It remains an open question whether pH-regulated assembly is an inherent feature of budding yeast CTPS, or whether other cellular factors are required for assembly. To address this, we first confirmed prior work that showed yeast CTPS assembles in cells upon cytoplasmic acidification (Petrovska et al., 2014); Ura7-GFP tagged at the endogenous locus forms foci upon nutrient deprivation (Figure 2a), or when cells are permeabilized with DNP (2,4-dinitrophenol) to manipulate cytoplasmic pH (Figure 2b). We next examined polymerization of purified recombinant Ura7 and Ura8 at different pHs by negative stain EM. Apo CTPS did not form polymers at any pH. At pH 7.4 both isoforms assembled short, single filaments on binding substrates or product, but pH 6.0 promoted assembly of much larger polymers that appeared to be bundled filaments (Figure 2c, Figure 2—figure supplement 1). Thus, like other species (Barry et al., 2014; Lynch et al., 2017; Zhou et al., 2019), yeast CTPS assembly is dependent on ligand binding, and the pH sensitivity is intrinsic to the enzyme itself.

We wondered whether the intrinsic pH sensitivity of yeast CTPS was conserved among other species. Previous studies have not described the pH-sensitivity of human CTPS filament assembly, and have shown robust assembly at pH 8.0 (Lynch et al., 2017; Lynch and Kollman, 2020). The overall sequence conservation between human and yeast CTPS, and the presence of a histidine residue at the assembly interface in human CTPS filaments, led us to consider whether human CTPS assembly is also pH-dependent (Figure 1d). We observed equally robust polymerization of purified human CTPS2 (hCTPS2) at pH 6.0 and 8.0, and negative stain EM averages show that filaments formed at different pHs have similar architectures, suggesting that there is no direct effect of pH on hCTPS2 assembly or structure (Figure 2d, inserts). Thus, pH-driven assembly of CTPS appears to be specific to yeast, raising the question of what unique structural features might confer pH-sensitivity.

To assess the kinetics of pH-dependent CTPS assembly, we monitored right angle light scattering by CTPS after addition of CTP (Figure 2e). Both isoforms had very low signal at pH 7, consistent with our negative stain imaging. At pH 6, Ura8 assembly was much faster than Ura7, but inspection at early time points showed that both exhibited biphasic assembly which could not be fit with a single equation (Figure 2—figure supplement 2). Instead, early and late assembly kinetics fit well to separate four-parameter curves, indicating potentially distinct assembly phenomena with different kinetics in the early and late phases. We examined the growth of Ura7 filaments by negative stain EM over time, and found that single filaments appear at early time points and the thicker bundles at later time points, suggesting that the biphasic scattering kinetics can be explained by initial linear polymerization followed by lateral aggregation (Figure 2—figure supplement 2).

We next determined structures of individual CTPS filaments by cryo-EM. To enable structure determination, we assembled filaments at pH 6.5, where single filaments predominate over larger bundles. We solved filament structures of both Ura7 and Ura8 in substrate- (ATP and UTP) and product- (CTP) bound states. Ura8 yielded the highest resolution structures, at 2.8 Å (substrates) and 3.8 Å (products) (Figure 3a, Table 1). In the Ura8 structures, all ligands were clearly visible and bound as previously described for other CTPS homologs, including an imino-phosphate reaction intermediate recently observed in Drosophila CTPS, and the presence of two distinct CTP-binding sites per monomer recently reported in human and Drosophila CTPS (Endrizzi et al., 2004; Goto et al., 2004; Lynch et al., 2021; Lynch et al., 2017; Lynch and Kollman, 2020; Zhou et al., 2021; Figure 3—figure supplement 1c). Ura7 filaments reached lower resolutions, 7.3 Å (substrates) and 3.7 Å (products), but the overall structures were indistinguishable from Ura8 at these resolutions (Figure 3—figure supplement 2, Table 2). Because of their higher resolution which enabled building of atomic models, we focus subsequent structural interpretation on the Ura8 filaments.

Like other previously reported structures of human, Drosophila, and E. coli CTPS, the yeast enzyme assembles as stacked tetramers. However, it does so using a completely different interaction interface, which was the same in both substrate- and product-bound yeast structures (Figure 3b). Helix 356–370 in the glutaminase domain mediates interactions between tetramers, so that each protomer is involved in assembly contacts (Figure 3b). His360 interacts with D370 which would be stabilized by protonation of His360 at low pH, and likely explains the pH-sensitivity of the assembly interaction (Figure 3b, Figure 3—figure supplement 3). We see density consistent with two rotamer positions of H360 that may reflect partial protonation of the sidechain at pH 6.5. One rotamer that likely reflects the protonated state interacts with D370, and the other rotamer that likely reflects an unprotonated state forms a hydrogen bond to a backbone carbonyl across the interface (Figure 3—figure supplement 3a–c). Additional contacts include hydrogen bonding between an asparagine pair on the two-fold symmetry axis (Asn364), hydrogen bonding (Gln352) to a backbone carbonyl across the interface, hydrophobic interactions of a cluster of tryptophans (Trp363, Trp392), and a pair of salt bridges (Lys391, Glu395) (Figure 3b, Figure 3—figure supplement 2). Helix 356–370 also forms the assembly interface of animal CTPS filaments, but the yeast interface is shifted by two turns of the helix, resulting in completely different residue contacts (Figure 3c). This leads to a larger interaction interface in yeast CTPS filaments (706 Ų per monomer in product-bound filaments) than in the human homolog (492 Ų per monomer in product-bound filaments) (Figure 3—figure supplement 4). Sequence differences between human and yeast CTPS at residues 364, 391, and 395 may explain how the unique yeast interface arose, which shifted H360 into a position to mediate a pH-sensitive interaction.

The primary difference between the substrate- and product-bound filament structures is the conformation of individual CTPS protomers. Active and inactive conformations of CTPS are characterized by a ~ 7° rotation of the glutaminase domain relative to the amido-ligase domain, which opens an ammonia channel between the two active sites in a single protomer (Lynch and Kollman, 2020). The Ura8 product-bound filament is in the canonical inhibited conformation, but the substrate-bound structure is in a conformation intermediate between canonical active and inhibited states, in which the ammonia channel remains closed. This suggested to us that one function of yeast CTPS filaments is to constrain the enzyme in a low activity conformation. To test this hypothesis, we determined the structure of free Ura8 tetramers at pH 7.4 bound to substrates at 2.8 Å resolution. In this unassembled state, Ura8 adopts the canonical active conformation, very similar to hCTPS2, including opening of the ammonia channel and rearrangement at the tetramerization interface (Figure 3d–h, Figure 3—figure supplement 5, Figure 3—video 1).

The active state we observe in free tetramers is incompatible with filament assembly, because in this conformation filament interfaces cannot be occupied on both sides of the tetramer simultaneously (Figure 3—figure supplement 6). Indeed, the major source of heterogeneity in single Ura7-substrate filaments, which were flexible and short, are tetramers that only make one contact at each paired assembly interface. This leads to tetramers in the active, substrate-bound conformation tethered to each other through single glutaminase domain interactions (Figure 3—figure supplement 6d). Thus, filament assembly with symmetric contacts acts as a steric constraint to trap yeast CTPS in a low activity conformation, while tethering through a single glutaminase domain supports short, flexible filaments that accommodate a fully active conformation.

To test whether CTPS filament assembly has a direct effect on enzyme activity, we generated mutations at H360 in the assembly interface of Ura7 that either disrupted or stabilized filaments. First, we generated H360A, which disrupts polymerization of the human enzymes, and which our structural analysis suggests is the pH sensor in assembly of the yeast enzyme (Lynch et al., 2017; Lynch and Kollman, 2020). Consistent with those prior results, Ura7-H360A did not assembly filaments with substrates or products at low or high pH (Figure 4a). To mimic protonation at H360, we mutated it to arginine, with the expectation that a constitutive positive charge might support polymerization at neutral pH. Indeed, we found that Ura7-H360R robustly assembles filaments at pH 7.4 which, like the wild-type protein, assemble only in the presence of substrate or product ligands (Figure 4a, Figure 4—figure supplement 1). Recombinant wildtype and both mutant proteins had similar purity and yield (Figure 4—figure supplement 2a). To confirm that H360R does not alter filament structure, we determined a 6.7 Å cryo-EM structure of substrate-bound Ura7, which is indistinguishable from wildtype at this resolution (Figure 4—figure supplement 2b–f, Figure 4—figure supplement 3). Thus, reversible assembly of H360R into wiltype-like filaments is pH-insensitive.

Decoupling polymerization from pH in engineered mutations provides tools to determine the functional consequences of polymerization on enzyme activity. Yeast CTPS activity has a strong intrinsic pH-dependence that peaks around pH 8.0 (Nadkarni et al., 1995; Yang et al., 1994), likely due to pH sensitivity of the active site cysteine in the glutaminase domain (Trotta et al., 1973). Consistent with this, Ura7 wildtype, H360A, and H360R all have low activity at pH 6.0. However, H360A had approximately threefold higher activity, indicating that unassembled CTPS maintains residual activity even at low pH (Figure 4b). Wild-type CTPS and mutants retain activity at pH 7.4, where we measured the substrate kinetics of all three. The K_M values were similar for wildtype and both mutants, indicating no change in affinity for the substrate UTP. However, V_max varied inversely with the degree of filament assembly (Figure 4c, Table 3). At pH 7.4 Ura7-H360A is completely tetrameric and had the highest activity, wild-type enzyme has a background of single filaments and had slightly lower activity, and Ura7-H360R had robust filament assembly and the lowest observed activity. This suggests that filament assembly in itself acts as an allosteric inhibitor of enzyme activity, with kcat reduced likely as a consequence of the constricted ammonia channel we observe in the filament structure.

To investigate the in vivo consequences of CTPS assembly, we generated yeast strains with filament assembly mutations at the endogenous locus, either with or without fluorescent protein tags, and tested both enzyme localization and cell growth (Figure 5a). We did not observe differences in growth between wild-type strains with and without fluorescent tags; nonetheless, the growth assays described below were all performed with untagged strains.

Based on our in vitro findings, we predicted that the pH-insensitive Ura7-H360R would constitutively assemble polymers in cells. But surprisingly, its localization was very similar to wildtype, diffuse during log phase growth and assembled into foci upon nutritional deprivation. However, Ura7-H360R-GFP assembly and disassembly kinetics are dysregulated (Figure 5b–c, Figure 5—figure supplement 1, Figure 5—videos 1–4). Upon transition to starvation media, wildtype Ura7-GFP typically assembles foci over the course of 30 min, but Ura7-H360R-GFP assembles more rapidly, with virtually all cells having foci at 5 min. Even more striking, upon recovery from starvation, in which wild-type Ura7 disassembles in most cells after 30 min, at 1 hr we did not observe significant depolymerization of Ura7-H360R-GFP foci (Figure 5—figure supplement 1). The rapid assembly and slow disassembly of Ura7-H360R in cells are consistent with the in vitro pH-insensitive phenotype, but suggest that the starvation-triggered pH change is necessary but not sufficient to enable cellular Ura7 assembly. In addition to disrupted assembly kinetics, Ura7-H360R grew more slowly than wildtype, when plated either from log phase cultures or from starved cultures (Figure 5d–e), suggesting that disruption of normal Ura7 assembly and disassembly is generally detrimental to growth.

Consistent with our in vitro findings, cells expressing non-assembling Ura7-H360A-GFP or Ura8-H360A-GFP did not form foci, even under nutritional stress when the cytoplasm is acidified (Figure 5a). The non-assembly single mutants of Ura7 or Ura8 grew indistinguishably from wild type during log phase and upon recovery from starvation (Figure 5d–e, Figure 5—figure supplement 2). This suggested that polymerization of either CTPS isoform is sufficient to maintain normal growth. To test this, we generated the double mutant URA7-H360A/URA8-H360A, and found that it had a severe growth defect in log phase and upon starvation recovery (Figure 5d–e), indicating an important role for CTPS polymerization in proliferation.

We wondered whether filaments might play a role in protecting CTPS from degradation, and that perhaps the growth defect in the double H360A mutants is due to loss of the enzyme during starvation (Petrovska et al., 2014). To test this, we starved Ura7-GFP, Ura7-H360A-GFP, and Ura7-H360R-GFP strains and measured the concentration of CTPS in total cell lysate. Wildtype Ura7 levels are decreased during starvation, but there was no difference in protein levels between wild type and the mutants at 4 or 24 hr of starvation (Figure 5—figure supplement 3). This indicates that filament assembly does not play a significant role in protecting CTPS from degradation, and suggests that growth defects observed upon starvation recovery for URA7-H360A/URA8-H360A arise from other effects of defective polymerization.

It was surprising that strains with H360A or H360R mutations experienced slow growth relative to wildtype during log phase, when the mutant strains have the same diffuse localization seen in the wild-type strain. This may be consistent with the observation that wildtype enzymes have a low level of background assembly at neutral pH (Figure 2c), and indicates that small, transient assemblies may play a role in regulating activity during log phase growth.

In wildtype cells at 30 min after nutrient deprivation, cells that had CTPS foci almost all had just a single structure (Figure 2b), consistent with prior reports of CTPS polymerization in yeast (Noree et al., 2010; Petrovska et al., 2014; Shen et al., 2016). However, during the live cell imaging experiments described above, in cells expressing wildtype Ura7-GFP we frequently observed the appearance of multiple small puncta at early time points that later coalesced into a single large structure (Figure 6a). This in vivo behavior is reminiscent of the rapid linear polymerization followed by lateral aggregation into larger bundles that we observed in vitro (Figure 2—figure supplement 2).

We determined structures of CTPS filament bundles formed at pH 6.0 to better understand the mechanisms of lateral assembly. Our initial question was whether the bundles were aggregating non-specifically, or whether there were defined assembly contacts. Reference-free two-dimensional averages of bundle segments from cryo-EM images of Ura7 and Ura8 were strikingly regular, suggesting ordered assembly contacts. This high degree of order allowed us to determine three-dimensional structures of both homologs in the presence of substrates or product using a single particle reconstruction approach by focused refinement of interacting pairs of filaments (Figure 6b, Figure 6—figure supplement 1). As we found for the single filament structures, Ura8 reconstructions went to higher resolution (3.3 Å for substrates and 3.7 Å for product) than Ura7 (6.6 Å substrate and 7.0 Å product). The individual filaments that pack laterally in the bundles closely resemble the corresponding single filament structures (Figure 6—figure supplement 2, Table 2). We observed two different bundle architectures. Ura7 formed very similar bundles in both ligand conditions, with adjacent filaments staggered relative to each other with a half-tetramer offset (Figure 6c). The Ura8 substrate-bound structure had the same architecture as both Ura7 bundles, but the product-bound Ura8 structure had very different interfaces between filaments, giving rise to filaments in register (Figure 6d, Figure 6—videos 1–4).

To better understand the nature of lateral assembly contacts, we built atomic models into the two Ura8 bundle structures. Ligands were clearly visible, and bound as in the single filament structures, and the longitudinal assembly interfaces were the same as observed in the single filaments (Figure 6—figure supplement 2). Domain rotations and the state of the closed ammonia channel were nearly identical between filaments and bundles (Figure 6—figure supplement 3). Unlike single Ura8 filaments determined at an intermediate pH (6.5) and with two rotamers for His360, the bundle map appeared to fit a single rotamer which points to D370 (Figure 3—figure supplement 3d–e). Lateral associations between filaments result in tightly packed bundles. The buried surface area per tetramer at lateral interfaces, 2202 Ų (substrates-bound) or 1676 Ų (product-bound), is comparable to the 2460 Ų involved in longitudinal filament assembly contacts, suggesting that lateral association contributes to the overall stability of the assembly (Figure 6—figure supplement 4, Table 4). Yeast CTPS has a 7-residue insert (residues 273–279) in the linker region that mediates the bulk of the lateral interaction in both bundle types (Figure 6e). In the substrate-bound structure, Leu277 nestles into a hydrophobic pocket formed by Ile383/Ile387/Ile412/Phe429 (Figure 6f, Figure 6—figure supplement 5a,b). In the product-bound structure the linker inserts near the tetramerization interface and packs against Pro236 and Ile225 (Figure 6—figure supplement 5c–d). Although at lower resolution, the Ura7 structures appear to make the same lateral contacts as substrate-bound Ura8 (Figure 6—figure supplement 1). Similar filament bundles have not been reported for other species, despite extensive structural characterization of human, Drosophila, and E. coli CTPS filaments (Barry et al., 2014; Lynch et al., 2017; Lynch and Kollman, 2020; Zhou et al., 2021), suggesting that the unique yeast insert promotes lateral assemblies that are specific to yeast.

Although individual filament-to-filament lateral interactions are all identical, there is variability of the assembly architecture at longer scales. The dihedral symmetry of the CTPS tetramer presents two potential lateral interfaces on each face of single filaments, but steric constraints limit occupancy to just a single lateral interaction per face. 3D classification of bundles yielded multiple structures with 3–5 associated filaments, with mixtures of cis or trans configurations of laterally associated filaments accounting for the variation (Figure 6—figure supplement 8). To envision potential larger bundle architectures, we extrapolated the lateral contacts by propagating assemblies in silico (Figure 6—figure supplement 6). For the staggered bundle architecture, propagation of cis or trans interactions results in curved sheets. Propagation of trans interactions in the Ura8 product bound structure with filaments in register also results in a curved sheet, but propagation of cis interactions results in a closed tube with nine filaments. Mixed cis and trans interactions are also possible and observed in some of our 3-D classes, and give rise to increasingly complex structures.

URA7 and URA8 wild type genes with ribosomal binding site were cloned into pet28b-6His (Addgene, Massachusetts; Kan resistance, C-terminal 6 X his tags) at XhoI and XbaI sites. Mutants were generated by designing semi-overlapping primers following quickchange guidelines which incorporated the mutation (Liu and Naismith, 2008). Plasmids were transformed into E. coli BL21 (DE3) RIL cells for expression. Cultures were grown in Luria broth (LB) at 37 °C until reaching OD₆₀₀ of 0.8, then temperature reduced to 18°C for induction with 1 mM IPTG overnight. The next morning, cultures were pelleted and either stored at –80°C or protein purification carried out immediately. All subsequent steps were carried out on ice or in a 4°C cold room. Pellets from 2 L of culture were resuspended in 40 mL ice cold lysis buffer (50 mM HEPES, 1 M NaCl, 20 mM Imidizole, 10 % glycerol, 1 mM PMSF, pH 7.8) and lysed with an Emulsiflex-05 homogenizer (Avestin, Ottawa, Canada) for approximately 5 min at 15,000 PSI. Lysate was then cleared by centrifugation at 33,764 g for 30 min at 4 °C in a Thermo Scientific Fiberlite F14−14 × 50 cy rotor. Clarified lysate was loaded onto a 5 ml HisTrap FF Crude column (GE, Massachusetts) on an ÄKTA start chromatography system (GE) pre-equilibrated with lysis buffer. Unbound material was washed away with 15 column volumes of lysis buffer before an isocratic elution with five column volumes of elution buffer (50 mM HEPES, 1 M NaCl, 250 mM Imidizole, 30 % glycerol, pH 7.8). Peak fractions were combined and concentrated approximately 3-fold using a 30 kDa cut-off Amicon centrifugal filter unit (Millipore, Massachusetts). Approximately 5 mL of concentrated protein was subjected to size-exclusion chromatography using the Äkta Pure system and a Superdex 200 increase 10/300 gl pre-equilibrated with running buffer (50 mM HEPES, 1 M NaCl, 10 % glycerol, pH 7.8). Peak fractions were collected and glycerol added to a final concentration of 30 % before again concentrating using a 30 kDa cut-off Amicon centrifugal filter unit. Protein at ~2 mg/ml was flash frozen in liquid nitrogen and stored at –80 °C. Recombinant human CTPS2 was purified as described in Lynch et al., 2020.

To reduce variation between wild type and mutants (Ura7 WT, H360A and H360R), all three were purified in tandem and flash frozen. Aliquots were thawed, desalted into glycerol-free buffer (50 mM TrisHCl pH 7.4, 200 mM NaCl, 10 mM MgCl₂), and concentrations measured at A₂₈₀. Reactions were set up in 96 well clear plastic flat bottom corning plates (Corning, New York) with a 100 μl final reaction volume. For low pH kinetics, 2 μM protein was incubated with activity buffer (50 mM MES pH 6.0, 200 mM NaCl, 10 mM MgCl2, 10 mM BME) and nucleotide (2 mM ATP, 2 mM UTP, 0.2 mM GTP) for 15 min at 30 °C. Reaction was initiated by addition of 10 mM glutamine and absorbance measured at 291 nm in a plate reader (Varioskan lux) set at 30 °C. Early activity traces were noisy, likely due to bundled CTPS, so slope was measured between 350 and 500 s where rate of activity stabilized. A linear regression was fit to determine slope, which was used to determine CTP concentration. Replicates were performed in triplicate and averaged. Substrate kinetics were run under less optimal conditions in order to capture rates at lowest substrate concentrations. Ura7 at a final concentration of 1.5 μM was added to reaction buffer (50 mM TrisHCl 7.4, 200 mM NaCl, 10 mM MgCl2) and nucleotides (1 mM ATP, 20 μM GTP, with a range of 20μM to 1000 μM UTP). Mixture was incubated at room temperature for 5 min then inserted into a plate reader set to 22 °C and allowed to equilibrate for an additional 5–8 min, while taking readings at 291 nm until readouts were stable. The plate was briefly ejected from the reader to manually initiate the reaction by addition of glutamine (1 mM final concentration, prepared in 20 mM TrisHCl pH 7.0). CTP production was measured at 291 nm for 5 min, capturing the early linear phase of the curve. Assays were performed in triplicate then averaged, and kinetics data were fit by four parameter logistic regression y = min + ((min-max)/(1+((max/S50)ⁿ))), solving for maximum rate, minimum rate, S50, and hill number using the solver plugin in Microsoft Excel 16.51.

Frozen CTPS was thawed and desalted into a glycerol-free buffer (200 mM NaCl, 20 mM HEPES 7.8). Sample was set up in a total volume of 120 μl for assembly at a final protein concentration of 2 μM with 1 mM CTP. Assembly buffer was 50 mM MES 6.0 or 50 mM TrisHCl 7.0–7.4, 500 mM NaCl, and 10 mM MgCl2. Protein was added to the buffer in a quartz cuvette and inserted into a Horiba Fluorolog three fluorometer (Horiba, Kyoto, Japan) set to 30 °C and allowed to incubate until signal stabilized (approximately 3–5 min). Assembly was initiated by addition of nucleotide then readings began immediately, usually with a 5-second lag between addition and commencing readings. A total of 350 nm excitation wavelength was used and emission spectra from 350 nm were collected with 0.65 nm slit width for both. Raw data was normalized by subtracting baseline signal from initial incubation step. Sample was run in triplicate for each condition, then averaged and plotted with standard deviation in microsoft excel 16.51. Curves in Figure 2—figure supplement 2 were fit using the four parameter logistic regression y = min + ((min-max)/(1+(max/S50)ⁿ)) solving for maximum rate, minimum rate, hill number, and S50. R² values were determined using the built-in Pearson function tool in Microsoft Excel 16.51.

CTPS was assembled in reaction buffer (50 mM TrisHCl or MES, 100 mM NaCl, 10 mM MgCl2) at 30 °C for 15 min prior to applying 5 µl of sample to glow-discharged carbon-coated grids and incubated for 1 min at room temperature. Grids were washed 3 x in H₂O followed by 3 x in 0.7 % uranyl formate with blotting in between all steps to remove excess liquid. Imaging was done on an FEI Morgagni electron microscope operating at an accelerating voltage of 100 kV. Datasets for hCTPS2 were collected on a Tecnai G2 Spirit (FEI) operating at 120 kV. Images were acquired at ×67,000 magnification (pixel size 1.6 Å/px) on a Ultrascan 4000 4k × 4k CCD camera (Gatan). Contrast transfer function (CTF) was estimated using CTFFIND4 (Rohou and Grigorieff, 2015), Cryosparc v2.1 (Punjani et al., 2017) was used for automatic blob picking and 2D classification.

Frozen URA7/URA8 was desalted into minimal buffer (5 mM MES 5.9, 50 mM NaCl) prior to assembly at 8 µM with either substrates (2 µM UTP, 2 µM ATP) or product (2 µM CTP). Reaction buffer contained 400 mM NaCl, 5 mM MgCl2, 3 mM DTT, and 50 mM MES either at pH 6.0 (predominantly bundles) or pH 6.5 (predominantly single filaments). Protein was assembled for 15 minutes at 30 °C then 3 μl of the reaction material was double blotted (Snijder et al., 2017) onto glow-discharged C-FLAT 2/2 holey-carbon grids (Protochips), allowing a 1 min room temperature incubation between sample depositions. Grids were subsequently blotted for 4.5 s using a Vitrobot MarkIV (ThermoFisher Scientific) with chamber conditions set to 100 % humidity and at 4 °C. For Ura7 H360R sample was assembled at 5 μM final concentration in 50 mM NaCl, 50 mM HEPES 7.5, and 10 mM MgCl2 then 3 μl deposited onto lacey carbon grids which had thin carbon floated on top. Data were acquired using an FEI Titan Krios transmission electron microscope operating at 300kV and equipped with a Bioquantum GIF energy filter (Gatan) set to zero-loss mode with a slit width of 20 eV. Movies were collected on a K2 Summit Direct Detector camera (Gatan) in super-resolution mode at a magnification of 130 K (pixel size 0.525 Å/px). Automatic data acquisition was done using the Leginon Software Package (Suloway et al., 2005) with a defocus range specified in Table 1. Movies were acquired containing 50 frames having an exposure rate of 8.9 e^-/Ų/s and a total dose of 89e^-/Ų. For product-bound URA8, which exhibited a severe preferred orientation, we collected data with a stage tilt of both 20 and 40 degrees and combined the data for processing.

Assembly conditions for filaments also contained some bundles, and vice versa, therefore these datasets were combined early on and bundles/filaments were separated through 2D classification and processed independently afterwards. See supplemental methods for further details. Briefly, images were manually curated to remove poor quality acquisitions such as bad ice or large regions of carbon. Dose-weighting and image alignment of all 50 frames was carried out using MotionCor2 (Zheng et al., 2017) with binning by a factor of 2 (final pixel size 1.05 Å/px). Initial CTF parameters were estimated using GCTF (Zhang, 2016). Particle picking for bundles was done using the crYOLO (Wagner et al., 2019) helical pickier tool trained on both filaments and bundles and using a box size covering one tetramer, followed by manual picking to further improve the quality of the particle picks. Helical picking was chosen for the ease at the particle picking step, despite not processing any of the data with helical refinement. For tilted data, we used the local GCTF per particle CTF estimation tool to improve per particle defocus values. To classify bundles from filaments, particles were extracted with a large box size (512 pixels). Particle stacks were exported to either cisTEM (Grant et al., 2018) or Cryosparc (Punjani et al., 2017) for iterative 2D reference-free classification. Starting models for all maps were always obtained ab initio, and for bundle data the process was repeated in cisTEM and later cryosparc as independent validation. All four bundle starting models were obtained in this way, and no symmetry was imposed. 3D classification of bundles in C1 yielded subsets differing in their arrangement of strands. Classes sharing a common core assembly pattern were combined for further processing. In all cases, the highest resolution bundle maps were obtained by C2 refinement of a masked central segment after performing signal subtraction of density outside this region. Early processing without imposing symmetry suggested overall twofold symmetry in both bundle types. Therefore, we used a mask for focused alignment on three-strands (staggered bundle; Ura7 product- and substrate-bound & Ura8 substrate-bound) or two strands (in-register bundle; Ura8 product-bound) as these contained all the relevant contacts. FSC were calculated using Relion (Scheres, 2012) post-process or from the Phenix density modification (Terwilliger et al., 2020) output. Directional FSC was calculated using online FSC calculator (https://3dfsc.salk.edu/).

Initial models for Ura7 and Ura8 were obtained by threading their sequences onto the hCTPS1 substrate-bound structure from Lynch et al., 2017. Where inserts existed models were built in manually using Coot (Emsley and Cowtan, 2004). Models for individual monomers were rigid body fit using Chimera (Pettersen et al., 2004) by domain (0–280 amido-ligase domain; 281–300 linker; 301–570 glutaminase domain) then backbone and side-chain positions refined with ISOLDE (Croll, 2018). C-termini were built manually in coot using the best available Ura7 map (product-bound) and Ura8 map (substrate-bound). ATP was modeled using ISOLDE, and UTP/CTP using Coot after all ISOLDE refinements. Ligands for product-bound Ura8 were rigid body fit in coot. The density corresponding to Ura8 residues 417–457 was found to be weaker and have relatively lower resolution than the rest of the structure. To build this loop the RosettaES (Frenz et al., 2017) loop modeling protocol in Rosetta (Alford et al., 2017) was used with a beamwidth of 256. The top scoring result of the RosettaES protocol was selected and its geometries were refined with ISOLDE. Residues 444–455 were later removed due to poor angles and unsupported map density. After building the monomer, it was replicated at all symmetry equivalent positions to create the full tetramer, and a single monomer at the assembly interface. Residues at the tetramerization interface and assembly interface were relaxed using a full simulation in ISOLDE. Side chains and backbone angles were adjusted for a single monomer, which was again replicated to all four symmetry equivalent sites to generate the final tetramer model with identical subunits. Models were built into the high resolution product-bound Ura7 map, which was rigid body fit by domains (residues 0–273,276-299,302-C-ter) into lower resolution Ura7 product-bound bundle map based on local map quality. Junctions between domains were deleted (residues 274–275,300-301). For substrate-bound Ura7 filament, the Ura7-product-bound filament model was rigid body fit by domain and junctions deleted (domains 0–280,281-300,301-Cter; deleted 280–284 and 300–302). Clashes for this Ura7-substrate-bound model were removed using ISOLDE. Substrate-bound Ura7 bundle was built from a rigid body fit of the substrate-bound Ura7 filament model as described above. Ura8 bundle model building began with models from their corresponding ligand-state filament, rigid body fit by domain, then refined in ISOLDE. This monomer was duplicated at all relevant contact sites and overall relaxed in ISOLDE to refine positions at tetramerization interface, filament assembly interface, and lateral bundle contacts simultaneously. The monomer was then replicated at all symmetry sites and because not each monomer experiences the same lateral contacts, it was docked into different spots and an ISOLDE simulation was run with only the residues at the lateral interface with restraints on everything else. The final monomer was duplicated into all sites to fill the map density and ligands docked in and refined using coot. Model statistics were assessed using molprobity online server () and the RCSB PDB Submission validation report. Buried surface area for lateral bundle contacts and filament interface were calculated using PDBePISA tool (https://www.ebi.ac.uk/msd-srv/prot_int/cgi-bin/piserver) with default parameters.

Caver 3.0.3 plugin (Chovancova et al., 2012) was used for Pymol 2.4.1 (Schrödinger, LLC, 2020) with default parameters. We set catalytic Cys404 as the start point for tunnel search of an individual monomer. Probe radius was set at 0.7 Å and the tunnel from Cys404 to the UTP base was identified. Inspection of radii along the length of the tunnel revealed that radii near the P52/C58/H55 constriction point in the Ura8 substrate-bound tetramer was between 1.2 Å and 1.9 Å, consistent with relieved ammonia channel constriction (Lynch and Kollman, 2020).

Yeast were maintained in standard YPD (Lab Express) or synthetic media containing 2 % D-Glucose (Fisher). Background strains were W303 {leu2-3,112 trp1-1 can1-100 ura3-1 ade2-1 his3-11,15}. Deletion and mutation strains were made by a PCR approach as described previously (Wendland, 2003), and C-terminal tagging of yeast were made as described previously (Sheff and Thorn, 2004). List of yeast strains used can be found in Table 5.

To induce polymers, yeast were grown in YPD until mid-late log phase, washed in 1 X PBS and resuspended in starvation media- 0.1 M Phosphate-citrate buffer (pH 5,6,7) and grown shaking at 30 °C for 3–4 hrs. To manipulate intracellular pH in the presence of 2 % Glucose, 2 mM of 2,4-Dinitrophenol (Sigma) was added to the media as described previously (Petrovska et al., 2014).

Liquid growth curves were made by diluting mid-late log phase and/or starved yeast cells to OD600 0.05 in YPD or SD +0.5 % Glucose and growth was monitored on a Varioskan Lux plate reader at 600 nm, shaking at 30 °C for 16 hr. Solid growth assays were done by making five 5-fold serial dilutions of mid-late log phase or staved yeast cells and plated on YPD or SD +0.25 % Glucose plates +2 % Agar and grown at 30 degrees for 48 hr.

Fixed and live fluorescence microscopy was done at ×100 objective magnification on a DeltaVision Elite microscope (GE) equipped with DIC optics, using a 60 × 1.42 NA objective, and a sCMOS 5.4 PCle air-cooled camera (PCO-TECH). Deconvolution was performed with SoftWorx (API, Issaquah, WA) and images were analyzed using Fiji, ImageJ (Scheres, 2012). Figures were assembled using Adobe Photoshop. Representative images shown of experiments done on three independent replicates. Fluorophores used were GFP and mCherry.