SNAP29 mediates the assembly of histidine-induced CTP synthase filaments in proximity to the cytokeratin network

Metabolic pathways are often compartmentalized to enhance specificity and efficiency. Although membrane-bound organelles are the classical compartments for various metabolic processes, membraneless supramolecular assemblies such as stress granules, p-bodies (Anderson and Kedersha, 2008; Buchan, 2014) and purinosomes (An et al., 2008) are often implicated in regulating specific metabolic pathways in response to cellular stress (Mitrea and Kriwacki, 2016). It has been suggested that phase-separation mechanisms are important for the dynamic assembly of these membraneless organelles (Boeynaems et al., 2018; Mitrea and Kriwacki, 2016; Uversky, 2017).

Cytidine 5′-triphosphate synthase (CTPS, for which there are CTPS1 and CTPS2 forms in mammals) is a rate-determining enzyme in the de novo synthesis of CTP, which serves as a substrate for the synthesis of structural components of DNA and RNA and is involved in the formation of phospholipids (Ostrander et al., 1998). CTPS compartmentalizes into filamentous structures under various conditions across species (Carcamo et al., 2011; Ingerson-Mahar et al., 2010; Liu, 2010; Noree et al., 2010). However, the cellular locations of these filamentous compartments remain unidentified, while the Golgi, centrosomes, actin, tubulin and vimentin have been examined for their association with CTPS in human cells (Carcamo et al., 2011). In yeast, it has been suggested that CTPS filaments are composed of inactive dimers (Noree et al., 2014), although cryo-electron cryomicroscopy (cryo-EM)-based studies in E. coli and humans have shown that tetramers are the units of CTPS filament formation (Barry et al., 2014; Lynch et al., 2017). CTPS in its active homotetramer configuration converts UTP into CTP by adding an amino group from glutamine (Gln) (Endrizzi et al., 2004; Goto et al., 2004; Kursula et al., 2006; Weng and Zalkin, 1987), which serves as a key regulator for the reversible assembly of CTPS filaments (Calise et al., 2014; Pai et al., 2016). In human cancer cells, Gln starvation induces CTPS filaments, which are known as ‘rods and rings’ (RR) structures that also contain inosine monophosphate dehydrogenase 2 (IMPDH2) (Calise et al., 2014; Carcamo et al., 2011; Chen et al., 2011; Gou et al., 2014). RR structures can also be induced by CTPS inhibitors, such as 6-diazo-5-oxo-L-norleucine (DON), azaserine and acivicin (Calise et al., 2014; Carcamo et al., 2011; Chen et al., 2011), which are Gln analogs.

It has been proposed that cytoophidium or CTPS filament formation could be an adaptive mechanism to compromise and regulate cell metabolism (Aughey et al., 2014; Petrovska et al., 2014; Wang et al., 2015). In Drosophila, activated Cdc42 kinase (Ack), Casitas B-lineage lymphoma (Cbl) and Myc have been shown to regulate the CTPS filament structure in egg chambers (Aughey et al., 2016; Strochlic et al., 2014; Wang et al., 2015). Given that filament formation of CTPS is dynamic, post translational modifications, such as ubiquitination and methylation, are required for CTPS filament formation in human cancer cells (Lin et al., 2018; Pai et al., 2016). In human cancer cells, we previously demonstrated that histidine (His) catabolism contributed to the folate cycle and methyl cycle, which are required for CTPS filament formation under Gln deprivation (Lin et al., 2018). The dynamic behavior of CTPS intrigued us, and hence we sought to investigate how filament assembly happens. Conventional methods like co-immunoprecipitation assays have been challenging for the identification of proteins interacting with CTPS filaments. Therefore, we used APEX2-mediated proximity labeling (Lam et al., 2015; Martell et al., 2012) of CTPS1 to identify proteins that are associated with CTPS1 filaments and/or involved in the process of CTPS1 filament formation.

Here, we demonstrate that in human cancer cells during Gln starvation, CTPS is assembled into filament-like structures along the cytokeratin network. Our findings suggest that disrupting the cytokeratin network or knocking down a specific keratin, KRT8, can affect the CTPS filament formation process. Furthermore, we found a SNARE-binding protein, synaptosome-associated protein 29 (SNAP29), that interacts dynamically with the cytokeratin network, which is involved in CTPS filament formation. In summary, we revealed that cytokeratin is an important compartment for the assembly of metabolic enzymes, such as CTPS and IMPDH, into filamentous structures under Gln starvation stress.

Our previous study revealed that in Gln- and serum-depleted [G(−)/S(−)] conditional medium, His is essential for CTPS filament formation, and His under Earle's buffered salt solution (EBSS) conditions robustly induced CTPS filaments in a dose-dependent manner in HEp-2 cells, thus providing an excellent tool for investigating the process of CTPS filament formation (Fig. 1A; Fig. S1A,B) (Lin et al., 2018). Therefore, we used APEX2-mediated proximity labeling to identify proteins related to CTPS1 filament formation (Fig. 1B) (Lam et al., 2015; Martell et al., 2012). We exogenously C-terminally tagged CTPS1 with the 28-kDa peroxidase APEX2 and N-terminally with a small 3×FLAG tag. The double-tagged CTPS1 assembled into filaments and dissembled with the addition of Gln; thus, these filaments were phenotypically similar to endogenous CTPS filaments (Fig. S2A,B) (Calise et al., 2014). As 90% of the cells already induced CTPS filaments within 6 h when treated with 200 µM His in EBSS medium, we chose this condition for proteomic analysis (Fig. S1A,B). FLAG–CTPS1–APEX2 formed filaments within 6 h, and the filaments were also efficiently biotinylated (Fig. 1C). Consistent with this, 15 min of Gln treatment led to the complete disassembly of the FLAG–CTPS1–APEX2 filaments and generated a cytosolic biotin signal (Fig. 1C). Moreover, endogenous CTPS filaments were not biotinylated in cells expressing FLAG–APEX2 alone (Fig. 1C). To detect biotinylated CTPS protein by western blotting, four biological samples [FLAG–CTPS1–APEX2 in EBSS (group 1), EBSS+His (group 2) and EBSS+His+Gln (group 3), and FLAG–APEX2 in EBSS+His (group 4)], were subjected to immunoprecipitation with streptavidin-coated magnetic beads and then probed with anti-CTPS antibody. Exogenous CTPS1 (∼100 kDa) was detected in groups 1–3 but not in the control group 4 (Fig. S2C). Furthermore, endogenous CTPS (∼70 kDa) was enriched in the EBSS+His group (group 2) (Fig. S2C). These results suggested that APEX2-tagged CTPS1 was assembled with endogenous CTPS into a filamentous structure.

We used isobaric tags for relative and absolute quantification (iTRAQ) followed by two-dimensional liquid chromatography-tandem mass spectrometry (2D LC-MS/MS) (Fig. 1D) (Wiese et al., 2007). FLAG–APEX2 (group 4) served as a non-specific interaction control. Group 2 served as the experimental group, as the presence of FLAG–CTPS1–APEX2 and His together facilitated biotinylation of CTPS filament-interacting proteins. Groups 1 and 3 served as the non-filament and filament-disassembly controls, respectively. Following APEX2 proximity biotinylation and immunoprecipitation, 10% of the immunoprecipitation product from each of the four groups was used to detect the immunoprecipitation efficiency, through silver staining, for quality control (Fig. S2D). The remaining samples were labeled with iTRAQ reagent followed by detection using 2D LC-MS/MS (Fig. 1D). From the iTRAQ analysis, we found 26 shared candidates when comparing the biotinylated filament group 2 to the other three non-biotinylated filament controls (Fig. S2E,F; Table S1). We identified CTPS2, which was previously reported to form filaments in both yeast and humans (Gou et al., 2014; Noree et al., 2010; Shen et al., 2016). Indeed, immunofluorescence staining of FLAG–CTPS2 showed filaments under EBSS+His conditions (Fig. S2G). We then tested the effects of most of the 26 candidate genes on filament formation by performing shRNA-mediated stable cell knockdown. The following three genes showed reduced filaments when incubated with His in the EBSS condition: SNAP29, ubiquitin-like 5 (UBL5) and capping actin protein (CAPG) (Fig. 1E,F). Gene knockdown of these three candidates was also confirmed using quantitative (q)PCR (Fig. 1G).

We performed 6-plex Tandem Mass Tag (TMT) labeling followed by 2D LC-MS/MS with three repeats of comparison between EBSS+His and EBSS to identify His-induced CTPS filament-interacting proteins (Fig. 2A; Fig. S2H–J and Table S2). We used P<0.05 as the initial cutoff and further used mean+2s.d. to select 123 proteins, which included CTPS1, CTPS2 and SNAP29 (Fig. 2A). As SNAP29, a SNARE-binding protein, is known to be involved in multiple protein trafficking processes (Guo et al., 2014; Morelli et al., 2014; Rapaport et al., 2010; Steegmaier et al., 1998), we further studied its involvement in CTPS filament assembly. SNAP29 enrichment during CTPS filament formation was also confirmed by western blot analysis of samples from the CTPS–APEX2-mediated proximity labeling (Fig. S3A). A second set of shRNAs targeting SNAP29 also reduced the proportion of cells with CTPS filaments (Fig. 2B) without affecting CTPS1 protein levels (Fig. 2C). A proximity ligation assay (PLA) confirmed the close proximity of CTPS1 and SNAP29 during filament formation, which yields a signal when two proteins of interest are within a 40-nm distance (Fig. S3B). In our previous study, we found that the in vivo enzymatic activity of CTPS was reduced with filament formation induced by His (Lin et al., 2018). However, in SNAP29-knockdown HEp-2 cells, the activity of CTPS was less affected by His addition, which might be due to the reduced filament formation (Fig. 2D; Figs S3C and S2D). Furthermore, the effect of SNAP29 siRNA, which relaxed the suppression of CTPS enzymatic activity, was reversed by overexpression of an siRNA-resistant SNAP29 construct (Fig. S3E,F). Collectively, the formation of CTPS filament mediated by SNAP29 is required to control the enzymatic activity of CTPS for adapting to Gln depletion stress.

We further used electron microscopy (EM) to examine CTPS filaments by anti-Flag gold labeling in HEp-2 cells expressing FLAG–CTPS–APEX2 under His induction, and some of the filament-like structures were labeled with gold particles (Fig. 2Ea–c; Fig. S3Ga). Further attempts to understand the interaction between CTPS and SNAP29 using immunoEM revealed that some SNAP29 signal was detected near the filamentous structures (Fig. 2Ed,e; Fig. S3Gb). Other membranous organelles also showed positive signals for SNAP29 (Fig. 2Ef). Approximately 36% and 40% of gold particles were labeled on filament-like structures for FLAG–CTPS1–APEX2 and FLAG–SNAP29, respectively (Fig. S3H). Even though filament-like structures were distinct under immunoEM, and some of them were positive for CTPS signals, we cannot exclude the possibility of the presence of other polymers in the same area.

SNAP29 mutation in humans leads to cerebral dysgenesis, neuropathy, ichthyosis and keratoderma (CEDNIK) syndrome, which is related to the defective transportation of components in keratinocytes during epidermal differentiation (Fuchs-Telem et al., 2011; Sprecher et al., 2005). Intriguingly, we found many keratin proteins were identified in our iTRAQ analysis comparisons between HIS and APEX, HIS and EBSS, and HIS and GLN (Fig. S2F, Table S1). Arguably, keratin proteins are well known contaminants in mass spectrometry. However, Gene Set Enrichment Analysis (GSEA) for proteins identified using a TMT labeling assay also showed enrichment for keratin filament proteins (Fig. S3I, Table S3) (Mootha et al., 2003; Subramanian et al., 2005). Taken together, a pan-cytokeratin antibody was initially used to observe whether CTPS filaments were located along the cytokeratin network in HEp-2 cells. Indeed, we observed CTPS filaments along the cytokeratin track on super-resolution images (Fig. 3A,B; Fig. S4A). Imaging at different time points, CTPS displayed a spatiotemporal association along the cytokeratin network (Fig. 3B). Under G(−)S(−) conditions, CTPS and IMPDH2, an enzyme of the purine biosynthesis pathway, assembled into filamentous structures along the cytokeratin network (Fig. S4B,C). However, it is unclear whether IMPDH2 filament assembly was dependent on CTPS as partial knockdown of CTPS did not affect IMPDH2 filament assembly in glutamine depletion medium (Fig. S4D,E). CTPS filaments were also found on the cytokeratin network in HeLa cells (Fig. S4F). There are more than 50 isomers of keratin that are subdivided into type I (K9–K28, K31–K40) and type II (K1–K8, K71–K80, K81–K86) intermediate filaments, and they form obligate heterodimers (Loschke et al., 2015). Keratins have been shown to be involved in the stress response, such as the role of keratin 8 (KRT8) in autophagy under oxidative stress (Baek et al., 2017). We decided to knockdown a few of the epithelial keratins identified in our proteomic analysis and found that shRNA-mediated knockdown of KRT8 significantly reduced the formation of CTPS filaments (Fig. 3C,D; Fig. S5A,C). K8 and K18 are most common in simple epithelia; they form a network via the assembly of heterodimers into non-polar unit-length filaments (ULF) and into intermediate filaments (Snider and Omary, 2014; Windoffer et al., 2011). However, knockdown of KRT18 in HEp-2 cells did not affect CTPS filament formation (Fig. 3C,D; Fig. S5B,C). Partial knockdown of KRT8 significantly reduced the cytokeratin network in HEp-2 cells when immunostained using a pan-cytokeratin antibody (Fig. S5D). By contrast, knockdown of KRT18 did not reduce the fluorescence intensity of the cytokeratin network (Fig. S5D), suggesting that KRT18 might be redundant, as KRT8 can interact with other keratins. The fold changes for KRT8 and KRT18 in the TMT labeling assay were not significant enough for them to be identified as candidates for being His-regulated CTPS interactors, possibly due to their abundant interaction with CTPS under both EBSS and EBSS+His conditions.

To further understand the proximity of CTPS to cytokeratin, we used a PLA. We found that both the keratin isotypes KRT8 and KRT18, were close to CTPS (Fig. 3E), and SNAP29 was also close to KRT8 (Fig. 3E), suggesting a possible interaction of SNAP29 and CTPS on cytokeratin. Immunofluorescence imaging confirmed the colocalization of GFP-tagged CTPS with mCherry-tagged KRT8 and KRT18 (Fig. 3F; Fig. S5E) and also endogenous CTPS with KRT8 in HEp-2 cells (Fig. S5F). Similarly, in HEK 293T cells, FLAG–CTPS1–GFP filaments were colocalized with mCherry-tagged KRT18 under DON treatment (Fig. S5G). We used mCherry–KRT18 or mCherry–KRT8 together with FLAG–CTPS1–GFP to monitor CTPS filament assembly in live cells under EBSS+His conditions. Live imaging of these cells revealed an association between CTPS1 and the KRT8–KRT18 network (Movies 1–3). A longer imaging time showed that the CTPS signal grew along the cytokeratin to give thicker filaments (Fig. 3G; Fig. S5H, Movies 2 and 3). Furthermore, using EM, we found that gold labeling of endogenous KRT8 can be detected on DAB-stained FLAG–CTPS–APEX2 filaments in HEp-2 cells (Fig. 3H; Fig. S6A). As expected, the percentage of KRT8 immunogold on filament-like structures did not change between DMEM and EBSS+His conditions (Fig. S6B).

Moreover, we found that disrupting the cytokeratin network with 8% 1,6-hexanediol (1,6-HD) dissembled the CTPS filaments completely within a few minutes (Fig. 4A–C) (Lin et al., 2016). However, the possibility that 1,6-HD dissembles CTPS filaments directly was not excluded based on our results. It was interesting that when cells recovered for an hour in EBSS+His medium after 10 min of treatment with 1,6-HD, CTPS filaments were observed adjacent to the reformed keratin filaments (Fig. 4C). Other cytoskeletal proteins, such as actin and tubulin, showed no colocalization with CTPS filaments, and their disruption had no effect on CTPS filament formation (Fig. 4D–F), which is consistent with results from a previous study (Carcamo et al., 2011).

Since PLA signals were detected for both CTPS and SNAP29 with cytokeratin (Fig. 3E), we tested whether SNAP29 affects the assembly of CTPS on cytokeratin. In SNAP29-knockdown cells, dotted CTPS signals were detected after 1 h of His induction, whereas in control cells, numerous CTPS filaments were already assembled along the cytokeratin network (Fig. 5A). Furthermore, single Z-section imaging showed that the PLA signal of CTPS and Flag–SNAP29 was located proximal to the cytokeratin network (Fig. 5B). To understand this association, we C-terminally tagged KRT18 with APEX2 to biotinylate the cytokeratin network, which is assumed to be the platform where CTPS filament assembly-related events occur (Fig. S6C). ImmunoEM anti-Flag gold labeling detected FLAG–CTPS–GFP along DAB-stained KRT18–APEX2 filaments in HEp-2 cells during the CTPS filament formation process (Fig. 5C; Fig. S6D). The percentage of Flag immunogold labeling on KRT18 DAB-stained filaments was significantly increased in EBSS+His condition when compared to the DMEM condition (Fig. S6E). Furthermore, KRT18–APEX2 revealed that the proximity of SNAP29 to the cytokeratin network increased spatiotemporally in response to stress (Fig. 5D; Fig. S6F). To confirm that SNAP29 availability can affect the process of CTPS filament assembly, we used N-ethylmaleimide (NEM), a non-selective thiol alkylator that has been reported to inhibit SNARE recycling and trap SNAP29 in the SNARE complex (Abada et al., 2017; Glick and Rothman, 1987). Indeed, we found that NEM treatment in HEp-2 cells led to the rapid fragmentation of CTPS filaments within 5 min without disrupting the cytokeratin network (Fig. 5E; Fig. S6G), suggesting that SNAP29 availability might be important for CTPS filament formation and maintenance. KRT18–APEX2-mediated proximity labeling showed that SNAP29 proximity to the cytokeratin network increased with NEM treatment, which was added during the last 10 min of biotin labeling (Fig. 5F). Using super-resolution imaging, we found a considerable amount of SNAP29 signal associated with the cytokeratin network, which increased with NEM treatment (Fig. 5G,H). Consistent with this, more SNAP29 was co-immunoprecipitated with KRT8 upon NEM treatment (Fig. 5I). Super-resolution imaging suggests that during CTPS filament formation, SNAP29 interacts with CTPS along cytokeratin network (Fig. 5Ja–e). It is plausible that with NEM treatment, SNAP29 recycling along the cytokeratin network is impaired, leading to its reduced availability for CTPS filament assembly and maintenance (Fig. 5Jf–j). Taken together, our data suggest that CTPS filament formation is dynamic and might be regulated by a balance of assembly and disassembly, in which NEM could intervene. Moreover, this assembly along cytokeratin might require proper conformation of the CTPS protein, because the G148A CTPS mutant, which cannot form tetramers could not assemble on the cytokeratin network (Fig. S6H), which is consistent with results from previous studies (Barry et al., 2014; Huang et al., 2017; Lin et al., 2018; Noree et al., 2014).

Here, we report that human CTPS filaments are compartmentalized along the cytokeratin network during nutrient starvation stress. We found that in response to stress, SNAP29 interacts spatiotemporally with the cytokeratin network. This availability of SNAP29 near the cytokeratin network might play a key role in CTPS filament formation and/or maintenance. In addition, SNAP29 is required for the regulation of CTPS enzymatic activity through protein assembly. We also found that IMPDH2 compartmentalizes along the cytokeratin network under glutamine and serum starvation, suggesting that during stress, cytokeratin might serve as a site for the compartmentalization of important metabolic enzymes, modulating their activity by polymerization (Lin et al., 2018).

In Caulobacter crescentus, CtpS has been reported to serve as a cytoskeletal filament to regulate cell curvature through interacting with a cell-shape regulator protein, crescentin, which has intermediate filament-like properties (Ingerson-Mahar et al., 2010). In humans, there are six types of intermediate filament that are widely categorized on the basis of sequence identity (Snider and Omary, 2014). Cytokeratins, which represent the largest family among all types of intermediate filaments, are the most diversified intermediate filament members. They have been reported to be involved in stress responses (Baek et al., 2017; Maruthappu et al., 2017; Snider and Omary, 2014), and now we have shown that knockdown of KRT8 interferes with the assembly of CTPS filaments under Gln deprivation stress. Immunofluorescence imaging and inhibitor treatment indicated that actin and microtubules do not associate with CTPS filaments (Fig. 4D–F). Moreover, simultaneous assembly and disassembly of the cytokeratin network and CTPS filaments were observed with the addition and removal of 1,6-HD, respectively (Fig. 4C) (Lin et al., 2016). These results suggested a possibility that CTPS filament formation is related to the cytokeratin network.

The combination of the APEX approach and His induction system allowed us to identify proteins that are not only adjacent to the CTPS filaments but also interact dynamically with CTPS, and hence could be involved in the process of filament formation. However, the enrichment of these two types of proteins depends on the dynamic nature of filament formation, given that CTPS filament formation is not synchronized in every cell. SNAP29, one of three candidates identified to regulate CTPS filament formation from our initial screen (Fig. 1E–G; Fig. S2F), was also verified in the repeat TMT analysis (Fig. 2A) and was found to co-immunoprecipitate with KRT8 (Fig. 5I). According to the results of KRT18–APEX2-based proximity analysis, immunofluorescence imaging of the SNAP29–CTPS filament–cytokeratin network, and the effects of NEM treatment, we proposed a model in which stress-dependent availability of SNAP29 on the cytokeratin network facilitates the assembly of CTPS into filamentous structures along these networks (Fig. 6). Indeed, several roles of intermediate filament proteins have recently emerged in the regulation of vesicle trafficking (Margiotta and Bucci, 2016). The contribution of histidine to the methionine cycle, and that methylation is required for filament formation, were demonstrated in our previous study (Lin et al., 2018). We think that SNAP29 function in CTPS filament formation requires histidine-mediated effects, which could be post-translational modifications of the CTPS protein or other filament-related proteins. It is well known that post-translational modification is critical for functions of the cytokeratin network (Snider and Omary, 2014). Interestingly, according to the KRT18–APEX2 results, the proximity between keratin 18 and CTPS was not significantly altered by stress during filament formation, suggesting that SNAP29 does not directly affect the relative location of CTPS in the cytokeratin network. The detailed molecular mechanism of CTPS filament assembly needs further investigation. Our present study might also further shed light on the CEDNIK condition, given that loss of SNAP29 function leads to defective skin development (Mastrodonato et al., 2018).

Human HEp-2, HeLa and HEK 293 cells were cultured in in Dulbecco's modified Eagle's Medium (DMEM) (GIBCO) supplemented with 10% fetal bovine serum (FBS) and 1× antibiotic-antimycotic (GIBCO) at 37°C under 5% CO₂.

The antibodies were anti-pan-cytokeratin (Abcam, cat. no. ab86734), anti-tubulin antibody (Abcam, cat. no. ab6160), SNAP29 antibody (GeneTex, cat. no. GTX131028), anti-SNAP29 antibody (Abcam, cat. no. ab181151), anti-IMPDH2 (Proteintech, cat. no: 12948-1-AP), anti-CTP synthase (Santa Cruz Biotechnology, cat. no. sc-134457), anti-CTPS antibody (GeneTex, cat. no. GTX105265), anti-KRT18 antibody (Santa Cruz Biotechnology, cat. no. sc-6259), anti-KRT8 antibody (Santa Cruz Biotechnology, cat.no. sc-8020), anti-KRT8 antibody (Proteintech, cat.no. 10384-1-AP), donkey polyclonal secondary antibody to mouse IgG - H&L (Abcam, cat. no. ab105278), monoclonal ANTI-FLAG^® M2 antibody (Sigma-Aldrich, cat. nos F3165 and F1804), streptavidin–Alexa Fluor 488 conjugate (Thermo Fisher Scientific, cat.no. S11223), Alexa Fluor 488–Phalloidin (Thermo Fisher Scientific, cat. no. A12379) and EasyBlot anti rabbit IgG conjugated to HRP (GeneTex, cat. no. GTX221666-01). Dilutions used are provided in Table S4. The pharmacological inhibitors used were nocodazole (Sigma-Aldrich, cat.no. M1404), cytochalasin D (Thermo Fisher Scientific, cat.no. PHZ1063), 1,6-hexanediol (Sigma-Aldrich, cat.no. 240117), 6-diazo-5-oxo-L-norleucine (DON) (Sigma-Aldrich, cat.no. D2141) and N-ethylmaleimide (NEM) (Sigma-Aldrich, cat.no. E3876) at concentration and duration as indicated in figure legends.

As mentioned in Lin et al. (2018), APEX2 was cloned from ‘APEX2-Actin in pEGFP’ (Addgene plasmid # 66172, deposited by Alice Ting) and inserted into the BamHI site of the p3xFlag-CTPS1-CMV26 vector to generate the p3xFlag-CTPS1-APEX2-CMV26 vector. For the p3xFlag-APEX2-CMV26 vector, APEX2 was cloned from p3xFlag-CTPS1-APEX2-CMV26 and inserted into an empty p3xFlag-Myc-CMV26 with NotI and BamHI to generate the p3xFlag-APEX2-CMV26 vector. For the generation of the p3xFlag-CTPS1-GFP-CMV26 vector, GFP was cloned from the pLVX-EF1alpha-CTPS-AcGFP-N1 vector [A206K was mutated in AcGFP to prevent dimer formation (Chang et al., 2018; Zacharias et al., 2002)] and inserted into p3xFlag-CTPS1-APEX2-CMV26 with BamHI sites to replace APEX2 with GFP. Human SNAP29, KRT18 and KRT8 were amplified from RNA extracted from HEp-2 cells. SNAP29 was inserted into p3xFlag-Myc-CMV26 with NotI and XbaI sites to generate the p3xFlag-SNAP29-CMV26 vector. For the generation of the siRNA-resistant SNAP29 construct, three nucleotides were point mutated (A678G, T681C and G684A) on the siRNA recognition site. Primers for site-directed mutagenesis were designed using QuikChange primer design tool (Agilent). KRT18 and KRT8 were inserted into pmCherry-N1 (Addgene #54517) with NheI and HindIII sites to generate pmCherry-KRT18 and pmCherry-KRT8. CTPS2 was cloned from HEp-2 cell RNA extract and inserted into empty p3xFlag-Myc-CMV26 vector with XbaI and BamHI to generate p3xFlag-CTPS2-CMV26. For the generation of the CTPS1-APEX2-MIGR1 vector, CTPS-APEX2 was cloned from the p3xFlag-CTPS1-APEX2-CMV26 vector using the XhoI site and inserted into the empty MIGRI vector (a gift from Chien-Kuo Lee, Graduate Institute of Immunology, National Taiwan University, Taiwan). For the generation of the KRT18-APEX2 vector, APEX2, including the stop codon, was cloned from the p3xFlag-CTPS1-APEX2-CMV26 vector using the HindIII site and inserted into the pmCherry-KRT18 vector. For exogenous expression, 5×10⁵ cells were transfected at 60–70% confluence using Lipofectamine 2000 (Invitrogen) following the manufacturer's instructions.

Lentiviral shRNAs targeting LacZ (TRCN0000072224), SNAP29 (TRCN0000231852, TRCN0000381551), CAPG (TRCN0000029463), UBL5 (TRCN0000011115), KRT8-(TRCN0000062384), KRT18 (TRCN0000299482) and other candidate genes were purchased from RNAi core (NCR), Academia Sinica, Taiwan. A total of 6×10⁵ cells were infected with 3.6×10⁶ colony forming units (CFU)/ml recombinant viral fluid together with 8 µg/ml of polybrene. At 24 h post infection, cells were replaced with fresh medium containing 2 µg/ml puromycin. The selection process was repeated one more time before establishing the stable gene knockdown cell line. RT-qPCR primers were as follows: SNAP29 (FW, 5′-CCTGAACAGAATGGCACCCT-3′; REV, 5′-TGGGGACAGGGTCTGTATCA-3′); UBL5 (FW, 5′-AGCTGATTGCAGCCCAAACT-3′; REV, 5′-TCGTGTACCACTTCTTCAGGACAA-3′), and CAPG (FW, 5′-CCTGAACAGAATGGCACCCT-3′; REV, 5′-TGGGGACAGGGTCTGTATCA-3′).

For immunostaining, 6×10⁴ HEp-2 cells were seeded in 100-mm non-treated glass cover slips in 24-well plates for 24 h. To induce CTPS filaments, the cell culture medium was replaced with conditional medium [EBSS, EBSS+His (200 µM concentration of His, unless otherwise mentioned); denoted G(−)S(−)]. Cells were washed twice with 1× PBS and fixed with fixation buffer (4% formaldehyde and 4% sucrose diluted in 1× PBS) for 10 min. The fixed cells were then washed twice with 1× PBS and permeabilized with 100% ice cold (−20°C) methanol or acetone at room temperature for 2 min. Subsequently, the cells were again washed twice with 1× PBS and blocked in blocking buffer (3% BSA and 0.2% Triton X-100 in 1× PBS) at room temperature for 20 min before incubating with primary antibody diluted in blocking buffer (without 0.2% Triton X-100) overnight at 4°C. The cells were washed three times for 5 min each with ‘Wash buffer’ (1× PBS containing 0.2% Tween-20) and incubated with secondary antibody for 1 h at room temperature. Subsequently, the cells were washed three times with wash buffer for 5 min each, and then the cells were mounted on glass plates with multi medium containing DAPI.

Cells transfected with FLAG–CTPS1–APEX2 or FLAG–APEX2 were incubated in EBSS or EBSS+His (200 µM) for 6 h at 36 h post transfection. For the EBSS+His+Gln group, cells were initially incubated under EBSS+His conditions, and during the last 15 min of the 6-h incubation, 4 mM Gln was added dropwise to the medium. After 6 h of incubation, each group was replaced with their corresponding conditional medium containing 500 µM biotin phenol (Iris Biotech, LS3500). Cells were further incubated at 37°C under 5% CO₂ for 30 min and then treated with H₂O₂ (at a final concentration of 1 mM) for 1 min. The reaction was quenched using a quencher solution containing 10 mM sodium ascorbate (A7631, Sigma-Aldrich), 10 mM sodium azide (S2002, Sigma-Aldrich) and 5 mM Trolox (238813, Sigma-Aldrich). Cells were subsequently processed for immunostaining or immunoprecipitation.

After biotin labeling, cells were scraped using denaturing lysis buffer (1% SDS, 50 mM Tris-HCl pH 7.4 and 5 mM EDTA) containing 1× protease inhibitor cocktail (Complete Mini, Roche) at room temperature to completely denature the proteins. Afterwards, non-denaturing buffer (1% Triton X-100, 50 mM Tris-HCl pH 7.4, 300 mM NaCl and 5 mM EDTA) was added to the lysed samples to dilute the SDS concentration to 0.1%. Subsequently, experimental groups were sonicated, and protein concentrations were measured using the Bradford method (Protein Assay, Bio-Rad). Biotinylated proteins were then pulled down using streptavidin-coated magnetic beads (Pierce, cat. no. 88816). To detect biotinylated proteins by western blot analysis, the blot was blocked with 5% BSA in Tris-buffered saline containing 0.1% Tween 20 overnight at 4°C and later incubated with streptavidin–HRP (Thermo Scientific, cat. no. 21126) for 1 h at room temperature. The blot was further washed in blocking buffer five times for 5 min each before developing with Western Lightning ECL Pro (PerkinElmer).

Immunoprecipitated biotinylated proteins were eluted from the streptavidin-coated magnetic beads using 0.1% trifluoroacetic acid (TFA)/50% acetonitrile (ACN) and dried in a Speed-Vac followed by in-solution digestion. Samples were dissolved in 250 mM triethylammonium bicarbonate (TEABC) and then reduced with 5 mM tris(2-carboxyethyl)-phosphine (TCEP) at 60°C for 1 h. Further samples were alkylated with 10 mM methyl methanethiosulfonate (MMTS) for 30 min at room temperature and subsequently digested with sequencing grade modified porcine trypsin (20 µg/ml) (Promega) overnight at 37°C. Tryptic digested peptides were dried by Speed-Vac for further iTRAQ labeling. Tryptic digested peptides were dissolved in 50 mM TEABC and then labeled with iTRAQ reagent (Applied Biosystems) for 1 h at room temperature. Labeled peptides were mixed at a 1:1:1 ratio and further desalted for 2D LC-MS/MS analysis.

In Lin et al. (2018), we previously described the method for LC-MS/MS analysis and database searching. The peptide mixture for iTRAQ was separated and analyzed by 2D LC-MS/MS using a strong cation exchange (SCX) and reverse-phase C18 (RP18) liquid chromatography system on a Dionex UltiMate 3000 nano LC system coupled to a LTQ-Orbitrap Elite mass spectrometer (Thermo Fisher Scientific). The peptide mixture was reconstituted in HPLC buffer A [30% ACN and 0.1% formic acid (FA)] and loaded onto a homemade column (Luna SCX, bead size, 5 µm; column dimensions, 180×0.5 mm for iTRAQ experiment; 130×0.5 mm for label-free experiment) (Phenomenex Inc., Torrance, CA) at a flow rate of 5 µl/min for 30 min. The peptides were then fractionated into 44 fractions for iTRAQ, using a continuous HPLC buffer B gradient (0–100% of 0.5 M ammonium chloride in the presence of 30% CAN and 0.1% FA). Each fraction was then mixed with a stream of 0.1% FA in H₂O, and the peptides were trapped on a Zorbax C18 column (bead size, 5 µm; pore size, 30 nm; column dimensions, 5×0.3 mm) (Agilent Technologies Inc., Santa Clara, CA) and separated on a 60-min (for iTRAQ) linear gradient of 99.9% CAN and 0.1% FA on a Hydro RP chromatography column (bead size, 2.5 µm; pore size, 10 nm; column dimensions, 200×0.075 mm) (Phenomenex Inc.). MS/MS analysis was performed on an LTQ-Orbitrap Elite mass spectrometer. Full-scan MS spectra (m/z 400 to m/z 2000) were acquired on the mass analyzer at a resolution of 60,000 at m/z 400, followed by MS/MS of the six most intense precursor ions, above a threshold of 5000, selected for fragmentation by collision induced dissociation (CID), in addition to high-energy collision dissociation (HCD) for the iTRAQ experiment in parallel acquisition mode with a normalized collision energy setting of 35% and an activation time of 10 ms for CID and 0.1 ms for HCD. The dynamic exclusion function was set as: repeat count, 1; repeat duration, 30 s; and exclusion duration, 40 s. Proteome Discover (version1.4) (Thermo Fisher Scientific) was used to analyze and quantify MS and MS/MS data. Swiss-Prot database containing 20,205 entries for Homo sapiens (download in September 2015) was used for identifying the proteins. Parent and fragment ion mass tolerance for CID were set to 10 ppm and 0.5 Da and for HCD to 10 ppm and 0.05 Da, respectively. Two missed cleavages were allowed for tryptic digestion. For protein identification, oxidation of methionine, protein N-terminal acetylation and pyro-glutamination for N-terminal glutamine were set as variable modifications, whereas methylthio modification of cysteine was set as a fixed modification. Additionally, iTRAQ labeling of lysine and the N-termini of peptides were also added. Peptide identification criteria were set as: peptide confidence, high; peptide length, 7–100; peptide maximum rank, 1; search engine rank, 1; minimal number of peptides, 2 for proteins; count only rank 1 peptide; count peptides only in top-scoring proteins; and false discovery rate (FDR) <0.01. In total, 3417 proteins were quantified. To correctly identify CTPS filament-interacting proteins, iTRAQ ratios between the experimental groups and control groups were calculated for each protein to generate three different datasets of iTRAQ ratios (i.e. 115/117, 115/114 and 115/116). We set the cutoff at the mean+1.8 s.d. to minimize the false positives (Table S1).

Cells transfected with FLAG–CTPS1–APEX2 were incubated in EBSS or EBSS+His (200 µM) for 6 h, after 24 h of transfection in triplicate. During the last 30 min of the 6-h incubation, biotin phenol (500 µM) was directly added to the medium. Cells were further incubated at 37°C under 5% CO₂ for 30 min and then treated with H₂O₂ (at a final concentration of 1 mM) for 1 min. The reaction was quenched using a quencher solution containing 10 mM sodium ascorbate, 10 mM sodium azide and 5 mM Trolox. Cells were subsequently processed for immunostaining or immunoprecipitation.

After biotin labeling, cells were scrapped using denaturing lysis buffer (1% SDS, 50 mM Tris-HCl pH 7.4 and 5 mM EDTA) containing 1× protease inhibitor cocktail (Complete Mini, Roche) at room temperature to completely denature the proteins. Afterwards, modified non-denaturing buffer (0.2% NP40, 50 mM Tris-HCl pH 7.4, 137 mM NaCl and 5 mM EDTA) was added to the lysed samples to dilute the SDS concentration to 0.1%. Subsequently, experimental groups were sonicated, and the protein concentration was measured using the Bradford method. Biotinylated proteins were then pulled down using streptavidin-coated magnetic beads. Beads were subsequently washed six times with RIPA lysis buffer (0.8% NP40, 50 mM Tris-HCl pH 7.4, 137 mM NaCl, 0.1% SDS and 5 mM EDTA), twice with 1× PBS and transferred to new tubes. One-tenth of the beads were retained for checking via western blotting.

Biotinylated proteins from the EBSS and His groups were immunoprecipitated using streptavidin-coated magnetic beads. The immunoprecipitates were eluted using 80% trifluoroethanol (TFE)/0.1% trifluoroacetic acid (TFA) and dried in a Speed-Vac. Dried samples were dissolved in 250 mM triethylammonium bicarbonate (TEABC), reduced with 5 mM tris(2-carboxyethyl)phosphine (TCEP) at 60°C for 1 h, alkylated with 10 mM methyl methanethiosulfonate (MMTS) for 30 min at room temperature and digested with sequencing grade modified porcine trypsin (20 µg/ml) (Promega, Madison, WI) overnight at 37°C. Tryptic digested peptides were dried in a Speed-Vac, dissolved in 100 mM TEABC and subjected to TMT labeling according to the manufacturer's instruction (Thermo-Fisher Scientific). Labeled peptides were mixed and further desalted for 2D LC-MS/MS analysis.

The peptide mixture was separated and analyzed by 2DLC-MS/MS using a strong cation exchange (SCX) and reverse-phase C18 liquid chromatography system on a Dionex UltiMate 3000 nano LC system coupled to a Orbitrap Fusion™ Lumos™ Tribrid™ Mass Spectrometer (Thermo-Fisher Scientific). The peptide mixture was reconstituted in HPLC buffer A (30% ACN and 0.1% FA) and loaded onto a homemade column (Luna SCX, bead size, 5 µm; column dimensions, 200×0.254 mm) (Phenomenex Inc., Torrance, CA) at flow rate of 2.5 µl/min for 20 min. The peptides were then fractionated into 26 fractions, using a continuous HPLC buffer B gradient (0–100% of 1 M ammonium nitrate in the presence of 25% ACN and 0.1% FA). Each fraction was then mixed with a stream of 0.1% FA in H₂O, and the peptides were trapped on a Zorbax C18 column (bead size, 5 µm; pore size, 30 nm; column dimensions, 5×0.3 mm) (Agilent Technologies Inc., Santa Clara, CA) and separated on a 60-min linear gradient of 99.9% ACN and 0.1% FA on a Hydro RP chromatography column (bead size, 2.5 µm; pore size, 10 nm; column dimensions, 200×0.075 mm) (Phenomenex Inc.). MS/MS analysis was performed on a Orbitrap Fusion Lumos mass spectrometer. Five full-scan MS ranges, including m/z 420–566, 562–652, 648–734, 730–844 and 840–1500, were acquired on the mass analyzer at a resolution of 60,000 at m/z 200, followed by MS/MS of the eight most-intense precursor ions using high-energy collision dissociation (HCD). Masses selected for MS/MS were isolated at a width of 0.7 m/z and fragmented with a normalized collision energy setting of 35%. The dynamic exclusion function was set as: repeat count, 1, and exclusion duration, 40 s. Proteome Discoverer (version 2.2, Thermo Fisher Scientific) was used to analyze the MS and MS/MS data. The Swiss-Prot database containing 20,259 entries for Homo sapiens (downloaded in March 2018) was used for identifying the proteins. Parent and fragment ion mass tolerance were set to 10 ppm and 0.03 Da, respectively. Two missed cleavages were allowed for tryptic digestion. For protein identification, oxidation of methionine, protein N-terminal acetylation and pyro-glutamination for N-terminal glutamine were set as variable modifications, whereas methylthiolation of cysteine was set as a fixed modification. Additionally, TMT labeling of lysine and the N-termini of peptides was added. Peptide identification criteria were set as: peptide confidence, high; minimum peptide length, 6; and FDR <0.01. In total, 3404 high confidence proteins containing at least two unique peptides were quantified. The median ratio of the quantified proteins between the experimental group (His) and control group (EBSS) in triplicate and corresponding P-value were calculated (Table S2). We set the cutoff at mean+2 s.d. and P-value ≤0.05 for selecting candidate proteins that might be significant interacting proteins of CTPS filaments.

Proteins identified in TMT labeling assay were used for Gene Ontology analysis. GSEA for biological process was performed using GESA_4.0.1 software utilizing MSigDB version 7.0 (Table S3) (Mootha et al., 2003; Subramanian et al., 2005).

For routine protein detection using western blotting, cells were lysed using ice-cold RIPA lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.1% SDS, 1% TritonX-100 and 5 mM EDTA) containing 1× protease inhibitor on ice for 30 min. Samples were sonicated, and the protein concentration was measured before running SDS-PAGE. Proteins were transferred from SDS-PAGE gels to PVDF membranes and then blocked with blocking buffer [7% nonfat milk in Tris-buffered saline with 0.1% Tween 20 (TBST)] for 1 h. Afterwards, membranes were incubated overnight with primary antibody in blocking buffer at 4°C and then washed thrice with TBST for 5, 10 and 15 min. Membranes were then incubated with secondary antibody in blocking buffer at room temperature and subsequently washed thrice in TBST before development using ECL Pro.

A total of 8×10⁵ cells were transfected with SNAP29 and scrambled siRNA using Lipofectamine 2000 (Thermo Fisher Scientific, cat. no. 11668019). The siRNA sequence for SNAP29 was 5′-AGACAGAAAUUGAGGAGCA-3′. At 24 h post transfection, cells were re-seeded into three groups (DMEM, EBSS and EBSS+His). SNAP29 knockdown was confirmed using western blotting, and simultaneous immunostaining with anti-CTPS antibody was performed to reconfirm that SNAP29 knockdown cells had reduced CTPS filaments in EBSS+His (200 µM) at 6 h. For UTP and CTP measurement, cells were incubated with ¹³C¹⁵N-uridine (100 µM) for 1 h before harvesting with methanol. Supernatants were further dried and re-suspended in 2 mM dibutylamine and 1.5 mM formic acid. Cell extracts were analyzed by LC-MS (LTQ-orbitrap, Thermo). Details of the method were previously described in Lin et al. (2018). For the overexpression experiment, a total of 5×10⁵ cells were transfected with SNAP29 and scrambled siRNA during seeding. At 16 h post transfection, cells were again transfected with WT-SNAP29 and siRNA-resistant SNAP29 constructs. After 24 h, cells were subjected to EBSS+His condition for 5 h followed by treatment with ¹³C¹⁵N-uridine (100 µM) for 1 h in the same medium.

A total of 6×10⁴ HEp-2 cells were seeded for PLA experiments. After fixation (4% formaldehyde and 4% sucrose in 1× PBS) and permeabilization (100% ice-cold methanol), PLA (Duolink PLA, DUO92101 Sigma) was performed based on the manufacturer's protocol.

Transfected cells were pre-fixed in 2% paraformaldehyde and 1.25% glutaraldehyde followed by post-fixation in 1% osmium tetroxide solution for 1 h (EMS Microscopy Academy). After dehydration in a graded series of ethanol, cell pellets were embedded by Spurr's resin and then polymerized in an oven at 70°C for 8 h. Ultrathin sections (70 nm) were cut using a Leica UC7 ultramicrotome and collected onto a nickel grid for further immunogold labeling. These sections were first treated with 10% H₂O₂ for 10 min, followed by 1% BSA to block non-specific binding. Subsequently, samples were incubated with an anti-Flag antibody (Sigma-Aldrich, cat.no. F3165) for 1 h. The 18-nm immunogold-conjugated antibody (Abcam) was used to detect the Flag-tagged protein. Finally, the sections were post-stained with 4% uranyl acetate for 10 min and rinsed several times with H₂O followed by 4% Reynolds lead citrate for 10 min. Micrographs were obtained at 100 kV in a JEM-1230 transmission electron microscopy (JEOL) with a Gatan Model 832 digital camera.

Cells transfected with FLAG–CTPS1–APEX2 or KRT18–APEX2 were incubated in EBSS+His (200 µM) or DMEM for 4 h. The protocol for DAB staining from a previous study was followed (Martell et al., 2017). The transfected cells were fixed in 2% glutaraldehyde followed by 0.5 mg/ml 3,3-diaminobenzidine (DAB) staining with 10 mM H₂O₂. Osmium tetroxide (1%) was used for post-fixation at 4°C for 30 min. Further dehydration and resin embedding were processed by standard EM sample preparation. Ultrathin sections (70 nm) were obtained by a Leica UC7 ultramicrotome. Uranyl acetate and Reynolds lead citrate (both at 4%) were used for negative staining of sections.

A total of 2×10⁶ cells were seeded overnight in 10-cm plates before replacing with conditional medium (EBSS+His) for 2 h. NEM at 1 mM was added to one of the plates for 10 min, and subsequently cells were scratched with KRT lysis buffer (20 mM Tris-HCl pH 7.4, 137 mM NaCl, 10% glycerol, 0.5% Triton X-100, 2 mM EDTA, 1× protease inhibitor cocktail, 1× phosphatase inhibitor cocktail) on ice. After 30 min of incubation on ice, lysed cells were sonicated and spun down at 15,000 g for 15 min to collect the supernatant. Protein A agarose beads (20 μl; Millipore 16-125) were added to each sample and incubated for 1 h at 4°C on a rotor for the pre-clearing assay. Samples were then spun at 13,200 g for 5 min to collect the supernatant. Next, the protein concentration was measured, and equal amounts of protein were used for the immunoprecipitation assay. Samples were rotated at 4°C overnight with anti-KRT8 antibody (Proteintech, cat.no. 10384-1-AP). Subsequently, samples were incubated with protein A beads for 1 h and then washed with KRT lysis buffer five times (1 min each). Beads were washed twice with 1× PBS and transferred to a new tube and further boiled with 2× SDS sample buffer.

Confocal images were acquired using a Zeiss Laser Scanning Confocal Microscope (LSM) 780 using Plan-Apochromat 100×1.40 Oil DIC M27 and Plan-Apochromat 20×/0.8 M27 lenses. For super-resolution images, ELYRA PS.1 super-resolution microscopy was used (Imaging Core, Academia Sinica, Taiwan). For live-cell imaging, images were acquired on a DeltaVision Ultra microscope (GE Healthcare) using a 60×1.42NA PlanApo N objective (Olympus) and an sCMOS camera; on a Nikon Ti2 Dragonfly High Speed confocal platform using a Nikon 100×/1.49 oil objective; and on LSM 780 confocal microscope using an Alpha Plan-Apochromat 100×/1.40 oil DIC M27 objective. Images and videos were processed using Fiji (ImageJ) (Schindelin et al., 2012) and Imaris (Bitplane).

For experimental data, a Student's two-tailed unpaired t-test was used for the analysis. Figure legends include all statistical details of the experiments.

We thank the Proteomics Core Laboratory at Chang Gung University for assistance with proteomic analysis. We thank the Imaging Core at Academia Sinica and Chang Gung University for assistance in super-resolution, live-cell and confocal imaging. We thank Sue-Ping Lee at Academia Sinica for helping us with super-resolution imaging. We thank the National RNAi Core Facility at Academia Sinica for providing shRNA reagents and related services. We thank FlyBase for providing information regarding the SNAP29 Drosophila gene. We thank Bloomington Stock Center and Fly Core Taiwan for providing fly stocks. We are grateful to the comments from Drs Mark Peifer, Robert Duronio, Laura Nilson, Juli Wu and Joerg Grosshans, Chien-Kuo Lee on the manuscript.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.240200.reviewer-comments.pdf