Phosphorylation, disorder, and phase separation govern the behavior of Frequency in the fungal circadian clock

Analysis of an AlphaFold 2.0 (Jumper et al., 2021) model of full-length FRQ indicates that FRQ lacks a defined arrangement of large folded domains; instead, most of the molecule is unstructured (Figure 1B), in agreement with previous studies predicting that FRQ mostly comprises intrinsically disordered regions (Hurley et al., 2013). However, FRQ is also predicted to contain a small, clustered core, with several secondary structural elements that collapse together to form a compact nucleus surrounded by long flexible disordered regions (Figure 1B). The AlphaFold pLDDT scores, which represent the confidence in the structural prediction (Jumper et al., 2021), are very low across the model, except within the core (Figure 1B, Figure 1—figure supplement 1), which contains the CK1-binding region. The FRH-binding region, FFD, falls within a more flexible region with a lower pLDDT score than that of CK1. Helices compose the majority of secondary structural elements predicted by AlphaFold (Figure 1B). The assignments of the helical regions were well supported by the Agadir program that predicts helical propensity based on the helix/coil transition theory (Muñoz and Serrano, 1994). Four regions distributed throughout the length of FRQ (residues 194–212, 312–329, 709–719, and 801–817) have at least 10% helical propensity, and these helical regions mostly agree with those of the AlphaFold model (Figure 1B). Together, these analyses primarily indicate that FRQ assumes a broad conformational ensemble containing an ordered nucleus that interacts extensively with partially ordered and flexible regions. However, much of the structural details of the model, including the positioning and interactions between secondary structural elements, are predicted with low confidence, thereby underscoring the need for a multi-pronged biophysical study of the protein.

The size and intrinsic disorder of FRQ have made it challenging to recombinantly express and purify, thereby limiting investigations of its biophysical properties. Previous purifications have characterized FRQ as an IDP, but suffered from low yields and aggregated species (Hurley et al., 2013; Lauinger et al., 2014). We also encountered extensive protein degradation and aggregation with conventional purification methods and affinity-tagging strategies; however, methods proven effective for IDPs (Pedersen et al., 2020) yielded comparatively well-behaved protein. Briefly, we carried out nickel affinity chromatography and size-exclusion chromatography (SEC) under denaturing conditions to maximize yield and reduce aggregation of full-length L-FRQ (residues 1–989; Figure 2A; Seal et al., 2021). Solubilization with high concentrations of guanidinium-HCl prevented proteolysis, aggregation, and limited co-purification of impurities that were otherwise difficult to remove. Once FRQ was isolated, the protein remained soluble, monodisperse, and could be concentrated after the denaturant was removed by dialysis (Figure 2A and B). To our knowledge, FRQ is one of the largest native IDPs produced in amounts amenable to structural and biophysical characterization.

To generate phosphorylated FRQ, the protein was co-expressed with (untagged) CK1, which forms a stable complex with FRQ and acts as its primary kinase (Querfurth et al., 2011). The same denaturing purification procedure was followed as for FRQ expressed without CK1. Phosphorylation was confirmed visually by altered migration on SDS-PAGE (Figure 2A) and mass spectrometry (Figure 1—figure supplement 2). The latter revealed a network of >80 phosphorylation sites (Figure 1—figure supplement 2), most of which agree with those found on natively expressed FRQ (Baker et al., 2009; Tang et al., 2009). Hereafter, we refer to this preparation as phosphorylated FRQ (p-FRQ), whereas non-phosphorylated FRQ (np-FRQ) refers to FRQ expressed without CK1. In general, we found that co-expression of FRQ with CK1 increased protein yield and stability, while limiting aggregation and degradation.

Size-exclusion chromatography coupled to multiangle light scattering (SEC-MALS) of p-FRQ gives a molecular weight measurement consistent with a primarily monomeric state (molecular weight [MW] of ~143 kDa versus a predicted MW of ~110 kDa; Figure 2B). In contrast, the np-FRQ measured by MALS indicates a dimer and its measured mass distribution increases upon concentration (Figure 2B) Hence, purified np-FRQ dimerizes, whereas phosphorylation destabilizes this association. SEC-coupled small-angle X-ray scattering (SAXS) analyses revealed that both p-FRQ and np-FRQ shared characteristics of highly flexible proteins, but that p-FRQ was more extended than np-FRQ (Table 1). The dimensionless Kratky plots of p-FRQ and np-FRQ, which report on compactness and conformational dispersity (Durand et al., 2010; Martin et al., 2021,Bateman et al., 2021; Sagar et al., 2020; Trewhella et al., 2017), revealed high degrees of flexibility for both proteins (Figure 2C). The peak values of the plot were shifted from those expected for compact structures but were also not indicative of fully disordered proteins (Durand et al., 2010; Rambo and Tainer, 2011; Sagar et al., 2020). Furthermore, p-FRQ appears more disordered than np-FRQ, a characteristic also reflected in their pairwise distance distributions (Figure 2—figure supplement 1). These data then indicate that FRQ generally adopts a relatively monodisperse conformational state that is partially compact but highly flexible and then increases expanse and flexibility upon phosphorylation. This general picture is consistent with the increased susceptibility to proteolysis that FRQ experiences with phosphorylation (Liu et al., 2019; Srivastava and Freed, 2019; Pelham et al., 2020; Querfurth et al., 2011).

The properties of FRQ that include its intrinsic disorder and propensity to form heterogeneous ensembles are not well suited to commonly applied structural characterization methods. We applied site-specific spin labeling (Figure 3A) and electron spin resonance (ESR) spectroscopy to monitor FRQ local structural properties. FRQ has four native cysteine residues that provide reactive sites for nitroxide spin labels (SLs) (Figure 3A and B; Berliner et al., 1982). The Cys residues distribute relatively evenly throughout the protein sequence and fall within areas of varying structural order that include the binding regions for the core clock proteins (FRH, CK1, WC1, and WC2; Tables 2 and 3, Figure 3B). Furthermore, we developed and applied enzymatic SortaseA and AEP1 peptide-coupling methods (Chandrasekaran et al., 2021; Nguyen et al., 2015) to spin-label FRQ at its N- and C- termini (Figure 3A, Figure 3—figure supplement 1).

Line-shaped analysis from continuous wave (cw)-ESR spectroscopy reports on local dynamics (Franck et al., 2015; Freed, 2005). For rapidly rotating nitroxide SLs, the characteristic spectrum consists of three narrow lines of equal width and intensity. As the rotational mobility decreases, the lines broaden, especially the high-field line (Bonucci et al., 2020; Emmanouilidis et al., 2021). cw-ESR reveals that FRQ has regions of varying order that agree with computational predictions and that some of these sites undergo phosphorylation-dependent conformational changes (Figure 3C). SLs attached to Cys490 and Cys658 (490SL and 658SL) are predicted to be within an α-helix and a β-strand, respectively. These sites correspondingly showed a broad low-field (left-most line) component and a small high-field component (right-most line) indicative of slower tumbling compared to sites predicted to be in flexible regions, such as 746SL (Figure 3C). The 490SL site had a broader profile compared to that of the 658SL site, although both are relatively well ordered. We observed a similar trend with δ, which is the peak-to-peak width of the central line (Hubbell et al., 2000). 490SL and 658SL had similar δ values that were higher than that of 746SL, indicating that they are less flexible than 746SL (Table 4). Furthermore, the spectra also highlighted structural differences between the phospho forms: p-FRQ and np-FRQ gave very different δ values at 490SL and the C-terminus, thereby demonstrating how phosphorylation can influence conformational properties site specifically. The C-terminus may have become more dynamic in p-FRQ compared to np-FRQ because the C-terminus contains a cluster of phosphosites (Figure 3C). In contrast, the N-terminus and especially 490SL, which locates to the CK1-interaction region of the FCD₂ domain, are more ordered in p-FRQ. Such differences in mobility between p-FRQ and np-FRQ in the FCD (490SL) may reflect changes in FRQ:CK1 interactions as FRQ becomes progressively phosphorylated across the circadian period (Liu et al., 2019).

Double electron–electron resonance (DEER) spectroscopy provides distance distribution between two nitroxide labels in the range of 15 Å to ~100 Å. We produced six cysteine pair variants that each contained only two of the four native FRQ Cys residues to measure pairwise distances and one single Cys variant to test for oligomerization. Owing to association and aggregation when concentrated, np-FRQ was not amenable to DEER measurements; however, such issues were not prohibitive with p-FRQ. Nevertheless, due to the challenges of spin-labeling appreciable concentrations of p-FRQ, the modulation depths of the DEER data, which reflect the number of spins in range to produce specific dipolar interactions, are small (~2–4%). These values compare to ~10–12% for a well-ordered, highly labeled pair of sites under these instrument conditions (Dunleavy et al., 2023). The single Cys spin-labeled variant 490SL produced log-linear time-domain traces indicative of a uniform background with little evidence of dimerization, consistent with the MALS data showing p-FRQ to be monomeric (Figure 3—figure supplement 2). A second single Cys variant, 746C, likewise produced little dipolar signal, although for this variant, solubility issues limited spin concentration (Figure 3—figure supplement 2). The double 370SL:746SL p-FRQ variant also gave no dipolar signals and hence no dimerization as reported by those two sites (Figure 3D). When at least one of the two SLs resides outside of the predicted core, the DEER experiments of the spin pairs generally revealed small, albeit significantly slow decaying time-domain traces that deviate from background (Figure 3D, Figure 3—figure supplement 2, Figure 4—figure supplement 1). Such traces are indicative of long inter-spin distances and conformational flexibility, which would be consistent with the large mobility and intrinsic disorder of these peripheral regions (Figure 3—figure supplement 2, Figure 4—figure supplement 1). Only the variant 490SL:658SL gave a fast-decaying time-domain signal, indicative of short inter-spin distances and conformational rigidity (Figure 3—figure supplement 2, Figure 4—figure supplement 1). Despite the overall uncertainty in the AlphaFold model (Figure 1B), it is notable that 490SL and 658SL both fall within the predicted ordered core.

The FFC was generated by mixing p-FRQ or np-FRQ with excess FRH and CK1, followed by SEC purification to give p-FFC or np-FFC, respectively. FRHΔN (residues 100–1106), which binds FRQ, was used to improve expression and enhance stability (Conrad et al., 2016; Hurley et al., 2013). The resulting symmetric SEC traces and SDS-PAGE gels revealed pure and homogeneous ternary complexes in each case (Figure 4A). As expected, CK1 phosphorylates FRQ in np-FFC (Figure 4A, Figure 4—figure supplement 2), and this activity does not require priming by other kinases (Marzoll et al., 2022). SEC-SAXS analysis of the FFC showed that both FRQ phospho-forms produced complexes that were more globular than FRQ alone, but still somewhat flexible, with the p-FFC being more flexible than the np-FFC (Figure 4B, Figure 2—figure supplement 1). MW estimates of the p-FFC and np-FFC from SAXS indicate a 1:1 stoichiometry of FRQ:FRH, with a likely single component of CK1 also contained within the complex (Table 1).

DEER measurements from selectively spin-labeled components within the complex were used to probe its overall architecture (Figure 5, Figure 5—figure supplements 1–4). An ADP-βSL probe (Muok et al., 2018) bound to both FRH and CK1 and revealed that the FRH and CK1 ATP-binding pockets are closely positioned within the FFC complex (Figure 5A, Figure 5—figure supplement 1). Nonetheless, the presence of broad and multiple peaks in the distance distributions indicated conformational heterogeneity. SLs placed within the FCD2 and FFD domains of FRQ (490C and 746C), which bind CK1 and FRH, respectively, produced a distance distribution similar to that of the labels that target the ATP-binding sites of the two enzymes (Figure 5A; Cheng et al., 2005). These independent observations support the conclusion that FRQ organizes FRH and CK1 in close proximity within the FFC. Notably, purified FRH and CK1 do not interact in the absence of FRQ (Figure 5—figure supplement 1).

SL separations within p-FRQ changed substantially when FRH and CK1 bound (Figure 5, Figure 3—figure supplement 2, Figure 4—figure supplement 1, Figure 5—figure supplements 2–4). In general, the time-domain decays were more rapid in the FFC compared to p-FRQ alone, and the resulting distance distributions had smaller mean distances and less breadth (Figure 5B–H). However, the changes were not uniform; for example, the 658SL-to-746SL distance distribution increased in the complex compared to free FRQ (Figure 5G). To further assess how the distributions of p-FRQ conformations changed when the p-FFC formed, the distance data was fit to a model of one near and one far Gaussian functions (Figure 5—figure supplement 4, Table 5). Shifts in population between the near and far components were most prominent for the 370–658, 490–658, and 490–746 variants, which contain sites involved in core clock interactions (Figure 5—figure supplement 4). The proportion of the closer component increased from ~30% to ~70% in the 370–658 variant (Figure 5—figure supplement 4) and from ~10% to ~70% in the 490–746 variant (Figure 5F) when bound to FRH and CK1. Similar trends were observed when we compared the percentage change of the average separation observed in FRQ alone versus when it is bound to FRH and CK1 (Figure 5H). The 490, 658, and 746 sites reside in regions important for WCC interactions (Guo et al., 2010) and FRH may organize such binding determinants. Notably, the 370SL:746SL spin pair showed little dipolar coupling in free p-FRQ but resolved into an observable mid-range separation in the presence of CK1 and FRH (Figure 5C), perhaps because position 746 is within the FRH-binding region. Thus, CK1 and FRH order p-FRQ to an appreciable degree and the two enzymes are adjacently positioned within the conformationally heterogeneous complex.

The unstructured regions and multivalent interactions of IDPs can lead to LLPS (Martin and Holehouse, 2020a; Martin et al., 2020b). Such de-mixing often manifests in the formation of biomolecular condensates and MLOs within cells (Posey et al., 2018). As has been previously noted (Pelham et al., 2020), FRQ may be capable of LLPS owing to its extensive disorder. We found that FRQ has a high propensity to undergo LLPS based on a set of computational tools that evaluate charge distribution and clustering, long-range aromatic and non-aromatic π:π contacts, and nucleic acid-binding propensities (Bolognesi et al., 2016; Vernon et al., 2018). The overall FRQ PScore, which grades LLPS propensity on these properties (Vernon et al., 2018), was 6.39, with three regions scoring above a value of 2, the threshold for favorable LLPS (Figure 6—figure supplement 1). The region with the highest PScore also accounted for the majority of LLPS propensity as determined by the catGRANULE prediction algorithm (Bolognesi et al., 2016). Furthermore, we found that these methods also predicted the functional analogs dPER and hPER1 to phase separate (Figure 6A, Figure 6—figure supplement 2). The FRQ functional analogs share physicochemical characteristics with FRQ, particularly an enrichment of hydrophilic residues, increases in Gly and Pro content, and a depletion of hydrophobic residues, even though their positional sequence conservation with FRQ is low (Figure 6—figure supplements 3 and 4; see also Jankowski et al., 2022; Pelham et al., 2021). Similar amino acid composition patterns among FRQ, dPER, and hPER1 include the most common residues in each protein and its UniRef50 cluster (sequences with at least 50% identity with the target protein) being serine, proline, and glycine (~30% of total sequence) accompanied by a high depletion in aromatic residues (only ~7.5%) (Figure 6—figure supplement 4). In addition, all of the sequences have a reduced proportion of hydrophobic residues such as A, I, L, and V and a slightly acidic isoelectric point. Thus, FRQ, dPER, and hPER1 all have low overall hydropathy (Figure 6—figure supplement 3). However, in addition, FRQ and its homologs are unusual in their high Arg and Asp content, which tend to separate in charge blocks throughout their sequences (as indicated by high kappa values, Figure 6—figure supplement 4; see also Jankowski et al., 2022). Such charge clustering correlates with LLPS (Somjee et al., 2020). Overall, this analysis suggests that like intrinsic disorder, LLPS driven by the physiochemical properties of the constituent residues may be a conserved feature in the negative arm of the circadian clock.

We tested whether both np-FRQ and p-FRQ can undergo LLPS in vitro. In each case, upon exchanging the purified protein into their respective phase separation buffers (which lack any crowding agents), we observed the appearance of a turbid solution that contained microscopic droplets (Figure 6B). These droplets scaled in size and number with increasing protein concentration, dissolved upon the addition of 1,6-hexanediol, and were shown to dock and fuse to one another, thereby confirming their liquid-like properties (Figure 6B, Figure 7—figure supplement 1). UV-Vis turbidity assays as a function of temperature revealed that FRQ undergoes a low critical solution temperature (LCST) phase transition that is concentration and phosphorylation status-dependent (Figure 6C, Figure 7—figure supplement 1, Tables 6 and 7). As expected, phase separation increased as temperature and protein concentration increased as indicated by growing amplitudes and earlier phase-separation initiation temperatures (Figure 6C, Figure 7—figure supplement 1, Tables 6 and 7). Lower salt concentrations favored the LLPS of p-FRQ, whereas higher salt disfavored LLPS, compared to higher salt concentrations favoring LLPS for np-FRQ (Figure 7—figure supplement 1). Although the NaCl concentration for np-FRQ LLPS conditions is non-physiological per se, the ionic strength conditions and effective solute concentrations in cells are comparatively high (Liu et al., 2017). Furthermore, the higher absorbance values of p-FRQ at lower temperatures relative to that of np-FRQ indicate complex phase behavior for the phosphorylated isoform that cannot be adequately explained by an LCST phase transition alone (Figure 7—figure supplement 1). At temperatures below the approximated phase boundary indicated by the temperature transitions (Figure 6C, Figure 7—figure supplement 1), some p-FRQ phase separation had already occurred in the phase-separation buffer. Thus, FRQ phosphorylation alters its behavior consistent with LLPS as evidenced by its non-zero, concentration-dependent absorbance at lower temperatures, most likely by increasing the relative negative charges, changing the charge patterns, and providing more sites for multivalent interactions within the molecule (Figure 6C, Figure 6—figure supplement 5, Figure 7—figure supplement 1; Somjee et al., 2020; Szabó et al., 2022).

Live-cell imaging of N. crassa expressing FRQ tagged with the fluorescent protein mNeonGreen at its C-terminus (FRQ^mNeonGreen) revealed heterogeneous patterning of FRQ in nuclei (Figure 6D and E). FRQ thus tagged and driven by its own promoter is expressed at physiologically normal levels, and strains bearing FRQ^mNeonGreen as the only source of FRQ are robustly rhythmic with a slightly longer than normal period length. Live-cell imaging in N. crassa offers atypical challenges because the mycelia grow as syncytia, with continuous rapid nuclei motion during the time of imaging. This constant movement of nuclei is compounded by the very low intranuclear abundance of FRQ and the small size of fungal nuclei, making not readily feasible visualization of intranuclear droplet fission/fusion cycles or intranuclear fluorescent photobleaching recovery experiments (FRAP) that could report on liquid-like properties. Nonetheless, bright and dynamic foci-like spots were observed well inside the nucleus and near the nuclear periphery, which is delineated by the cytoplasm-facing nucleoporin Son-1 tagged with mApple at its C-terminus (Figure 6D and E, Video 1). Such foci are characteristic of phase-separated IDPs (Bartholomai et al., 2022b; Caragliano et al., 2022; Gonzalez et al., 2021; Tatavosian et al., 2019) and share similar patterning to that seen for clock proteins in Drosophila (Meyer et al., 2006; Xiao et al., 2021), although the foci we observed are substantially more dynamic than those reported in Drosophila.

The PScore and catGRANULE LLPS predictors gave low overall scores for the FRQ partners, FRH and CK1 (Figure 6—figure supplement 2). This result is unsurprising because these proteins consist of predominantly folded domains. Nonetheless, FRH and CK1 may still act as clients that localize to the FRQ-formed LLPS-like droplets. We chose to investigate whether FRH and CK1 could be recruited into droplets with fluorescent microscopy. Structured proteins are not well tolerated in the temperature-dependent turbidity assays because of the large signal that they give upon domain unfolding. We found that fluorescently labeled FRH and CK1 were recruited into phase-separated FRQ droplets, but a control protein labeled with the same fluorophore, Escherichia coli CheY, was not (Figure 7A). The ability of FRQ to recruit its partner proteins into phase-separated droplets provides a means to sequester them and potentially control their availabilities and activities.

LLPS has been suggested to modulate the activities of enzymes by creating a microenvironment to alter substrate sequestration and product release (Shapiro et al., 2021). We tested the effect of FRQ LLPS on its phosphorylation by CK1, a key reaction for regulating the period of the clock (Liu et al., 2019). We compared the phosphorylation of FRQ by CK1 in a buffer that supports phase separation under different temperatures using the latter as a means to control the degree of LLPS without altering the solution composition. Both the UV-Vis turbidity assays and the differential interference contrast (DIC) micrographs of the protein at different temperatures establish that FRQ LLPS increases substantially with temperature (Figures 6C and 7C). Furthermore, LLPS of p-FRQ and np-FRQ both depend highly on buffer composition (Figure 7—figure supplement 1). CK1 autophosphorylation served as an internal control and its temperature dependence was consistent with previous studies that show increased catalytic activity at higher temperatures (Marzoll et al., 2022). As temperature increases drove more LLPS, the phosphorylation of FRQ decreased, even as CK1 autophosphorylation increased (Figures 7B and 8). Quantitatively, we observed a reduction in FRQ phosphorylation by twofold under LLPS conditions relative to conditions under which it does not phase separate (Figure 7B).

Through application of cw-ESR, we probed FRQ dynamics when it undergoes LLPS. We observed a striking rigidification of spin-labeled FRQ when LLPS was promoted by buffer conditions (Figure 7D). The reduction of CK1-mediated FRQ phosphorylation, which likely requires substantial dynamics given the distribution of phosphorylation sites throughout the FRQ sequence, may reflect this change in molecular dynamics. Thus, in addition to affecting the localization of FRQ and its partners, LLPS also impacts the enzymatic activity of the complex by perhaps modulating molecular encounters and motions essential for large-scale phosphorylation.

Full-length L-FRQ (UniProt ID:P19970↗), CK1 (CK1a; UniProt ID:V5IR38), and FRHΔN (residues 100–1106) (UniProt ID:Q1K502) were sub-cloned into a pET28a vector (Novagen) bearing an N-terminal His₆-tag and kanamycin resistance. Note that L-FRQ contained the S513A substitution to improve protein stability as noted in Querfurth et al., 2007. Cysteine to serine substitution mutants of FRQ were produced using standard mutagenesis, with the primers described in the Key Resources Table. The plasmids encoding the proteins were transformed into BL21 (DE3) E. coli cells, grown in LB media containing 50 μg/mL kanamycin to an optical density (OD₆₀₀) of ~0.6–0.8 at 37°C and protein expression was induced with 1 mM IPTG at 17°C or room temperature. Cells were harvested after 12–18 hr by centrifugation at 5000 rpm and 4°C. FRH and CK1 were purified by standard Ni-NTA chromatography followed by ion-exchange and size-exclusion chromatography. Briefly, cells were resuspended in lysis buffer (50 mM Tris pH 8.0, 500 mM NaCl, 10% glycerol, 1 mM MgCl₂, 5 mM imidazole) and sonicated on ice. The lysate was clarified by centrifugation at 20,000 rpm for 1 hr. The clarified lysate was applied to a gravity Ni-NTA column and washed with five column volumes (CVs) of wash buffer (50 mM Tris pH 8.0, 500 mM NaCl, 10% glycerol, 1 mM MgCl₂, 20 mM imidazole). The protein was eluted using the elution buffer (50 mM Tris pH 8.0, 500 mM NaCl, 10% glycerol, 1 mM MgCl₂, 500 mM imidazole). The protein was then diluted into ion-exchange (IEX) buffer A (50 mM HEPES pH 7.5, 10% glycerol), applied to an S cation exchange column (GE Healthcare) and eluted over a 20 CV gradient against IEX buffer B (50 mM HEPES pH 7.5, 1 M NaCl, 10% glycerol). Fractions containing the protein were pooled, concentrated, and run over an S200 26/60 gel-filtration column pre-equilibrated in gel-filtration buffer (GFB; 50 mM HEPES, 200 mM NaCl, 10% glycerol). L-FRQ was purified under denaturing conditions. Cells were resuspended in lysis buffer (50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 5 mM imidazole, 4 M guanidinium hydrochloride [GuHCl]) and sonicated on ice. The lysate was clarified by centrifugation at 20,000 rpm for 1 hr. The clarified lysate was applied to a gravity Ni-NTA column and washed with 3 CV of lysis buffer. The protein was eluted using the elution buffer (50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 500 mM imidazole, 4 M GuHCl). The protein was concentrated and applied to a S200 26/60 gel-filtration column pre-equilibrated in 50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 5 mM imidazole, and 4 M GuHCl. Fractions containing the protein were pooled and concentrated. The concentrated protein was flash-frozen and stored at –80°C. Prior to experiments, the purified L-FRQ was exchanged into a non-denaturing buffer (50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 200 mM arginine, 10% glycerol). For FFC preparations, GFB (50 mM HEPES pH 7.42, 200 mM NaCl, 10% glycerol) was used.

Protein samples between 50 and 75 μM were injected into a Superose-6 (GE Life Sciences), equilibrated with 50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 200 mM arginine, and 10% glycerol. The gel filtration column was coupled to a static 18-angle light scattering detector (DAWN HELEOS-II) and a refractive index detector (Optilab T-rEX) (Wyatt Technology, Goleta, CA). Data were collected every second at a flow rate of 0.5 mL/min and analysis was carried out using ASTRA VI, yielding the molar mass and mass distribution (polydispersity) of the sample. The monomeric fraction of BSA (Sigma, St. Louis, MO) was used to standardize the light-scattering detectors and optimize data quality control.

X-ray scattering experiments were performed at the Cornell High Energy Synchrotron Source (CHESS) G1 station. SEC–SAXS experiments were performed using a Superdex 200 5/150 GL or Superose-6 5/150 GL (3 mL) column operated by a GE AKTA Purifier at 4°C with the elution flowing directly into an in-vacuum X-ray sample cell. Samples were centrifuged at 14,000 × g for 10 min at 4°C before loading onto a column pre-equilibrated in a matched buffer (50 mM Na₂PO₄ pH 7.42, 300 mM NaCl, 200 mM arginine, 10% glycerol). Samples were eluted at flow rates of 0.05–0.1 mL/min. For each sample, 2 s exposures were collected throughout elution until the elution profile had returned to buffer baseline, and scattering profiles of the elution buffer were averaged to produce a background-subtracted SEC-SAXS dataset. The scattering data was processed in RAW (Hopkins et al., 2017) and regions with homogeneous R_G values were chosen for the Guinier and Kratky analyses.

Sort/AEP1-tagged FRQ was cloned into a pET28a (Novagen) vector as described in Chandrasekaran et al., 2021. The production of the SortaseA and AEP1 enzymes and their associated peptides (GGGGC for SortaseA and CNGL for AEP1) is also described in Chandrasekaran et al., 2021 and (Nguyen et al., 2015), respectively. To enzymatically label FRQ with either SortaseA or AEP1, 1:1 protein and SortaseA/AEP1 as well as 2× excess peptide were incubated overnight to allow the reaction to reach completion. The following day, the reaction was run on a Superpose 6 10/300 analytical SEC column to remove unreacted peptide and enzyme, and the protein-containing fractions were pooled and concentrated. To prepare the ADP-β-S-SL samples, ADP-β-S-SL was synthesized as previously reported in Muok et al., 2018. Then, 100 μL of buffer containing 100 μM of ADP-β-S-SL and 100 μM FRH or CK1 were incubated overnight and washed with GFB (50 mM HEPES pH 7.42, 200 mM NaCl, 10% glycerol) through 5× dilution followed by re-concentration to remove any unbound ADP-β-S-SL. Also, 30 μL of each sample were used for data collection. For MTSL labeling, the protein was incubated with 5× molar excess of MTSL overnight to allow the reaction to reach completion. The following day, the reaction was run on HiTrap Desalting column (GE Life Sciences) to remove any excess SL.

To prepare fluorophore-labeled FRH and CK1, the purified proteins were incubated with 2× molar excess Cy5-maleimide dye (GE Life Sciences) overnight to label any accessible cysteines on the target proteins. The excess dye was removed by running the reaction over a HiTrap desalting column equilibrated with GFB (50 mM HEPES pH 7.42, 200 mM NaCl, 10% glycerol).

The cw-ESR spectra were obtained on a Bruker E500 ESR spectrometer operating at X-band (~9.4 GHz). The ESR measurements were carried out at room temperature with a modulation amplitude of 0.5–2 G and modulation frequency of 100 kHz. All spectra were digitized to 1024 points with a sweep width of ~100 G and an average of 4–16 scans. The δ values were calculated using the display distance function on the Bruker Xepr 2.6b software.

For four-pulse DEER, the samples were exchanged into deuterated buffer containing 30% d8-glycerol, checked by cw-ESR, and plunge frozen into liquid N₂. The pulse ESR measurements were carried out at Q-band (~34 GHz) on a Bruker E580 spectrometer equipped with a 10 W solid-state amplifier (150 W equivalent TWTA) and an arbitrary waveform generator . DEER was carried out using four pulses (π/2-τ1-π-τ1-πpump-τ2-π-τ2-echo) with 16-step phase cycling at 60 K. The pump and probe pulses were separated by 56 MHz (~20 G). The DEER data was background subtracted with an exponential background and the distance reconstruction was carried out using denoising and the SVD method (Srivastava et al., 2017a; Srivastava and Freed, 2017b). Uncertainty in the distance distributions was calculated using the method described by Srivastava and Freed, 2019. Briefly, the uncertainty plotted shows the ‘range’ of singular values (SVs) over which the singular value decomposition (SVD) solution remains converged. For most of the data displayed, we only used the first few SVs and the solution remained converged for ±1 or 2 SVs near the optimum solution. For example, if the optimum solution was four SVs, then the range in which the solution remained converged is ~3–6 SVs. We plot three lines – lowest range of SVs, highest range of SVs, and optimum number of SVs – in the supplementary information figures; the optimum SV solution is shown in black and the region between the converged solutions with the highest and lowest number of SVs is shaded in red. Owing to the pointwise reconstruction of the distance distribution, the SVD method enables localized uncertainty at each distance value. Therefore, some points will have high uncertainty, whereas others low.

Cysteine-less (cys-less) FRQ was created in order to minimize disulfide cross-linking at the high protein concentrations needed for in vitro phase separation experiments. For this, all four of the native cysteine residues were substituted to serine using standard molecular biology techniques. For the phase separation experiments, this cys-less FRQ variant was purified and concentrated for dialysis into phase separation buffer. Cys-less FRQ was dialyzed overnight into 25 mM Tris pH 8.0, 150 mM NaCl, or 25 mM Tris pH 8.0, 500 mM NaCl for p-FRQ and np-FRQ, respectively, at 4°C. Phase separation was induced by dialyzing in either low or high salt and increasing temperature. The protein solution was observed as a clear solution before dialysis and then turbid after dialysis, indicating that LLPS had occurred. Microscopy (see ‘DIC and fluorescence microscopy’) confirmed the turbid solution is due to LLPS and not aggregation. Turbidity as measured by absorbance at 300–600 nm was then recorded with an Agilent Cary 3500 UV-Vis Multicell Peltier using a temperature ramp rate of 5°C/min increasing from 10°C to 80°C. A final protein concentration of 5–15 μM was used for all turbidity assays.

Phase separation was induced via overnight dialysis into 25 mM Tris pH 8.0, 150 mM NaCl, or 25 mM Tris (pH 8.0), 500 mM NaCl for p-FRQ and np-FRQ, respectively. The final protein concentration used for microscopy was 5–15 μM. Images were taken on the Zeiss Axio Imager Z1 Upright Trinocular Fluorescence Microscope at 100× magnification. To test the interaction of FRQ with FRH and CK1, the phase-separated FRQ was spiked with equimolar Cy5-labeled FRH or Cy5-labeled CK1 for final concentrations of 5 μM.

22 μL samples containing 10–15 μM LFRQ were incubated in the presence or absence of 2 μM CK1 and 2 μM FRH in 50 mM Na₂HPO₄ pH 7.5, 300 mM NaCl, 200 mM arginine, 10 mM MgCl₂. For the LLPS assay, FRQ was dialyzed into a buffer containing 50mM Na₂HPO₄ pH 7.5, 300mM NaCl and 10mM MgCl₂. Then, 2 μL of 2.5 mM cold ATP and 0.1 µM of [γ–32P] ATP (3000 Ci/mmol, PerkinElmer) was added to produce a total volume of ~25 μL. After ~1 hr of incubation, the reaction was quenched with 25 μL of 4× SDS with 10 mM EDTA pH 8.0 and then subjected to gel electrophoresis on a 4–20% gradient Tris-glycine gel. The gel was dried with a GelAir dryer (Bio-Rad) and placed in an imaging cassette for at least 16 hr and then imaged with a phosphorimager (GE, Typhoon FLA 7000). Band intensities were quantified using the FIJI software.

Phosphorylated FRQ was run on a 4–20% tris-glycine SDS-PAGE gel, and the corresponding band was excised and trypsin digested. The digests were reconstituted in 0.5% formic acid (FA) for nanoLC-ESI-MS/MS analysis. The analysis was carried out using an Orbitrap Fusion Tribrid (Thermo Fisher Scientific, San Jose, CA) mass spectrometer equipped with a nanospray Flex Ion Source and coupled with a Dionex UltiMate 3000 RSLCnano system (Thermo, Sunnyvale, CA). The peptide samples (5 μL) were injected into a PepMap C-18 RP nano-trapping column (5 µm, 100 µm i.d. × 20 mm) and then separated on a PepMap C-18 RP nano column (2 µm, 75 µm × 25 cm) at 35°C. The Orbitrap Fusion was operated in positive ion mode, with spray voltage set at 1.5 kV and source temperature at 275°C. External calibration for FT, IT, and quadrupole mass analyzers was performed. In data-dependent acquisition (DDA) analysis, the instrument was operated using FT mass analyzer in MS scan to select precursor ions followed by 3 s ‘Top Speed’ data-dependent CID ion trap MS/MS scans at 1.6 m/z quadrupole isolation for precursor peptides with multiple charged ions above a threshold ion count of 10,000 and normalized collision energy of 30%. MS survey scans at a resolving power of 120,000 (FWHM at m/z 200) for the mass range of m/z 375–1600. Dynamic exclusion parameters were set at 35 s of exclusion duration with ±10 ppm exclusion mass width. All data were acquired using Xcalibur 4.4 operation software (Thermo Fisher Scientific). The DDA raw files with MS and MS/MS were subjected to database searches using Proteome Discoverer (PD) 2.4 software (Thermo Fisher Scientific, Bremen, Germany) with the Sequest HT algorithm. The database search was conducted against a Neurospora crassa database and E. coli database by adding LFRQ sequence into the database. The peptide precursor tolerance was set to 10 ppm and fragment ion tolerance was set to 0.6 Da with 2 miss-cleavage allowed for either trypsin or chymotrypsin. Variable modification of methionine oxidation, deamidation of asparagines/glutamine, phosphorylation of serine/threonine/tyrosine, acetylation, M-loss and M-loss + acetylation on protein N-terminus, and fixed modification of cysteine carbamidomethylation were set for the database search. Only high-confidence peptides defined by Sequest HT with a 1% FDR by Percolator were considered for confident peptide identification. All the mass spectra of phosphorylated peptides were double-checked by manual inspection.

All canonical protein and UniRef50 sequences for FRQ, FRH, and CK1, and SWISS-Prot-obtained proteome sequences of N. crassa, Drosophila melanogaster, and Homo sapiens were downloaded from the UniProt database (Bateman et al., 2021) A combination of (a) in-house-written Python scripts to allow us to analyze large datasets, (b) localCIDER (Holehouse et al., 2017), (c) IUPred2A (Mészáros et al., 2018), (d) Agadir (Muñoz and Serrano, 1994), (e) AlphaFold 2.0 (Jumper et al., 2021), (f) PScore (Vernon et al., 2018), and (g) catGRANULE (Bolognesi et al., 2016) was used to perform integrated bioinformatics analyses and computational biological studies of our proteins of interest. Unless mentioned, all programs were used by applying the default parameters. All scripts used to generate the data in this article are provided as the supplementary information datasets.

Construction of the frq[mNeonGreen], son-1[mApple] strain used for imaging was previously described (Bartholomai et al., 2022b). Briefly, a plasmid was constructed to target the endogenous frq locus with DNA encoding N. crassa codon-optimized mNeonGreen appended to the C-terminus of the frq ORF with a flexible linker. A gene encoding hygromycin B phosphotransferase was introduced downstream from the 3′ UTR of frq for selection. A plasmid was constructed containing the N. crassa son-1 ORF with mApple appended to the C-terminus with a flexible linker for insertion at the csr-1 locus. Transformation cassettes were amplified by PCR using NEB Q5 High-Fidelity 2X Master Mix (New England Biolabs, Cat# M0492) and integrated into the genome by homologous recombination after electroporation of conidiophores. A homokaryotic strain containing the fluorescent fusion proteins was used for imaging.

N. crassa homokaryons containing frq[mNeonGreen] and csr-1::son-1[mApple] were grown overnight on 2% agar gel pads made with Vogel’s minimal medium (Vogel, 1956) in constant light at 25°C. Before imaging, a section near the edge of the growing mycelium was removed and inverted on a chambered cover glass (Thermo Scientific, Cat# 155360).

Images were acquired using a Zeiss AxioObserver laser scanning confocal microscope with a ×100/1.46 NA oil immersion plan-apochromat objective. 561 nm and 488 nm lasers were used at 2% power with 2× averaging and a gain of 800 for the detection of fluorescent signals by GaAsp PMTs. Images were processed for display using FIJI (Schindelin et al., 2012). The cropped images displaying a single nucleus were enlarged 10× using bicubic interpolation and contrasted to the minimum and maximum gray-level values. A multicolor lookup table was applied to the FRQ[mNeonGreen] channel to better observe heterogeneity of the signal, and the SON-1[mApple] channel was pseudo-colored cyan to better contrast with the lookup table applied to the FRQ[mNeonGreen] channel. The surface plot of FRQ[mNeonGreen] was generated from the raw image. Video 1 and Figure 6E were acquired using a Nikon Ti-E-based system, equipped with an Andor W1 spinning disk scan head, an Andor Zyla sCMOS camera, a 488 nm laser, and Chroma bandpass filters. Figure 6E is a still image from a single focal plane from a z-stack that was deconvolved using Nikon Elements and was acquired using a ×100/NA1.45 plan apo oil immersion objective. Figure 6E and Video 1 were acquired using a ×60/NA1.45 plan apo oil immersion objective. The 1 min movie was acquired with a 300 ms exposure per frame. Image processing for display was carried out using FIJI.