Dynamic interaction of spliceosomal snRNPs with coilin explains Cajal body characteristics

Cells and, specifically, the cell nucleus contain various inclusions and bodies in which specific proteins and RNAs concentrate. None of the nuclear bodies are bounded by a physical barrier, such as lipid membranes. In recent years, intensive research has been conducted to determine how nuclear inclusions form and are maintained in the dynamic nuclear environment. A model based on liquid–liquid phase separation (LLPS) has attracted a lot of attention because it can explain many properties of nuclear bodies, such as their spherical shape, their ability to merge and fission, and to exchange their components with the nucleoplasm (Courchaine et al., 2016; Staněk and Fox, 2017; Zhu and Brangwynne, 2015). The LLPS model was further supported by characterizing the properties of the proteins that drive phase separation. RNA-binding domains and disordered low-complexity regions were shown to be crucial for the formation of bodies (Banani et al., 2016; Guillén-Boixet et al., 2020; Kato et al., 2012; Lin et al., 2015; Mitrea and Kriwacki, 2016; Molliex et al., 2015; Patel et al., 2015). A second important factor that promotes phase separation is the presence of nucleic acids and, in particular, various noncoding RNAs, which are essential components of many nuclear compartments. However, it should also be mentioned that most studies describing the molecular principles of nuclear body formation are based on in vitro experiments with defined sets of factors under artificial conditions (ion concentration, volume exclusion, temperature, etc.), and only a limited number of studies have fully tested the predictions of the LLPS model in vivo (McSwiggen et al., 2019b). While some in vivo experiments were consistent with the LLPS predictions, an alternative model describing a spontaneous concentration of RNA polymerase II on viral DNA has been also proposed (McSwiggen et al., 2019a).

Cajal body (CB) is often mentioned as a classical example of a non-membrane–bound structure whose formation is based on LLPS (Mitrea and Kriwacki, 2016; Zhu and Brangwynne, 2015). CBs are capable of merging and splitting, and their components are constantly cycling between the body and the nucleoplasm (Dundr et al., 2004; Novotný et al., 2011; Platani et al., 2000). However, their LLPS properties have not been rigorously tested. Coilin is a conserved architectural protein of CBs, and its oligomerization is essential for CB formation (Courchaine et al., 2022; Hebert and Matera, 2000; Machyna et al., 2013; Machyna et al., 2015). Coilin deletion and subsequent CB disappearance is lethal for developing fish embryos and causes sever reduction in mouse fertility (Strzelecka et al., 2010; Tucker et al., 2001). In addition, coilin interacts with short noncoding snRNAs, scaRNAs, and snoRNAs and contains several low-complexity domains (Enwerem et al., 2014; Machyna et al., 2014; Machyna et al., 2015). All of these properties strongly resemble “classical” LLPS-inducing factors and make coilin a candidate protein that drives phase separation and CB formation (Courchaine et al., 2016; Kato et al., 2012; Lin et al., 2015; Mitrea and Kriwacki, 2016). However, a study analyzing principles of CB formation suggested that CB assembly is driven by heterotypic interactions and proposed that additional, as yet unidentified factor(s) are important for CB formation (Riback et al., 2020). Indeed, a recent study revealed a long list of proteins whose downregulation affects the number and structure of CBs (Arias Escayola et al., 2025).

CBs are conserved structures involved in biogenesis and quality control of small nuclear RNP particles (snRNPs) (Staněk, 2017). CBs also efficiently respond to intracellular conditions. They disappear after inhibition of snRNP biogenesis and RNA transcription (Ferreira et al., 1994; Haaf and Ward, 1996; Lemm et al., 2006) and appear when concentration of partially assembled snRNPs increases (Novotný et al., 2015; Roithová et al., 2018). These data indicate that snRNPs, and especially the snRNP assembly intermediates, could be the additional factors essential for the CB integrity.

In this project, we tested a hypothesis that CB formation is driven by LLPS and that the coilin–snRNP interaction is important for CB assembly. First, we manipulated intracellular coilin concentration and cell incubation temperature to specifically test some predictions of the LLPS model. We further generated several coilin variants containing point mutations or deletions that prevented coilin oligomerization or interaction with snRNPs. We monitored movement of the mutated coilin proteins inside and outside CBs using FRAP, single-point fluorescence correlation microscopy (spFCS), and the 2D pair correlation function (2D-pCF) to determine the main factors determining coilin dynamics inside the nucleus and CBs. We further analyzed a role of RNA, coilin oligomerization, and coilin interaction with snRNPs in CB formation. Our results are not fully compatible with a one-component LLPS model and show that besides coilin self-interaction, RNA presence and binding between coilin and snRNP are crucial for CB integrity. We propose and test a model based on the assembly/disassembly dynamics of the coilin–snRNP complexes that adequately describes experimentally determined coilin dynamics and is consistent with the observed role of coilin and snRNPs in CB formation.

First, we decided to perform basic characterization of CBs and coilin. To avoid the contribution of endogenous coilin in CB formation, we utilized a coilin knockout HeLa cell line (HeLa^coilinKO [Basello et al., 2022]). Expression of human coilin tagged at the N terminus with EGFP (EGFP-coilin^WT) in HeLa^coilinKO cells rescued the formation of CBs, and these CBs accumulated snRNPs and fibrillarin (Basello et al., 2022). First, we tested whether increased EGFP-coilin^WT concentration (measured as EGFP fluorescence intensity) enhanced CB area (determined as a combined parameter of CB size and number Fig. 1 A, left column) or affected EGFP-coilin^WT accumulation in CBs (Fig. 1 A, middle column). Because LLPS is temperature sensitive, we determined CB area and EGFP-coilin^WT concentration in CBs in cells incubated at three different temperatures (30°C, 37°C, and 42°C). In neither case did we observe that EGFP-coilin^WT accumulation in CBs and CB area scales with EGFP-coilin cellular concentration. Moreover, the size of individual CBs did not vary, and CBs showed similar areas in all three tested temperatures (Fig. 1 A, right column). The only parameter that was temperature-dependent was the number of CBs, which increased in cells incubated at 30°C (Fig. 1 B). A similar behavior was previously observed in primary fibroblasts (Carmo-Fonseca et al., 1993). While the temperature dependence of CB number is consistent with the LLPS predictions, the lack of correlation between coilin concentration and CB area indicates that CBs do not follow a one-component LLPS model, which predicts that increased coilin concentration in the cell nucleus should be buffered by higher coilin accumulation in CBs. See Table 1 that summarizes and compares experimental findings with predictions of the LLPS model.

Next, we studied the coilin distribution and dynamics inside CBs. While epifluorescence and confocal microscopy show CBs as round homogenous inclusions, electron microscopy revealed that the coilin localization inside CBs was not homogenous, and coilin accumulated in electron-dense substructures (Pena et al., 2001; Raska et al., 1991). Here, we monitored the distribution of endogenous coilin in CBs by indirect immunofluorescence, followed by 3D structural illumination microscopy (3D-SIM) and stimulated emission depletion microscopy (STED) (Fig. 2 A). Both super-resolution microscopy approaches revealed a similar pattern of non-homogenous coilin localization within CB resembling the electron micrographs of CB. These data indicated that CB substructures were not an artifact of the sample preparation for electron microscopy, but CB is genuinely a non-homogenous structure. To determine, whether coilin-rich CB substructures are dynamic or static, we examined CB substructures in live cells by fast-acquisition 3D-SIM using a stable cell line–expressing coilin tagged with EGFP (HeLa^coilin-EGFP; [Machyna et al., 2014] [Fig. 2 B and Video 1]). This observation showed that CBs are intrinsically dynamic structures with constant internal movement of its sub-compartments. To further analyze coilin dynamics inside CBs, we photobleached a half of CB and observed recovery of the fluorescence. We observed a rapid recovery in the bleached area, while fluorescence in the unbleached part of CB reciprocally dropped (Fig. 2 C, left panel). When the entire CB area was photobleached, fluorescence recovery was an order of magnitude slower (Fig. 2 C, right panel), which is in agreement with previous reports (Dundr et al., 2004; Sleeman et al., 2003). The FRAP results are consistent with live-cell 3D-SIM observations and together show a rapid movement of coilin within CBs and a much slower exchange of coilin between CBs and the nucleoplasm. However, with the current resolution of microscopy techniques, we are unable to distinguish whether the fast dynamics within CB are caused by the movement of larger coilin-rich domains, as indicated by live 3D-SIM, or by the movement of individual coilin molecules.

To further examine coilin flux across the CB-nucleoplasm boundary, we applied fast acquisition confocal microscopy followed by 2D-pCF analysis. This method identifies anisotropic tracks at different spatial locations in the image and allows creation of high-resolution maps of molecule movement and identification of barriers for proteins movement inside a living cell (Di Rienzo et al., 2016; Malacrida et al., 2020). First, we applied the method on a control sample containing a solution of 20-nm fluorescence beads mixed with much bigger nonfluorescent agarose beads. While the fluorescent particles freely diffuse in the solution, they are excluded from the area occupied by agarose beads, creating a clear border at the agarose beads surface (Fig. 3 A). Next, we analyzed EGFP-coilin^WT motion inside the nucleus and observed a strong movement anisotropy around CBs (Fig. 3 B), which is consistent with FRAP data and indicates a presence of a diffusion barrier for coilin at the CB-nucleoplasm boundary.

The diffusion constraints for coilin movement in and out of CB are consistent with the role of coilin as the key assembly factor of CBs. However, increased coilin concentration does not induce CBs (Fig. 1), which indicates that other cofactors are required for CB formation (Riback et al., 2020). Therefore, we searched for additional factors that modulate coilin properties and influence CB. First, we tested whether RNA, which is a common component of LLPS-driven bodies, is important for CB. As was already shown, coilin directly interacts with numerous short noncoding RNAs (Machyna et al., 2014), but the role of these RNAs in CB formation is unclear. To investigate whether RNA is critical for CB formation, we microinjected RNase A into the nucleus of HeLa^coilin-EGFP and monitored CBs under a confocal microscope. Within tens of seconds after RNase A injection, CBs started dissolving and disappeared within 2–3 min (Fig. 4 A and Video 2). To control whether the microinjection itself had any effect on the CB integrity, we microinjected DNase I into the nucleus of HeLa^coilin-EGFP but did not observe any CB disassembly (Video 3). To further control the effect of RNase injection, we injected RNase A into the nucleus together with 70-kD dextran-TRICT, incubated for 2–3 min at 37°C with 5% CO₂, fixed, and stained for fibrillarin, which is located in CBs and nucleoli and SMN, which is found in gems (Fig. 4 B). Fibrillarin localization to nucleoli is RNase sensitive (Decker et al., 2022; Koo et al., 2016; Ochs et al., 1985). Similarly to previous reports, we observed delocalization of fibrillarin into the nucleoplasm. In addition, we did not detect any CB-like structures positive for the fibrillarin staining. In contrast, SMN-positive gems were largely unaffected by the RNase injection, which suggest that RNase injection does not disrupt all nuclear membraneless structures.

To better understand the role of RNA in CB formation, we tested whether coilin self-interaction, which is critical for CB assembly (Courchaine et al., 2022; Hebert and Matera, 2000; Machyna et al., 2015), is RNA dependent. Cell extracts were prepared from HeLa^coilin-EGFP and treated with RNase A for 30 min. Coilin-EGFP was immunoprecipitated by anti-EGFP antibodies, and monitoring of the co-purified endogenous coilin revealed that RNA degradation significantly reduced coilin self-association (Fig. 4 C). To further test the effect of RNA on coilin self-association, we expressed EGFP-coilin^WT in HeLa^coilinKO, and cross-linked coilin and RNA using UV light followed by immunoprecipitation of cross-linked complexes by anti-EGFP antibodies (cross-link and immunoprecipitation—CLIP). The immunoprecipitated complexes were analyzed by western blotting using the anti-GFP antibody. Covalent cross-linking fixed higher coilin complexes that were detectable in denaturing SDS-PAGE conditions (Fig. 4 D, left line). RNA digestion reduced the presence of higher coilin complexes, while the amount of monomeric coilin and a ∼40-kD EGFP-coilin fragment increased (Fig. 4 D, right line). In both treated and untreated extracts, we observed a band above 250 kD, which likely represents EGFP-coilin dimers, and its presence was also partially reduced by the RNase treatment. These data indicate that RNA is critical component that promotes formation of higher coilin complexes, either providing a scaffold for binding of multiple coilin molecules or stimulating coilin self-interaction.

We recently described a point mutation K496E in the conserved C-terminal Tudor domain that prevents CB formation in cells lacking endogenous coilin (Basello et al., 2022). Because it has been shown that coilin C-terminal part binds snRNPs (Xu et al., 2005), we decided to determine whether snRNPs and, specifically, the coilin–snRNP interaction are important for CB assembly. We aligned the last 156 aa at the C terminus to identify conserved residues (Fig. 5 A). Next, we mutated several conserved aa forming the hydrophobic core of the Tudor domain, specifically I477G, F479G, W554G, and L557G (Fig. 5 B). We further replaced conserved loops 1 and 2 with β-turns and created EGFP-coilin variants named Δβ1β2 (the aa sequence LELTSSYSPDV was replaced by DPSG) and Δβ3β4 (aa sequence LREPGKFDLVYHNENGAEVVEY was replaced by DPNG), respectively. It has been previously shown that the conserved loop substitution did not disrupt the Tudor domain structure (Shanbhag et al., 2010). All the mutations/deletions in the Tudor domain did not block accumulation in preformed CB in HeLa^WT cells but failed to rescue CBs in HeLa^coilinKO cells, showing that intact C-terminal Tudor domain is necessary for CB formation (Fig. 5 C). These data together with previous experiments show that both the self-interacting N-terminal and Tudor C-terminal domains are critical for CB integrity (Hebert and Matera, 2000).

Because Tudor domain was proposed to interact with snRNPs and Sm proteins (Xu et al., 2005), we tested whether the mutations affect coilin–snRNP association. We included previously characterized substitutions K496E and R8A, which also prevent CB formation in HeLa^coilinKO cells and act as dominant-negative variants that disrupt CB when overexpressed in HeLa^WT ([Basello et al., 2022; Courchaine et al., 2022] and Figs. 6 A and S1 A). We further included coilin variants with the mutation in the hydrophobic core (W554G), the loop I deletion (Δβ1β2) that prevents CB assembly and tested them for interaction with Sm proteins as common markers of snRNPs. All of the tested mutations negatively affected the association of coilin with Sm proteins when expressed HeLa^coilinKO cells (Figs. 5 D and 6 B). To gain more insight into the identity of snRNPs associated with coilin, we analyzed co-precipitated snRNAs and snRNP-specific proteins with EGFP-coilin and its variants (Fig. 6, B and C). U2 snRNA was co-purified with EGFP-coilin to the highest extent, followed by U1 and U4 snRNAs. Analysis of selected snRNP markers (SNRNPB2/U2B″ for U2, SART3 for U4/U6, and PRPF6 for U5 snRNPs) confirmed that coilin was associated with all major snRNPs and that the K496E mutation in the Tudor domain disrupted this interaction. As a control, we used the R8A substitution, which prevents coilin self-association but does not prevent interaction with snRNPs (Fig. 6, B–D) (Courchaine et al., 2022). The double-mutant variant containing both R8A and K496E substitutions combined the effect of both mutations, and this variant did not interact with snRNPs, lost coilin–coilin association, failed to accumulate in preformed body, and did not reconstitute CBs in HeLa^coilinKO (Fig. 6). Collectively, these data showed that in addition to coilin self-interaction, association of snRNPs with coilin Tudor domain is necessary for CB formation.

In our experiments, we consistently observed a smaller EGFP-coilin fragment of apparent molecular weight around 40 kD reacting with anti-EGFP antibodies (Figs. 4, 5, and 6). A calpain cleavage site has been identified downstream of the nucleolar localization signal around arginine 136 (Velma et al., 2012) (Fig. S1 B). To probe whether the observed fragment could represent the N-terminal 1–136 aa, we cloned coilin N-terminal part spanning aa 1–136 and tagged it with EGFP at the N terminus (EGFP-coilin^1–136). The EGFP-coilin^1–136 was transiently expressed in Hela^coilinKO cells and detected by western blotting (Fig. S1 C). EGFP-coilin^1–136 migrated at the same position as the observed cleavage coilin product, which strongly indicates that the cleavage occurs around position 136. The 1–136 peptide localized to nucleoli and CBs in parental WT cells, and only to nucleoli in cells deleted of endogenous coilin (Fig. S1 D), which is consistent with the previous findings that the N-terminal domain is responsible for coilin self-interaction and targets coilin to CBs (Courchaine et al., 2022; Hebert and Matera, 2000).

It has been previously shown that CB formation can be induced by restricting the movement of various CB components by tethering them to chromatin (Kaiser et al., 2008). We therefore analyzed whether snRNPs binding reduces coilin movement in the nuclear environment and, thus, triggers coilin aggregation and formation of CBs. To do so, we performed spFCS measurements in the nucleoplasm of HeLa^coilinKO cells expressing EGFP-coilin^WT, as well as variants containing a single substitution EGFP-coilin^R8A or EGFP-coilin^K496E, and the construct with both mutations (EGFP-coilin^R8A+K496E) (Fig. 7 A). EGFP alone, which freely diffuses throughout the nucleus, was used as a control. Its diffusion coefficient was determined to be 32.26 ± 11.646 μm²/s. The autocorrelation curve of EGFP-coilin^WT was fitted well with a two-component model with one fast-diffusing fraction, with a diffusion coefficient similar to EGFP (40.03 ± 27.63 μm²/s) and a second fraction that moved >50 times slower with diffusion coefficient of 0.62 ± 0.279 μm²/s. The large difference in the diffusion coefficients of the two fractions strongly suggested that the fast component represented coilin in a monomeric interaction-free state, while the slow component contained a mixture of EGFP-coilin molecules involved in multiple interactions (e.g., self-oligomerization and interactions with other protein and RNA partners). However, it is unlikely that monomeric EGFP-coilin would move faster than EGFP alone. We assume that the diffusion coefficient of the fast-moving fraction of EGFP-coilin is artificially high, caused by fitting two components with large differences in diffusion coefficients. Using this assumption, we fixed the diffusion coefficient of the fast-moving fraction to the diffusion coefficient measured for EGFP (32.26 μm²/s) and refitted the autocorrelation curves to obtain the diffusion coefficient of the slow coilin fraction. The refitting slightly changed the diffusion coefficient of the slow moving EGFP-coilin^WT molecules, which was determined to be 0.70 ± 0.244 μm²/s (Fig. 7 B).

We then used the same approach to fit the movement of the other three coilin variants. All coilin mutants displayed the presence of two components but differed in their respective slow-moving fractions. EGFP-coilin^WT and EGFP-coilin^K496E clustered together, and their diffusion in the nucleus was the slowest. In detail, EGFP-coilin^K496E protein showed very similar dynamics to EGFP-coilin^WT, with a diffusion coefficient of 0.80 ± 0.255 μm²/s, while EGFP-coilin^R8A and EGFP-coilin^R8A+K496E variants were faster, with diffusion coefficients of 1.56 ± 0.735 and 1.99 ± 1.666 μm²/s, respectively (Fig. 7 B).

Next, we compared the ratio of the amplitudes of slow- and fast-moving fractions and observed that EGFP-coilin^R8A and EGFP-coilin^R8A+K496E displayed a ratio of ∼1:1 meaning that ∼50% of molecules are found in the fast-moving fraction and ∼50% of the molecules diffuse slowly. We did not find any significant difference between dynamics of EGFP-coilin^R8A, which can interact with snRNPs and EGFP-coilin^R8A+K496E, where the interaction with snRNPs is prevented. We therefore conclude that the interaction with snRNPs is not the dominant factor determining coilin dynamics in the nucleoplasm. The presence of a slow component in EGFP-coilin^R8A+K496E is likely due to other coilin interactions (e.g., with Nopp140, WRAP53, SMN, and others) (Arias Escayola et al., 2025; Courchaine et al., 2022).

Particle brightness analysis suggested that only EGFP-coilin^WT and EGFP-coilin^K496E oligomerize (Fig. 7 C), which is consistent with immunoprecipitation and CLIP experiments (Figs. 4 and 6). The data also show that EGFP-coilin^WT or EGFP-coilin^K496E are two times brighter than EGFP alone or coilin mutants with the R8A mutations that prevent self-interaction. This would indicate that coilin preferentially forms dimers and higher oligomers are rare. However, EGFP fluorophores in oligomers can quench each other, which would lead to an underestimation of the amount of coilin oligomers. In addition, larger oligomers move more slowly and may be quenched during the measurement and would not be detected. Therefore, we conclude that the coilin does not form large oligomers containing a high number of coilin monomers, but the exact number needs to be further determined. Unfortunately, the difference in particle brightness precluded direct comparison of components ratio between the two groups: EGFP-coilin^WT and EGFP-coilin^K496E on one side and EGFP-coilin^R8A and EGFP-coilin^R8A+K496E on the other side. In addition, due to higher particle brightness, the observed distribution of fast/slow fractions for EGFP-coilin^WT and EGFP-coilin^K496E should be taken with caution (Fig. 7 B). However, the fact that coilin^WT and coilin^K496E dynamic behavior is similar and cluster together strongly indicates that coilin oligomerization is the main factor that determines coilin dynamics in the cell nucleoplasm.

To gain further insight into the dynamics of coilin in the cell nucleus, we performed FRAP using the same cell line and constructs used for spFCS. These two methods are complementary because spFCS is able to detect fast diffusion, whereas FRAP is more suitable for tracking slower moving molecules. Similarly to spFCS, FRAP curves of EGFP-coilin^WT and EGFP-coilin^K496E were almost identical, indicating the same dynamics in the nucleoplasm. EGFP-coilin^R8A and EGFP-coilin^R8A+K496E mutants moved significantly faster than EGFP-coilin^WT, and their recovery curves were similar to each other (Fig. 7, D and E). FRAP analysis of two other coilin variants with mutations in the C terminus (Δβ1β2 and W554A) revealed almost identical FRAP curve to the EGFP-coilin^K496E protein (Fig. S1 E). We observed slower recovery of the EGFP-coilin^R8A+K496E double mutant than EGFP alone, which is again consistent with spFCS and suggests that the central part of coilin and the RG box are able to interact with other partners. To eliminate the effect of potential interaction partners and further test the effect of coilin self-association on the dynamics of coilin in the nucleoplasm, we introduced R8A mutation into the EGFP-coilin^1–136, which lacks the central part as well as the RG box and the C-terminal Tudor domain. FRAP measurements of EGFP-coilin^1–136/R8A revealed a similar diffusion to EGFP alone, while EGFP-coilin^1–136 recovery was slower, confirming the importance of coilin self-interaction and the coilin interaction network for coilin dynamics in the nucleoplasm (Fig. S1 F). Together, FRAP and spFCS measurements ruled out the possibility that snRNP binding significantly changes coilin dynamics and provided evidence that coilin oligomerization is the dominant feature that determines coilin dynamics inside the nucleoplasm. These data also show that the interaction with snRNPs does not change coilin dynamics in the nucleoplasm and indicate that snRNPs do not induce CB via slowing down coilin movement.

Next, we decided to build a mathematical model that would describe coilin dynamics inside the nucleus and exchange between CBs and the nucleoplasm. Based on previous results describing coilin self-interaction as the key factor in CB assembly and our current observations showing the importance of coilin–snRNP interaction for CB integrity, we considered the oligomer–coilin–snRNP (coilin_[N]–snRNP) complex as a fundamental building block of CB. To describe dynamic nature of coilin_[N]–snRNP complexes, we build a model based on reaction rates Michaelis–Menten equations (Fig. 8) and aimed to test whether this model is able to describe coilin dynamics in the CB measured by FRAP. The key feature of the model is that coilin concentration inside CBs depends on the rate of coilin oligomerization defined by the rate constant k₁* (apparent constant rate of the assembly reaction) and kinetics of the interaction between coilin oligomers and snRNPs in the nucleoplasm (constant rates k₂* and k₋₂) and inside CB (constant rates k₃* and k₋₃). The model is coilin centric and does not predict concentration of snRNPs, which are included in k₂* and k₃* constants. We also postulated that only coilin associated with a snRNP is able to enter CB, a condition that is based on the observed diffusion barrier for coilin molecules at the edge of CBs (Fig. 3). The model also includes the N-terminal fragment 1–136 aa, which we observed in all our experiments with EGFP-coilin transfection into HeLa^coilinKO cells. EGFP-coilin^1–136 is able to form oligomers with full-length coilin but does not bind snRNPs and thus does not affect the overall dynamics of the system.

First, we applied the model to fit FRAP curves in the nucleoplasm. We employed a step-wise approach to sequentially identify individual parameters of the model. In all cases, we applied global fitting to determine the same parameters for all FRAP curves. This approach considerably decreases the number of free parameters and thus reduces the freedom of the fit algorithm and the uncertainty of the determined parameters. The list of fitted parameters are summarized in Table S1. We started with the simplest situation and fitted FRAP curves of EGFP alone, which freely diffuses through the nucleoplasm and served as internal control. Next, we modeled behavior of EGFP-coilin^1–136/R8A, which does not appear to interact with coilin partners and contains the R8A mutation that prevents oligomerization. Both EGFP and EGFP-coilin^1–136/R8A recovery can be modeled using single-component equations (see Materials and methods for equations [Eqs. 1 and 2]) (Fig. S2, A and B). We fitted only one parameter “s” (μm/s) that describes the rate of molecules movement across the boundary between bleached and unbleached areas and more generally the movement speed in the nucleoplasm. The fitting results are summarized in Table S2. Next, we fitted FRAP curves of EGFP-coilin^R8A+K496E (Fig. S2 C), which contains the two mutations that block both oligomerization and snRNPs binding. To fit EGFP-coilin^R8A+K496E FRAP curves, we had to employ a two-component Eq. 3. To model the coilin and further variants, we fixed parameters determined in previous fitting steps. For example, the s_f value determined by fitting of EGFP-coilin^1–136/R8A was used for modeling of other EGFP-coilin variant FRAPs. We globally fitted concentrations of the full-length EGFP-coilin^R8A+K496E (c_m) and the EGFP-coilin^1–136 fragment (c_f) and the rate of movement of EGFP-coilin^R8A+K496E (s_m) (Table S2). The fitting procedure predicted roughly equal initial concentrations of full-length coilin and EGFP-coilin^1–136/R8A (compare sum of c_{m_out} and c_{m_in} with sum of c_{f_out} and c_{f_in}), which is in a good agreement with western blotting analysis (Figs. 4, 5, and 6).

Biochemical interactions (coilin oligomerization and interaction with snRNPs) were characterized by reaction rates k_i* and k_−i (see Fig. 8). As in the case of the movement rate s, the determined values for k_i were fixed and used to model a more complex situations. It should be noted that k_i* constants are only apparent reaction constants and include the cellular concentration of snRNPs. First, we fit FRAP curves of the EGFP-coilin^R8A variant, which does not oligomerize but interacts with snRNPs (Fig. S3 A and Table S2) (Eq. 4) to determine the apparent binding constants k₂*and unbinding constant and k₋₂, respectively. Similarly, we determined the oligomerization/depolymerization rate constants k₁* and k₋₁ using EGFP-coilin^K496E (Fig. S3 B and Table S2) (Eq. 5). Finally, we proceeded to model the FRAP curves for EGFP-coilin^WT (Eq. 6), which exhibits the most complex behavior. We kept all the previously determined constants fixed as fitting constraints and fitted the rate of movement of coilin oligomer–snRNP complexes (s_o+s) (Fig. S3 C and Table S2).

In a next step, we applied the parameters determined for coilin movement in the nucleoplasm and fitted FRAP recovery curves for EGFP-coilin^WT after bleaching in the CB (Fig. S4). We fitted constants k₃* and k₋₃, which describe formation and disassembly of coilin–snRNP complexes in CBs, and the transfer rates “t_i” between CB and the nucleoplasm. Here, we applied the condition that coilin without snRNP cannot enter CBs, and the constant t_{o_in} was set to zero. After 100,000 iterations, the fitting program converted to parameters that described the dynamics of EGFP-coilin^WT transfer across CB boundary and overall behavior in the nucleus (Table S3 and Fig. S4).

The proposed mathematical model describes coilin dynamics inside the nucleus and is able to predict coilin transfer in and out of CBs and kinetics of coilin–snRNP interactions. The modeling offers some interesting implication for coilin dynamics in the nucleus, snRNP metabolism, and CBs. (1) The model predicts ∼eightfold higher k₋₃ than k₋₂ (Table S3). These constants can be viewed as a constant rate of a putative biochemical reaction that converts immature snRNPs, which is able to interact with coilin, to mature snRNPs, which do not associate with coilin. This indicates that snRNP maturation rate is considerably faster inside CBs than in the rest of the nucleus. (2) The predicted transfer rate across CB boundary for coilin_[N]–snRNP particle is three times higher for incoming particles (t_{o+s_in} = 0.048) than particles leaving CBs (t_{o+s_out} = 0.014). This is consistent with our prediction that coilin_[N]–snRNP complexes accumulate in CBs. (3) The rate that describes coilin oligomers moving from CBs (t_{o_out} = 0.033) is twofold higher than for coilin_[N]–snRNP complexes (t_{o+s_out} = 0.014). We cannot exclude that this is a result of our assumption that coilin oligomers without snRNP cannot enter CBs (t_{o_in} was set to 0), and the fitting program thus converts to high leaving constant. However, if the value is correct, then the observed slow coilin exchange between CB and the nucleoplasm would indicate that majority of coilin oligomers that leaves CBs is not bound by snRNPs. These values also indicate that the drop in the concentration of immature snRNPs in the nucleoplasm reduces influx of coilin_[N]–snRNP particles into CBs, which results in CB disassembly. (4) Finally, when we allowed the model to predict colin concentrations in and out of CB at the equilibrium, the program predicted approximately twofold higher concentrations of coilin inside CB than in the nucleoplasm (con. end values in Table S3).

When the whole CB was bleached, we observed significant differences in recovery curves, which correlated with the EGFP-coilin concentration in cells determined as EGFP fluorescence. The higher the total EGFP-coilin concentration inside the cells, the faster and more complete recovery was observed (Fig. 9, A and B). To test the effect of snRNP binding on coilin residence time in CBs, we expressed EGFP-coilin^WT or EGFP-coilin^K496E in parental HeLa cells, which contain normal CB where both coilin variant accumulate. EGFP-coilin^WT exhibited a similar recovery kinetics and correlation between concentration and apparent plateau as in HeLa^coilinKO (Fig. 9 B). In contrast, we obtained a weak correlation between expression and apparent plateau for EGFP-coilin^K496E variant. This result is consistent with the model that EGFP-coilin^K496E variant does not contribute to coilin ability to cross the CB boundary, and EGFP-coilin^K496E CB accumulation is driven solely by the interaction with endogenous coilin. Furthermore, when EGFP-coilin^K496E concentration reached a threshold, CBs disappeared, preventing the measurement of FRAP in these cells (as shown by the lack of dots in Fig. 9 B for high expression of EGFP-coilin^K496E). These data are consistent with our model proposing that coilin_[N]–snRNP complexes are critical for CB assembly. Overexpression of EGFP-coilin^WT promotes coilin_[N]–snRNP complex formation, whereas EGFP-coilin^K496E presence in coilin oligomers reduces their potential to interact with snRNP and thus has a dominant-negative effect on CB integrity.

Finally, we used the correlation between recovery curves and coilin expression to test robustness of our mathematical model. To fit the FRAP curves, the model has to estimate concentration of EGFP-coilin in the nucleoplasm. We compared EGFP-coilin^WT concentration predicted by our model with a real value measured in the nucleoplasm of the same cells where FRAP was measured. We observed that the model consistently predicted slightly higher concentration compared with the real values, but in general the correlation between the predicted and the real values are in a good agreement (Fig. 9 C).

Characterizing the molecular principles that govern the formation of cellular non-membrane structures in dynamic cellular environment is a challenging task. Here, we describe several characteristics of nuclear CBs and provided evidence that CBs exhibit some but not all properties of the LLPS condensate (Table 1). Specifically, we observed the diffusion barrier that prevents free and fast exchange of coilin between CB and surrounding nucleoplasm. However, the diffusion barrier is apparently specific to selected CB components. While coilin and several other proteins (SMN, Gemin3, and TGS1) show similar or slower exchange between the CB and the nucleoplasm, many other CB components, including snRNPs, snoRNPs, Nopp140, B23, and GAR1, move through the CB border much faster (Dundr et al., 2004; Novotný et al., 2011; Sleeman et al., 2003). Similarly, fluorescently labeled dextran easily penetrated into Xenopus oocyte CBs, which is consistent with the model that CBs are rather permeable and allow free exchange of components with surrounding environment. This result is partially consistent with the LLPS model and confirms the role of coilin as the dynamic core molecule of CBs. On the other hand, the one component LLPS model is unable to explain the internal CB substructures that are observed by electron microscopy (Pena et al., 2001; Raska et al., 1991) and super-resolution fluorescence microscopy (Fig. 2). In addition, CB dependencies on temperature and coilin concentration were not fully consistent with predictions of the LLPS model (Fig. 1), suggesting that CBs cannot be fully described by a simple one-component LLPS model, where coilin is the only factor that drive CB formation. The ability of coilin to form higher complexes was also suggested as an explanation of slow diffusion of coilin within CB isolated from Xenopus oocyte (Handwerger et al., 2003). In contrast to CBs in frog oocytes, we observed rather fast diffusion of coilin within human CB, and the diffusion barrier was identified only at CB borders (Figs. 2 and 3). It was also shown that protein density in amphibian CBs is only ∼30% higher than in the nucleoplasm, which suggests that the diffusion barrier is not caused by high CB density (Handwerger et al., 2005). Coilin is thus specifically tethered in CB, likely due to multiple interactions that are occurring in CBs and that are modulated by other factors like snRNPs.

We further provide evidence that RNA is an essential component of CBs that promotes coilin oligomerization by showing that microinjection of RNase A disrupts CBs and that RNase treatment disrupts coilin–coilin interactions (Fig. 4). Coilin directly interacts with numerous short noncoding RNAs, namely snRNAs essential for splicing and snoRNAs important for rRNA biogenesis (Machyna et al., 2014), but how these RNAs promote coilin oligomerization and CB formation has been unclear. Our findings are consistent with the notion that snRNAs and snoRNAs, possibly including the increased local concentration of snRNAs at the site of their transcription, can induce local coilin oligomerization and CB assembly (Makarov et al., 2013). This model is consistent with the preferential localization of CBs near active snRNA genes (Frey and Matera, 1995; Jacobs et al., 1999; Smith et al., 1995; Wang et al., 2016) and accumulation of pre-snRNAs in CBs (Frey and Matera, 2001; Smith and Lawrence, 2000; Suzuki et al., 2010).

Finally, we identified that the complex of oligomeric–coilin and snRNPs is crucial for CBs formation. A role for the C-terminal domain in CB assembly has been suggested, but the molecular reason was unknown (Basello et al., 2022; Shpargel et al., 2003). NMR analysis of the coilin C terminus identified the presence of a Tudor domain. Coilin Tudor domain, however, lacks the critical aromatic aa that recognize symmetrically dimethylated arginines found in Sm proteins, and thus the molecular basis of coilin–snRNP interaction is unclear (Shanbhag et al., 2010). We identified several aa substitutions in the Tudor C-terminal domain that prevent interaction with snRNPs and concomitantly CB formation. Similarly, the mutant Δβ1β2 lacking the conserved loop I lost ability to interact with snRNPs (Figs. 5 and 6). Taken together, our results suggest the cooperative action for the Tudor domain and conserved loops in snRNP binding. Our results are fully consistent with observed CB induction in cells where the concentration of various snRNP components was artificially increased (Kaiser et al., 2008; Novotný et al., 2015; Roithová et al., 2018; Sleeman et al., 2001). Similarly, CB disintegrates when snRNPs are disrupted or their formation is inhibited, which further stresses the essential role of snRNPs in CB formation (Arias Escayola et al., 2025; Cihlarova et al., 2022; Girard et al., 2006; Lemm et al., 2006; Roithová et al., 2018; Takata et al., 2012). It is currently unclear how coilin_[N]–snRNP complexes promote CB condensation. The RG repeats might play a role in this process because both coilin and three of seven snRNP-specific Sm proteins (SmB/SNRPB, SmD1/SNRPD1, and SmD3/SNPRD3) contain RG-rich domains, which are important for snRNP targeting to CBs (Roithová et al., 2018). Alternatively, snRNP binding might induce structural changes in coilin that exposes coilin domains, promoting condensation or interaction with other factors that drive the CB assembly, like Nopp140 (Courchaine et al., 2022).

Based on the observed coilin dynamics and its interaction with RNA and snRNP, we proposed a mathematical model that describes coilin exchange between CBs and the nucleoplasm. As far we know, this is the first model that links biochemical reactions with dynamics of components in non-membrane–bound organelles and postulates how basic biochemical processes drive formation of non-membrane structures. Despite we were not able to comprehend all coilin interactions, the model describes some of the observed characteristics of CBs. It was previously suggested that the assembly rate of U4/U6•U5 tri-snRNPs is 10-fold faster in CBs than in the nucleoplasm (Klingauf et al., 2006; Novotný et al., 2011). Our prediction based on coilin dynamics is consistent with these observations and predicts ∼8 faster assembly in the CB than in the nucleoplasm. Our estimate includes all putative snRNP maturation reactions (U4/U6 di-snRNP and U4/U6•U5 tri-snRNP assembly, U2 snRNP maturation) and thus might reflect an average value for different reactions. The model also predicts that a drop of incomplete snRNP concentration in the nucleus results in disintegration of CB, which has been previously observed (Ferreira et al., 1994; Haaf and Ward, 1996; Lemm et al., 2006). Finally, it has been shown that coilin self-interacts, but the number of coilin molecules per oligomer has not been determined (Hebert and Matera, 2000). Our FCS measurements indicate that the number of coilin proteins in an oligomer is rather low (Fig. 7), which is consistent with our mathematical model that provided the best fitting when number of monomers in oligomers were set to 3 or 4.

We propose that the concentration of coilin_[N]–snRNP complexes primarily depends on concentration of immature snRNPs, which is consistent with the fact that CB number and size are insensitive to increased coilin concentration ([Sleeman et al., 2001] and [Fig. 1]). The nuclear concentration of immature snRNPs is likely to vary dramatically because their presence depends on their biogenesis as well as the transcription and splicing activity of a cell. Incomplete snRNPs are produced also during splicing when the postsplicing complex disassembles into individual postsplicing snRNPs that resemble immature particles undergoing de novo biogenesis (Klimešová et al., 2021). Thus, all snRNPs except U1 snRNP need to be recycled, and snRNP recycling takes place preferentially in CBs (Stanek et al., 2008). The concentration of coilin_[N]–snRNP complexes and appearance of CBs thus reflects the physiological state of the cell. Coilin can act as a “sensor” that monitors free snRNAs and incomplete snRNPs and merges inputs from different checkpoints of snRNP biogenesis. Consistently, downregulation of coilin is lethal for developing zebrafish embryo and reduces mouse fertility (Strzelecka et al., 2010; Tucker et al., 2001). In addition, dissociation of CBs from snRNA genes reduces snRNA expression (Wang et al., 2016). Therefore, CB formation appears to be a multifunctional cellular response that balances and buffers snRNP metabolism, including coordination of snRNA gene expression (Wang et al., 2016), enhancing snRNP assembly and recycling (Klingauf et al., 2006; Novotný et al., 2011; Strzelecka et al., 2010), and the sequestration of incomplete and defective snRNPs from the nucleoplasm (Novotný et al., 2015; Roithová et al., 2018).

HeLa, HeLa^coilin-EGFP, and HeLa^coilinKO cells were cultured in high-glucose (4.5  g/liter) DMEM (Sigma-Aldrich) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco). Parental HeLa and HeLa^coilinKO cells originated from ATCC HeLa-CCL-2 strain. They were further characterized by Hebert et al. (2002) as HeLa KN. HeLa^coilin-EGFP originated from HeLa Kyoto (Cellosaurus database ID: CVCL_1922) and characterized by Machyna et al. (2014); Poser et al. (2008).

EGFP-coilin^K496E and EGFP-coilin^WT were retrieved from our plasmids stock and described by Basello et al. (2022). All coilin mutants used in this manuscript are derived from EGFP-coilin^WT or EGFP-coilin^K496E and were obtained by site-directed mutagenesis (primers list is in Table S4). All constructs were verified by DNA sequencing. All plasmids are based on EGFP-C3 (Clontech). Plasmids were transiently transfected into the cells with Lipofectamine LTX Transfection Reagent (Thermo Fisher Scientific) for microscopy analysis or Lipofectamine 3000 Transfection Reagent (Thermo Fisher Scientific), for biochemistry analysis. Transfection was carried out according to the manufacturer’s instructions, and cells were analyzed 24 h after.

Cells grown on coverslips were wash in PBS and fixed with 4% paraformaldehyde/PIPES for 10 min at 37°C, permeabilized with 0.5% Triton X-100 (Sigma-Aldrich) for 5 min at RT, and blocked in 5% Goat serum (Jackson ImmunoResearch). The following primary antibodies were used for immunofluorescence staining: mouse monoclonal antibody against 2,2,7-trimethylguanosine (clone K121, catalog # sc-32724; Santa Cruz Biotechnology), mouse monoclonal antibody against SMN (clone 2B1, catalog # S2944; Sigma-Aldrich/Merck), rabbit polyclonal against coilin (H300, catalog # sc-32860; Santa Cruz Biotechnology), and human autoantibodies against fibrillarin (Cvacková et al., 2008). All antibodies were diluted in PBS, and cells were incubated for 1 h at RT. After PBS wash, the coverslips were stained with secondary goat antibodies anti-rabbit Alexa Fluor 555 (catalog # A21429), anti-mouse Alexa Fluor 647 (catalog # A21236) (Life Technologies), and anti-human Alexa Fluor 594 (catalog # A11014; Thermo Fisher Scientific) diluted in PBS for 1 h at RT. Coverslip were then washed in PBS and rinsed in distilled water. After being dried, they were mounted in Fluoromount-G (Southern Biotech). Images were acquired using inverted microscope DMi8 with confocal head Leica TCS SP8 (Leica Microsystems) equipped with a 63×/1.40 NA oil immersion objective and acquisition software LAS X (Leica Microsystems). Stacks of 20–35 z sections with 150-nm z steps and 42-nm pixel size were taken per sample, and maximum intensity projections are presented. If not specified otherwise, all microscopy images were taken at RT.

HeLa^coilinKO cells were grown and transfected with EGFP-coilin^WT on coverslips. 24 h after transfection, samples were incubated at 30°C, 37°C, or 42°C for 30 min and subsequentially washed in PBS and fixed with 4% paraformaldehyde/PIPES for 10 min at the same temperatures. The coverslips were mounted in Fluoromount-G and scanned using automated acquisition driven by the Acquisition ScanˆR program using ScanˆR system (Olympus) equipped with an oil-immersion objective (60×/1.35 NA). Several hundreds of cells were collected per sample. Cellular compartments were automatically identified based on fluorescence intensity combined with compartment edge detection. Cell nuclei were visualized using DAPI staining, and coilin and CBs were identified based on EGFP detection. Total intensities, areas, and counts for each cellular object were obtained using the Analysis ScanˆR software (Novotný et al., 2015). Analysis was performed in triplicate. Final graphs were made in Excel.

For both STED and 3D-SIM performed on fixed samples, immunostaining was done similarly as described above (“Immunofluorescence and image acquisition” section). Main differences were secondary antibodies for 3D-SIM, Alexa Fluor 488 goat anti-rabbit (catalog # A11008; Thermo Fisher Scientific), and mounting media, and 90% glycerol-based mounting media supplemented with 1,4-diazabicyclo[2.2.2]-octane (Sigma-Aldrich) for both techniques. 1.5 high-precision coverslips were utilized for all samples (Menzel-Glaser).

For STED, acquisition was performed on inverted microscope DMi8 with laser scanning confocal head Leica TCS SP8 equipped with pulsed white light laser (excitation) and STED 3X module (660 nm CW depletion laser). For acquisition, highly sensitive HyD detectors with time-resolved gating function were used in photon counting mode (gate start: 0.55 ns, gate end: 5 ns). z-stacks of 15–30 planes with 60 nm optical sections and 15 nm pixel size were acquired. 100×/1.40 NA oil immersion objective specifically designed for STED and acquisition software LAS X was used. Deconvolution was carried over utilizing Huygens Deconvolution Software works (Scientific Volume Imaging).

For 3D-SIM, acquisition was performed on the Delta Vision OMX (Applied Precision) imaging system equipped with 3D-SIM module, pco.edge 5.5 sCMOS camera and a 60×/1.42 NA objective. z-stacks of at least 15 planes with 0.125 μm optical sections were acquired. For the image reconstruction, we used the SoftWorx (Applied Precision) software implemented with the 3D SI Reconstruction function.

For live-cell fast-acquisition 3D-SIM was performed using the same microscopy system and using the same objective. HeLa^coilin-EGFP cells were cultured on 1.5 high precision glass-bottom Petri dishes (MatTek Corporation). DMEM was replaced with FluoroBrite DMEM (Gibco) supplemented with 10% FBS prior to image acquisition. A time-lapse is composed of z-stacks of 12 planes with 0.125-μm optical sections was acquired at 37°C and 5% CO₂. Special care was taken to minimize the exposure time and maximize acquisition speed. Image reconstruction was performed as described above.

HeLa^coilin-EGFP cells were grown on glass-bottom Petri dishes for 24 h. DMEM was replaced with FluoroBrite DMEM supplemented with 10% FBS prior to image acquisition. RNase solution was prepared by mixing 1 μl of RNase A (Invitrogen) and 1 μl of 10x concentrated RNase buffer in 8 μl of distilled water. Optionally, the injection solution contained 70-kD dextran-TRITC to trace the injected cells. Similarly, DNase solution was prepared mixing 1 μl of TurboDnase and 1 μl of 10x concentrated TurboDNase buffer in 8 μl of distilled water (Invitrogen). Both solutions were microinjected using InjectMan coupled with FemtoJet (Eppendorf) (Roithová and Staněk, 2019). Microinjection equipment was mounted on inverted microscope DMi8 with confocal head Leica TCS SP8 equipped with a 20×/0.75 NA immersion objective with correction ring and acquisition software LAS X. The microscope was set for time-lapse acquisition (2.80 s/stack), and z-stacks of six planes with 1.0-μm optical sections and 114-nm pixel size were used. Both green channel and transmitted light were acquired using highly sensitive HyD detector or PMT, respectively. Cells were kept at 37°C with 5% CO₂ and microinjected during the recording. Alternatively, microinjected cells were incubated for 2–3 min at 37°C with 5% CO₂, fixed with 4% paraformaldehyde, and immunostained for fibrillarin and SMN.

FRAP and time-lapse images shown in Fig. 2 were acquired using Delta Vision OMX imaging system equipped with Photometrics CoolSNAP HQ CCD camera and a 60×/1.42 NA objective, utilizing SoftWorx software. Cells were kept at 37°C and 5% CO₂ for the whole imaging time. Acquisition speed was set on 1 frame/s. Bleaching was performed in a single pulse in a pre-defined ROI of 3 μm in diameter. ROI intensities were exported in Excel, where final FRAP curves were generated. Due to a consistent lower signal (then rationally expected) in the first time point after bleaching, a control experiment was performed on a paraformaldehyde-fixed sample that confirmed an apparent recovery from first to second image after bleach. Therefore, the first point after bleach is not shown in Fig. 2 C, and second point is considered as ground level for recovery. Subsequent FRAP experiment (Figs. 7, D and E; and S1, E and F) were performed on inverted microscope DMi8 with confocal head Leica TCS SP8 equipped with a 63×/1.3 NA glycerol immersion objective corrected for use at 37°C and acquisition software LAS X. Cells were kept at 37°C and 5% CO₂ for the whole imaging time. 488-nm solid-state laser was used for both acquisition (0.2% laser intensity) and bleaching (25% laser intensity, single pulse, zoom-in mode, 3 μm in diameter). The following parameters were used for time-lapse acquisition: single focal plane, 62-nm pixel size, 2 lane accumulation, and 152-ms frame interval. For data in Figs. 7, D and E; and S2, four adaptive acquisition was implemented, meaning that the same parameters were used, but frame interval was modified as such: 152-ms interval for the first 100 frames, 300-ms for 75 frames, and 600-ms for the last 80 frames.

The measured FRAP curves were corrected for photobleaching, normalized, and analyzed by means of compartmental analysis (Novotný et al., 2011). CBs and nucleoplasmic bleached areas were both modeled as a spherical volume V_in = 14 fl (corresponding to 1.5-µm radius of the bleached spot) surrounded by an isotropic homogeneous nucleoplasm of a volume V_out = 670 fl (Klingauf et al., 2006). Starting with the simplest model possible, free diffusion of EGFP and EGFP-coilin^1–136/R8A, respectively, a compartmental system was constructed with its components distributed between V_in and V_out and describing the diffusion of the compartments in the nucleoplasm. A transfer rate of the species i across the hypothetical boundary was described by a time-invariant transfer coefficient s_i. The corresponding evaluated values of the parameters of the fit were fixed and used further in the more complicated models, where there was a possibility of using the same parameter (e.g., value of s_f evaluated from EGFP-coilin^1–136/R8A model was fixed and used for R8A+K496E double mutant, R8A mutant, etc., where it was not fitted anymore). The models, or experimental systems, included coilin mutants (R8A+K496E, R8A, and K496E), their compartments diffusing and interacting in the nucleoplasm, and subsequently WT coilin in the nucleoplasm and WT coilin in CBs. Biochemical interactions taking place in these more complicated models were characterized by corresponding reaction rates k_j, j = 1, 2, 3 (see Fig. 6), and again these parameter values were fixed and used in more complicated models where possible, starting with the simpler models. Each compartmental system was described by a set of ordinary differential equations written in terms of component concentrations (see below “Mathematical description of the model” section). Initial conditions reflected the situation in FRAP experiments—due to the incomplete and variable depletion degree, concentrations immediately after the bleaching pulse at t = 0 had to be fitted and were kept specific (local) for each experiment. On the other hand, transfer rates and reaction rates were kept common for all experiments. Each experimental set contained 15 FRAP curves (mutants in nucleoplasm and WT in nucleoplasm), with exception of WT in CBs (26 FRAP curves, also used for prediction of the local coilin concentrations).

For estimation of all parameters, each modeled compartmental system was fitted to normalized FRAP curves by a method called LARES: an artificial chemical process approach for optimization (Irizarry, 2004) using MATLAB (The MathWorks). The LARES approach was selected for optimization because the algorithm was shown to be very fast and robust when tested with different classes of problems, including different degrees of multi-modality, flatness, and discontinuities (Irizarry, 2004). Goodness of fit was evaluated by a value of reduced χ2 and distribution of residuals (Tables S2 and S3). To increase overdetermination of the model and the accuracy of recovered parameters, we used a global fitting (Beechem et al., 1983; Eisenfeld and Ford, 1979; Knutson et al., 1983) of multiple FRAP curves of the same experimental system (15 or 26 curves, respectively). During this analysis, the transfer rates k_i and rate constants k_j were common for all analyzed curves, while the concentrations were fitted locally for individual curves. To avoid being trapped in a local minimum of the multidimensional χ2 surface, the global fitting was performed with numerous sets of the initial parameter guesses, and also high number of iterations (up to 100,000) were used when optimizing with the LARES method to find parameters describing each of the experimental systems. Moreover, nonlinear least squares (lsqcurvefit function in MATLAB) was used to confirm the determined solution by slightly modifying the calculated parameters and eventually looking for even more precise solution.

Dynamics of the proposed model shown in Fig. 6 can be described by a system of ordinary differential equations written in terms of individual compartment concentrations inside the bleached volume in nucleoplasm or CB, respectively, (indexed in) and in the surrounding nucleoplasm (indexed out). Terms V_in and V_out stand for the bleached volume, and S represents surface of this volume.

The differential equations are listed according to the order in which the global parameter values have been evaluated and fixed (transfer rates s_i and t_i, respectively, and reaction rates k_i), starting with simple experimental systems (free diffusion) and moving to the more complicated ones (several compartments, including interactions).

HeLa^coilinko cells were grown on glass-bottom Petri dishes and transfected 24 h after seeding. 24 h after transfection, DMEM was replaced with FluoroBrite DMEM supplemented with 10% FBS prior to microscopy acquisition.

spFCS measurements were performed using an Abberior QUAD Scanning microscope (Abberior Instruments GmbH) equipped with Ti-E Nikon body and water objective HC Plan-Apochromat 63× NA 1.2. Cells were kept at 37°C and 5% CO₂. Samples were illuminated by pulsed 488-nm laser using pinhole size 1 Airy unit and detector range 505–600 nm. The dimension of the observed volume was 1.2 fL. Data acquisition was performed using simultaneously the Abberior software for operating the microscope and Symphotime64 software (PicoQuant) to online visualize FCS curves and save TTTR data. Illumination spots in the nucleoplasm were selected based on the reference images of the cells. The instrument was aligned, and the detection volume was calibrated using 20 nM Abberior STAR 488 carboxylic acid (Abberior GmbH) calibration dye. TTTR offline data processing and fitting was performed in a custom made “TTTR Data Analysis” software. The analysis pipeline starts with splitting each recording into 10 equivalent parts, calculation of autocorrelation curve for each part, estimation of standard deviation for each time point of the correlation curve, and weighted nonlinear least-square fitting by a standard one- or two-component 3D diffusion in a Gaussian shape detection volume model with triplet state fraction (Schwille et al., 1999; Widengren et al., 1995).

For 2D-pCF, cells were imaged on Leica sp8 equipped with a 63×/1.3 NA glycerol immersion objective corrected for use at 37°C and acquisition software LAS X. Cells were kept at 37°C and 5% CO₂ for the whole imaging time. 488-nm solid-state laser was used for acquisition. 2,048 time points were acquired, and the analysis was carried out using SimFCS 4 software (G-SOFT INC) according to Malacrida et al. (2020). For Fig. 3 A, a solution of 20-nm fluorescence beads (catalog # T-7280; Molecular Probes) was mix with far bigger nonfluorescent agarose beads (catalog # sc-2002; Santa Cruz Biotechnology) in PBS. 1 ml of solution was loaded into a glass-bottom Petri dish and imaged on DeltaVision microscope system (Applied Precision Ltd.) coupled to the Olympus IX70 microscope equipped with a 60×/1.42 NA oil immersion objective, a CoolSNAP HQ2 camera (Photometrics; Princeton Instruments), and the acquisition software SoftWoRx (Applied Precision Ltd.). In Fig. 3 B, a mask excluding the region outside the cell nucleus that does not contain EGFP signal was drawn to exclude near-zero values produced by noise. In fact, those small random values will give rise to large anisotropy values and random angles as reported by Hedde et al. (2019).

Cells cultivated in 10-cm Petri dishes were washed twice with ice-cold PBS, scraped from the dish, and centrifuged at 1,000 g for 10 min at 4°C. Harvested cells were resuspended in NET-2 buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.05% Nonidet P-40) supplemented with protease inhibitor cocktail (EMD) and RNase inhibitor (RNasin plus, Promega) and then pulse sonicated on ice (30 pulses; 0.5 s for each pulse at 40% of maximum energy). The cell lysate was centrifuged at 14,000 g for 10 min at 4°C. A sample of the supernatant was taken (for input), and the remaining immunoprecipitated overnight at 4°C with 7.5 µg of goat anti-GFP antibodies ([Machyna et al., 2014]; a gift from P. Tomancak, Max-Planck Institute for Molecular Cell Biology and Genetics, Dresden, Germany) prebound to 30 μl of protein G agarose PLUS beads (catalog # sc-2002; Santa Cruz Biotechnology). After five washings with NET-2 buffer, both the immunoprecipitated proteins as well as the input were resuspended in 30 μl of 2× sample buffer. Subsequently, proteins were separated by 10% SDS-PAGE and transferred to a nitrocellulose membrane (Protran). Membranes were blocked with 5% nonfat milk (wt/vol) in 0.05% Tween-20 in PBS (PBST) and incubated with the appropriate primary antibodies: coilin (5P10, [Almeida et al., 1998]; kindly provided by M. Carmo-Fonseca, Institute of Molecular Medicine, Lisboa, Portugal), Y12 ([Lerner et al., 1981]; produced from hybridoma cell line, a gift from K. Neugebauer, Yale University, New Heaven, CT, USA), SART3 (rabbit polyclonal, catalog # ab176822; Abcam), PRPF6 (mouse monoclonal, catalog # sc-166889; Santa Cruz Biotechnologies), and U2B'' (mouse monoclonal, catalog # 57036; Progen) and mouse monoclonal anti-GFP diluted in 1% nonfat milk in PBST, followed by washing steps in PBST and incubation with peroxidase-conjugated anti-mouse IgG (catalog # 115-035-003; Jackson ImmunoResearch Laboratories) and peroxidase-conjugated anti-rabbit IgG (catalog # 111-035-003; Jackson ImmunoResearch Laboratories) for western blotting. Enzymatic activity was detected using the SuperSignal West Pico/Femto Chemiluminescent Substrate (Thermo Fisher Scientific). When indicated, samples were incubated with 1 μl RNase A (10 mg/ml, Thermo Fisher Scientific) or BSA (supplemented with RNasin plus) for 10 min at 37°C prior to immunoprecipitation.

CLIP was performed according to Krchnáková et al. (2019). Briefly, cells grown on the 14 cm in diameter Petri dish were transfected for EGFP-coilin (WT, K496E, and R8A). Cells were washed with ice-cold PBS and irradiated with 150 mJ/cm² UV light (254 nm). After UV cross-linking, cells were lysed with 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, 1% IGEPAL, 0.1% SDS, 0.5% sodium deoxycholate containing a protease inhibitor cocktail, and EDTA, sonicated, and RNA was partly digested with RNase I. Immunoprecipitation was done with Dynabeads protein G coupled with goat anti-GFP antibody. After radioactive labeling of 5′ RNA ends with T4 polynucleotide kinase, RNA–protein complexes were separated on SDS-PAGE, transferred onto nitrocellulose membrane, and visualized by autoradiography. Subsequently, membranes were blocked with 5% nonfat milk (wt/vol) in PBST and incubated with anti-coilin antibody (rabbit H-300; SantaCruz) diluted in 1% nonfat milk in PBST, followed by washing steps in PBST and incubation with secondary antibodies conjugated with horseradish peroxidase. Enzymatic activity was detected using the SuperSignal West Pico/Femto Chemiluminescent Substrate.

Fig. S1 provides the additional data about coilin^R8A mutant and the coilin^1–136 fragment. Figs. S2, S3, and S4 show the FRAP data of various coilin mutants, including the fitting of individual curves using the global fitting approach. Table S1 shows the parameters used for fitting the FRAP curves by our mathematical model. Table S2 shows the results of global fitting of various coilin mutants FRAP curves in the nucleoplasm. Table S3 shows the results of global fitting of WT coilin FRAP curves in the CBs. Table S4 shows a list of primers used for site-directed mutagenesis. Video 1 shows the movement of coilin-EGFP inside CB by live 3D-SIM. Video 2 shows the disintegration of CBs after intranuclear microinjection of RNase A. Video 3 shows that the intranuclear microinjection of DNase has no visible effect on CBs.

Data are available in the article itself and its supplementary materials.

The authors would like to thank Enrico Gratton, Aleš Benda, and Radek Macháň for their assistance with FCS and spFCS analysis and Zuzana Cvačková for help with sample preparation.

The work was supported by MEYS (LTAUSA18103 to D. Staněk), CSF (24-11157S to D. Staněk), and the National Institutes of Health (NIH) (NINDS-R01NS128358-01 to K.M. Neugebauer). The microscopy images were acquired at the Light Microscopy Core Facility, IMG in Prague, and FCS measurements at IMCF, BIOCEV in Vestec, supported by MEYS (LM2015062, CZ.02.1.01/0.0/0.0/16_013/0001775) and OPPK (CZ.2.16/3.1.00/21547). N. Radivojević was supported by the grant from Charles University grant #264123. This work is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Author contributions: D.A. Basello: conceptualization, data curation, formal analysis, investigation, methodology, validation, visualization, and writing—original draft, review, and editing. M. Blažíková: data curation, formal analysis, software, and writing—review and editing. A. Roithová: investigation, methodology, and writing—review and editing. M. Hálová: investigation. N. Radivojević: investigation and visualization. K.M. Neugebauer: conceptualization, funding acquisition, resources, and writing—review and editing. D. Staněk: conceptualization, funding acquisition, project administration, resources, supervision, validation, and writing—original draft, review, and editing.