Light-sensitive phosphorylation regulates retinal IMPDH1 activity and filament assembly

Photoreceptor cells in the retina require high levels of guanine nucleotides to meet the metabolic demands of cyclic guanosine monophosphate (cGMP) signaling. To maintain nucleotide levels, they activate the de novo purine synthesis pathway to supplement nucleotides supplied by salvage pathways. Purine synthesis is a critical metabolic hub, and enzymes in the pathway are under tight control by various modes of regulation, including feedback inhibition, allostery (Buey et al., 2015), biomolecular condensation (Pedley et al., 2022a), posttranslational modification (Liu et al., 2019), oligomerization and polymerization, and assembly of polymers into micron-scale subcellular structures known as metabolic filaments (Hvorecny and Kollman, 2023; Lynch et al., 2020; Park and Horton, 2019; Simonet et al., 2020).

The de novo purine synthesis pathway yields inosine monophosphate (IMP), the precursor for adenine and guanine nucleotides. IMP dehydrogenase (IMPDH) is the rate-limiting enzyme in guanosine triphosphate (GTP) synthesis (Fig. 1 A). In humans, IMPDH has two canonical 514-residue isoforms with 84% sequence identity and similar kinetics (Carr et al., 1993). IMPDH1 is expressed at low levels in most tissues, while IMPDH2 is upregulated in proliferating and transformed cells (Collart et al., 1992; Jackson et al., 1975; Nagai et al., 1992; Senda and Natsumeda, 1994). IMPDH is regulated by GTP and ATP, which compete for three nucleotide binding sites in the regulatory Bateman domain to inhibit or activate the enzyme, respectively (Buey et al., 2022). In solution, IMPDH forms a catalytic tetramer that reversibly dimerizes upon nucleotide binding to form a functional octamer, which can assemble end-on-end to form polymers (referred to as “filaments” here). GTP binds nucleotide binding sites 2 and 3, while ATP binds site 1 and competes with GTP for site 2 (Fig. 1 B) (Buey et al., 2015, 2017; Johnson and Kollman, 2020).

Binding of GTP and ATP controls the polymerization of IMPDH in vitro. The presence of ATP in sites 1 and 2 promotes filaments composed of octamers in an extended, high-activity state, while GTP in sites 2 and 3 leads to filaments containing compressed, less active octamers (Fig. 1, C and D) (Anthony et al., 2017; Burrell et al., 2022; Johnson and Kollman, 2020). For IMPDH2, filament assembly maintains a “flat” conformation of the catalytic tetramer that prevents full inhibition of the enzyme within the filament, allowing the enzyme to remain partially active in vitro (Fernández-Justel et al., 2019; Johnson and Kollman, 2020). In cells, single IMPDH filaments bundle together to form micron-scale ultrastructures (Juda et al., 2014; Schiavon et al., 2018; Thomas et al., 2012). IMPDH filament assembly in vivo correlates with metabolic conditions that require high GTP levels, like T cell activation and the rapid proliferation of stem cells (Calise and Chan, 2020; Calise et al., 2018; Carcamo et al., 2011; Duong-Ly et al., 2018; Keppeke et al., 2018).

IMPDH1 is the dominant isoform in the vertebrate retina and is expressed as two tissue-specific splice variants (Bowne et al., 2006a; Hedstrom, 2009). IMPDH1(546) has a disordered extension of 37 residues that replace the five canonical residues at the C-terminus. IMPDH1(595) contains the same C-terminal extension plus 49 residues at the N-terminus (Fig. 2 A) (Burrell and Kollman, 2022; Spellicy et al., 2007). IMPDH1(546) and IMPDH1(595) are less sensitive to GTP inhibition compared with the canonical enzyme IMPDH1(514), enabling increased GTP production to meet the significant demands of cGMP signaling in photoreceptor cells (Fig. 2 B) (Andashti et al., 2020, 2021; Burrell et al., 2022; Plana-Bonamaisó et al., 2020). Within IMPDH1(514) and IMPDH1(546) filaments, the assembly interface between two octamers can adopt two conformations referred to as the large and small interfaces, which correlate with high and low enzymatic activity, respectively (Burrell et al., 2022). For IMPDH1(595), the N-terminal extension locks it in the large interface, which stabilizes a more active conformation of the enzyme and contributes to increased resistance to GTP inhibition compared with the canonical enzyme (Fig. 2 C) (Burrell et al., 2022).

The importance of IMPDH1 in retinal function is reflected in the multiple missense mutations that have been linked to gradual vision loss caused by autosomal dominant retinitis pigmentosa (Bowne et al., 2002, 2006b; Grover et al., 2004; Kennan et al., 2002; Wada et al., 2005) and Leber congenital amaurosis (Bowne et al., 2006b). None of the mutations alter canonical IMPDH1 catalytic activity in vitro (Aherne et al., 2004; Mortimer and Hedstrom, 2005; Xu et al., 2008), but we recently showed that some mutations alter the architecture of IMPDH retinal variant filaments, disrupting regulation by GTP and providing clues as to why clinical manifestations are limited to the retina (Burrell et al., 2022).

Recent phosphoproteomic analyses of bovine retinas identified three phosphorylation sites in IMPDH1 associated with light and dark states (Plana-Bonamaisó et al., 2020). Phosphorylation occurs at T159/S160, residues directly involved with allosteric nucleotide binding site 1, in response to light. This desensitizes the enzyme to GTP regulation, allowing elevated GTP pools for phototransduction. Phosphorylation at S416, a residue on a mobile flap in the catalytic domain required for catalysis, was observed in both light and dark states. The third site S477 is preferentially phosphorylated in the dark, but its effects on reaction kinetics and GTP regulation are unclear. An S477D phosphomimetic mutant of IMPDH1(546) had similar reaction kinetics to wildtype (WT) in the absence of nucleotides, and GTP regulation of IMPDH1(514)-S477D appeared similar to WT canonical enzyme, although no data on the GTP inhibition of S477D mutant retinal variants were shown (Plana-Bonamaisó et al., 2020). Phosphorylation at S477 was speculated to disrupt filament formation (Plana-Bonamaisó et al., 2020).

Here, we show that S477D disrupts assembly of the large, high-activity filament interface, completely preventing IMPDH1(595) filament assembly and forcing IMPDH1(546) filaments into the small, lower-activity interface. These changes in higher-order structure correspond to increased sensitivity to GTP inhibition and lower reaction velocity in enzyme assays. This suggests that phosphorylation directly affects flux through IMPDH1 by preventing its ability to form high-activity filaments, introducing an additional mechanism for controlling enzyme filament formation to tune metabolic activity.

We examined purified S477D variants IMPDH1(595)-S477D and IMPDH1(546)-S477D by negative stain electron microscopy (EM) in the presence of no ligand (Apo) or either 1 mM ATP or 1 mM GTP, which promote polymerization of the WT enzyme (Burrell et al., 2022; Johnson and Kollman, 2020). S477D completely prevented assembly of IMPDH1(595) filaments, allowing only free octamer formation in the presence of ATP or GTP (Fig. 3 A). At this resolution, higher-order assembly of IMPDH1(546)-S477D appeared unaffected, assembling octamers and polymers similar to WT enzyme (Fig. 3 B). S477D also partially prevented polymerization of the canonical IMPDH1(514), although the effects were not as clear as in the retinal variants (Fig. S1). Previous structures of IMPDH1 filaments showed that S477 contacts the N-terminus of the opposing monomer in the large interface, but makes no contacts in the small interface, suggesting phosphorylation at S477 is likely to specifically break the large interface (Fig. 2 D). Thus, we suspected that the IMPDH1(546)-S477D filaments we observed by negative stain were in the small interface. Complete disassembly of IMPDH1(595) filaments supports our previous structural data showing that IMPDH1(595) filament assembly is restricted to the large interface (Burrell et al., 2022).

WT retinal variants IMPDH1(546)-WT and IMPDH1(595)-WT are significantly less sensitive to GTP inhibition compared with the canonical IMPDH1(514) (Burrell et al., 2022). Here, we wanted to test the S477D mutants for any effects on GTP regulation of activity. In both retinal variants, S477D resensitized the enzyme to GTP inhibition. S477D decreases the half-maximal inhibitory concentration (IC₅₀) for GTP in IMPDH1(595) and IMPDH1(546) to the same extent as the engineered mutation Y12A, which also disrupts filament assembly (Fig. 3, C and D) (Anthony et al., 2017; Burrell et al., 2022; Fernández-Justel et al., 2019; Johnson and Kollman, 2020). Enzyme assays are summarized in Table 1.

Inhibition of IMPDH1 involves two conformational changes: interdomain movements that switch from an extended to compressed octamer and transition of the catalytic tetramer from a flat conformation to a bowed conformation at the filament assembly interface (Burrell et al., 2022; Johnson and Kollman, 2020). The large interface is associated with stabilizing the flat conformation, leading to higher catalytic activity, while the small interface is associated with the bowed conformation and lower activity. In completely breaking IMPDH1(595) filament assembly, S477D removes the enzyme’s ability to engage the large interface and stabilize the flat conformation. Free IMPDH octamers adopt the compressed and bowed conformation upon GTP binding, leading to lower activity (Burrell et al., 2022; Johnson and Kollman, 2020), explaining the increased sensitivity to GTP inhibition of IMPDH1(595)-S477D in our assays. However, why IMPDH1(546)-S477D was resensitized to the levels of the engineered filament-disrupting mutation Y12A was not clear. To gain further insights into structural details, we turned to cryogenic EM (cryo-EM).

First, we solved a 3.1 Å cryo-EM structure of the IMPDH1(595)-S477D free octamer in the presence of GTP, ATP, IMP, and NAD⁺ (Fig. 4 A and Fig. S2). We built an atomic model and aligned it with a previous model (Protein Data Bank [PDB]: 7RGD) of IMPDH1(595)-WT under the same ligand conditions in a compressed state within a filament. We observed no significant changes in the conformation of the backbone, with a calculated root-mean-square deviation (RMSD) of alpha carbons when aligned on chain A of both structures of 0.6 Å. We then took a closer look at the catalytic tetramer of each model to determine if IMPDH1(595)-S477D was in the flat or bowed conformation (Fig. 4 B). We aligned S477D and WT tetramers on chain A and observed a 3° shift in chain C (the monomer positioned diagonally across the tetramer) of S477D relative to WT, showing that IMPDH1(595)-S477D free octamer adopts a bowed conformation similar to the inhibited canonical IMPDH1(514)-WT within a filament (RMSD of alpha carbons when aligned on chain A of IMPDH1(595)-S477D versus inhibited IMPDH1(514)-WT = 0.7 Å). IMPDH1(595)-WT is stabilized in the large interface and flat conformation in the filament by its extended N-terminus (Burrell et al., 2022). The extended N-terminus was unresolved in our map of the free octamer, suggesting it only becomes ordered upon interaction with a binding partner.

Next, we solved a 3.3 Å cryo-EM structure of the IMPDH1(546)-S477D compressed octamer within a filament in the presence of GTP, ATP, IMP, and NAD⁺ (Fig. 5 A and Fig. S3). We built an atomic model of the compressed octamer and compared it with a previous model (PDB: 7RGQ) of IMPDH1(546)-WT in a compressed state within a filament, which adopts the small interface (Burrell et al., 2022). Aligning these two models showed no significant differences between single protomers or octamers (RMSD of alpha carbons when aligned on chain A of both structures = 0.5 Å). From the same dataset, we separately determined a structure of the filament assembly interface by centering the reconstruction on the D4 symmetry center between octamers in the filament, an approach we have used in the past to yield the highest resolution reconstructions of the interface (Burrell et al., 2022; Johnson and Kollman, 2020). This approach yielded a 3.3 Å structure of the assembly contacts, which is essentially identical to the IMPDH1(546)-WT assembly interface (RMSD of alpha carbons when aligned on chain A of both structures = 0.6 Å) (Fig. 5 C; and Fig. 6, A and B). Thus, at the level of protomer, octamer, and filament, the S477D mutation does not affect the structure of fully inhibited IMDPH1(546).

In the active state in the presence of ATP, IMP, and NAD⁺, IMPDH1(546)-WT assembles a filament with a large interface (Burrell et al., 2022). We solved a 2.4 Å cryo-EM structure of IMPDH1(546)-S477D under the same conditions (Fig. 5 B and Fig. S4) and aligned the model to IMPDH1(546)-WT in an extended state within a filament (PDB: 7RGM). Individual protomers appeared highly similar (RMSD of alpha carbons when aligned on chain A of both structures = 0.6 Å), but when we aligned octamers on chain A, we detected a 3° shift in the protomer on the opposite side of the octamer. We solved a 2.1 Å interface-centered structure of IMPDH1(546)-S477D and compared the filament assembly interface with IMPDH1(546)-WT (Fig. 6, C and D). The IMPDH1(546)-S477D extended, active filament adopts the small interface driven by contacts between Y12 and E487 (Fig. 5 D, top). This is significantly different from the IMPDH1(546)-WT large interface, which has a relative 41° rotation with R356 contacting Y12 (Fig. 5 D, bottom). We then aligned active IMPDH1(546)-WT, inhibited IMPDH1(546)-WT, and active IMPDH1(546)-S477D to compare their catalytic tetramer conformations. The catalytic tetramer of active IMPDH1(546)-S477D adopts a unique intermediate conformation between bowed and flat (Fig. 5 E, inset). Forming the small interface with an intermediate tetramer conformation suggests the enzyme may be primed to enter the compressed, bent state when GTP is present, lowering the transition barrier and effectively raising the affinity for GTP. In a kinetics assay, this would mean IMPDH1(546)-S477D is more sensitive to GTP inhibition, as we detected in Fig. 3 D.

Given the obvious disruption to higher-order assembly of IMPDH1(595)-S477D, we wanted to investigate the impact of the phosphomimetic on filament assembly in cells. We transfected HeLa cells with expression constructs encoding FLAG-tagged WT, S477D mutant, and S477A mutant retinal variants (Fig. 7). After 48 h, we treated cells with 1 mM ribavirin for 2 h, which led to robust IMPDH filament assembly in nearly 100% of cells (Calise et al., 2016; Carcamo et al., 2011; Covini et al., 2012; Keppeke et al., 2015). We then fixed and stained cells with an anti-IMPDH2 antibody to label endogenous IMPDH and anti-FLAG antibody to label transfected protein.

Both WT and S477A variants assembled typical micron-scale filaments in a high percentage of transfected cells, as expected (595-WT: 98%, 595-S477A: 98%, 546-WT: 98%, 546-S477A: 100%). However, S477D exerted a dominant-negative effect, blocking essentially all filament assembly by endogenous IMPDH in transfected cells (595-S477D: 2%, 546-S477D: 2%). In cell populations transfected with S477D mutants, filaments can be observed in adjacent non-transfected cells as an internal control (white arrows, Fig. 7). Since overexpression alone also causes IMPDH1 filament assembly (Gunter et al., 2008; Keppeke et al., 2018, 2023), we performed the same experiment without ribavirin treatment and observed similar results (Fig. S5).

These data suggest an important point: IMPDH1(546)-S477D forms filaments in vitro but only with the small interface, so its inability to form filaments at all in cells implies that IMPDH filaments predominantly assemble using the large interface in vivo. Together, our data suggest that phosphorylation at S477 is a molecular mechanism for regulating flux through de novo GTP synthesis by controlling higher-order assembly of the pathway’s rate-limiting enzyme.

Despite the increasing interest in higher-order assembly of IMPDH, posttranslational modifications of the enzyme have received little attention. However, a recent phosphoproteomic analysis of bovine retinas adapted to light and dark states shed light on three novel phosphorylation sites in IMPDH1: T159/S160, S416, and S477 (Plana-Bonamaisó et al., 2020). The study focused mainly on light-dependent phosphorylation at T159/S160, which desensitizes IMPDH1 to GTP inhibition to boost GTP levels in support of phototransduction. On the other hand, phosphorylation at S477 was correlated with the dark, when less GTP production is needed (Plana-Bonamaisó et al., 2020). S477D seemed to have no effect on basal IMPDH1(546) activity in vitro, suggesting that S477D might instead play a structural role in disrupting filament assembly.

Here, we showed that S477D resensitizes human retinal IMPDH1 variants to GTP inhibition and prevents filament assembly in living cells in a dominant-negative manner, which would lead to overall decrease in cellular GTP pools. Our data demonstrate a mechanism for control of IMPDH filament assembly dynamics and GTP production by phosphorylation at a single site. We propose that phosphorylation at S477 reduces the extent of IMPDH filaments in photoreceptor cells in the dark, leading to a global decrease in GTP levels when nucleotide turnover rate is reduced. In the light, an unknown phosphatase removes the modification, allowing filaments to reassemble and boost GTP levels to support cGMP signaling again (Fig. 8).

The S477D mutation in either the 546 or 595 variant has a dominant-negative effect on assembly of endogenous WT IMPDH. IMPDH2 is the dominant isoform in most cell types and only assembles filaments via the large interface (Burrell et al., 2022; Johnson and Kollman, 2020). The dominant effect of IMPDH1(595)-S477D can be explained by its incorporation into mixed tetramers or octamers with the WT enzyme. Given our in vitro data that IMPDH1(546)-S477D assembles filaments in both active and inhibited states, we were initially surprised that it failed to form filaments at all in cells. However, this may again be explained by the likely coassembly of IMPDH1(546)-S477D into mixed oligomers with endogenous IMPDH (Keppeke et al., 2023). Mixed oligomers of IMPDH2 (large interface only) and IMPDH1(546)-S477D (small interface only) would prevent assembly of either filament type and likely explain the dominant-negative effect of IMPDH1(546)-S477D. It is possible that IMPDH1 assembles short filaments in cells using the small interface, but the incompatibility with IMPDH2 would render them short-lived. In the retina, which does not express appreciable levels of IMPDH2 or canonical IMPDH1 (Aherne et al., 2004; Bowne et al., 2006a; Hedstrom, 2009; Spellicy et al., 2007), mixed oligomers of WT IMPDH1(595) (large interface only) and WT IMPDH1(546) (small or large interface) would likely only assemble filaments with the large interface. Overall, this is consistent with the observation that IMPDH filaments form in cells and tissues under conditions of high demand for IMPDH activity (Calise et al., 2018; Chang et al., 2015; Duong-Ly et al., 2018; Keppeke et al., 2018, 2020; Peng et al., 2021; Ruan et al., 2020), which is promoted by assembly into filaments with the large interface. Further studies in genome-edited cells expressing single IMPDH variants will provide insight into in vivo assembly of the different forms of IMPDH.

We recently characterized IMPDH1 filaments in the zebrafish retina, finding a circadian pattern to filament assembly, with elongated filaments forming in photoreceptors during the day and smaller punctate structures at night (Cleghorn et al., 2022). We also showed that the zebrafish IMPDH1 retinal variants are very similar to human IMPDH1 in terms of structure, filament assembly, and regulation (Cleghorn et al., 2022). This is consistent with our finding here that the S477D mutation, which mimics dark-dependent phosphorylation, disrupts filament assembly and reduces enzyme activity in the presence of GTP. Besides zebrafish, IMPDH filament assembly has also been observed in mouse retinas, although the functional significance is yet to be described (Fernández-Justel et al., 2019).

Studies in cultured cells and mice have demonstrated a strong link between IMPDH filament assembly and cells that consume high levels of GTP. Robust IMPDH filament assembly occurs in stem cells (Carcamo et al., 2011; Keppeke et al., 2018) and lymphocytes (Calise et al., 2018; Duong-Ly et al., 2018; Peng et al., 2021), which require increased nucleotides to support rapid proliferation (Dayton et al., 1994; Fairbanks et al., 1995; Quéméneur et al., 2003). IMPDH filaments have also been identified in aggressive tumors (Keppeke et al., 2020; Ruan et al., 2020). Polymerization of IMPDH1(546), IMPDH1(595), and IMPDH2 partially prevents feedback inhibition by guanine nucleotides compared with the unpolymerized state (Burrell et al., 2022; Fernández-Justel et al., 2019; Johnson and Kollman, 2020). In cells, this would allow accumulation of GTP to higher steady state levels, supporting rapid proliferation (IMPDH2) or phototransduction in photoreceptor cells (IMPDH1 retinal variants).

More than 20 different metabolic enzymes form filaments in cells, suggesting that filament assembly is a general mechanism for regulating multiple metabolic pathways (Hvorecny and Kollman, 2023; Lynch et al., 2020; Narayanaswamy et al., 2009; Noree et al., 2010; Park and Horton, 2019, 2020; Shen et al., 2016; Simonet et al., 2020). For many metabolic enzymes, filament formation is an important mechanism for fine-tuning catalytic activity (Barry et al., 2014; Hansen et al., 2021; Hu et al., 2022; Hunkeler et al., 2018; Hvorecny et al., 2023; Lu et al., 2023; Lynch and Kollman, 2020; Lynch et al., 2017; Polley et al., 2019; Stoddard et al., 2020; Zhong et al., 2022). Here, we demonstrate that filament assembly itself can be directly tuned by phosphorylation, adding an additional layer to an already complex mode of regulation for IMPDH that occurs at different levels. Activity of IMPDH tetramers is modulated by feedback inhibition and oligomerization into octamers. IMPDH octamers are then regulated further by allosteric binding of nucleotides to the Bateman domain, which compresses or extends the conformation of the octamer. Polymerization of octamers into filaments can then modulate sensitivity to GTP inhibition. Filament assembly itself is then regulated by phosphorylation at S477 in response to changing metabolic conditions of the cell.

IMPDH localizes to the purinosome (Zhao et al., 2015), a biomolecular condensate assembled by the six enzymes directly upstream of IMPDH in de novo purine synthesis (An et al., 2008; Pedley et al., 2022b). Purinosome assembly correlates with increased flux through the pathway (Pareek et al., 2020; Zhao et al., 2015) and has been hypothesized to be regulated by posttranslational modifications (Liu et al., 2019). In fact, purinosome assembly is modulated by casein kinase II, AMP-activated protein kinase, and 3-phosphoinositide–dependent protein kinase 1, suggesting various signaling cascades spatially regulate purinosome enzymes (An et al., 2010; Schmitt et al., 2016, 2018). It is enticing to speculate that a network of phosphorylation and/or other modifications may spatiotemporally coordinate assembly of IMPDH into various subcellular compartments depending on the metabolic state of the cell.

Given the widespread utilization of filament assembly as a regulatory mechanism, it is likely that other polymerizing enzymes are regulated by posttranslational modification in the same way. Some correlations between posttranslational modification and IMPDH or cytidine triphosphate synthetase (CTPS) filament assembly in cells have been reported. Ankyrin repeat domain-containing protein 9 (ANKRD9) is a component of a cullin-RING superfamily E3 ligase complex that ubiquitinates both isoforms of IMPDH for degradation (Lee et al., 2018). ANKRD9 also localizes to and stabilizes IMPDH filaments induced by serum starvation in HeLa cells, suggesting ubiquitination plays some role in filament assembly (Hayward et al., 2019). The E3 ubiquitin ligase activity of Cbl is required for CTPS filament assembly in Drosophila, although it is not clear if the ubiquitination is direct (Wang et al., 2015). In a similar fashion, protein methylation is required for CTPS filament assembly in human cells (Lin et al., 2018). The mechanisms underlying these modifications remain unclear, but our current study opens the possibility for direct posttranslational modification as a general mechanism for modulating filament assembly dynamics.

The data presented here establish a molecular mechanism for reversible phosphorylation controlling assembly of IMPDH filaments and regulating GTP levels in the retina. This mechanism is likely to apply to a variety of filament-forming enzymes coordinating many different metabolic pathways in the cell. Future studies will determine if phosphorylation or other posttranslational modifications are the mechanisms of control underlying general metabolic filament assembly in vivo.

Kanamycin-resistant pSMT3 bacterial expression vectors containing coding sequences for human IMPDH1(546) and IMPDH1(595) with N-terminal 6xHis-SUMO tags were generated previously (Burrell et al., 2022). To generate S477D mutant versions, overlapping primers 5′-CGC​AGC​CTGGATGT​CCT​TCG​GTC​CA-3′ and 5′-TGG​ACC​GAA​GGA​CATCCAG​GCT​GCG-3′ containing a TC > GA change to replace a serine with an aspartic acid codon were used to amplify each plasmid. Plasmids were ligated at 50°C for 1 h in a Gibson assembly reaction. The reaction product was treated with DpnI enzyme at 37°C for 30 min to degrade template DNA and then used for transformation of competent TOP10 E. coli cells. Transformed cells were cultured overnight on lysogeny broth (LB) agar plates containing 50 µg/ml kanamycin. Single colonies were picked and cultured overnight in LB medium + 50 µg/ml kanamycin, and recombinant plasmids were isolated using the GeneJET Plasmid Miniprep Kit (#K0503; Thermo Fisher Scientific). Insert sequences in each plasmid were confirmed by Sanger sequencing. Vector sequence was previously confirmed by Sanger sequencing of parental constructs.

For mammalian expression, coding sequences for WT and S477D retinal variants were subcloned from the above constructs into a pCMV3 vector (generously provided by Dr. Gerson Keppeke, Universidad Católica del Norte, Coquimbo, Chile). S477D constructs were then used as templates for PCR reactions to generate full-length S477A constructs, which were then self-ligated by Gibson assembly. Primers used for cloning are listed in Table 2. The same procedure was used for cloning mammalian constructs as bacterial expression constructs (see above), except that different primers were used. Full-length plasmid sequencing was performed on all mammalian constructs by Nanopore sequencing. Insert regions were also validated by Sanger sequencing.

Recombinant human IMPDH1 retinal variant proteins were expressed and purified as described previously (Anthony et al., 2017; Burrell et al., 2022; Cleghorn et al., 2022). Briefly, BL21 (DE3) E. coli cells were transformed with WT or S477D mutant IMPDH1(546) or IMPDH1(595) expression vectors. Cells were cultured in LB medium at 37°C until reaching an optical density (OD₆₀₀) of 0.9, and then expression was induced by addition of isopropyl-β-d-thiogalactoside (IPTG) to 1 mM for 4 h at 30°C. Cells were pelleted and flash-frozen in liquid nitrogen and stored at −80°C until use.

Pellets were thawed, resuspended in lysis buffer (50 mM KPO₄, 300 mM KCl, 10 mM imidazole, 800 mM urea, pH 8.0), and lysed with an Emulsiflex-05 homogenizer. Lysate was clarified by centrifugation at 33,746 × g for 30 min at 4°C and 6xHis-SUMO-tagged IMPDH1 was purified by immobilized metal ion affinity chromatography using HisTrap FF nickel sepharose columns (Cytiva) on an ӒKTA Start chromatography system. After on-column washing with lysis buffer, protein was eluted in buffer (50 mM KPO₄, 300 mM KCl, 500 mM imidazole, pH 8.0) and fractionated into 1-ml volumes. Peak fractions were combined and incubated with 1 mg ULP1 protease (Malakhov et al., 2004) per 100 mg IMPDH1 for 1 h on ice, then supplemented with 1 mM dithiothreitol (DTT) and 800 mM urea. Protein was concentrated with a 30,000 molecular weight cutoff filter (Amicon) and size-exclusion chromatography was performed on an ӒKTA Pure chromatography system using a Superose 6 column (Cytiva) pre-equilibrated in gel filtration buffer (20 mM HEPES, 100 mM KCl, 800 mM urea, 1 mM DTT, pH 8.0). Peak fractions were concentrated using a 10,000 molecular weight cutoff filter and then flash-frozen in liquid nitrogen and stored at −80°C.

Recombinant human IMPDH1(514) WT and S477D proteins were purified as described previously (Plana-Bonamaisó et al., 2020; Thomas et al., 2012). Briefly, BL21 (DE3) E. coli cells were transformed with pETEV15b constructs (Alonso-García et al., 2009) and expression was induced at 16°C for 12–14 h with 1 mM IPTG. Cell pellets were resuspended in 20 ml binding buffer (20 mM Tris, pH 8.0, 500 mM KCl, 20 mM imidazole, 1.5 M urea, and 1 mM DTT) containing protease inhibitors (Pierce protease inhibitor tablets, EDTA-free) per liter of cell culture. Lysates were sonicated on ice (8 W power, cycles of 30 s on and 30 s off) and then clarified at 17,000 × g for 30 min at 4°C. His-tagged IMPDH1 proteins were purified immediately after centrifugation by immobilized metal ion affinity chromatography on a HisTrap crude affinity resin (GE Healthcare) according to the manufacturer’s instructions. Protein was eluted with binding buffer containing 500 mM imidazole and the His tag was cleaved overnight at room temperature (RT) with recombinant tobacco etch virus protease (0.2 mg per liter of culture) purified in-house while dialyzing against storage buffer (20 mM Tris, pH 8.0, 500 mM KCl, and 1 mM DTT). Cleaved proteins were concentrated and injected into a Sephacryl S300 HR 16/60 column (GE Healthcare) equilibrated in storage buffer. Protein fractions were concentrated, glycerol was added up to 10%, and aliquots were frozen at −80°C until use.

Aliquots of IMPDH1 protein were diluted at least 100-fold from gel filtration buffer (20 mM HEPES, 100 mM KCl, 800 mM urea, 1 mM DTT, pH 8.0) into activity buffer (20 mM HEPES, 100 mM KCl, 1 mM DTT, pH 7.0) to 1 µM concentration and incubated with 1 mM ATP, 1 mM IMP, 1 mM MgCl₂, and varying concentrations of GTP for 30 min at 20°C in 96-well ultraviolet transparent plates (#3679; Corning). 100 μl reactions were initiated by addition of 300 µM NAD⁺. NADH production was measured by optical absorbance at 340 nm at 25°C once per minute for 15 min with a Varioskan Lux microplate reader (Thermo Fisher Scientific). Absorbance was correlated with NADH concentration using a standard curve. Specific activity was calculated by linear interpretation of the reaction slope for a 5-min window beginning 2 min after reaction initiation. All data points are an average of three measurements from the same protein preparation. Error bars represent SD. Fits for activity assays were calculated in GraphPad Prism v10.1.1 using a variable slope four-parameter model with equation $V=Vmin+(Vmax−Vmin)/(1+(IC50/[I])n)$ where n = Hill coefficient. Fitting method used was least squares regression and no weighting or constraints were applied. Confidence intervals were computed using the likelihood ratio asymmetric method. The IC₅₀ and Hill coefficient best-fit values and 90% confidence intervals are reported in Table 1.

Protein samples were diluted at least 100-fold from gel filtration buffer (20 mM HEPES, 100 mM KCl, 800 mM urea, 1 mM DTT, pH 8.0) into assembly buffer (20 mM HEPES, 100 mM KCl, 1 mM DTT, pH 7.0) to 1 µM concentration, and 2.5 μl volume was applied to glow-discharged continuous carbon EM grids (Protochips) and stained with 0.7% uranyl formate as described previously (Johnson and Kollman, 2020). Grids were imaged by transmission EM at 100 kV accelerating voltage on an FEI Morgagni microscope equipped with a Gatan Orius CCD. Micrographs were collected at a nominal 22,000× magnification with a 3.9 Å pixel size.

Protein samples were diluted from 20- to 100-fold from gel filtration buffer (20 mM HEPES, 100 mM KCl, 800 mM urea, 1 mM DTT, pH 8.0) into assembly buffer (20 mM HEPES, 100 mM KCl, 1 mM DTT, pH 7.0) to 1–5 µM concentration, and 2.5 μl volume was applied to glow-discharged C-flat holey carbon EM grids (Protochips), blotted at 4°C with 100% relative humidity, and plunge-frozen in liquid ethane using a Vitrobot vitrification system (Thermo Fisher Scientific) as described previously (Johnson and Kollman, 2020). High-throughput data collection was performed with Leginon (Suloway et al., 2009) or SerialEM (Mastronarde, 2005) software packages controlling a Thermo Fisher Scientific Glacios TEM operating at 200 kV equipped with a Gatan K2 Summit direct electron detector or a Thermo Fisher Scientific Titan Krios G3 TEM at 300 kV equipped with a Gatan image filter and K3 Summit direct electron detector.

Movies were collected in super-resolution mode, then aligned and corrected for full-frame motion and sample deformation using the patch motion correction algorithm in CryoSPARC Live with 2× Fourier binning and dose compensation applied during motion correction (Punjani et al., 2017). Contrast transfer function (CTF) estimation, initial particle picking, and two-dimensional (2D) classification were performed in CryoSPARC Live. CryoSPARC v3 or v4 was used for all subsequent image processing. Each dataset was processed individually using a similar workflow.

After several rounds of 2D classification to select the best-resolved classes, particles were boxed and re-extracted to generate a quality particle stack for further processing. Either ab initio reconstruction was used to generate an initial volume from these particles or a map from a previous IMPDH filament structure lowpass-filtered to 30 Å was used as a starting volume for refinement. The D4 point-group symmetry of IMPDH filaments means there are two points along the filament that can be used as symmetry origins: the center of an octamer or at the assembly interface between octamers. Octamer-centered reconstructions of filaments were done with homogeneous or non-uniform refinement with D4 symmetry, per-particle defocus refinement, and per-group CTF refinement with a previously solved octamer-centered filament map (either extended or compressed) as the starting volume (Punjani et al., 2020). Interface-centered reconstructions were performed the same, except with a previously solved interface-centered filament map as a starting volume. Image processing workflows are provided for each structure (Figs. S2, S3, and S4). Directional Fourier shell correlation (FSC) was calculated using the remote 3DFSC processing server (https://3dfsc.salk.edu) (Tan et al., 2017).

Existing structures were used as templates for model building. Octamer-centered compressed IMPDH1(595)-WT (PDB: 7RGD) was a template for the IMPDH1(595)-S477D model. Octamer-centered compressed IMPDH1(546)-WT (PDB: 7RGQ), interface-centered compressed IMPDH1(546)-WT (PDB: 7RGI), octamer-centered extended IMPDH1(546)-WT (PDB: 7RGM), and interface-centered extended IMPDH1(546)-WT (PDB: 7RGL) were used as templates for IMPDH1(546)-S477D models. Initial templates were placed into cryo-EM densities by rigid-body fitting in UCSF ChimeraX v1.6.1 (Goddard et al., 2018). Then, initial models were examined and manually fit residue-by-residue using ISOLDE v1.6.0 (Croll, 2018). Manually adjusted models were then passed through real-space refinement in Phenix v1.20.1-4487 (Liebschner et al., 2019), and this process of manual and automated fitting was iterated several times to improve metrics like clash score, rotamer outliers, and Ramachandran outliers. Final models were then submitted through the PDB validation service, and remaining erroneous residues and ligands were manually adjusted in Coot v0.9.8.8 (Emsley et al., 2010) to generate deposited models. Fits of deposited models were consistent with the resolution of the maps. Data collection parameters and refinement statistics are summarized in Table 3. Figures were prepared with UCSF ChimeraX v1.6.1.

HeLa cells (human cervical adenocarcinoma) were acquired from ATCC (#CCL-2) and maintained in DMEM medium (#10569010; Gibco) supplemented with 10% fetal bovine serum (#10437028; Gibco) in a 37°C incubator set to 5% atmospheric CO₂. Cells were subcultured 2–3 times per week with subcultivation ratios between 1:4 and 1:8 to maintain 50–80% confluence.

For transfection, HeLa cells were seeded at a density of 4 × 10⁴ cells per well in 8-well chambered glass culture slides (#229168; Celltreat) and incubated overnight. The next day, culture medium was replaced with 345 μl fresh DMEM per well and mammalian expression plasmids were transfected into cells using the TransIT-HeLaMONSTER Transfection Kit (#MIR2904; Mirus Bio) according to the manufacturer’s protocol. Briefly, transfections were prepared for each well as follows: 328 ng plasmid was mixed with 34 μl of Opti-MEM I Reduced Serum Medium (#31985070; Thermo Fisher Scientific) prior to the addition of 1 μl TransIT-HeLa Reagent, gentle mixing, and then addition of 0.67 μl MONSTER Reagent and further mixing. This transfection mixture was incubated for 20 min at RT prior to dropwise addition to the appropriate well. Cells were incubated for 48 h without media change, then treated (or not) with 1 mM ribavirin (#R9644; Sigma-Aldrich) for 2 h to induce filament assembly prior to fixation and immunofluorescence staining.

HeLa cells were fixed with 4% paraformaldehyde for 15 min at RT, washed 3× with PBS, then permeabilized with 0.1% Triton X-100 for 5 min at RT. Cells were washed 3× with PBS between all further steps. Cells were incubated with primary antibodies diluted in PBS for 1 h at RT, then with secondary antibodies diluted in PBS for 1 h at RT in the dark, and finally with 5 µM DAPI (ab228549; Abcam) in PBS for 20 min at RT in the dark. The slide was mounted with ProLong Glass Antifade Mountant (#P36984; Thermo Fisher Scientific) on #1.5 thickness coverglass (#48393-251; VWR) and allowed to cure for 24 h prior to imaging. Slides were imaged at RT on a Leica DM5500B widefield fluorescence microscope equipped with a Leica HCX PL Fluotar 40X 0.75 numerical aperture objective, Lumencor SOLA Light Engine solid-state light source, Leica DAPI ET, GFP ET, and TXR ET filters, and Hamamatsu ORCA-Flash 4.0 LT sCMOS camera controlled by Leica Application Suite X acquisition software. Composite images were generated from single-channel micrographs using Fiji/ImageJ v1.54f (Schindelin et al., 2012).

Primary antibodies used were rabbit polyclonal IgG anti-IMPDH2 (1:500, #12948-1-AP; Proteintech) and mouse monoclonal IgG1 anti-FLAG epitope tag (1:200, #66008-3-Ig; Proteintech). Secondary antibodies used were Alexa Fluor 488-conjugated goat anti-rabbit IgG (1:400, #ab150077; Abcam) and Alexa Fluor 594-conjugated goat anti-mouse IgG (1:400, #ab150120; Abcam).

For quantification of the percentage of cells containing filaments (Figs. 7 and S5), two biological replicates were performed and a minimum of 100 cells were counted in each transfection condition.

Fig. S1 demonstrates by negative stain EM that S477D partially disrupts filament assembly of the canonical IMPDH1(514) under the same assay conditions as the retinal variants. Figs. S2, S3, and S4 summarize the cryo-EM data processing workflows for the inhibited IMPDH1(595)-S477D, inhibited IMPDH1(546)-S477D, and active IMPDH1(546)-S477D structures, respectively. They also contain example micrographs, 2D classes, FSC curves, and densities for bound ligands. Fig. S5 shows the same experiment as Fig. 7 but without ribavirin treatment, as overexpression of IMPDH1 alone leads to filament assembly.

Coordinates for atomic models are deposited in the PDB with accession codes: compressed IMPDH1(595)-S477D free octamer (8U7M); compressed IMPDH1(546)-S477D filament, octamer-centered (8U7Q); compressed IMPDH1(546)-S477D filament, interface-centered (8U7V); extended IMPDH1(546)-S477D filament, octamer-centered (8U8O); and extended IMPDH1(546)-S477D filament, interface-centered (8U8Y). Cryo-EM maps are deposited in the Electron Microscopy Data Bank with accession codes: compressed IMPDH1(595)-S477D free octamer (EMD-41986); compressed IMPDH1(546)-S477D filament, octamer-centered (EMD-41989); compressed IMPDH1(546)-S477D filament, interface-centered (EMD-42012); extended IMPDH1(546)-S477D filament, octamer-centered (EMD-42026); and extended IMPDH1(546)-S477D filament, interface-centered (EMD-42029).

The authors would like to thank the Arnold and Mabel Beckman Cryo-EM Center at the University of Washington for access to electron microscopes. We also thank Alex Merz for the use of a biosafety cabinet and incubators for human cell culture; Suzanne Hoppins and Andrea Wills for fluorescence microscope use; and Kelli Hvorecny, Eric Lynch, Richard Muniz, and Lauren Salay for fruitful discussion.

This work was supported by a Helen Hay Whitney Foundation postdoctoral fellowship to S.J. Calise; the U.S. National Institutes of Health grant nos. T32GM008268 to A.G. O’Neill and A.L. Burrell, F31EY030732 to A.L. Burrell, R01GM118396, R35GM149542, R21EY031546, and S10OD032290 to J.M. Kollman; and the Spanish Ministerio de Ciencia e Innovación-FEDER-Fondo Social Europeo grant PID2019-109671GB-I00 to R.M. Buey.

Author contributions: S.J. Calise: Conceptualization, Methodology, Validation, Formal Analysis, Investigation, Resources, Writing—Original Draft, Writing—Review & Editing, Funding Acquisition, Visualization, and Project Administration. A.G. O’Neill: Formal Analysis, Funding Acquisition, Investigation, and Resources. A.L. Burrell: Conceptualization, Funding Acquisition, Methodology, Investigation, Project Administration, and Resources. M.S. Dickinson: Formal Analysis and Investigation. J. Molfino: Investigation and Resources. C. Clarke: Investigation and Resources. J. Quispe: Investigation. D. Sokolov: Investigation. R.M. Buey: Funding Acquisition and Resources. J.M. Kollman: Conceptualization, Methodology, Resources, Writing—Review & Editing, Visualization, Supervision, and Funding Acquisition.