Pervasive Protein Thermal Stability Variation during the Cell Cycle

1Mass spectrometry (MS)-based proteomics is a powerful method to accurately quantify distinct features of proteomes and has provided comprehensive insight into the variability of protein abundance under distinct biological conditions, including different cell types, protein complexes, and subcellular fractions. Recent advances in MS multiplexing technologies enable the global measurement of protein melting curves (thermal proteome profiling; TPP). TPP and a partial digestion-based methodology for assessing protein thermal stability add another dimension to MS-based proteomics to quantify various states of the proteome. TPP can detect individual proteins that are stabilized in situ by binding to a ligand and, therefore, is useful for profiling drug targets and off-targets. However, its utility to uncover global changes in protein thermal stability in distinct biological contexts is unclear.

2The eukaryotic cell cycle is the key regulatory circuit that controls the temporal separation of fundamental processes that facilitate cell proliferation. It is well established that various aspects of proteome organization, including protein abundance and post-translational modifications, vary during cell-cycle progression. We hypothesized that cell-cycle-dependent post-translational modifications, protein-protein interactions, and spatial rearrangements to distinct biophysical environments globally influence protein thermal stability. Here, we systematically measured in situ protein thermal stability, abundance, and solubility during cell-cycle progression on a proteome-wide scale.

3We report the pervasive variation of protein thermal stability during the cell cycle and link it to various biological processes including transcription, spindle formation and key metabolic pathways. Further, intrinsically disordered proteins are stabilized during mitosis, coinciding with fundamental rearrangements of the proteome and the spatial outline of the cell. These changes coincide with extensive sumoylation and mitotic phosphorylation, suggesting that post-translational modifications might promote thermal stability and, in turn, prevent protein aggregation during mitotic spindle formation and chromosomal separation. Protein stabilization serves as a proxy for biological activity and complex formation, thereby revealing new players in the cell cycle. Our comprehensive analysis of cell-cycle-dependent variation of protein thermal stability, abundance, and solubility provides a valuable resource to advance the fields of transcription, structural biology, intrinsically disordered proteins, metabolism, and the cell cycle.

4To investigate proteome variation across different stages of the cell cycle in situ, we used thermal proteome profiling (TPP) to measure protein thermal stability and abundance. We synchronized and collected HeLa cells in six different cell-cycle stages: G1/S transition (no release after double thymidine block), early S phase (2 hr after release), late S phase (4 hr after release), S/G2 transition (6 hr after release), mitosis (0.5 hr after release from nocodazole block), and early G1 phase (4.5 hr after release from nocodazole block). We also collected asynchronous cells (Figures 1 and S1A–S1D; STAR Methods). Intact cells were heated and lysed with a mild detergent (NP-40), and soluble proteins were identified by multiplexed quantitative MS based on tandem mass tag (TMT) labeling.

5To derive changes in protein thermal stability and abundance across cell-cycle stages, we used two-dimensional TPP (2D-TPP), which multiplexes different cell-cycle stages at each temperature (see Figure 1; STAR Methods). This analysis yields protein abundance ratios to the reference condition, G1/S phase. We obtained quantitative data across the different cell-cycle stages and temperatures for each of the 4,970 proteins that passed our quality control requirements (see STAR Methods), out of the total set of 10,064 identified proteins (Table S1). We calculated abundance and stability scores for each protein and determined the significance based on three biological replicates (see Figures 1 and 2A; Table S2; STAR Methods).

6Changes in protein abundance can reflect altered protein levels or altered solubility, with the latter capturing e.g., a transition from a polymerized to a depolymerized state. To disentangle these two possibilities, we also quantified the total amount of protein in the cell after solubilizing with a strong detergent (SDS) (Table S3).

7To capture the shape of protein melting curves, we applied the TPP temperature range (TPP-TR) approach to mitotic and G1/S cells; doing so multiplexes ten different temperatures for each stage (Figure 1; Table S4). Hereafter, we will use the term “stability” to refer to the thermal stability of proteins, i.e., a protein becomes more stable if exposure to a higher temperature is required in order to cause this protein to aggregate inside the cell. We will use the term “solubility” to refer to differences in the observed abundance of a protein in mild (NP40) versus strong (SDS) detergent.

8Overall, we generated a comprehensive catalog of protein abundance, stability, and solubility, with detailed temporal cell-cycle resolution.

9We explored our datasets for well-characterized cell-cycle factors. For instance, cyclin-A2 (CCNA2) varied in abundance, but not in stability (Figures 2A and S1E). In contrast, cyclin-dependent kinase 1 (CDK1) did not show significant changes in abundance across the cell cycle, but its thermal stability dramatically declined after G2 (Figures 2A and S1E).

10Strikingly, several proteins with functions relevant to the cell cycle displayed altered stability rather than abundance during the cell cycle (Figure 2B). Among these, we identified S-adenosylhomocysteine hydrolase-like protein 1 (AHCYL1), which inhibits ribonucleotide reductase large subunit (RRM1) in a cell-cycle-dependent manner; the ATP-dependent DNA helicase Q1 (RECQL); and inosine-5′-monophosphate dehydrogenase 1 and 2 (IMPDH1, IMPDH2), which maintain the intracellular balance of A and G nucleotides and thus play a key role in cell proliferation. Consequently, the stability changes of other proteins so far not implicated in the cell cycle should be indicative of their cell-cycle-related roles, as will be discussed and shown below.

11Overall, we uncovered significant cell-cycle-dependent changes in stability and abundance for 1,592 and 1,044 unique proteins, respectively (Figure 2C; Table S2). Most abundance changes occurred in early G1, whereas most stability changes occurred in mitosis.

12We used gene ontology (GO) analysis to obtain an overview of which cellular processes and substructures are most prominently affected by cell-cycle-dependent changes in stability. We identified various processes related to the cell cycle (Figure 2D; Table S5), such as nuclear envelope breakdown and the purine salvage pathway. During mitosis, we observed stabilization of signaling-, DNA-, and chromatin-associated proteins and destabilization of proteins involved in metabolic processes. The subcellular distribution of proteins displaying altered stability was relatively even (Figure S2A); however, the endoplasmic reticulum contained more proteins with variable stability (p value 3.04 × 10−5, Fisher’s exact test) (Figures S2A and S2B). This might reflect the morphological changes in the endoplasmic reticulum during mitosis.

13To explore a potential relationship between stability and abundance, we clustered the combined abundance and stability scores based on their Euclidean distance (Figure 2E; Table S2). The resultant 21 distinct clusters primarily consisted of proteins affected in either their stability or their abundance (Figures 2E and S2C), strongly suggesting that these proteomic traits are controlled independently. Comparison of our relatively quantified data with previously determined protein half-lives revealed that proteins changing in abundance across the cell cycle turn over significantly faster (Figure 2F). In contrast, proteins changing in stability tend to have longer half-lives, but with a smaller effect size (Figure 2F). Further, protein melting points were not correlated with protein half-lives (Figure 2G). We conclude that the cell-cycle-dependent variance of stability is controlled through mechanisms that do not affect protein abundance.

14Cell-cycle progression is tightly coupled to intracellular metabolism. However, it has been challenging to identify metabolic enzymes that are more active because of increased substrate availability, rather than increased enzyme abundance. Since substrate binding increases the stability of enzymes, our data can serve as a proxy for enzyme activity. Indeed, the variation of enzyme stability coincided with cell-cycle progression for several interconnected metabolic pathways (Figure 3A; Table S5). The fatty acid biosynthesis pathway is required for mitotic exit, and its key enzymes were stabilized in mitosis and early G1. The non-oxidative branch of the pentose phosphate pathway (PPP) has an increased metabolic flux at G1/S that is reflected in the higher stability of the respective enzymes. This provides evidence for cell-cycle-dependent changes in the non-oxidative branch of the PPP not only on the metabolic but also on the proteome level.

15Furthermore, we depleted three metabolic enzymes with cell-cycle-dependent stability patterns: catechol-O-methyltransferase (COMT), cystathionine beta-synthase (CBS), and choline-phosphate cytidylyltransferase A, (PCYT1A). Doing so significantly reduced both the frequency of cells in S phase and proliferation (Figures S3A–3F).

16In summary, our data not only recapitulate the importance of previously characterized enzymes contributing to cell-cycle progression but they also can be mined to identify the time-resolved metabolic activity of enzymes.

17Next, we investigated whether our data capture changes in protein-chromatin interactions. We found that RNA polymerase II (Pol II) subunits POLR2A and POLR2B were significantly decreased in both abundance and stability in mitosis (Figure 4A). The melting profiles of POLR2A and POLR2B displayed biphasic behaviors in interphase cells, but were monophasic in mitosis (Figure 4B; Table S4), suggesting the existence of two distinct populations of RNA Pol II subunits in interphase.

18Transcriptional arrest occurs in mitosis, and it has been reported that RNA Pol II is cleared from chromatin. We hypothesized that the distinct RNA Pol II populations in interphase reflect a transcriptionally inactive subpopulation not bound to DNA and a more stable, transcriptionally active subpopulation bound to DNA. To test this, we treated asynchronous cells with the bromodomain inhibitor JQ1, which causes transcription elongation defects. JQ1 treatment led to reduced RNA Pol II stability (Figures 4C and S4A) and a slight increase in abundance (Figure 4D), suggesting that RNA Pol II stability correlates with transcriptional activity.

19The phosphorylation state of the POLR2A carboxy-terminal domain (CTD) changes at different stages of transcription, although both phosphorylated and unphosphorylated forms occur in the subpopulation that is not bound to DNA. To rule out that the change in POLR2A stability is solely due to a change in phosphorylation state, we used a phospho-specific antibody to monitor POLR2A stability. JQ1 treatment destabilized phosphorylated POLR2A (Figures S4C and S4D), indicating that the reduced stability was not driven by dephosphorylation. We also treated HeLa cells with THZ1, a CDK7 inhibitor that has been reported to reduce RNA Pol II occupancy at promoters and gene bodies. THZ1 treatment led to reduced POLR2A stability (Figures 4C and S4B) and reduced POLR2A levels (Figure 4D), both of which likely contribute to the potent transcriptional arrest observed with this inhibitor.

20To confirm that DNA-binding stabilizes RNA Pol II, we performed thermal shift experiments on purified yeast RNA Pol II in the presence of DNA. Consistent with the above observations, the addition of DNA stabilized RNA Pol II (Figures 4E and S4E). These findings suggest that the decreased stability in mitosis compared to interphase is due to the clearance of a subpopulation of RNA Pol II from chromatin and coincides with a downregulation of the overall RNA Pol II levels. In general, our results suggest that the measurement of RNA Pol II stability in situ correlates with the transcriptional activity in cells.

21We calculated the correlation of the abundance and stability values of proteins that are subunits of the same annotated complex and compared the resultant distribution to correlation values stemming from all other proteins that are not part of annotated complexes (Figure 5A). The abundance profiles of protein complex subunits across the cell cycle were highly correlated (Figure 5B), even stronger correlation was observed for stability (Figure 5C), suggesting that protein complexes mostly melt as a whole unit in situ once a critical temperature is reached. Indeed, protein complex subunits have a significant tendency toward coherent melting behavior (Figure S5A). Combined stability and abundance values yielded the best discrimination between proteins that are part of complexes from those that are not, including complexes with temporally regulated assembly (Figures 5D and S5B).

22We found that subunit correlation patterns are indicative of stable subcomplexes, or of subunits that are loosely associated or potentially moonlight in other complexes. For example, in the exosome complex (Figure S5C) the core components displayed a high degree of correlation, while the two catalytic subunits, DIS3 and EXOSC10, did not follow this pattern. While the core subunits were collectively destabilized in mitosis, DIS3 and EXOSC10, which are uniquely required for mitotic progression in fission yeast and Drosophila, displayed increased stability in mitosis. Likewise, the 26S proteasome, a complex linked to cell-cycle progression, showed a different stability and abundance variation in a subset of proteins belonging to the 19S regulatory subcomplex (Figure S5D). Furthermore, we observe that the core SMC subunits of condensin II have a distinct behavior from the rest of the complex (Figure S5E).

23Also in case of the nuclear pore complex (NPC), the combined nucleoporin (Nup) stability and abundance profiles clustered into known subcomplexes (Figures 5E and 5F). The NPC has a cell-cycle-dependent assembly status: it is membrane bound in interphase but resolves into soluble subcomplexes during mitosis, as triggered by phosphorylation. Surprisingly, the stability behavior of the distinct subcomplexes was strikingly different, even with opposite sign. Major scaffolding subcomplexes showed a minor destabilization in mitosis, indeed mirroring the NPC assembly status. The effect sizes for the Y and inner ring complexes are slightly different, possibly reflecting architectural disparities that impact on their extractability (Figures 5F and S5F). In striking contrast, most intrinsically disordered Nups, such as NUP358 (RANBP2), remained stable in mitosis or were even strongly stabilized. This is unexpected, because many of the Nups with high intrinsically disordered protein (IDP) content are aggregation prone proteins.

24NUP358 critically contributes to Y complex oligomerization in intact NPCs during interphase. In mitosis, however, it dissociates from the Y complex and localizes to the spindle region, and depletion of NUP358 leads to spindle defects. NUP358 is a large, multifunctional protein that contains several folded domains linked by long, intrinsically disordered regions that are important for nucleocytoplasmic exchange in interphase.

25Similarly to NUP358 (Figure 6A), several other confirmed spindle binding proteins displayed strong mitotic stabilization, such as CHD4, KIF4A, and SMARCA4 (Figures 6B, S6A, and S6B). Compared to the entire proteome, annotated mitotic spindle binding proteins had a significantly lower stability (Figure S6C; Table S6). Since the aggregation of several of these proteins in mitosis was already substantial at 47°C (30% on average as in the case of NUP358) (Figure 6A), one would expect this to have a catastrophic effect on the spindle integrity. Indeed, heating mitotic cells to 47°C resulted in a severely damaged mitotic spindle, and at 50°C no traces of the spindle were detectable (Figure 6C).

26We further examined proteins that were stabilized in mitosis and found common functional and biophysical characteristics: such proteins were significantly enriched (false discovery rate [FDR] < 5%) for the GO terms, DNA binding, chromosome and chromatin organization, and transcription-related processes (Figures S6D and S6E; Table S5). Systematic computational partitioning of proteins according to the fraction of their sequence predicted as disordered revealed that the mitotic stabilization behavior is specific to proteins with large proportions of disordered regions (Figure 6D).

27The synchronized stability change of the disordered proteins during mitosis suggests the presence of a regulatory switch. We considered mitotic phosphorylation, given that it is known to occur in disordered regions, and that phosphorylation-mediated folding of disordered proteins has been reported. We found that proteins with annotated mitotic phosphorylation sites (Table S6) were stabilized in mitosis, and that there is a considerable overlap of disordered and mitotically phosphorylated proteins that exhibit the strongest stabilization (Figures 6E and 6F). A similar trend is observed for sumoylation, a modification linked to cell-cycle progression and mitosis and crosstalk with phosphorylation (Figure S6F). The trend is not observed when non-mitotic phosphorylation or ubiquitination, acetylation or methylation sites are considered (Figure S6F).

28Taken together we observe that chromatin and spindle related proteins with low stability and disordered regions are stabilized in mitosis. This effect is most pronounced for proteins that contain annotated mitotic phosphorylation sites.

29The control measurements of protein abundance in mild and strong detergent that we had originally introduced to disentangle protein expression from solubility, indicated striking solubility transitions in well-defined subproteomes at specific cell-cycle stages. For example, ribosomal proteins, nuclear lamins, vimentin and plectin, all displayed a strong transition from a non-soluble to soluble state in mitosis (clusters 19, 20, and 21; Figures 2E, 7A, and S2). This observation reflects known properties of cellular substructures: nuclear lamins and vimentin are both phosphorylated, leading to depolymerization of filaments in mitosis, and plectin undergoes a phosphorylation-triggered transition from an insoluble to soluble state. Strikingly, the nuclear lamins showed a delay in the transition back to the insoluble state in early G1 (Figures 7A and 7B).

30Further, we observed that such solubility transitions captured components of phase-separated, membraneless organelles that are dissolved in mitosis. For instance, various nucleolar proteins are more soluble in mitosis when the nucleolus is dissolved (Figure 7A). Our above-described data capture such transitions only if they occur during the cell cycle, but they are not suitable to quantify the fractions of a protein in the soluble and insoluble state. Therefore, we quantified proteome solubility under mild and strong detergent conditions for the G1/S stage as compared to mitosis in a dedicated TMT experiment (Figure S7A; Table S7).

31The analysis revealed that a small chromatin-associated fraction of the proteome is consistently insoluble in G1/S and mitosis. This fraction includes CDCA8, INCENP, and AURKB, which are core members of the chromosomal passenger complex (Figure S7A) and key regulators of several mitotic events, and SUPT16H and SSRP1 of the FACT complex, which is essential for microtubule growth and bundling in mitosis (Table S7). We did not observe a single protein becoming more than 2-fold less soluble in mitosis compared to G1/S (Table S7), indicating that there is no pronounced transition from a soluble to insoluble state upon mitotic entry. In striking contrast, various proteins became significantly more soluble in mitosis. Similarly, but not identical, to the mitotically stabilized IDP proteome discussed above, those proteins were strongly enriched in RNA binding functions (Figure S7B), mitotic phosphorylation sites (Figures 7C and 7D; Table S6) and sumoylation (Figure S7C). Many of them were components of the nuclear envelope and also membraneless organelles.

32The proteins identified to undergo a solubility transition in mitosis comprise only a fraction of the previously annotated proteomes of these membraneless organelles, indicating that their core proteome might be less extensive than previously thought (Figures 7E and 7F). However, the subproteomes that did undergo a solubility transition were strongly enriched in disordered regions (Figure 7C) and captured key phase separating proteins of membraneless organelles. Among those were proteins with reported roles for nucleolar morphology, such as nucleostemin, GNL3, fibrillarin, FBL, nucleolar protein 7, NOL7, and the protein SON, one of the most strongly transitioning proteins in nuclear speckles and recently proposed to be a key scaffolding factor for these membraneless organelles (Figure 7F). An exception to this trend was the transcription factor 1, UBTF (Figures 7F and S7A), which acts as a molecular beacon to recruit additional proteins during the de novo biogenesis of the nucleolus. Our data show that UBTF has a very rare behavior among the nucleolar proteins, in that it remains insoluble during mitosis.

33The nucleolus is the site for ribosome assembly. When it dissolves in mitosis, non and partially assembled ribosomal proteins are set free and their exact fate remains understudied. The stability of many ribosomal proteins was significantly decreased in mitosis (Figure 7A). A subpopulation of ribosomal proteins that is insoluble in G1/S becomes soluble in mitosis (Figure 7F) and might be attributed to non-assembled species. In line with this view, the insoluble interphase subpopulations of the 40S and 60S components have significantly different effect sizes (Figures 7F and 7G), reflecting the shorter residence time of 40S proteins in the nucleoulus as compared to 60S proteins. The ribosomal associated complex that participates in maturation of the large 60S ribosomal subunit, as determined by (BOP1, RRS1, GNL3, EBNA1BP2, FTSJ3, and MKI67), exhibited the largest effect size in its coherent solubility transition (Figures 7F and 7G), indicative of a predominant nucleolar localization in G1/S.

34Taken together, our data delineate the insoluble and the cell-cycle-dependent, solubility transitioning proteome. Our experimental strategy enables unbiased assessment of key phase transitioning proteins responsible for the altered solubility state of the mitotic cell and provides molecular insight into the mitosis specific morphological changes.

35We have performed an in-depth, proteome-wide stability profiling of the eukaryotic cell cycle. We discovered pervasive protein thermal stability variation during the cell cycle, representing a substantial layer of proteome organization that is on par with the variability of protein abundance.

36Our results with RNA Pol II suggest that relative changes in transcriptional activity can be assessed in situ by measuring the stability of POLR2A. The disappearance of the more stable POLR2A population during mitosis coincides with the overall reduction of POLR2A levels. A speculative interpretation could be that the stable DNA-bound RNA Pol II fraction is specifically degraded upon mitotic entry.

37The combined stability and abundance variation of NPC and other complexes across the cell cycle serves in delineating major subcomplexes and moonlighting subunits. This highlights the utility of the present dataset for structural biologists, as it can be used for hypothesis generation of cell-cycle-dependent complex composition and subunit function.

38Mitotically phosphorylated IDPs are stabilized in mitosis. The NPC illustrates the significance of this finding. Although intrinsically disordered Nups are notoriously aggregation prone, scaffold Nups provide a grafting surface that restrains their local concentration in interphase. NPC disassembly, however, does not render intrinsically disordered Nups aggregation prone. Strikingly, NUP358 that facilitates scaffold Nup oligomerization in interphase is stabilized in mitosis. These insights exemplify the challenge that cells have to deal with during the morphological changes occurring in mitosis and apparently address by major transitions in their biophysical environment. It is possible that IDPs are integrated in specific protein complexes or subcellular structures in interphase that restrain their local concentration, such is the case, for instance, for NUP358 and the NPC scaffold. In mitosis, these restraints on local concentration could potentially disappear and render IDPs aggregation prone. Phosphorylation of these proteins might be important to prevent aggregation in mitosis. Their similar biophysical milieu might phase separate them into the spindle region, where they are less exposed to aggregation triggers, such as membranes.

39We also mapped the subset of the proteome undergoing solubility transitions, from an insoluble to a soluble state as well as the proteins that are permanently insoluble, during the cell cycle. The molecular signature of RNA binding enriched proteins that are solubilized in mitosis is very similar to the DNA binding enriched subproteome that gets stabilized in mitosis. The two subproteomes have an overlap of 15% and are enriched in disordered regions and mitotic phosphorylation. The solubility transition of the cell in mitosis is often obvious on a morphological scale, e.g., the dissolution of the nucleolus is apparent by light microscopy. Our data categorize the proteins involved in this process, several of which play key roles in the architecture of membraneless organelles. The vast majority of all proteins that are insoluble during interphase are solubilized in mitosis. We conclude that the mitotic cell is more soluble and less aggregation prone as compared to the interphase cell.

40In conclusion, our study reveals that the biophysical properties of proteins extensively vary across the cell cycle, reflecting changes in metabolic activity, complex composition, and transcriptional activity. It can thus be argued that changes in stability represent a robust way to identify proteins whose activity and interaction patterns vary across the cell cycle. In contrast, changes in abundance will typically not reflect such variations. We show the utility of our dataset for biological hypothesis generation and expect it to become widely used in metabolomics, transcription, structural biology, IDPs, and cell-cycle-related studies.

41Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Mikhail M. Savitski (savitski@embl.de).

42HeLa Kyoto cells (female) were cultured at 37°C, 5% CO2, in DMEM including 1mg/ml glucose, supplemented with 10% FBS, 1mM glutamine. Cells were synchronized in G1-S by a double thymidine block (addition of 2mM thymidine (Calbiochem, #6060) overnight, release cells for 8 hours, second block with 2mM thymidine overnight). After the second release cells were collected at four different time points: 0 hours (G1-S phase), 2 hours (early S phase), 4 hours (late S phase), and 6 hours (S-G2 phase). For synchronization in mitosis cells were treated overnight with 2mM thymidine, released for 4 hours and treated with 100 ng/ml nocodazole (Millipore, #487928) overnight. After 0.5 hour release cells were collected (M phase). For synchronizing cells in G1, 100 ng/ml nocodazole was added overnight, then mitotic cells were collected by shake-off and released into fresh medium, letting them grow for 4.5 hours, when only attached cells were collected. Also, an asynchronous culture was used as a control.

43The arrest was assessed by flow cytometry, and at the same time thermal proteome profiling experiments were performed. An aliquot of the cells was kept to perform a SDS-lysis to determine protein expression changes. All experiments were done in biological triplicates.

44Cell cycle analysis was performed by flow cytometry with two bivariate analyses: DNA/ Proliferating Cell Nuclear Antigen (PCNA) and DNA/ phosphorylated Serine 10 of histone H3 (pH3)

45PCNA staining was adapted from. 1x106 cells were permeabilized for 10 min on ice with 0.1% Triton X-100/1% bovine serum albumin (BSA)/PBS, and then fixed with methanol at −20°C for 3 min. After pelleting, cells were resuspended and stored in PBS at 4°C. For pH3 staining, 2x106 cells were fixed with 1% formaldehyde for 15 min at RT. Cells were then resuspended in PBS and stored in 70% ethanol at 4°C.

46Both samples were spun and the cell pellet was resuspended and incubated for 15 min in PBS at room temperature before continuing with blocking in 1% BSA in PBS for 10 min. Cells were then incubated with primary antibodies (anti-PCNA PC10 conjugated with Alexa Fluor 488, Cell Signaling Technology #8580 or anti-Phospho-Histone H3 (Ser10) conjugated with Alexa Fluor 647, Cell Signaling Technology #9716S) for 1 hour at room temperature. Cells were washed and DNA was stained by addition of 20 μg/ml propidium iodide (PI) in 0.1% Triton X-100 containing 0.2 mg/ml RNase.

47Samples were analyzed on a BD LSR Fortessa instrument. For PI detection, the laser line of 561 nm and 75 mV with a band pass (BP) filter of 610/20nm was used. Alexa Flour 488 was detected with a laser line of 488 nm, 50 mW, and a BP 530/30 nm. In the case of Alexa Fluor 647, a laser line of 640 nm, 80 mV and BP 670/14 was used. Cell cycle stage analysis was performed with FlowJo v10 software. Bivariate analysis of PCNA versus DNA content presented an inverted U shape; cells with the higher signal were considered to be in S-phase. Cells with a G2/M DNA content, and positive for pH3 antibody were considered mitotic. The % of cells in G1 was calculated by fitting the PI signal to a cell cycle distribution using the Watson pragmatic model approach.

48Thermal protein profiling was done as previously described. In brief, cells were harvested, washed with PBS and 10 aliquots, each of 0.5 × 106 cells, were distributed in PCR tubes. Each aliquot was heated for three minutes to a different temperature (37.0-40.4-44.0-46.9-49.8-52.9-55.5-58.6-62.0-66.3°C). Lysis buffer (final concentration 0.8% NP-40, 1.5mM MgCl2, protease inhibitor, phosphatase inhibitor, 0.4 U/μl benzonase) was added and cells were incubated at 4°C for one hour. Protein aggregates were removed. From the soluble fraction the protein concentration was determined and for sample preparation 10 μg protein (based on two lowest temperatures) were taken further. Proteins were reduced, alkylated and digested with trypsin/Lys-C with the SP3 protocol. Peptides were labeled with TMT10plex (Thermo, #90110). Here, samples were combined in two ways: either all temperatures of one cell cycle arrest were combined allowing the determination of melting curves, or all cell cycle arrests were combined for each temperature resulting in seven conditions analyzed together. Pooled samples were fractionated on a reversed phase C18 system running under high pH conditions, resulting in twelve fractions. In addition, both samples from NP-40 lysis (37°C) and SDS lysis from G1/S or M arrested samples were combined, allowing the determination of protein solubility differences.

49Peptides were separated using an UltiMate 3000 RSLC nano LC system (Thermo Fisher Scientific) equipped with a trapping cartridge (Precolumn C18 PepMap 100, 5μm, 300 μm i.d. x 5 mm, 100 Å) and an analytical column (Acclaim PepMap 100, 75 μm x 50 cm C18, 3 μm, 100 Å). The LC system was directly coupled to a Q Exactive Plus mass spectrometer (Thermo Fisher Scientific) using a Nanospray-Flex ion source. Solvent A was 99.9% LC-MS grade water (Fisher Scientific) with 0.1% formic acid and solvent “B” was 99.9% LC-MS grade acetonitrile (Fisher Scientific) with 0.1% formic acid. Peptides were loaded onto the trapping cartridge using a flow of 30 μL/min of solvent A for 3 min. Peptide elution was afterward performed with a constant flow of 0.3 μL/min using a total gradient time of 120 min. During the elution step, the percentage of solvent B was increased stepwise: 2% to 4% B in 4 min, from 4% to 8% in 2 min, 8% to 28% in 96 min, and from 28% to 40% in another 10 min. A column cleaning step using 80% B for 3 min was applied before the system was set again to its initial conditions (2% B) for re-equilibration for 10 minutes.

50The peptides were introduced into the mass spectrometer Q Exactive Plus (Thermo Fisher Scientific) via a Pico-Tip Emitter 360 μm OD x 20 μm ID; 10 μm tip (New Objective). A spray voltage of 2.3 kV was applied and the mass spectrometer was operated in positive ion mode. The capillary temperature was 320°C. Full scan MS spectra with a mass range of 375-1200 m/z were acquired in profile mode in the Orbitrap using a resolution of 70,000. The filling time was maximum 250 ms and/or a maximum of 3e6 ions (automatic gain control, AGC) was collected. The instrument was cycling between MS and MS/MS acquisition in a data-dependent mode and consecutively fragmenting the Top 10 peaks of the MS scan. MS/MS spectra were acquired in profile mode in the Orbitrap with a resolution of 35,000, a maximum fill time of 120 ms and an AGC target of 2e5 ions. The quadrupole isolation window was set to 1.0 m/z and the first mass was fixed to 100 m/z. Normalized collision energy was 32 and the minimum AGC trigger was 2e2 ions (intensity threshold 1e3). Dynamic exclusion was applied and set to 30 s. The peptide match algorithm was set to ‘preferred’ and charge states ‘unassigned’, 1, 5 - 8 were excluded.

51For knockdown experiments HeLa Kyoto cells were transiently transfected using siRNA spotted cell culture plates and CELLviewTM slides following a previously described reverse transfection protocol using pre-designed Silencer® Select siRNAs (Ambion). In brief, the transfection solution composed of 20 μl siRNA (3 μM), 7 μl Lipofectamine 2000 (Invitrogen), 12 μl Opti-MEM (Invitrogen) containing 0.4 M sucrose, and 12 μl water was prepared. After incubation for 20 min at room temperature, 29 μl of 0.2% gelatin and 3.5x10%–4% fibronectin were added. Finally, the transfection cocktail was diluted 1:50 and 300 μl were spotted on 24-well cell culture plates, or 50 μl on CELLviewTM slides respectively, followed by immediate lyophilization using the concentrator miVAC (Genevac). Dependent on the incubation time, 12500 (2d), 6250 (3d) or 3125 (4d) cells/well were seeded on the spotted 24-well plates. Additionally, a non-transfected control (NTC) was included for each experiment, where cells were seeded on non-spotted wells. The knockdown of target genes was validated on the level of mRNA by quantitative real-time PCR (qRT-PCR) analysis after 2d of transfection, as well as on the level of protein by LC-MS/MS analysis after 2, 3 and 4d of transfection (MS analysis as described before).

52For qRT-PCR analysis, mRNA extraction was performed using the RNeasy Mini Kit (QIAGEN) according to the manufacturer’s instructions and including an on-column DNase I digestion. Subsequently, 700 ng total RNA were converted into cDNA using the SuperScriptTM III Reverse Transcriptase (Invitrogen) following the manual for first-strand cDNA synthesis. qRT-PCR reactions were performed on a StepOnePlus Real-Time PCR System (Applied Biosystems) using 5 ng cDNA, 0.2 μM gene-specific primers and 1x Power SYBR® Green PCR Master Mix (Life Technologies). For each condition technical triplicates were prepared and the mean threshold cycle (CT) calculated. For normalization, ACTB levels were used as an internal reference and mRNA levels relative to siCTRL were calculated according to the 2-ΔΔCt method.

53Cell cycle analysis was performed by flow cytometry using PI staining. After 4d of transfection cells were harvested, washed and fixed in 66% Ethanol at 4°C for several hours. After washing, DNA was stained with PI and the cell cycle analyzed as described above. In addition, apoptosis was measured using the Annexin V-FITC Detection kit (Promokine) according to the manual.

54For assessment of live and dead cells a commercial assay was used (Promega, MultiTox-Fluor Multiplex Cytotoxicity Assay) and the manufacturer’s protocol was followed.

55HeLa Kyoto cells were treated with either 5 μM JQ1 (Sigma Aldrich), 5 μM THZ1 (Cayman chemicals) or DMSO (vehicle) for 3 h at 37°C, 5% CO2, in DMEM supplemented with 2% FBS and 1mM glutamine. Heat treatment, lysis and removal of aggregates were performed as described above. Abundance and melting curve of POLR2A were determined by SDS-PAGE and western blot analysis. In brief, 10 μg protein (based on the lowest temperature) were separated on 4%–15% precast polyacrylamide gels (Bio-Rad) following the manual. Blotting was performed using the Trans-Blot® Turbo™ Transfer System (Bio-Rad) according to the instructions. For detection of total and phosphorylated POLR2A, Rpb1-specific antibodies (#2629 and #4735 for Ser2/5 phosphorylation, Cell Signaling) were used and the manufacturer’s protocol followed. Chemiluminescence was recorded using the ChemiDoc imaging system (Bio-Rad) and analyzed via Image Lab 5.2.1.

56HeLa Kyoto EGFP-alpha-tubulin/H2B-mCherry were synchronized at the G1/S transition as done for the parental cell line. After the second thymidine block, cells were released for 8 to 10 hours until most of them were in mitosis as determined by phase contrast microscopy. They were collected, washed, resuspended in PBS at 0.5 × 106 cells, and subjected to a heat treatment as done for the thermal profiling and MS. Afterward they were fixed with 4% PFA for 15 min at room temperature. An aliquot of each sample was let settle down onto a well of a CELLview slide (Greiner Bio-One) previously coated with poly-L-Lys. Cells were imaged on an inverted Leica TCS SP8 STED3x microscope (Leica Microsystems, Mannheim, Germany), using a HC PL APO CS2 63x/1.40 OIL objective and a tuneable and pulsed White Light Laser (WLL, emission from 470-670nm). Standard, diffraction limited confocal images were recorded for mRFP (594 nm, BP 600-690 nm) and EGFP (488 nm, BP 495-541 nm). z stacks were acquired to cover the whole cell with a z-step size of 1 μm. Images were processed using the Fiji distribution of ImageJ (https://fiji.sc/).

57RNAP II yeast strain was a gift from Mikhail Kashlev at the NIH/National Cancer Institute (NCI), Bethesda. Pol II was purified from S. cerevisiae using a His-tag purification protocol as previously described. When RNA pol II was bound to nucleic acids, then a 5-fold molar excess of a pre-annealed 38-bp transcription scaffold with a 20-nt RNA (5′-UAUAUGCAUAAAGACCAGGC-3′) as previously described was used. Briefly, the template (T) and non-template (NT) strands were mixed and annealed by heating to 95°C in RNase-free water, and then slowly cooled to 25°C in 1 hr. Then, an equimolar amount of RNA was added and annealed by incubation at RT before addition of RNA pol II.

58For the thermal profiling, aliquots of purified Pol II ± DNA/RNA scaffold were incubated for 3 minutes at the indicated temperature, left at room temperature for 3 more minutes and the aggregated protein was removed by ultracentrifugation (20min at 100.000 xg at 4°C). The supernatant was used for Coomassie gel.

59Raw MS data was processed with IsobarQuant and peptide and protein identification was performed with the Mascot 2.4 (Matrix Science) search engine. Data was searched against a human database from Uniprot including known contaminants and the reversed protein sequences. Search parameters: trypsin, missed cleavages 3, peptide tolerance 10ppm, 0.02Da for MS/MS tolerance. Fixed modifications were carbamidomethyl on cysteines and TMT10plex on lysine. Variable modifications included acetylation on protein N terminus, oxidation of methionine and TMT10plex on peptide N-termini. The programming language R (https://www.r-project.org/) was used to analyze the raw output data from IsobarQuant.

60Three independent datasets were processed in total (2D-TPP data, SDS data and TPP-TR data) using a similar base workflow as described in the following: First, the output files of IsobarQuant were loaded into R and merged. The raw data was then saved in an ExpressionSet R-object. Subsequently, potential batch effects were removed using limma and data was normalized using variance stabilization, vsn strategy. For each temperature, an independent normalization run was used in order to account for the decreasing intensity of signal sums with increasing temperature. To calculate the abundance and stability scores, we used a bootstrap algorithm; for calculating the expression scores, the results of the concordant limma analysis were modified. For all scores, two different false discovery rates were estimated, a local and a global FDR. The global FDR was calculated based on the score distribution (z-distribution) and thereby correlates with the effect size of the score. The local FDR, on the other hand, describes the quality and the reproducibility of the score values and takes into account the variance between replicates.

61After normalization of the raw TPP-TR data, the TPP package was used to identify melting points. After normalization of the 2D-TPP dataset, fold changes were calculated against the respective 37°C and the G1/S data points. These were used for the calculation of abundance and stability scores.

62For the SDS data, a limma analysis was performed on the normalized data by comparing each cell-cycle to the G1/S cell cycle. The resulting fold changes (logFC) were transformed into z-scores using the Cox method in order to determine the expression scores for each protein. The global FDR for the expression score was calculated from the z-values using fdrtool R package taking the q-value as the output. The local FDR for the expression score was calculated by adjusting the limma p values for multiple testing by Benjamini & Hochberg with the p.adjust method (method = ”fdr”).

63In order to calculate the abundance and stability scores, a bootstrap method was used with 500 iterations and a prerequisite of at least 2 measured data points for the two lowest temperatures. In the bootstrap loop, one G1/S-ratio point out of the three replicates for each temperature was chosen and used for the calculation. The abundance value was simply the average of the log2(G1/S.ratio) for the first two temperatures (37 and 40.4°C). The stability value was calculated by first subtracting the average log2(G1/S.ratio) of the first temperatures from all log2(G1/S.ratios) of all temperatures and then calculating the sum of the resulting values. After repeating this 500 times, each protein yields a distribution of abundance and stability values. From these distributions we calculated the standard deviation and the average. Together with the number of data points that were considered for bootstrapping, we could calculate a p value (transformation into z-scores and then assuming Student’s t-distribution) to estimate the likelihood that the distribution is different from zero (no change).

64The averages were transformed into the z-distribution which represent the new stability and abundance scores. The global FDR was then calculated using the fdrtool package.

65In order to make a call on significant hits based on these different scores, we chose a cut-off of 0.01 for the local, as well as the global FDR for each cell-cycle and score.

66Abundance, expression and stability score data were clustered using proteins that were detected with a significant change in at least one of the different scores in one cell cycle stage. For each protein, each score for each cell cycle stage (18 features) were used as features to calculate the euclidean distance. Using the Ward.D2 clustering method the results were divided into 21 different clusters which captured most of the variability in the data.

67For solubility experiments limma calculated p values for comparison between solubility in G1/S (abundance difference between NP40 and SDS) and mitosis (abundance difference between NP40 and SDS) were calculated, and corrected using Benjamin-Hochberg. An FDR cut-off of 0.05 was used, additionally an effect size of > 0.322 (25% regulation) was required for the log2 fold-changes, and the solubility in G1/S had to be < 0.

68The lower and the upper hinges of the boxes correspond to the 25% and the 75% percentile, and the bar in the box the median. The upper and upper whiskers represent the largest and lowest values, respectively (but at maximum 1.5 times the IQR). Points outside the whiskers are plotted individually. Group tests were performed using a nonparametric Wilcoxon test and statistical significance is indicated by: ∗ p ≤ 0.05, ∗∗ p ≤ 0.01 and ∗∗∗ p ≤ 0.001 if not stated otherwise.

69All quantified proteins from this study were mapped to Reactome pathways (https://reactome.org/download-data/, February 2017) using the mygene-package in Python (https://pypi.python.org/pypi/mygene). For each pathway, corrected p values (local FDR) of all its quantified protein components were combined using the Empirical Brown method (EBM) to account for dependent p values. These combined p values were corrected using the Benjamini-Hochberg procedure. For pathway selection and visualization, we considered those that were sufficiently covered (> 60% pathway coverage with 5-150 quantified protein members), and had at least 30% of its protein components changing significantly in at least one cell cycle stage (local.FDR (stability) < 0.01). 106 pathways that passed these criteria were then manually curated for redundant proteins that changed significantly in several pathways, with 30 pathways retained in the end (visualized in Figure 2D; Table S5). If the combined adjusted p value for a pathway in a given stage was less than 0.05, full opacity is given; otherwise the bubble remains transparent. Overall tendency of a pathway to be stabilized or destabilized was calculated as the median stability score of its significantly changing protein components (local.FDR (stability) < 0.01) in each cell cycle stage.

70Gene ontology analysis on proteins significantly affected during the cell cycle in either stability or abundance has been performed with DAVID (version 6.7) (https://david.ncifcrf.gov/), based on GO terms “Biological Process,” “Molecular Function,” “Cellular Compartment” and SP-PIR Keywords. All quantified proteins represented the corresponding background. Benjamini-Hochberg corrected p values were used. For visualization purposes only GO-hits with an FDR < 0.05 were taken into account, and shown in Figure S2C. For Figure 2E only the fold enrichment of the top-hits in each GO-category is shown for each of the 21 clusters.

71Proteins were assigned to complexes defined by COMPLEAT and CORUM, as curated and published. Stability- as well as abundance z-scores were concatenated for each protein, or remained separate vectors, respectively. Subunits of complexes were correlated (Pearson method) with subunits of the same complex; as a control, proteins not assigned to any complex were correlated with other non-complex associated proteins.

72For clustering complexes assigned as above stability as well as abundance z-scores were considered. The clustering was performed using the k-medoids algorithm iterating from 2 to 5 potential clusters per complex, and the Euclidean distance between the calculated medoids was monitored at each iteration step. For each complex then the maximum distance reflecting the best possible clustering, is recovered.

73To assess the nature and origin of the shift in melting points in mitosis relative to the reference, we took into account the disordered state of a protein. The disordered state of each protein was described using the d2p2-database, which combines predictions of several algorithms and makes calls on regions if they have been found to be disordered in at least 75% of those libraries (PONDR VL-XT PONDR VSL2b, PrDOS, PV2, IUPred (+sub-versions of it), Espritz (+sub-versions if it)) (http://d2p2.pro/, February 2017). The protein was ranked according to the relative number of amino acid residues that lie within supposed disordered regions spanning a minimum of 5 amino acids. The relative disordered rank was also corrected taking into account known structural motifs that might overlap with the predicted disordered region using the PDB (see Tables S6 and S7).

74Modifications were extracted from the PhosphoSitePlus (https://www.phosphosite.org/homeAction.action, March 2017), containing information on known acetylation, methylation, ubiquitination, sumoylation and phosphorylation sites. Proteins containing such sites were scanned for stability and abundance patterns across the cell cycle stages. 512 mitotically regulated phosphorylation sites were extracted from (Tables S6 and S7: List of regulated mitotic phosphorylation sites with high occupancy).

75Localization of a protein was defined by the GO-term (only Experimental Evidence is allowed, no inference from computational methods) and the humanProteinAtlas (http://www.proteinatlas.org/, May 2017) using the data supported, validated or approved by antibody-analysis. For the evaluation of the solubility transition data the humanProteinAtlas annotation was used as well as the data from for nuclear envelope annotation.

76All analyzed data is made available in supplemental data files. Supplementary tables and clustering of individual complexes can be found at Mendeley: https://doi.org/10.17632/xrbmvv5srs.2. The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD008646. All code used for data analysis is available at https://github.com/fstein/cellcycle_TPP.