcium levels, and other potential regulators influence cilia-loss dynamics induced by a cell cycle-independent stimulus.
Ciliary-loss behaviors varied both in the dynamics (10 −3 to 10 1 μm/min) and physical process (resorption versus deciliation). The coexistence of resorption and deciliation in the same cilium (Combined dynamics) is intriguing and may suggest independent or differential regulation of the distal and proximal portions of the cilium [36]. It might be, for example, that structural features of the axoneme, such as doublet-singlet microtubule interface or posttranslational modifications, contribute to differential regulation of ciliary regions. We were unable to directly and specifically observe the axoneme in our live-cell imaging experiments. However, in fixed serum-stimulated cells, we identified several noncanonical structures that hint at the fate of the axoneme during ciliary loss. These included discontinuous axoneme staining that might reflect partial breaks away from the base, short axoneme stubs that could represent a remnant of a severed or resorbed cilium [95,96], and even axonemes without corresponding ciliary membrane staining, possibly representing a portion of axoneme retracted into the cell, as has been previously reported [39]. While these interpretations are speculative due to the markers used and the nature of static representations of this dynamic process, the relatively low abundance of these structures in starved and tubacin-treated conditions indicates that they may represent ciliary loss intermediates.
We emphasize that loss of the cilium involves a complex interplay between general ciliary trafficking and regulation, cytoskeletal dynamics, and intracellular signaling ( [4,9] Fig 5) and that our manipulations of calcium and katanin are best viewed as initial probing into the molecular nature of these decisions. Further work is required to understand the molecular mechanisms and must take into account that many key regulators of ciliary loss have additional roles in cytoskeletal regulation [22,32,54] and other cellular functions, including calcium [97,98], AurA [99,100], HDAC6 [101][102][103][104], and katanin [70,71,105].
We observed deciliation as a means for cell cycle-linked ciliary loss, whereas previous descriptions of such behavior in mammalian cells have only been under conditions of experimentally induced toxicity and stress [50,52]. The morphology and composition of isolated cilia confirmed two major points supporting the interpretation that the observed Instant loss events represent deciliation: 1) the entire ciliary membrane can be shed from cells as an intact structure, and 2) shed ciliary membranes contain tubulin, suggesting that the axoneme is severed and shed along with the ciliary membrane. Interestingly, a related phenomenon-the release of a small membrane segment from the ciliary tip, referred to as apical abscission, decapitation, or release of ciliary ectosomes-has been described in several contexts [88,106,107], but it is unclear whether axoneme components are present in these structures. Ciliary decapitation was previously reported as an initiating step in the cilia-loss process [88] as well as in assembling cilia [9]. It would be particularly interesting to test whether there is a mechanistic link between these events by investigating whether PI(4,5)P 2 -and F-actin-mediated membrane constriction that drives decapitation participates in deciliation from the ciliary base as well as other locations along the cilium. Finally, the implications of our findings for understanding cilia loss should be considered in the context of the general features of cultured kidney-derived epithelial cells, in comparison to and contrast with cilia in other experimental approaches, cell types, tissues, and organisms. Although most cilia share a similar core structure and associated machinery, differences in structure and function (i.e., primary versus motile/specialized), and the relative size of cilia to the cell body, are likely to be relevant to the mechanisms of ciliary assembly and loss.

Cell culture
IMCD3 cells were grown in DMEM-F12 medium with 10% fetal bovine serum and 1% penicillin-streptomycin-kanamycin antibiotic cocktail at normoxic conditions. Cells were passaged every 2-3 days at a dilution of 1:10-1:20. Cells were tested for mycoplasma with Sigma Look-Out Mycoplasma PCR Detection Kit (Cat# MP0035) as directed by the manufacturer, and incidences of mycoplasma contamination were treated with Mycoplasma Removal Agent (MP Biomedicals, #093050044). Following decontamination, experiments that were potentially affected by mycoplasma contamination were repeated at least three times to determine any difference in results, and no significant differences were observed.

Serum starvation and stimulation
Cells were seeded in 24-or 6-well dishes with glass coverslips for imaging following fixation or 35 mm glass-bottomed MatTek dishes (#P35G-0-10-C) for live imaging. Cells were seeded in 24-well dishes at a density of 1.5 x 10 4 cells and 6-well and 35 mm MatTek dishes at 1-1.5 x 10 5 to achieve 50%-70% confluence next day. For serum starvation, cells were washed once with 0.2% DMEM-F12 + PSK, then grown in 0.2% DMEM-F12 + PSK for 24 hrs. Serum stimulation was by either readdition of FBS directly to dishes to 10% final concentration or replacement with 10% FBS DMEM-F12.

Generation of stable cell lines
IMCD3-SSTR3::GFP-mCherry::PACT: mCherry::PACT was cloned from a pLV plasmid (pTS3488, created by multisite Gateway cloning by Christian Hoerner) onto a pLV-Puro-EF1a construct using Gibson cloning. Lentivirus with the cloned construct was generated in HEK293T and used to infect IMCD3-SSTR3::GFP (gift from Nachury laboratory, [55]) under selection with 800 ng/μL puromycin for 4-5 days. Infected cells were FACS sorted into polyclonal populations by mCherry fluorescence intensity, and a pool of low-expressing cells was selected to prevent overexpression phenotypes of a centrosomal protein.

Immunofluorescence microscopy
Generally, fixation for immunofluorescence microscopy was done with 100% methanol for 5 minutes at −20˚C, followed by washes with 0.1% Triton X-100 in PBS at room temperature for 2 minutes, and washed three times in PBS. Samples were blocked for 1 hr at RT or overnight at 4˚C in 2% BSA, 1% goat serum, 75 mM NaN 3 . Antibodies were diluted to the indicated concentrations in blocking buffer. Primary antibody incubations were performed for 1 hr at RT or overnight at 4˚C. Secondary antibody incubations were performed for 1-2 hrs at RT. Following each antibody incubation, samples were washed three times in PBS + 0.05% Tween-20 for 5 minutes each at RT.
Images were acquired with a Zeiss Axiovert 200 inverted epifluorescence microscope and a 63x objective, or a Leica SP8 scanning laser confocal microscope with LASX Software, using mercury or argon lamps with white light laser excitation, and a 63x 1.4 NA objective. Exposure times were constant during each experiment. For imaging of serum-starved and serum-stimulated cells, fields of view were selected based on DAPI staining by two criteria: 1) to select for moderate cell density, in order to avoid effects of high density on cell cycle and ciliation and 2) to eliminate bias in percent cilia quantifications from scanning by ciliary markers.

Live-cell confocal microscopy
Cells were cultured on glass-bottomed Mattek dishes and imaged in DMEM-F12 media with 15 mM HEPES without phenol red. Movies were acquired 4-12 hrs after serum stimulation with a Leica SP8 scanning laser confocal microscope using 0.5-μm z-slices, 30-90-second intervals, autofocus (Best Focus function), in a 37˚C incubator, and red and green channels were acquired simultaneously. The video file was saved as .lif from LASX software and opened in Imaris x64 8.0.2 as a 3D render for analysis of cilia disassembly dynamics and basal body positioning.

Data analysis
Cilia counts and length measurements were performed either manually in Fiji or Imaris x64 8.0.2 and 9.2.1 or through semiautomated detection in Imaris. Manual analysis involved detecting ciliary membrane, marked by an enrichment of SSTR3::GFP above background threshold, that were adjacent to a centriole (mCherry::PACT in dual-fluorescent cells or pericentrin immunofluorescence in single [SSTR3::GFP-expressing] or nonfluorescent cells), to distinguish from accumulations of SSTR3+ membrane elsewhere in the cell. Manual length measurements in Fiji [108] were made with the line function, and in Imaris (Bitplane, version 8.0.2 and up) with the Measurement tool. Generally, single z-plane images were analyzed in Fiji or Imaris, while confocal z-stacks were analyzed in Imaris which allowed more accurate length measurement due to the 3D render (Surpass) capability. When possible, length measurements in confocal images were semi-automated in Imaris using the Surfaces function to create an artificial object encompassing the ciliary membrane, and exporting Bounding Box data as a proxy for length (the longest dimension of the object).
For live cell serum stimulation experiments, movies were visually scanned in Imaris for examples of disassembling cilia. Images of each disassembling cilium were cropped by time (from t0 to several mins after complete loss) and position (restricted to area of occupancy during the that time window) and then saved in a separate file. To generate ciliary length curves, the ciliary membrane was isolated as an artificial object using the Surface function. When possible, the object was automatically tracked over consecutive time points with length data generated at each time point. In cases where automatic tracking was not possible due to low signalto-noise of ciliary membrane fluorescence, measurements were taken manually at 15-30 minute intervals until the initiation of ciliary disassembly, and at each time point during the disassembly event.
Matlab. Raw length measurement data from disassembling cilia movies were compiled into an Excel spreadsheet. The data were imported into a Matlab algorithm which performed Gaussian smoothing using a moving average and normalization to starved control cilia dataset. The start point of the cilium loss event was defined by scanning backwards from the final point (time at which length = 0 μm) until the derivative (length difference between consecutive time points) is not different in magnitude than the mean control slope (−0.005 μm/min).

Cilia isolation
Cell culture: clones of IMCD3 cells, either unsorted and stably expressing GFP-SSTR3 or FACS-sorted for medium expression of GFP-SSTR3, were grown on 15-cm dishes at 3 x 10 6 cells/dish in DMEM/F12 with 10% FBS and antibiotics for 24 hrs. The cells were washed 3x with HDF wash buffer (136 mM NaCl, 5 mM KCl, 5 mM glucose, 4 mM NaHCO 3 , 0.7 mM EDTA), and medium was replaced with DMEM/F12 and 0.2% FBS and antibiotics (serumstarved) for 24 hrs. Then, all dishes were washed 3x with HDF buffer and half received phenolred free DMEM/F12 with 0.2% FBS (serum-starved), and the other half received phenol red free DMEM/F12 with 10% FBS (serum-stimulated) for 24 hrs. Total serum starved time was 48 hrs and total serum stimulated time was 24 hrs.
Sample preparation. Culture medium was collected and subjected to centrifugation for 10 minutes at 1,000 x g at 4˚C to remove large cell debris. Samples were then kept on ice until plating on treated dishes or stored at 4˚C for a maximum of 1 day. 4 mL serum-stimulated or -starved medium was incubated on a treated MatTek dish overnight at 4˚C, followed by three gentle PBS washes. Samples were then imaged directly, without fixation with a Leica SP8 confocal microscope.

SEM. Antibody-immobilized
MatTek dishes incubated with serum stimulation medium were fixed for SEM in 4% PFA, 2% glutaraldehyde, and 0.1M Na cacodylate. Glass bottoms were removed, processed for imaging, and imaged with a Hitachi S-3400N VP SEM scope in the Beckman Imaging Facility, Stanford University.
Filter-spin concentration method. Harvest of cilia: Deciliation of starved IMCD3 cells (positive control): Serum-stimulated or -starved culture medium, or fresh culture medium (with 10% FBS, an additional control) was removed from six 150-cm dishes, combined, and centrifuged at 1,000 x g at 4˚C for 10 minutes in the A-4-81 rotor, Eppendorf 5810R centrifuge to remove large cellular debris. Cells were washed two times with warm PBS containing 0.4% EDTA. 10 mL was added to a MatTek dish and incubated for 10 minutes at 37˚C, followed by gentle up and down pipetting to remove cells from dish. An aliquot of cell suspension was removed for cell count. Cells were centrifuged at 13,000 x g for 5 minutes at RT. The cell pellet was resuspended in 5 mL ice cold deciliation buffer (62) (112 mM NaCl, 3.4 mM KCl,10 mM CaCl2, 2.4 mM NaHcO3, 2 mM HEPES, pH7.0, and a protease inhibitor tablet [Roche]). The cell suspension was incubated at 4˚C for 15 minutes with end-over-end rotation, and then centrifuged at 1,000 x g for 5 mins at 4˚C in an Eppendorf centrifuge. The resulting supernatant was used for biochemistry and immunostaining.
Biochemistry. Half of the supernatant material from deciliated, serum-starved or -stimulated cells was centrifuged at 21,000 x g for 15 minutes in JA25.5 rotor in Beckman Coulter Avanti J-25I centrifuge at 4˚C. The supernatant was carefully removed, and pellets were resuspended in 160 μl of sample buffer (1% SDS, 10 mM Tris-HCl, pH 7.5, 2 mM EDTA). Samples were boiled at 95˚C for 8 mins, and equal volumes were separated by 10% PAGE and transferred to PVDF. Blots were blocked (2% BSA, 1% normal donkey and goat serum in TBS, pH 7.4) for 1 hr at RT or overnight at 4˚C. Membranes were blotted with YL1/2 (1:1,000), mouse acetylated-tubulin antibody (1:1,000), and IFT88 rabbit antibody (1:500) in blocking buffer for 1 hr at RT. Blots were washed 5x with TBST. Secondary anti-rabbit, anti-mouse, or anti-rat antibodies labeled with either IRDye680CW or IRDye800CW (Li-Cor Biosciences, #926-32213), at 1:30,000 dilution were incubated with blots for 30 minutes at RT. Blots were washed five times with TBST and scanned on Licor Odyssey scanner (Li-Cor BioSciences).
Immunofluorescence. Half of the supernatant material from deciliated, serum-starved, and serum-stimulated cells was concentrated using a 250 ml 0.2 μm PES filter unit with lowpower house vacuum to reduce the volume (20 or 40 mL from one or two 15-cm dishes, respectively) to 2 mL, and finally a Millipore Ultrafree-MC filter (PVDF 0.2 μm size #UFC30GV100) to reduce the volume to approximately 0.5 mL. 5 μl of concentrated supernatant was pipetted onto an acid-treated glass slide. A 22-mm acid-treated circular glass coverslip was placed on the sample, and the slide was immediately plunged into liquid nitrogen for approximately 5 seconds. After removing the slide, the coverslip was removed and fixed in −20˚C 100% methanol for 5 minutes. Alternatively, 400 μm of the sample were loaded into an 8-well glass-bottomed LabTek dish (Sigma-Aldrich, # Z734853), and centrifuged at 3,500 x g for 20 minutes, 4˚C. Immunofluorescence staining was performed as described above.

Statistics
All analyses were performed in GraphPad Prism. Statistical tests used for each analysis are indicated in the figure legends. No explicit power analysis was used to determine sample size. All experiments were performed with at least three biological replicates, i.e., samples from independent cell culture passages. When used, technical replicates (i.e., repeats from the same cell culture passage) were averaged for each biological replicate. In brief, comparisons of mean values such as mean percent cilia across replicate experiments were compared using an unpaired t test. Analyses of individual measurements such as cilia length were subjected to normality tests (Kolmogorov-Smirnoff, D'Agostino and Pearson, and Shapiro-Wilk). If data passed all normality tests, unpaired t test was used; if not, the Mann-Whitney U test was used. If data passed normality by some tests but not others, both types of analyses were performed. Results were similar between parametric and nonparamentric tests unless stated otherwise.