1. Biochemistry and Chemical Biology
Download icon

Evolution of (p)ppGpp-HPRT regulation through diversification of an allosteric oligomeric interaction

Research Article
  • Cited 0
  • Views 507
  • Annotations
Cite this article as: eLife 2019;8:e47534 doi: 10.7554/eLife.47534

Abstract

The alarmone (p)ppGpp regulates diverse targets, yet its target specificity and evolution remain poorly understood. Here, we elucidate the mechanism by which basal (p)ppGpp inhibits the purine salvage enzyme HPRT by sharing a conserved motif with its substrate PRPP. Intriguingly, HPRT regulation by (p)ppGpp varies across organisms and correlates with HPRT oligomeric forms. (p)ppGpp-sensitive HPRT exists as a PRPP-bound dimer or an apo- and (p)ppGpp-bound tetramer, where a dimer-dimer interface triggers allosteric structural rearrangements to enhance (p)ppGpp inhibition. Loss of this oligomeric interface results in weakened (p)ppGpp regulation. Our results reveal an evolutionary principle whereby protein oligomerization allows evolutionary change to accumulate away from a conserved binding pocket to allosterically alter specificity of ligand interaction. This principle also explains how another (p)ppGpp target GMK is variably regulated across species. Since most ligands bind near protein interfaces, we propose that this principle extends to many other protein–ligand interactions.

https://doi.org/10.7554/eLife.47534.001

Introduction

Regulation by signaling ligands is a universal mechanism for adaptation and homeostasis (Chubukov et al., 2014; Traut, 2008). How proteins evolve to be differentially regulated by their signaling ligands is under ongoing investigation (Najmanovich, 2017; Taute et al., 2014). A key signaling ligand in bacteria is the nucleotide (p)ppGpp. (p)ppGpp directly binds and regulates diverse protein targets at positions ranging from protein active sites to protein–protein interfaces (Corrigan et al., 2016; Gourse et al., 2018; Liu et al., 2015a; Sherlock et al., 2018; Wang et al., 2019; Zhang et al., 2018). However, how these diverse themes evolve across bacterial species have not been systematically elucidated. In addition, while most of the (p)ppGpp regulation is studied for high (p)ppGpp concentrations upon stress and starvation (Cashel et al., 1996; Gourse et al., 2018; Liu et al., 2015a), (p)ppGpp at basal levels also has important protective roles (Gaca et al., 2013; Gaca et al., 2015a; Kriel et al., 2012; Potrykus et al., 2011; Puszynska and O'Shea, 2017). However, molecular targets and mechanisms of basal (p)ppGpp regulation have not been delineated.

Here, we examined the specificity of (p)ppGpp binding by characterizing (p)ppGpp regulation of the enzyme hypoxanthine phosphoribosyltransferase (HPRT), one of the earliest identified targets of (p)ppGpp (Hochstadt-Ozer and Cashel, 1972; Kriel et al., 2012). Using biochemical, structural and evolutionary analyses of HPRT homologs across diverse bacterial phyla, we identify a conserved (p)ppGpp binding motif and identified a HPRT dimer–dimer interaction that allosterically positions a flexible loop to promote strong (p)ppGpp binding. The bacterial HPRT homologs lacking this dimer–dimer interaction are largely refractory to (p)ppGpp regulation. We propose that changes in protein oligomerization enable evolution of regulation by (p)ppGpp, despite evolutionary constraint imposed on the direct (p)ppGpp-binding site. We examined this principle using another (p)ppGpp target GMK and discuss its broad implications in evolutionary diversification of signaling ligand specificity.

Results

Regulation of HPRT activity by basal levels of (p)ppGpp is important for GTP homeostasis

HPRT is a purine salvage enzyme that converts purine bases and PRPP to GMP and IMP, which are precursors of the essential nucleotide GTP (Figure 1A, upper panel). HPRT activity was previously found to be inhibited by (p)ppGpp in several bacteria including Bacillus subtilis (Gaca et al., 2015b; Hochstadt-Ozer and Cashel, 1972; Kriel et al., 2012). In B. subtilis, (p)ppGpp ranges from <30 μM at basal concentrations to millimolar levels upon starvation. (p)ppGpp inhibits B. subtilis HPRT activity at an IC50 of ≈10 μM, allowing (p)ppGpp to regulate purine salvage at its basal concentrations (Figure 1B). The absence of (p)ppGpp in B. subtilis results in an uncontrolled GTP increase and toxicity upon guanine addition (Kriel et al., 2012) (Figure 1A, middle panel).

Regulation of HPRT by basal levels of (p)ppGpp is important for GTP homeostasis.

(A) Pathways showing the effect of extracellular guanine on B. subtilis GTP homeostasis. In WT, (p)ppGpp regulates HPRT and GMK. (p)ppGpp0 cannot produce (p)ppGpp (see B) and ecgmk ecgpt has enzymes resistant to (p)ppGpp regulation (see F). (B) E. coli XGPRT more weakly inhibited by pppGpp than B. subtilis HPRT. The IC50 for E. coli XGPRT is 45 µM compared to 10 µM for B. subtilis HPRT. Error bars represent SEM of triplicate. (C) Expression of E. coli XGPRT leads to imbalanced GTP/ATP homeostasis in B. subtilis. GTP/ATP ratio in B. subtilis treated with 1 mM guanosine as determined by thin layer chromatography of 32P-labeled cells. Time is minutes after guanosine treatment. In ecgmk, (p)ppGpp-insensitive E. coli gmk replaces B. subtilis gmk at its endogenous locus. ecgmk amyE::bsuhprT and ecgmk amyE::ecgpt express B. subtilis hprT and E. coli gpt, respectively, from an IPTG-dependent promoter at an exogenous locus. All strains grown with 1 mM IPTG. (D) Initial velocities of B. subtilis HPRT at varied PRPP and pppGpp concentrations. Data are fitted to a global competitive inhibition equation (r2 = 0.975), and Ki = 1.7 µM. (E) Data from (D) in a Hanes–Woolf transformation. Parallel lines indicate equal maximum velocities for each pppGpp concentration. Error bars represent SEM of at least three replicates.

https://doi.org/10.7554/eLife.47534.002

To demonstrate the physiological relevance of basal (p)ppGpp regulation of HPRT, we introduced the E. coli enzyme XGPRT (ecgpt) into B. subtilis. E. coli XGPRT produces GMP similarly to B. subtilis HPRT but is only modestly inhibited by (p)ppGpp at induced concentrations (Figure 1B). To visualize the effect of this regulation on GTP homeostasis, we also introduced a (p)ppGpp-resistant E. coli GMK to replace the (p)ppGpp-sensitive B. subtilis GMK, allowing the subsequent conversion of GMP to GDP to bypass (p)ppGpp regulation (Figure 1A, bottom panel). When these strains were treated with guanosine, GTP levels increased far more in cells with the modestly regulated ecgpt than in cells with the basal (p)ppGpp-regulated B. subtilis hprT (Figure 1C), demonstrating that basal regulation of HPRT by (p)ppGpp is important for preventing strong fluctuations in GTP.

We performed Michaelis Menten kinetics on B. subtilis HPRT and found (p)ppGpp is a strong competitive inhibitor of its substrate PRPP (Figure 1D–E). The Ki for pppGpp is 1.7 μM, while the Km for PRPP is 166 μM, allowing (p)ppGpp at a basal level to effectively compete with physiological concentrations (0.1–1 mM) of PRPP (Bennett et al., 2009; Hove-Jensen et al., 2017) and keep GTP under surveillance not just during starvation, but also during unstarved growth.

Conservation and variation of (p)ppGpp regulation of HPRT across species

It has been shown for the (p)ppGpp targets RNA polymerase and guanylate kinase that conservation of their binding to (p)ppGpp is limited within distinct bacterial phyla (Liu et al., 2015b; Ross et al., 2016). We therefore tested the conservation of HPRT regulation by conducting a broad biochemical survey of HPRT enzymes from free-living bacteria, human commensals, and pathogens. We revealed (p)ppGpp can bind and inhibit activities of HPRTs from every phylum of bacteria we tested, including Actinobacteria, Bacteroidetes, Firmicutes, Deinococcus-Thermus, and Proteobacteria, and even eukaryotic HPRTs (Figure 2). To do so, we purified 32 recombinant HPRTs and tested their enzymatic activities and found that most of them were strongly inhibited by ppGpp and pppGpp at 25 μM (Figure 2A–B and Table 1). Considering that (p)ppGpp is induced to millimolar concentrations during starvation and its uninduced levels are ≈10–30 μM during exponential growth in B. subtilis and E. coli (Cashel et al., 1996; Kriel et al., 2012), these results suggest significant inhibition of HPRT under physiological, basal levels of (p)ppGpp.

Figure 2 with 1 supplement see all
Conservation and variation in (p)ppGpp regulation of HPRTs across species.

(A–B) (p)ppGpp inhibits HPRTs across species. Representative phylogenetic tree constructed from 16S rRNA sequences of species shown with eukaryotic outgroup hidden. Branch length scale represents 0.5 substitutions per site. See Materials and methods for tree construction. (A) Relative activities of HPRTs with 25 μM ppGpp. (B) Relative activities of HPRTs with 25 μM pppGpp. Error bars represent SEM of triplicates. (C–D) pppGpp binds HPRTs across species. (C) Representative DRaCALA between B. anthracis Hpt-1 in cell lysate serially diluted 1:2 and 32P-labeled pppGpp. The central signal is proportional to pppGpp – HPRT interaction. Binding isotherm from isothermal titration calorimetry between B. anthracis Hpt-1 and pppGpp. See Figure 2—figure supplement 1 for energy isotherm and parameters. (D) The Kd between pppGpp and HPRTs obtained with DRaCALA from serially diluted cell lysates containing overexpressed HPRTs (see Materials and methods). Error bars represent SEM derived from one binding curve. P. phosphoreum and C. burnetii HPRTs with * have affinities too weak to calculate but are estimated to be >15 μM (see Figure 2—figure supplement 1). (E) Relative activities of HPRTs with increasing concentrations of pppGpp. See Table 1 for partial datasets for HPRTs from M. tuberculosis, C. crescentus, S. meliloti, R. torques, S. mutans, and C. gilvus.

https://doi.org/10.7554/eLife.47534.003
Table 1
(p)ppGpp binding and inhibition of HPRT homologs.
https://doi.org/10.7554/eLife.47534.005
OrganismRelative activitydKd (pppGpp) (μM)IC50 (pppGpp)
(μM)
25 μM ppGpp25 μM pppGpp
Actinobacteria
Streptomyces coelicolor−0.01 ± 0.010.12 ± 0.011.77 ± 0.313.8 ± 0.2
Collinsella aerofaciens0.31 ± 0.040.42 ± 0.051.46 ± 0.56
Cellulomonas gilvusNAbNAb0.6 ± 0.11
Mycobacterium tuberculosisNAaNAaNAc
Bacteroidetes
Bacteroides intestinalis0.49 ± 0.060.05 ± 0.010.11 ± 0.03
Bacteroides caccae0.34 ± 0.030.01 ± 0.020.11 ± 0.02
Bacteroides ovatus0.43 ± 0.04−0.03 ± 0.020.15 ± 0.04
Bacteroides uniformis0.4 ± 0.010.01 ± 0.010.1 ± 0.04
Bacteroides thetaiotaomicron0.41 ± 00 ± 0.010.11 ± 0.04
Bacteroides finegoldii0.46 ± 0.03−0.01 ± 0.010.15 ± 0.06
Firmicutes
Eubacterium ventriosum−0.03 ± 0.03−0.07 ± 0.030.66 ± 0.1
Ruminococcus lactaris−0.01 ± 00 ± 00.11 ± 0.03
Ruminococcus torquesNAbNAb0.22 ± 0.07
Clostridium leptum−0.01 ± 0.010 ± 03.29 ± 0.94
Eubacterium eligens−0.03 ± 0.030.02 ± 0.010.79 ± 0.18
Blautia hansenii0.07 ± 00.14 ± 0.010.83 ± 0.14
Eubacterium hallii0.17 ± 0.05−0.06 ± 0.050.07 ± 0.02
Clostridium spiroforme0.07 ± 0.010.17 ± 09.54 ± 1.17
Listeria monocytogenes0.13 ± 00.32 ± 0.020.32 ± 0.06
Bacillus subtilis0.08 ± 0.010.41 ± 0.010.75 ± 0.0710.1 ± 1.3
Staphylococcus aureus0.31 ± 0.010.48 ± 0.012.83 ± 0.3810.6 ± 0.3
Streptococcus mutansNAbNAb5.12 ± 1.45
Bacillus anthracis 10.25 ± 0.010.5 ± 0.011.54 ± 0.1829.5 ± 1.2
Enterococcus faecalis0.31 ± 0.010.49 ± 0.021.12 ± 0.11
Holdemania filiformis0.66 ± 0.010.43 ± 00.17 ± 0.11
Bacillus anthracis 20.81 ± 0.040.7 ± 0.056.18 ± 1.77
Deinococcus-Thermus
Thermus thermophilus0.52 ± 0.07−0.01 ± 0.010.04 ± 0.01
Proteobacteria
Caulobacter crescentus0.33 ± 0.050.47 ± 0.070.24 ± 0.09
Escherichia coli0.33 ± 0.010.48 ± 0.023.38 ± 1.4726.2 ± 2.0
Klebsiella pneumoniae0.47 ± 0.030.49 ± 0.013.38 ± 0.89
Sinorhizobium melilotiNAbNAb3.08 ± 0.92
Yersinia enterocolitica0.7 ± 0.010.7 ± 0.034.44 ± 1.25
Photobacterium phosphoreum0.97 ± 0.010.96 ± 0.03NAc>100
Pseudomonas syringae0.6 ± 0.010.6 ± 0.015.85 ± 1.76
Pseudomonas aeruginosa0.84 ± 0.010.83 ± 0.025.28 ± 1.52
Legionella pneumophila0.78 ± 0.010.69 ± 0.031.72 ± 0.38>100
Coxiella burnetii0.79 ± 0.140.72 ± 0.16NAc>100
Neisseria meningitidis0.9 ± 0.020.8 ± 0.031.64 ± 0.55
Eukarya
Saccharomyces cerevisiae0.64 ± 0.020.96 ± 0.018.08 ± 8.58
Caenorhabditis elegansNAbNAb12.77 ± 7.6829.6 ± 2.8
Homo sapiensNANANA66.6 ± 5.4
  1. ± standard error of the mean.

    NA = no data obtained.

  2. aProtein not purified.

    bProtein did not have activity.

  3. cInteraction with pppGpp too weak to estimate Kd.

    dRelative to activity without ppGpp or pppGpp.

We also quantified the HPRT homologs’ binding affinity (Kd) for pppGpp with differential radial capillary action of ligand assay (DRaCALA) (Figure 2C–D) (Roelofs et al., 2011), results of which were highly comparable with well-established quantitative methods such as isothermal titration calorimetry (Figure 2C and Figure 2—figure supplement 1). Using this method, we found that most HPRTs bind pppGpp with Kd values ranging from ≈0.1 to ≈10 μM (Figure 2D and Table 1).

Closer inspection revealed an additional quantitative difference in regulatory potency among these HPRTs. HPRTs from human microbiota commensals, including all species from Bacteroidetes along with the Firmicutes Eubacterium spp. and Ruminococcus spp., manifested the tightest interactions with pppGpp (Kd ≈ 0.1–1 μM) (Figure 2D and Table 1). This suggests a relationship between the intestinal environment and regulation of this purine salvage enzyme by (p)ppGpp, which may serve to buffer the intracellular environment against fluctuating extracellular purine concentrations. Beyond microbiota, HPRTs from soil-dwelling bacteria (e.g. Streptomyces coelicolor, B. subtilis) and pathogens (e.g. Bacillus anthracis, Staphylococcus aureus, E. coli), occupying the phyla Actinobacteria, Firmicutes and Proteobacteria, were also inhibited by (p)ppGpp with Kd values below 10 μM.

Interestingly, a few HPRTs were only weakly inhibited by 25 μM (p)ppGpp (Figure 2A–B). These weakly regulated HPRTs all clustered in β- and γ-Proteobacteria (e.g. Pseudomonas aeruginosa and Neisseria meningitidis), with the exception of Mycobacterium tuberculosis (Figure 2—figure supplement 1C). To test whether the HPRTs weakly inhibited at basal levels of (p)ppGpp (25 μM) can be inhibited by higher levels of (p)ppGpp, we examined select HPRT homologs at a range of pppGpp concentrations and at physiological substrate concentrations (1 mM PRPP and 50 μM guanine) (Bennett et al., 2009) (Figure 2E). Most bacterial HPRTs (e.g. B. subtilis, E. coli, S. aureus) were sensitive to basal pppGpp. However, several bacterial HPRTs that were weakly inhibited by 25 μM pppGpp (e.g. C. burnetii, L. pneumophila) are also almost refractory to high concentration of pppGpp. In fact, they are less inhibited by pppGpp than eukaryotic HPRTs from human and C. elegans, for which (p)ppGpp is not a physiological regulatory ligand. Therefore, we can broadly classify bacterial HPRTs by their sensitivity to (p)ppGpp: those that are sensitive to basal (p)ppGpp and a few that are resistant to basal (p)ppGpp, although may still be mildly sensitive to induced (p)ppGpp.

In addition to differences between HPRT homologs, we also noticed a strong, species-dependent difference between pppGpp and ppGpp inhibition. All HPRTs in Bacteroidetes were potently inhibited by pppGpp but only weakly by ppGpp (Figure 2A–B). In contrast, nearly all other HPRTs displayed stronger inhibition by ppGpp than pppGpp. It is noted that Bacteroides spp., whose HPRTs are more sensitive to pppGpp than ppGpp, contain a GppA homolog, and they make mostly ppGpp in vivo (Glass et al., 1979). On the other hand, Firmicutes, whose HPRTs are more sensitive to ppGpp than pppGpp, lack GppA homologs and produce predominantly pppGpp. Therefore, although pppGpp and ppGpp are often regarded as similar, they can have marked differences for certain cellular targets.

(p)ppGpp binds the conserved active site of HPRT and closely mimics substrate binding

To examine the molecular determinants underlying (p)ppGpp regulation of HPRT, we turned to Hpt-1 from the pathogenic bacterium B. anthracis, which pppGpp competitively inhibits similarly to B. subtilis HPRT (Figure 3—figure supplement 1). We crystallized Hpt-1 with and without ppGpp (Figure 3A and Table 2), and for comparison, we also crystallized Hpt-1 with its two substrates, PRPP and the non-reactive guanine analog 9-deazaguanine (Figure 3B and Table 2) (Héroux et al., 2000). Apo HPRT diffracted to 2.06 Å resolution with four molecules in the asymmetric unit and was nearly identical to a deposited apo structure of B. anthracis Hpt-1 (PDB ID 3H83) (Figure 3—figure supplement 2). The HPRT-substrates structure diffracted to 1.64 Å containing two molecules in the asymmetric unit (Figure 3—figure supplement 2), and PRPP and 9-deazaguanine were coordinated by two Mg2+ ions with each monomer (Figure 3B and Figure 3—figure supplement 3). The HPRT-ppGpp complex diffracted to 2.1 Å resolution with two molecules in the asymmetric unit (Figure 3—figure supplement 2), and each monomer contained one ppGpp coordinated with Mg2+, in agreement with near 1:1 stoichiometry measured via ITC (Figure 2—figure supplement 1 and Figure 3—figure supplement 3). These crystals formed in drops containing pppGpp, but there was insufficient density to completely model the 5′ γ-phosphate (Figure 3—figure supplement 4). With LC-MS/MS, we verified that the pppGpp was not contaminated with ppGpp. Although it is possible that the γ-phosphate was hydrolyzed during crystallization, the presence of waters and unassigned density around the 5′ phosphates allow for the possibility that the γ-phosphate is present but dynamic.

Figure 3 with 8 supplements see all
(p)ppGpp binds the HPRT active site and shares an almost identical binding pocket with substrates.

(A) B. anthracis Hpt-1 crystallized with ppGpp. Five loops (I, II, III, III′, IV) form the HPRT active site (Sinha and Smith, 2001). ppGpp and Mg2+ shown with omit electron density contoured at 2σ. Figure 3—figure supplement 1 shows that pppGpp competitively inhibits Hpt-1. Figure 3—figure supplement 4 shows omit electron density for the two ppGpp molecules crystallized in the asymmetric unit. (B) B. anthracis Hpt-1 crystallized with 9-deazaguanine, PRPP, and two Mg2+. Ligands shown with omit electron density contoured at 2.5σ. Residues Tyr70 – Ser76 in loop II were not resolved. See Figure 3—figure supplement 2 for asymmetric units of Hpt-1 crystallized with ppGpp, substrates, and sulfates. See Figure 3—figure supplement 3 for coordination of Mg2+ with ppGpp and PRPP. (C) Overlay of substrates (PRPP and 9-deazaguanine; yellow) and inhibitor (ppGpp; purple) bound to HPRT. Spheres represent Mg2+ and are colored according to their coordinating ligand. See Figure 3—figure supplement 6 for binding pocket comparison. (D) The ppGpp binding pocket on HPRT. Black dotted lines indicate select hydrogen bonds. The peptide backbone is shown for residues where the interactions are relevant. See Figure 3—figure supplement 5 for complete interaction maps. (E) DRaCALA of B. subtilis HPRT variants binding to 32P-labeled pppGpp, determined with cell lysates containing overexpressed HPRT variants. Disruption of side chain interactions weakened binding. For E and F, the residues are colored according to their interaction with ppGpp. Red = side chain interaction, blue = backbone amide interaction, and gray = both side chain and backbone interactions. Error bars represent SEM of three replicates. (F) Sequence frequency logo of the (p)ppGpp binding pocket from 99 bacterial HPRTs. Logo created using WebLogo from UC-Berkeley. See Figure 3—figure supplement 8 for binding residues from select bacterial and eukaryotic HPRTs.

https://doi.org/10.7554/eLife.47534.006
Table 2
X-ray data collection and structure determination statistics.
https://doi.org/10.7554/eLife.47534.015
Data collection
StructureHPRT with sulfates
(PDB 6D9Q)
HPRT with substrates
(PDB 6D9R)
HPRT with ppGpp
(PDB 6D9S)
Wavelength0.97860.978720.9786
Resolution range (highest resolution bin) (Å)41.3–2.06 (2.13–2.06)50–1.64 (1.67–1.64)40.2–2.11 (2.18–2.11)
Space groupP 31 2 1P 32 2 1P 31 2 1
Unit cell
a, b, c (Å)82.597, 82.597, 242.416113.758, 113.758, 56.73182.61, 82.61, 174.92
α, β, γ (°)90, 90, 12090, 90, 12090, 90, 120
Completeness (%)99.16 (95.52)99.8 (99.0)98.92 (94.51)
Unique reflections602645175040413
Redundancy10.8 (6.4)21.5 (14.4)8.1 (6.2)
I/σI24.8 (1.74)31.09 (2.93)16.41 (1.22)
Rmeas0.1240.137 (1.376)0.164
Rpim0.0370.029 (0.35)0.06
CC 1/2(0.791)(0.81)(0.724)
Refinement
Resolution range (highest resolution bin) (Å)41.3–2.06 (2.11–2.06)37.19–1.64 (1.70–1.64)40.2–2.11 (2.16–2.11)
Rwork/Rfree a (%)18.2/21.716.5/19.620.2/24.4
r.m.s.b deviations
Bonds (Å)0.0040.010.003
Angles (Å)1.031.410.812
Ramachandran statistics (%)
Favored97.9098.5296.63
Allowed2.101.483.37
Disallowed0.000.000.00
Rotamer outliers (%)0.780.300.63
No. of atoms
Macromolecules574228812837
Ligands46161116
Solvent440317118
B factor (Å2)
Macromolecules50.6324.0676.03
Ligands68.8337.76161.17
Solvent52.1535.5370.22
  1. aRwork/Rfree = Σ||Fobs|-|Fcalc||/|Fobs|, where the working the free R factors are calculated by using the working and free reflection sets, respectively. The free R reflections were held aside throughout refinement. bRoot mean square. See Table 2—source datas 13 for PDB validation reports.

Comparison between the HPRT-ppGpp and HPRT-substrates structures revealed that ppGpp binds the HPRT active site and closely mimics the conformation of the two substrates (Figure 3C and Figure 3—figure supplements 5 and 6). The purine ring of ppGpp overlaps with the purine base substrate, and the two phosphate arms of ppGpp and PRPP both spread across the active site between loop I and loop III (Figure 3C and Figure 3—figure supplement 6). Generally speaking, in ppGpp–protein interactions, the phosphates of ppGpp are either elongated or compacted in a ring-like conformation (Steinchen and Bange, 2016). With the phosphates of ppGpp spread across the binding pocket, the HPRT-ppGpp interaction represents an elongated conformation (Figure 3A).

The (p)ppGpp-binding site in HPRT manifests a novel (p)ppGpp–protein interaction. The 5′ phosphates and ribose of ppGpp interact with loop III (EDIIDSGLT), a well-characterized PRPP binding motif (Sinha and Smith, 2001), through side chain interactions with Glu99 and Asp100 and backbone amide interactions with Asp103 – Thr107 (Figure 3D and Figure 3—figure supplement 7). The 3′ phosphates of ppGpp are coordinated by backbone amides of loop I, the side chain of Arg165, and the Mg2+ ion (Figure 3D and Figure 3—figure supplement 7). The guanine ring of ppGpp is surrounded by a hydrophobic cleft formed by Ile101, Phe152, and Leu158, and Lys131 hydrogen bonds the guanine’s exocyclic oxygen (Figure 3D and Figure 3—figure supplement 7). We validated the (p)ppGpp-binding residues by mutation analyses which showed that altering residues with side chain interactions with (p)ppGpp greatly weakened pppGpp binding (Figure 3E). In sum, this binding pocket illustrates the first example of a (p)ppGpp binding motif that shares the PRPP-binding motif. Since many proteins bind PRPP (Hove-Jensen et al., 2017), this (p)ppGpp motif may represent a new class of (p)ppGpp targets.

A frequency logo of the binding pocket from 99 bacterial HPRTs shows that most (p)ppGpp-interacting residues are highly conserved across bacteria (Figure 3F and Figure 3—figure supplement 8), and nearly all binding residues are also conserved in the eukaryotic HPRTs (Figure 3—figure supplement 8). The strong conservation across species is not surprising given the close overlap between (p)ppGpp and substrates in the HPRT structures. Since all residues involved in binding ppGpp are also involved in binding substrates, altering the site to affect inhibitor binding would also impact enzyme activity.

The nearly identical recognition of (p)ppGpp and substrates, along with the conservation of the active site, raised the following question: how can some HPRTs be strongly inhibited by basal levels of (p)ppGpp and other HPRTs be almost refractory to (p)ppGpp control despite sharing a conserved binding pocket?

(p)ppGpp prevents PRPP-induced dissociation of HPRT dimer-of-dimers

We noticed a significant difference between the ppGpp- and substrates-bound tertiary structures in the conformation of a flexible loop. This loop, also called loop II, is common to all HPRTs and covers the active site during catalysis (Shi et al., 1999). In the HPRT-ppGpp complex, one side of loop II faces the active site while the other side is a critical part of the interface between two HPRT dimers (Figure 4A). We applied PISA analysis (‘Protein interfaces, surfaces, and assemblies’ service at European Bioinformatics Institute) to predict that this dimer-dimer interaction results in a tetrameric HPRT (Krissinel and Henrick, 2007). In the substrates-bound state, loop II is instead shifted ≈4 Å toward the active site from its position in the dimer-dimer interface (Figure 4A). Loop II also pulls its flanking dimer interface components β3 and α3 toward the active site (Figure 4A). While crystal contacts likely prevent loop II from completely moving over the active site (Figure 4—figure supplement 1), PISA analysis predicts that these changes abolish the dimer-dimer interaction. Using size-exclusion chromatography, we confirmed that apo B. subtilis HPRT is a tetramer (Figure 4B). In contrast, when PRPP, the first substrate to bind HPRT (Yuan et al., 1992), is added at a high concentration (500 μM) in the mobile phase, HPRT tetramers dissociate to dimers (Figure 4B and Figure 4—figure supplements 2, 3 and 4).

Figure 4 with 7 supplements see all
(p)ppGpp counteracts substrate-induced HPRT dimerization and HPRT tetramerization potentiates (p)ppGpp binding.

(A) Overlay of HPRT tetramer crystallized with ppGpp (silver) and HPRT dimer crystallized with substrates (salmon). ppGpp is purple and the substrates are yellow. Inset, View of the dimer–dimer interface within an HPRT tetramer. Secondary structure components at the interface are labeled β3, α3, and loop II. Figure 4—figure supplement 1 shows that changes in loop II are not induced by crystal contacts. (B) Size-exclusion chromatography of B. subtilis HPRT without ligand (left) and with PRPP (right). PRPP addition shifts the oligomeric state from tetramer to dimer. B. anthracis Hpt-1 is also a dimer with PRPP (see Figure 4—figure supplement 2). In B and D, solid line shows B. subtilis HPRT tetramer peak and dotted line shows dimer peak. PRPP must be in mobile phase to cause the shift (Figure 4—figure supplement 3), and the purine base does not affect the oligomeric state with PRPP (Figure 4—figure supplement 4). For molecular weight standards, see Figure 4—figure supplement 7. (C) B. subtilis HPRT crosslinked with dimethyl adipimidate (DMA). Crosslinked multimers were separated on SDS–PAGE. Predicted multimers are shown on the left. Shaded circles represent predicted to be crosslinked monomers, and white circles represent incomplete crosslinking. 1, 2 indicates order of incubation. GTP competition with PRPP did not recapitulate pppGpp blockage of PRPP-induced dimer–dimer dissociation which could be due to difference in affinity, whereby pppGpp binds strongly enough to outcompete PRPP but GTP does not. See Figure 4—figure supplement 5 for effect of crosslinked tetramer (Y117C) on HPRT activity and PRPP competition with pppGpp. See Figure 4—figure supplement 6 for ITC with PRPP. (D) Size-exclusion chromatography of B. subtilis HPRT with a partial loop II deletion (∆70–76) or a complete loop II deletion (∆ 69–77). The complete deletion results in dimerization of the protein. (E) Binding curves between 32P-labeled pppGpp and B. subtilis HPRT variants obtained with DRaCALA. The tetrameric Δloop II (70-76) variant (orange squares) binds as well as wild type HPRT. The dimeric Δloop II (69-77) variant (purple triangles) displays weaker binding. Error bars represent SEM of three replicates.

https://doi.org/10.7554/eLife.47534.019

We next interrogated the effect of (p)ppGpp on the oligomeric state of HPRT using the protein crosslinker dimethyl adipimidate (DMA). Crosslinked apo HPRT was resolved by SDS–PAGE as bands corresponding to monomers, dimers, trimers, or tetramers (Figure 4C, lane 2). All but the tetramer is likely formed by incomplete crosslinking, since dynamic light scattering showed apo B. anthracis Hpt-1 to be a homogeneous population with a hydrodynamic radius (RH) consistent with a tetramer in solution (Table 3). HPRT with pppGpp remained a tetramer according to both crosslinking and dynamic light scattering experiments (Figure 4C and Table 3). Incubation of HPRT with PRPP resulted in loss of trimer and tetramer bands in SDS–PAGE (Figure 4C, lane 3) and a corresponding shift in tetramer to homogeneous dimer with dynamic light scattering (Table 3), confirming that PRPP-bound HPRT is a dimer.

Table 3
Dynamic light scattering of B. anthracis Hpt-1 with pppGpp and PRPP.
https://doi.org/10.7554/eLife.47534.027
SampleRHa (nm)Polydb% PolydEst. MW (kDa)
HPRT4.10.923.190
HPRT + PRPP3.10.722.547
HPRT + pppGpp4.11.024.991
HPRT + both4.21.227.897
  1. PRPP = phosphoribosyl pyrophosphate.

    aRH = hydrodynamic radius.

  2. bPolyd = polydispersity.

Importantly, with both PRPP and pppGpp present, HPRT was tetrameric (Figure 4C, lanes 6 and 7 and Table 3), indicating that pppGpp prevents PRPP-induced dimer–dimer dissociation. Thus, pppGpp appears to selectively stabilize HPRT tetramers against PRPP-induced dissociation.

These data also suggest that preventing HPRT from dimer–dimer dissociation and keeping it as a tetramer will disfavor PRPP binding. Indeed, a cysteine substitution across the dimer–dimer interface (Y117C) forcing HPRT to stay as a tetramer via disulfide crosslink results in lower enzymatic activity, and PRPP cannot outcompete pppGpp for binding the tetrameric HPRT variant (Figure 4—figure supplement 5).

Taken together, these data reveal that HPRT binds PRPP as catalytically active dimers, whereas (p)ppGpp maintains HPRT as tetramers.

Dimer–dimer interaction allosterically positions loop II for potentiated (p)ppGpp binding

Given HPRT’s oligomerization, we performed ITC experiments with pppGpp (Figure 2—figure supplement 1) and PRPP (Figure 4—figure supplement 6) to examine ligand binding cooperativity. In the conditions we tested, within physiological ranges of pppGpp and PRPP, we did not observe significant cooperativity. This is expected as cooperativity has not been observed in native bacterial and human HPRTs and has been reported only in mutant human HPRTs (Balendiran et al., 1999; Guddat et al., 2002; Lightfoot et al., 1994; Patta et al., 2015).

To understand the importance of HPRT’s differential oligomeric state with substrates and (p)ppGpp, we performed interface mutational analyses, and we found that, unexpectedly, the dimer–dimer interaction is critical for HPRT’s regulation by (p)ppGpp. We disrupted the dimer–dimer interface of B. subtilis HPRT by constructing two loop II deletion variants of different lengths that resulted in different apo oligomeric states (Figure 4D). While the tetrameric Δloop II (70-76) bound pppGpp as tightly as wild-type HPRT, the dimeric Δloop II (69-77) displayed strongly ablated binding to pppGpp (Kd too weak to estimate; Figure 4E).

Since an engineered dimeric HPRT variant has weakened binding to (p)ppGpp, we next examined the oligomeric state of naturally occurring HPRT homologs with weakened inhibition by (p)ppGpp (Figure 2). Strikingly, in nearly all cases, the (p)ppGpp-insensitive HPRTs were also constitutive dimers (Figure 5A). These homologs have normal enzymatic activity (Table 4) and only differ in their ability to be regulated by (p)ppGpp. Our results suggest that tetrameric HPRT, but not dimeric HPRT, allows (p)ppGpp at basal levels to bind to and inhibit its activity.

Figure 5 with 3 supplements see all
Dimer–dimer interaction holds loop II away from the (p)ppGpp binding pocket to potentiate (p)ppGpp binding.

(A) Size-exclusion chromatograms show that C. burnetii, L. pneumophila, N. meningitidis, P. aeruginosa, and P. syringae HPRTs are non-tetrameric without ligands. Vertical lines represent B. subtilis HPRT tetramer and dimer peaks. (B) Overlay of apo B. anthracis Hpt-1 (teal) and the L. pneumophila HPRT interface (wheat; PDB ID 5ESW). In L. pneumophila HPRT, loop II and β3 are further from the dimer-dimer interface and closer to the active site. See Figure 5—figure supplement 1 for overlay with substrates-bound HPRT. (C) Overlay of B. anthracis Hpt-1 – ppGpp (silver; ppGpp in purple) and L. pneumophila HPRT (wheat; PDB ID 5ESX) shows that the 5′ phosphate binding pocket is compressed by the conformation of loop II in the dimeric HPRT. Arg76 (side chain shown as sticks) in L. pneumophila HPRT forms part of the compressed pocket. Asp109 in L. pneumophila HPRT is hidden due to poor electron density. (D) Size-exclusion chromatogram of the dimeric Bsu-Lpn chimera. The chimera has 21 residues at the B. subtilis HPRT interface replaced with L. pneumophila HPRT residues (see Figure 5—figure supplement 2). (E) DRaCALA shows that the Bsu-Lpn chimera binds 32P-labeled pppGpp more weakly. Error bars represent SEM of three replicates. Bsu-Lpn chimera performed in duplicate. Figure 5—figure supplement 3 shows that stability of dimeric HPRTs is not compromised. (F) Competition between PRPP and 32P-labeled pppGpp shows that the dimeric Bsu-Lpn chimera is more sensitive to PRPP than the tetrameric wild type. Error bars are range of duplicate.

https://doi.org/10.7554/eLife.47534.028
Table 4
HPRT kinetic parameters.
https://doi.org/10.7554/eLife.47534.032
HPRTKm (guanine)
(μM)
Km (PRPP)
(μM)
kcat
(s−1)
B. anthracis 15.2 ± 1.8111.6 ± 16.822
B. subtilis--165.6 ± 11.925.4
C. burnetii17.6 ± 2.044 ± 3.312.9
P. phosphoreum--81 ± 12.5a30.4a
N. meningitidis68 ± 7.0103 ± 129.6
E. colib--192 ± 7.0a59a
L. pneumophilac10.5 ± 1.160 ± 9--
M. tuberculosisd10 ± 1650 ± 700.193
  1. PRPP = phosphoribosyl pyrophosphate.

    aobtained with hypoxanthine as the purine base.

  2. dPatta et al. (2015).

    ± standard error.

The structural basis of how the dimer–dimer interaction promotes (p)ppGpp binding in tetrameric HPRTs can be seen from comparative structures of apo B. anthracis Hpt-1 and apo L. pneumophila HPRT (PDB ID 5ESW) (Zhang et al., 2016). Loop II in dimeric L. pneumophila HPRT is positioned closer to the active site (Figure 5B), mimicking substrates-bound HPRT even in the absence of substrates (Figure 5—figure supplement 1), and compresses the cavity surrounding the 5′ phosphates of ppGpp (Figure 5C). This conformation does not fully accommodate (p)ppGpp, but it should accommodate PRPP since its 5′ monophosphate fits in the cavity compressed by loop II (Figure 5—figure supplement 1). In contrast, in tetrameric HPRTs, loop II is pulled away from the active site by the dimer–dimer interaction and is positioned for optimal (p)ppGpp binding. There are many examples of the interdependence of protein oligomerization and ligand binding (Traut, 1994). By discovering and comparing two different evolved oligomeric states of HPRT, we have been able to demonstrate how a structural rearrangement through oligomerization allosterically affects ligand binding to a non-interface pocket.

To identify the determinant for the oligomeric state of naturally occurring HPRTs, we constructed a chimera of B. subtilis HPRT with 21 dimer–dimer interface residues replaced with their corresponding L. pneumophila HPRT residues (Figure 5—figure supplement 2). The chimera is a stable dimer (Figure 5D and Figure 5—figure supplement 3), indicating that the determinants for oligomerization lie in the dimer-dimer interface residues. The chimera also had a > 20 fold lower affinity for pppGpp than the tetramer (Kd ≈ 24 μM versus ≈ 1 μM) (Figure 5E), and PRPP more potently competed with pppGpp binding to the dimeric chimera than to tetrameric WT (Figure 5F), suggesting that there may be an evolved linkage between the dimer–dimer interface of HPRT tetramers and (p)ppGpp regulation.

Dimer-dimer interface coevolved with strong (p)ppGpp binding across species

The relationship between HPRT oligomeric state and sensitivity to (p)ppGpp prompted us to examine whether HPRT oligomerization has coevolved with (p)ppGpp regulation. To test this, we turned to ancestral protein sequence reconstruction to infer the evolution of HPRT (Hochberg and Thornton, 2017). With a phylogenetic tree derived from an alignment of 430 bacterial and eukyarotic HPRTs, we used maximum likelihood to infer the most likely ancestral HPRT protein sequences based on the phylogeny of the extant HPRT sequences (Yang et al., 1995) (Figure 6A). From the first ancestral HPRT (Anc1), the phylogenetic tree bifurcated into two broad lineages. One lineage contained the known dimeric HPRTs with weakened (p)ppGpp inhibition (Figure 6A) (Hug et al., 2016). The second lineage contained the vast majority of HPRTs, including (p)ppGpp-sensitive HPRTs in the Bacteroidetes, Firmicutes, and γ-proteobacteria. There are three interface residues conserved only in the dimeric lineage (Trp82, Pro86, and Ala110) (Figure 6B). Varying these residues in B. subtilis HPRT produces non-tetrameric variants (Figure 6—figure supplement 1). Notably, Anc1 HPRT was likely tetrameric, since it lacks the residues found in the dimeric lineage (Figure 6B).

Figure 6 with 2 supplements see all
Coevolution of the HPRT dimer-dimer interface and regulation by basal levels of (p)ppGpp.

(A) A maximum likelihood phylogenetic tree of 430 HPRT amino acid sequences rooted on eukaryotic HPRTs. Bootstrap values from 100 replicates are shown. Anc1-3 refer to ancestral HPRTs. The tree reveals two main lineages: one with dimer-dimer interaction motifs (Anc2; purple) and one lacking the interaction motifs (Anc3; blue). Branch length scale indicates 0.5 substitutions per site. See Materials and methods for tree construction. See Source Data one for the full alignment of HPRTs. (B) Alignment of HPRT dimer-dimer interface from Anc1, Anc2, Anc3, and select extant HPRTs. Nearly all (p)ppGpp-regulated HPRTs share Lys81 found in Anc2. Dimeric HPRTs insensitive to basal (p)ppGpp lack Lys81, but share Trp82, Pro86, and Ala110 (B. anthracis Hpt-1 numbering). Figure 6—figure supplement 1 shows the role of these residues at the interface. For the ancestors, red indicates high (>90%), yellow moderate (50–90%), and no color low (<50%) confidence level in the residue identity. (C) B. subtilis K81E and K81A HPRTs have weakened tetramerization according to size-exclusion chromatography. Figure 6—figure supplement 2 shows Lys81 at the dimer–dimer interface. (D) Binding of 32P-labeled pppGpp to wild type (black circles), K81A (red squares), and K81E (blue triangles) HPRTs using DRaCALA. Error bars represent SEM of three replicates. (E) PRPP competition with 32P-labeled pppGpp for binding wild type (black circles) and K81A HPRTs (red squares). K81A is more sensitive to PRPP competition than the tetrameric wild type. Error bars are range of duplicate.

https://doi.org/10.7554/eLife.47534.033

A tracing of the dimer-dimer interface pinpointed residues that coevolve with (p)ppGpp regulation. We followed the change in interface residues from Anc1 to Anc2, the ancestor common to the (p)ppGpp-inhibited HPRTs (Figure 6A). One residue in particular, Lys81, increased in confidence between Anc1 and Anc2 (Figure 6B). Lys81 is part of a conserved β-strand and loop motif (residues 81–87) that interact with one another across the dimer–dimer interface, so their evolution is likely important for HPRT tetramerization (Figure 6—figure supplement 2). Lys81 is highly conserved in (p)ppGpp-regulated HPRTs but not in (p)ppGpp-insensitive HPRTs (Figure 6B), suggesting that (p)ppGpp regulation is associated with the evolution of this residue. Therefore, we constructed Lys81 variants with weakened dimer–dimer interfaces (Figure 6C). One variant with a charge reversal (K81E) resulted in a mostly dimeric HPRT and a less disruptive K81A HPRT exhibited a rapid equilibrium between tetramer and dimer (Figure 6C). Importantly, pppGpp bound less well to both K81A and K81E HPRT variants relative to wild-type HPRT (Figure 6D), and PRPP more potently competed with pppGpp binding to the K81A variant relative to the tetrameric wild type (Figure 6E). We conclude that the HPRT dimer-dimer interface has coevolved with strong (p)ppGpp regulation by sequestering loop II at the dimer–dimer interface and opening the active site for (p)ppGpp binding (Figure 7), whereas evolution of interface residues associated with losing this dimer-dimer interaction weakens (p)ppGpp regulation.

Model for how evolution of protein oligomerization affects ligand-mediated regulation.

For an enzyme such as HPRT, where the inhibitor (p)ppGpp binds to nearly identical sites as the substrates (PRPP), evolutionary plasticity of inhibition can be mediated through subunit oligomerization. Dimeric HPRTs present a smaller binding pocket that preferentially bind to the substrate than to (p)ppGpp. On the other hand, HPRTs with a dimer–dimer interface motif would allosterically open the pocket, thus favoring (p)ppGpp binding and inhibition than substrate-binding and activity.

https://doi.org/10.7554/eLife.47534.039

Protein–protein interface allosterically alters the binding of (p)ppGpp to guanylate kinase

Intrigued by finding the determinants of (p)ppGpp regulation of HPRT, we turned to guanylate kinase (GMK), another (p)ppGpp-regulated enzyme (Kriel et al., 2012; Liu et al., 2015b). We previously found that GMK is inhibited by (p)ppGpp in multiple phyla of bacteria but not in Proteobacteria despite the fact (p)ppGpp binds the conserved active site (Liu et al., 2015b). This puzzle may now be explained by a protein–protein interaction affecting (p)ppGpp binding: GMK is a dimer. In the (p)ppGpp-sensitive GMK, a lid domain opens the active site for (p)ppGpp binding through pulling of the lid by the C-terminal helix of the adjoining monomer in a GMK dimer (Figure 8A). In the (p)ppGpp-insensitive E. coli GMK, the lid domain is closed, with the C-terminal helix from the adjoining monomer perhaps responsible for stabilizing the closed position (Figure 8A). We had previously swapped the lid domains to test their role in (p)ppGpp inhibition, but this had led to inactive enzymes (Liu et al., 2015b). To identify whether the C-terminal tail is holding the lid domain open or closed, we deleted 10 residues from the C-terminus in both the (p)ppGpp-sensitive B. subtilis GMK and the (p)ppGpp-resistant E. coli GMK with striking observations (Figure 8B C). Remarkably, while these residues are away from the (p)ppGpp binding pocket of the adjoining monomer, their deletion led to weakened binding between B. subtilis GMK and pppGpp and reduced inhibitory effect on its enzymatic activity. On the other hand, deleting the C-terminus of the (p)ppGpp-resistant E. coli GMK allowed a weak binding between E. coli GMK and pppGpp, and rendered E. coli GMK sensitive to pppGpp inhibition. This demonstrates that the allosteric interaction between the lid domain and C-terminus of two adjoining GMK monomers is the determining factor for (p)ppGpp specificity.

Protein–protein interface allosterically alters the binding of (p)ppGpp to guanylate kinase.

(A) In S. aureus GMK (left; PDB ID 4QRH), an open lid domain (red) allows (p)ppGpp binding at the active site, but in E. coli GMK (right; PDB ID 2ANB), the closed lid domain occludes (p)ppGpp binding at the active site (bound to GMP). A C-terminal tail in the second monomer of a GMK dimer interacts with the lid domain in both proteins. Nineteen residues in the lid domain and six residues in the C-terminal tail of S. aureus GMK are unresolved. (B) DRaCALA of full-length and ∆C-terminal tail (∆CT tail) B. subtilis and E. coli GMKs. Error bars are mean of duplicate. (C) Inhibition of GMKs from (D) relative to uninhibited activity. Error bars are SEM of triplicate. (D) Maximum likelihood phylogenetic tree of 41 GMKs rooted on eukaryotic GMKs as the outgroup. Bootstrap values from 1000 replicates are shown. Anc1-3 refer to ancestral HPRTs. See Materials and methods for tree construction. See Source Data one for the full alignment of GMKs. (E) Alignment of the lid domain and C-terminal tail of ancestral and select extant GMKs. Boxed residues are part of the lid domain - C-terminal tail interaction.

https://doi.org/10.7554/eLife.47534.037

We constructed a phylogenetic tree of 41 GMKs (Figure 8D). As with HPRT, the tree showed that the (p)ppGpp-insensitive and (p)ppGpp-sensitive GMKs bifurcated from the last common ancestor. Importantly, these (p)ppGpp-insensitive and (p)ppGpp-sensitive GMKs use different residues to form the interaction between the lid domain and the C-terminal helix (Figure 8E). (p)ppGpp-insensitive GMKs have a hydrophobic interaction between the lid domain and C-terminal tail, whereas (p)ppGpp-sensitive GMKs contain more charged residues at this interface. Ancestral protein reconstruction suggests that the last common ancestor (Anc1) is more similar to the (p)ppGpp-sensitive GMKs (Anc3) (Figure 8E). Altogether, this suggests that the protein-protein interface of GMK has also evolved to affect (p)ppGpp binding.

Discussion

(p)ppGpp is a stress-induced signaling molecule in bacteria that is also critical for cellular fitness and homeostasis even at basal levels. However, (p)ppGpp targets and their evolution in different bacteria remain poorly understood. Here, we have described a mechanism explaining how basal levels of (p)ppGpp potently regulate the activity of the housekeeping enzyme HPRT through a novel binding site that may represent a new class of (p)ppGpp effectors. Intriguingly, this site overlaps completely with the active site and is conserved among bacteria, yet differential regulation by (p)ppGpp can be achieved through variation of an allosteric component: the interaction between dimeric subunits of the HPRT tetramer. This interaction tethers a flexible loop at the interface and away from the active site, allowing the binding pocket to accomodate (p)ppGpp. Lack of the dimer–dimer interaction causes HPRT to occlude (p)ppGpp and favor substrate binding. This dimer–dimer interaction is due to an interface motif that appears to have co-evolved with (p)ppGpp binding in the majority of bacterial HPRTs sensitive to (p)ppGpp, whereas dimeric HPRTs without the motif are resistant to (p)ppGpp. We conclude that evolution of the dimer-dimer interface in tetrameric HPRTs has potentiated (p)ppGpp binding to enable basal (p)ppGpp modulation of metabolism, thus increasing bacterial fitness in fluctuating environments.

A novel, high-affinity (p)ppGpp binding motif that shares the PRPP motif

Our HPRT-ppGpp structure revealed a binding motif distinct from known (p)ppGpp-protein interactions. There are over 50 known (p)ppGpp targets across bacteria, but identifying motifs associated with (p)ppGpp binding has been difficult (Corrigan et al., 2016; Wang et al., 2019; Zhang et al., 2018). In a few cases, (p)ppGpp binds allosteric sites at protein interfaces (Kanjee et al., 2011; Ross et al., 2013; Ross et al., 2016; Steinchen et al., 2015; Wang et al., 2019). For many other targets, (p)ppGpp binds at a GTP binding site, leading to overlapping (p)ppGpp and GTP binding motifs (Fan et al., 2015; Kihira et al., 2012; Liu et al., 2015b; Pausch et al., 2018; Rymer et al., 2012). In the case of HPRT, however, the (p)ppGpp binding site is not at an interface between two proteins, nor is it a GTP binding site. Instead, it shares a well-characterized motif associated with PRPP binding (EDIIDSGLT in B. anthracis Hpt-1) (Sinha and Smith, 2001). Other PRPP-binding proteins, including UPRT and APRT, have been shown to bind (p)ppGpp (Wang et al., 2019; Zhang et al., 2018), and it is likely that (p)ppGpp also binds to their PRPP motif. Identification of this motif may provide a new class of (p)ppGpp-binding proteins, allowing us to predict additional targets.

(p)ppGpp interacts with proteins in two main conformations: elongated, with the phosphate arms extended away from one another in a T shape, and ring-like, with the phosphate arms near one another in a Y shape. The compact, ring-like conformation has been associated with higher affinity interactions than the elongated conformation (Steinchen and Bange, 2016). However, in the HPRT–ppGpp interaction, ppGpp takes an elongated conformation, but exhibits strikingly tight affinities as high as Kd ≈ 0.1 μM for some species (Figure 2D and Table 1). This high affinity may be due to extensive backbone amide interactions with both phosphate arms as well as (p)ppGpp’s close mimicry of substrate binding (Figure 3D). It is likely that higher affinity interactions with the elongated (p)ppGpp conformation will become more common as additional targets are characterized.

HPRT tetramerization enables basal (p)ppGpp inhibition

Our data show that the dimer–dimer interface of HPRT potentiates (p)ppGpp regulation by allosterically promoting a conformation conducive to ligand binding (Figure 7). HPRT’s tetramerization harnesses the flexible loop II to open the binding pocket (Figure 4A) for (p)ppGpp to bind with high affinity. Loop II’s influence on (p)ppGpp binding may be the reason that it has evolved to be part of the interface of bacterial HPRTs, whereas in eukaryotic (e.g. human) HPRTs, which likely do not interact with (p)ppGpp in nature, loop II is on the outside of the oligomer facing solvent (Eng et al., 2015).

Our model can also explain the evolutionary significance of bacterial HPRTs functioning as dimers without dissociating to monomers. Strength of protein–protein interactions has been correlated with increased buried surface area at the interface (Nooren and Thornton, 2003). In a hypothetical monomer–monomer interaction, the smaller loop II interface may be too weak and too transient to keep loop II away from the active site. In the interaction between two homodimers, the loop II interface is duplicated, which increases the surface area and provides anchor points to hold loop II at the interface and away from the (p)ppGpp binding pocket (Figure 7).

HPRT tetramerization also affects a key component of basal (p)ppGpp’s ability to regulate HPRT: its competition with PRPP. Most bacterial HPRTs have a Km for PRPP in the range of 50–200 μM (Guddat et al., 2002; Zhang et al., 2016) (Table 4). Tetrameric HPRTs that bind (p)ppGpp with high affinity also tend to have higher Km values for PRPP, suggesting that they are optimized for (p)ppGpp binding rather than PRPP binding (Table 4). By forcing HPRT into a dimeric state independent of ligands, either through natural evolution in the case of the HPRT homologs or through mutating the dimer–dimer interface sequence, HPRT becomes both refractory to (p)ppGpp binding (Figure 5) and also preferably binds PRPP (Figures 5F and 6E, Table 4). The overall effect is an HPRT that is resistant to inhibition by basal levels of (p)ppGpp.

HPRT regulation allows maintenance of cellular homeostasis by basal levels of (p)ppGpp

Our characterization of HPRT explains how bacteria can be regulated by basal levels of (p)ppGpp. While (p)ppGpp is mostly chacterized as a regulator of gene expression that functions in starvation-induced concentrations, basal concentrations of (p)ppGpp are known to be responsible for sustaining antibiotic tolerance and virulence in E. faecalis (Gaca et al., 2013), maintaining cyanobacterial light/dark cycles (Puszynska and O'Shea, 2017), influencing rRNA expression and growth rate in E. coli (Potrykus et al., 2011), and regulating GTP synthesis in Firmicutes (Gaca et al., 2013; Kriel et al., 2012). However, the principles that determine how basal and induced (p)ppGpp regulate different cellular targets remained unclear. In the single species B. subtilis, for example, (p)ppGpp interacts with DNA primase (Ki, ppGpp = 250 μM), IMP dehydrogenase (Ki, ppGpp = 50 μM), and guanylate kinase (Ki, pppGpp = 14 μM), allowing them to be inhibited in vivo at induced (p)ppGpp levels (Liu et al., 2015b; Pao and Dyes, 1981; Wang et al., 2007). In contrast, (p)ppGpp’s interaction with B. subtilis HPRT is far stronger (Ki = 1.7 μM; Figure 1D), potentially allowing (p)ppGpp to modulate its activity at basal levels throughout growth. Our data suggest that the dimer-dimer interaction is a mechanism that strengthens (p)ppGpp binding affinity, but also favors the inhibitor over the substrate PRPP. Physiological PRPP levels in bacteria are roughly 0.1–1 mM (Bennett et al., 2009; Berlin and Stadtman, 1966; Jensen, 1983). At 1 mM PRPP, the high end of physiological levels, (p)ppGpp at a basal level (25 μM) can still strongly inhibit tetrameric HPRT activity but not the constitutive dimeric homologs (Figure 2E). Thus tetramerization allows basal (p)ppGpp to regulate HPRT and thus constantly regulate purine nucleotide synthesis via the salvage pathway.

The basal regulation of HPRT may be important for fitness in environmental niche. For example, for bacteria in the microbiota whose HPRTs were potently inhibited by (p)ppGpp, it may robustly maintain intracellular metabolism despite fluctuations in exogenous purines that could depend on the purine content of the diet (Choi et al., 2004; Kaneko et al., 2014; Zgaga et al., 2012).

Protein oligomerization allosterically alters ligand specificity

The mechanism we characterized for HPRT and GMK may represent a more broadly applicable principle by which evolution of protein oligomerization changes the conformation of a ligand binding pocket to alter ligand specificity. Oligomerization of proteins into homomers can provide multiple adaptive advantages, including mechanisms of allosteric regulation by a ligand binding to a site other than the enzyme's active, enabling cooperative substrate binding, promoting protein stability, providing complete active sites or ligand binding sites at oligomeric interfaces, maintaining proximity of signal transduction, and serving cytoskeletal or other structural roles (Ali and Imperiali, 2005; Bergendahl and Marsh, 2017; Matthews and Sunde, 2012Perutz, 1989; Perica et al., 2012; Traut, 2008; Traut, 1994). For HPRT, instead, the advantage of tetramerization is strengthening (p)ppGpp specificity at the active site for enhanced competitive regulation by basal levels of this nucleotide.

Similarly, we show oligomeric interaction also plays a major role in altering specificity for the (p)ppGpp target guanylate kinase (GMK). GMK is inhibited by (p)ppGpp in multiple phyla of bacteria but not in Proteobacteria despite the fact (p)ppGpp binds the conserved active site (Liu et al., 2015b). We now found that this is because GMK is a dimer and the monomer–monomer interaction in GMK allosterically affects the lid domain conformation and (p)ppGpp binding by occluding (p)ppGpp binding in E. coli and promoting (p)ppGpp binding in B. subtilis. By deleting the C-terminal helix from the adjoining monomer that is responsible for stabilizing the lid domain position, we can, for the first time, remove the differential specificity to (p)ppGpp between E. coli and B. subtilis GMK homologs while keeping the enzymes active (Figure 8).

It is striking to note that a majority of ligands bind within 6 Å of a protein–protein interface (Gao and Skolnick, 2012). Thus, the allosteric effect of oligomeric or other protein-protein interactions on ligand binding likely extends to many other regulations by signaling ligands.

Coevolution of ligand binding and protein oligomerization

What is the evolutionary advantage of regulating ligand-binding through protein-protein interactions? One potential advantage is that it provides evolutionary flexibility for ligand binding. Evolving different residues within ligand binding sites can alter ligand specificity, but active sites that bind substrates and inhibitors are under functional and evolutionary constraints to maintain enzymatic activity (Echave et al., 2016; Huang et al., 2015). On the other hand, protein–protein interfaces like the dimer–dimer interface of HPRT and the monomer–monomer interface of GMK are more evolutionarily flexible, particularly when they are not obligate interactions, as they allow mutations to accumulate away from a conserved active site (Echave et al., 2016; Mintseris and Weng, 2005). This suggests that changing oligomeric states could be an evolutionarily flexible mechanism for altering ligand specificity (Figure 7). Indeed, the coevolution of (p)ppGpp regulation and the HPRT and GMK protein–protein interactions has provided us with examples of how protein oligomerization and ligand binding can coevolve, demonstrating that organisms have already adopted this strategy.

In addition to (p)ppGpp, many small molecules serve as signaling ligands that regulate protein activities (Gao and Skolnick, 2012; Najmanovich, 2017). Our results suggest that ligand binding, even at non-interface binding pockets, influence evolutionary diversification of protein oligomers potentially through purifying selection of conformations that favor protein–ligand interactions. While homomeric evolution of some proteins has been implicated as physicochemical or stochastic processes (Abrusán and Marsh, 2018; André et al., 2008; Lukatsky et al., 2007; Lynch, 2013), our data provide evidence for ligand binding as an adaptive advantage driving the evolutionary diversification of protein homomers. Given the proximity of ligand binding sites to protein interfaces (Gao and Skolnick, 2012), and since it is easier to evolve protein–protein interactions (Perica et al., 2012) rather than evolving new sites for allosteric regulation, such an adaptive benefit is likely to exist more broadly beyond (p)ppGpp in other protein–ligand interactions.

Materials and methods

Key resources table
Reagent type
(species) or
resource
DesignationSource or referenceIdentifiersAdditional information
Strain, strain background (Bacillus subtilis)gmk::ecgmk(Liu et al., 2015b)N/A
Strain, strain background (B. subtilis)gmk::ecgmk amyE::Phyperspac-ecgptThis workN/A
Strain, strain background (B. subtilis)gmk::ecgmk amyE::Phyperspac-bshprTThis workN/A
Recombinant DNA reagentpLIC-trPC-HA (plasmid)(Stols et al., 2002)N/A
Recombinant DNA reagentpHM1381 (plasmid)(Mechold et al., 2002)N/A
Recombinant DNA reagentpDR90 (plasmid)(González-Pastor et al., 2003)N/A
Peptide, recombinant proteinhuman HPRT1NovoproteinNovoprotein #C294
Chemical compound, drug5-phospho-D
-ribose 1-diphosphate pentasodium salt (PRPP)
MilliporeSigmaMilliporeSigma # P8296
Chemical compound, drugdimethyl adipimidate (DMA)ThermoFisherThermoFisher # 5001417627
Chemical compound, drug9-deazaguanineSanta Cruz BiotechnologySanta Cruz #sc-217528
Chemical compound, drugSYPRO RubyBio-RadBio-Rad #1703125
Chemical compound, drugSYPRO OrangeMilliporeSigmaMilliporeSigma #S5692
Chemical compound, drugguanineMilliporeSigmaMilliporeSigma #G6779
Software, algorithmGraphPad Prism v. 5.02 (software)https://www.graphpad.com/RRID:SCR_002798
Software, algorithmImageQuant (software)https://www.gelifesciences.comRRID:SCR_014246
Software, algorithmMicroCal Origin 5.0 (software)https://www.malvernpanalytical.comRRID:SCR_002815
Software, algorithmMEGA X (software)https://www.megasoftware.net/RRID:SCR_000667
Software, algorithmPhenix (software)https://www.phenix-online.org/RRID:SCR_014224
Software, algorithmCoot (software)https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/RRID:SCR_014222
Software, algorithmDynamics v. 5.25.44 (software)https://www.wyatt.com/products/software/dynamics.htmlN/A

Strain and plasmid construction and mutagenesis

Request a detailed protocol

For purification and DRaCALA analyses, hprT coding sequences were cloned into the pLIC-trPC-HA vector (pJW269) using the ligation independent cloning (LIC) protocol (Eschenfeldt et al., 2009; Stols et al., 2002) and was scaled up to include more hprT sequences when necessary as described (Abdullah et al., 2009). The hprT sequences were identified in each organism by enzyme classification number (EC 2.4.2.8), by BLAST using B. subtilis or E. coli hprTs, or by database annotation (UniProt, EnsemblBacteria, KEGG Genome). Bacterial hprT homologs were amplified from genomic DNA, cloned using LIC, and transformed into E. coli BL21(DE3) (see Supplementary file 1 for primers and plasmids). Amino acid substitutions were performed either with QuikChange XL Site-Directed Mutagenesis (Agilent Technologies) or with a megaprimer site-directed mutagenesis protocol (Kirsch and Joly, 1998). Loop II deletions were made using a protocol as described (Hansson et al., 2008). pLIC-trPC-HA inserts were amplified and sequenced for confirmation using oJW1124 and oJW492.

To create strains overexpressing protein in B. subtilis gmk::ecgmk, B. subtilis hprT and E. coli gpt were amplified with oJW1569/1570 and oJW1550/1551, respectively, prior to restriction cloning into pDR90 using HindIII and SphI (New England Biolabs). Plasmids were linearized with XhoI and transformed into JDW2108 (B. subtilis gmk::ecgmk) by growing JDW2108 in 1x modified competence medium (Spizizen, 1958) until turbid (≈6 hr) prior to adding linearized plasmid and growing an additional 2 hr. Transformants were selected on 80 μg/ml spectinomycin (MilliporeSigma) and recombination into amyE was confirmed by patching transformants on LB-starch plates and checking for lack of starch utilization.

Protein purification

Request a detailed protocol

HPRTs were recombinantly expressed in E. coli BL21(DE3) (NEB) from a pLIC-trPC-HA plasmid with the gene inserted downstream of a sequence encoding a 6X histidine tag and a tobacco etch virus (TEV) protease recognition site. Seed cultures grown to mid-log phase were diluted 1:50 into batch culture of LB supplemented with 100 μg/mL carbenicillin (Research Products International). Protein synthesis was induced at OD600 ≈ 0.8 with 1 mM isopropyl β-D-1 thiogalactopyranoside (IPTG) (RPI) for 4 hr. Cells were pelleted and stored at −80° C until purification. GMKs and E. coli XGPRT were expressed in an identical manner.

For large-scale purifications, cells were resuspended in Lysis Buffer (50 mM Tris-HCl pH 7.5, 500 mM NaCl, 10 mM imidazole) and lysed with a French press. The lysate was centrifuged to obtain the soluble fraction, which was filtered through 0.45 μm filters. The sample was put over a HisTrap FF column (GE Healthcare) on an AktaPure FPLC (GE Healthcare), the column was washed with 20 column volumes of Lysis Buffer, and the protein was eluted with a gradient of increasing Elution Buffer (50 mM Tris-HCl pH 7.5, 500 mM NaCl, 250 mM imidazole). Recombinant protein fractions were dialyzed with 10 mM Tris-HCl (pH 7.5), 100 mM NaCl, 1 mM DTT, and 10% glycerol prior to concentrating and flash-freezing for storage. For small-scale purifications of HPRTs, GMKs, and XGPRT, Ni-NTA spin columns were used according to manufacturer’s instructions (Qiagen). Lysis buffer was 100 mM sodium phosphate (pH 8.0), 500 mM NaCl, and 10 mM imidazole. Wash and elution buffers were the same as the lysis buffer except with 20 mM and 500 mM imidazole, respectively. Protein purity was determined using SDS-PAGE, and concentrations were measured using the Bradford assay (Bio-Rad) or using A280 with extinction coefficients calculated by ProtParam (SIB ExPASy Bioinformatics Resource Portal).

To purify 35 HPRT homologs for activity assays (see Figure 2), a 96-well Capturem His-tagged purification kit (Clontech) was used according to the manufacturer’s instructions. The following buffers were used: lysis [xTractor (Clontech) + 1 μg/mL DNase I], wash [20 mM Na3PO4 pH 7.6, 200 mM NaCl, 10 mM imidazole], elution [20 mM Na3PO4 pH 7.6, 500 mM NaCl, 500 mM imidazole]. Following purification, the buffer was exchanged to 10 mM HEPES pH 8, 150 mM NaCl, 10 mM MgCl2, 1 mM DTT using Zeba spin desalting plates (Thermo Scientific) according to the manufacturer’s instructions. The Bradford assay was used to measure protein concentration, and the proteins were aliquoted at 8 μM and flash-frozen with liquid nitrogen.

For crystallography with ppGpp, recombinant B. anthracis Hpt-1 was purified as described above followed by dialysis with His-tagged TEV protease in 10 mM Tris-HCl (pH 7.5), 100 mM NaCl, and 1 mM DTT. The dialyzed protein was incubated with Ni-NTA beads for 30 min, the beads were centrifuged and the supernatant was run over a HiPrep Sephacryl 16/60 S-100 HR column (GE Healthcare) in 10 mM Tris-HCl pH 8.3, 250 mM NaCl, and 1 mM DTT. Relevant fractions were concentrated to ≈10 mg/mL. For crystallography with substrates, the same protocol was followed but with 10 mM Tris-HCl pH 8 and 100 mM NaCl as the size exclusion buffer.

Enzyme inhibition assays

Request a detailed protocol

HPRT activity assays were performed as described previously (Biazus et al., 2009; Kriel et al., 2012; Xu et al., 1997). The standard assay was performed at 25°C and contained 100 mM Tris-HCl (pH 7.4), 12 mM MgCl2, 1 mM PRPP (MilliporeSigma), 50 μM guanine or hypoxanthine (MilliporeSigma), and 20 nM HPRT (measured by monomer). Reactions were initiated with the purine base and monitored in a spectrophotometer (Shimadzu UV-2401PC) at 257.5 nm or 245 nm for conversion of guanine to GMP or hypoxanthine to IMP, respectively. A difference in extinction coefficients of 5900 M−1cm−1 was used for GMP and guanine and 1900 M−1cm−1 for IMP and hypoxanthine. For inhibition curves, assays were performed at the substrate concentrations listed above and at variable pppGpp concentrations. (p)ppGpp was synthesized as described (Liu et al., 2015b). Initial velocities of the inhibited reactions were normalized to the uninhibited initial velocity prior to fitting to the equation Y = 1/(1 + (x / IC50)s) to calculate IC50. GMK activity assays were performed as previously described (Liu et al., 2015b), except an enzyme concentration of 20 nM and a substrate concentration of 100 μM GMP were used. Due to lower activity of the E. coli ΔCT-tail variant, assays were performed at 200 nM enzyme rather than 20 nM enzyme for E. coli GMK.

To test inhibition of HPRT homologs (see Figure 1), reactions were performed in a Synergy Two microplate reader (BioTek). The assay was performed at 25°C with 50 μM hypoxanthine, 1 mM PRPP, and 100 nM HPRT (measured by monomer) in the same reaction buffer as above. For Coxiella burnetii HPRT, 50 μM guanine was used as the substrate since its activity was very low with hypoxanthine. Reactions were performed in triplicate without (p)ppGpp, with 25 μM ppGpp, and with 25 μM pppGpp. Reaction rates from the first-order kinetic curves were determined using R (v 3.4.3).

For kinetic studies, pppGpp concentrations were varied between 3.125 and 50 μM and PRPP concentrations were varied between 0.03125 and 2.5 mM. Data were analyzed and fitted to a global Michaelis-Menten competitive inhibition model using GraphPad Prism v5.02. PRPP provided by MilliporeSigma was listed as ≥75% pure. For the purposes of kinetic experiments, the purity was assumed to be 100%.

Isothermal titration calorimetry

Request a detailed protocol

Experiments were performed using the MicroCal iTC200 (GE Healthcare). B. anthracis Hpt-1 was dialyzed into the ITC buffer (10 mM HEPES pH 8, 150 mM NaCl, 10 mM MgCl2) with three buffer changes at 4°C. The concentration of protein was calculated using a molar extinction coefficient of 16390 M−1cm−1 and A280(true) (A280 – (1.96 × A330) (Pace et al., 1995). The experiments were performed at 25°C with 45.5 μM HPRT (measured by monomer), a reference power of 6 μCal/s, and a stirring speed of 1000 RPM. pppGpp was solubilized in dialysate from protein dialysis and its concentration was measured using a molar extinction coefficient of 13,700 M−1cm−1 and A253. pppGpp (500 μM) was titrated into B. anthracis Hpt-1 with the following: 1 × 1 μL (discarded), 19 × 2 μL. ITC between PRPP and Hpt-1 were performed in identical conditions. PRPP was solubilized in protein dialysate at a concentration of 1 mM. For data analysis, PRPP was assumed to be at a concentration of 820 μM, since the PRPP lot analysis from MilliporeSigma listed the PRPP as 82% pure. Data analysis and one-site binding modeling were performed using MicroCal Origin 5.0 software.

X-ray crystallography

Request a detailed protocol

Proteins were prepared for crystallography as described above. For crystals formed with pppGpp, B. anthracis Hpt-1 was concentrated to ≈10 mg/mL in 10 mM Tris pH 8.3, 250 mM NaCl, and 1 mM DTT. The pppGpp ligand was resuspended in ddH2O. MgCl2 was added to the protein at a final concentration of 1 mM and crystals were formed using hanging drop vapor diffusion with 900 μL of reservoir liquid in each well. Each drop contained 0.9 μL protein, 0.9 μL reservoir liquid, and 0.2 μL pppGpp (final concentration, 1.5 mM pppGpp). Crystals formed in 0.2 M ammonium tartrate dibasic pH 6.6% and 20% PEG 3350 in 3–6 months. Identical crystals formed in replicated conditions in 1–2 weeks. Crystals were soaked in reservoir liquid with 25% glycerol prior flash freezing in liquid nitrogen. Apo protein crystals formed within 1–2 days in multiple conditions with high sulfate concentrations. Using protein preparations from above, apo crystals formed in 0.01 M CoCl2, 0.1 M MES monohydrate pH 6.5, 1.8 M ammonium sulfate, and crystals were soaked in reservoir liquid with 25% glycerol for cryoprotection prior to freezing.

For crystals with substrates, B. anthracis Hpt-1 was concentrated to ≈10 mg/mL in 10 mM Tris pH 8 and 100 mM NaCl. Additives were diluted in the protein solution at final concentrations of 10 mM MgCl2, 2 mM PRPP, and 1 mM 9-deazaguanine (Santa Cruz Biotechnology) prior to crystallization. PRPP was resuspended in ddH2O and 9-deazaguanine was resuspended in 100% DMSO. Drops for hanging drop vapor diffusion comprised 1 μL crystal condition and 1 μL protein/ligand mixture. Crystals formed within 3 days in 0.2 M ammonium acetate, 0.1 M sodium acetate trihydrate pH 4.6, 30% PEG 4000. Reservoir solution with 25% ethylene glycol was added to the drops for cryoprotection prior to flash freezing in liquid nitrogen.

Diffraction data was collected at the Life-Science Collaborative Access Team (LS-CAT), beamline 21-ID-F (Hpt-1 with sulfates), 21-ID-G (Hpt-1 with ppGpp), and 21-ID-D (Hpt-1 with substrates) at the Advanced Photon Source (APS) at Argonne National Labs (Argonne, IL). Data was indexed and scaled using HKL2000 (Otwinowski and Minor, 1997). Phasing for Hpt-1 with sulfates and ppGpp was determined by molecular replacement with PDB ID 3H83 as a search model using Phenix (Adams et al., 2010). Phasing for Hpt-1 with substrates was determined by molecular replacement with Hpt-1-ppGpp as a search model using Phenix. Iterative model building with Coot and refinement with Phenix produced final models (Emsley and Cowtan, 2004).

Predicted oligomeric states of crystallographic assemblies were determined by PISA (‘Protein interfaces, surfaces, and assemblies’ service PISA at the European Bioinformatics Institute) (Krissinel and Henrick, 2007).

Size exclusion chromatography

Request a detailed protocol

Size exclusion chromatography was performed using AktaPure and a Superose 12 10/300 GL column (GE Healthcare, Inc). A buffer consisting of 10 mM HEPES pH 8.0, 100 mM NaCl, and 10 mM MgCl2 was used to run ≈10–20 μM (0.2–0.5 mg/mL) protein over the column at 0.1–0.25 mL/min. Additional 1 mM DTT was used for proteins containing cysteines. For gel filtration with ligands, 500 μM PRPP and 100 μM 9-deazaguanine were included in the mobile phase buffer where necessary. A gel filtration standard (Bio-Rad) was used to establish molecular weight, and bovine serum albumin (BSA) was included as an additional marker.

Dimethyl adipimidate crosslinking

Request a detailed protocol

Crosslinking was performed with ≈10 μM B. subtilis HPRT (measured by monomer), 20 mM dimethyl adipimidate (DMA) (Thermo Scientific), and 500 μM ligand. DMA was suspended in 25 mM HEPES pH 8.0, 100 mM NaCl, 10 mM MgCl2, and 10% glycerol, and the solution was buffered to pH 8.5. HPRT was dialyzed into 25 mM HEPES pH 7.5, 100 mM NaCl, and 10% glycerol. Ligands were incubated with protein for 10 min followed by a 15-min incubation with DMA at room temperature. Reactions were terminated with addition of 2X Laemmli buffer (Bio-Rad) for immediate analysis with SDS-PAGE (10% polyacrylamide gel). Gels were stained with SYPRO Ruby (Bio-Rad) according to the manufacturer’s protocol and imaged using a Typhoon FLA9000 (GE Healthcare).

Dynamic light scattering

Request a detailed protocol

Dynamic light scattering was performed using DynaPro99 (Protein Solutions/Wyatt Technologies). Readings were from a 20 μL solution (10 mM Tris-HCl pH 8, 100 mM NaCl, 10 mM MgCl2, 6.5% glycerol) with 4 mg/mL (≈175 μM) B. anthracis Hpt-1 and 2 mM ligands. Data were analyzed using Dynamics version 5.25.44 software.

DRaCALA

Request a detailed protocol

DRaCALA was performed with pure protein and radioactive ligand as described (Roelofs et al., 2011). [5′ α-32P] pppGpp was synthesized according to modified protocols of non-radioactive and radioactive pppGpp syntheses (Corrigan et al., 2016; Hogg et al., 2004; Mechold et al., 2002). The reaction contained 25 mM bis-Tris propane (pH 9.0), 15 mM MgCl2, 0.5 mM DTT, 2 mM ATP, 2 μM RelSeq (1-385), and 37.5 μCi [α-32P] GTP (Perkin Elmer). The reaction was incubated at 37°C for 1 hr. The reaction was diluted in 0.5 mL of Buffer A (0.1 mM LiCl, 0.5 mM EDTA, 25 mM Tris-HCl pH 7.5) prior to adding to a 1 mL HiTrap QFF strong anion exchange column (GE Healthcare) equilibrated with 10 column volumes (CV) of Buffer A. The column was washed with 10 CV of Buffer A followed by an additional wash with 10 CV of 83% Buffer A + 17% Buffer B (Buffer B: 1 M LiCl, 0.5 mM EDTA, 25 mM Tris-HCl pH 7.5). 32P-pppGpp was eluted with a mixture of 50% Buffer A + 50% Buffer B. Fractions of 1 mL were collected from the elution.

DRaCALA reactions (20 μL) using purified HPRT contained 100 mM Tris-HCl (pH 7.4), 12 mM MgCl2, protein diluted to appropriate concentrations in 20 mM HEPES pH 8 and 100 mM NaCl, and 32P-pppGpp (1:100 final dilution of first elution fraction from 32P-pppGpp purification). DRaCALA reactions using GMK contained 100 mM Tris-HCl (pH 7.4), 100 mM KCl, and 10 mM MgCl2. Reactions with B. subtilis GMK and ΔCT-tail were performed with 20 μM protein and reactions with E. coli GMK and ΔCT-tail were performed with 40 μM protein. All protein concentrations were measured by monomer. For competition experiments between 32P-labeled pppGpp and PRPP, non-radioactive PRPP was resuspended in ddH2O and added at the concentrations specified. Protein was dialyzed or diluted into buffer lacking glycerol, as glycerol interferes with diffusion of the aqueous phase. Reactions were incubated at room temperature for 10 min. Two microliters from each reaction were spotted in duplicate on Protran BA85 nitrocellulose (GE Healthcare) via pipette or a replicator pinning tool (VP 408FP6S2; V and P Scientific, Inc). Spots were allowed to dry and radioactivity was detected with phosphorimaging (Typhoon FLA9000). Fraction bound of 32P-pppGpp was calculated as described (Roelofs et al., 2011). Data were analyzed in GraphPad Prism v5.02 and fitted to the equation Y = (Bmax ×Kd) / (Kd + X) (Roelofs et al., 2011). PRPP competition curves were fitted with a four-parameter logistical (4PL) model.

DRaCALA using cell lysates was adapted from a previous protocol (Roelofs et al., 2015). One milliliter of cells containing overexpressed recombinant HPRT was pelleted and resuspended in 100 μL of binding buffer (20 mM Tris-HCl pH 8, 100 mM NaCl, 12 mM MgCl2, 0.1 mM DTT) supplemented with 1 μM PMSF, 250 μg/mL lysozyme, and 10 μg/mL DNase I. Cells were lysed with three freeze/thaw cycles. In a 20 μL DRaCALA reaction, 10 μL of cell lysate was added to binding buffer and 32P-pppGpp. Reactions were performed and analyzed as above. For measuring binding affinity (Kd) of proteins in cell lysates, recombinant protein expression level in cell lysates was determined by comparing expression to a standard of purified B. subtilis HPRT co-resolved with SDS-PAGE.

Differential scanning fluorimetry

Request a detailed protocol

Differential scanning fluorimetry was performed using 10 μM protein (measured by monomer) in a buffer containing 20 mM HEPES pH 8, 100 mM NaCl, 10 mM MgCl2, 1 mM DTT, and 10X SYPRO orange dye (diluted from 5000X stock; MilliporeSigma). Proteins were mixed in an optically clear quantitative PCR (qPCR) 96-well plate and sealed with plastic film. Relative fluorescence intensity was monitored in a Bio-Rad qPCR machine using FRET detection over a temperature increase of 1°C/min from 25°C to 90°C. Data were analyzed using GraphPad Prism.

Thin layer chromatography

Request a detailed protocol

Thin layer chromatography was performed as described previously (Bittner et al., 2014; Schneider et al., 2003). Cells were grown in a low-phosphate medium with casamino acids (S7 defined medium [Vasantha and Freese, 1980] supplemented with 0.1% glutamate, 1% glucose, and 0.5% casamino acids; low-phosphate medium contained one-tenth the phosphate concentration as S7 defined medium). JDW2121 and JDW2128 were grown with 1 mM IPTG to induce expression of exogenous phosphoribosyltransferases. Cells were labeled with 50 μCi/ml 32P orthophosphate (900 mCi/mmol; PerkinElmer) at an OD ≈ 0.02–0.05 and grown to an OD ≈ 0.15–0.2 prior to treating with 1 mM guanosine (MilliporeSigma). Guanosine stock was 100 mM diluted in 100% DMSO. At given timepoints following guanosine addition, nucleotides were extracted by adding 100 μl of sample to 20 μl 2 N formic acid and incubating on ice for at least 20 min. Samples were centrifuged at max speed for 15 min and supernatant (≈60 μl) was transferred to a new tube. Samples were spotted on PEI cellulose plates (MilliporeSigma) and developed in 1.5 M KH2PO4 pH 3.4. Plates were exposed to a phosphor screen and scanned on a Typhoon scanner. Nucleotide levels were quantified by subtracting background signal in each lane and expressing the level as a ratio to the untreated sample (t = 0).

Protein sequence analysis and ancestral sequence reconstruction

Request a detailed protocol

To construct a 16S rRNA tree, sequences were obtained from the Ribosomal Database Project (Cole et al., 2014) or from NCBI. Sequences were aligned with ClustalW in MEGA X. MEGA X was used to identify the best substitution model for evolution as the General Time Reversible model with gamma distributed substitution rates. The maximum likelihood phylogenetic tree was constructed with this model in MEGA X with 1000 bootstrap replicates. Eukaryotic 18S rRNA sequences for H. sapiens, C. elegans, and S. cerevisiae were used as the outgroup.

For HPRT phylogenetic analysis, protein sequences were obtained in UniProt either by searching for HPRT’s enzyme classification number (EC 2.4.2.8) in a given organism or by using NCBI’s and UniProt’s BLAST algorithms to find HPRTs similar (>50% identity) to a model organism representing different clades (e.g. B. subtilis, E. coli, S. aureus, S. coelicolor, B. thetaiotaomicron, L. pneumophila, P. aeruginosa, T. gondii, C. elegans, H. sapiens). Protein sequences were aligned using MUSCLE in MEGA X with a gap-opening penalty of −3.2 and a gap-extending penalty of −0.2. See Figure 6—source data 1 for the complete alignment. The most appropriate evolutionary model for amino acid substitution was identified using ProtTest 3 (Darriba et al., 2011) as the Le and Gascuel model (Le and Gascuel, 2008) with gamma distributed substitution rates (four categories) and invariant sites. The maximum likelihood phylogenetic tree was constructed using this substitution model in MEGA X with 100 bootstrap replicates and a nearest neighbor interchange topology search. The clade containing eukaryotic HPRTs was used as the outgroup. Ancestral HPRT sequences were reconstructed using a marginal probability reconstruction in MEGA X with the phylogenetic tree and the same substitution model used to obtain the tree (Yang et al., 1995).

For GMK phylogenetic analysis, protein sequences were obtained from UniProt by searching for EC 2.7.4.8 in the organisms studied in Liu et al. (2015b). Related organisms were identified with UniProt’s BLAST algorithm. Protein sequences were aligned using MUSCLE in MEGA X with default settings. See Figure 8—source data 1 for the complete alignment. ProtTest3 identified the most appropriate substitution model as the Le and Gascuel model (Le and Gascuel, 2008) with gamma distributed substitution rates (four categories) and invariant sites. The phylogenetic tree and ancestral reconstruction were performed as they were with HPRT.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
    A possible role of purine nucleotide pyrophorylases in the regulation of purine uptake by Bacillus subtilis
    1. RD Berlin
    2. ER Stadtman
    (1966)
    The Journal of Biological Chemistry 241:2679–2686.
  10. 10
  11. 11
  12. 12
    The Stringent Response
    1. M Cashel
    2. DR Gentry
    3. VJ Hernandez
    4. D Vinella
    (1996)
    In: FC Neidhardt, R Curtiss, JL Ingraham, ECC Lin, KB Low, B Magasanik, W Reznikoff, M Riley, M Schaechter, HE Umbarger, editors. Escherichia Coli and Salmonella Cellular and Molecular Biology. American Society for Microbiology. pp. 1458–1496.
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
    Coot: model-building tools for molecular graphics
    1. P Emsley
    2. K Cowtan
    (2004)
    Acta Crystallographica. Section D, Biological Crystallography 60:2126–2132.
    https://doi.org/10.1107/S0907444904019158
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
    Synthesis of Guanosine Tetra- and pentaphosphates by the obligately anaerobic bacterium Bacteroides thetaiotaomicron in response to molecular oxygen
    1. TL Glass
    2. WM Holmes
    3. PB Hylemon
    4. EJ Stellwag
    (1979)
    Journal of Bacteriology 137:956–962.
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
    The regulation of purine utilization in Bacteria. V. inhibition of purine phosphoribosyltransferase activities and purine uptake in isolated membrane vesicles by Guanosine tetraphosphate
    1. J Hochstadt-Ozer
    2. M Cashel
    (1972)
    The Journal of Biological Chemistry 247:7067–7072.
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
    Metabolism of 5-phosphoribosyl 1-pyrophosphate (PRPP) in Escherichia coli and Salmonella typhimurium
    1. KF Jensen
    (1983)
    In: A Munch-Petersen, editors. Metabolism of Nucleotides, Nucleosides and Nucleobases in Microorganisms. London: Academic Press. pp. 1–25.
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
    Protein Dimerization and Oligomerization in Biology
    1. JM Matthews
    2. M Sunde
    (2012)
    1–18, Dimers, oligomers, everywhere, Protein Dimerization and Oligomerization in Biology, New York, Springer.
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
    Enzyme changes during Bacillus subtilis sporulation caused by deprivation of guanine nucleotides
    1. N Vasantha
    2. E Freese
    (1980)
    Journal of Bacteriology 144:1119–1125.
  83. 83
  84. 84
  85. 85
  86. 86
    A new method of inference of ancestral nucleotide and amino acid sequences
    1. Z Yang
    2. S Kumar
    3. M Nei
    (1995)
    Genetics 141:1641–1650.
  87. 87
  88. 88
  89. 89
  90. 90

Decision letter

  1. Michael T Laub
    Reviewing Editor; Massachusetts Institute of Technology, United States
  2. Michael A Marletta
    Senior Editor; University of California, Berkeley, United States

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your article "Evolution of (p)ppGpp-HPRT regulation through diversification of an allosteric oligomeric interaction" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Michael Marletta as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

In this manuscript, the authors investigated the inhibition of HPRT, a key enzyme for the salvage of purine nucleotides, by the bacterial alarmone (p)ppGpp. They purified recombinant HPRT homologs from a broad spectrum of organisms for in vitro analysis, and revealed a significant variation of their affinities to (p)ppGpp. They then performed in-depth studies of pppGpp-sensitive HPRT homologs from Bacillus species using a combination of structural, biophysical and biochemical techniques. Surprisingly, although the binding site of pppGpp almost completely overlaps with that of HPRT substrates, substrate binding leads to the dissociation of HPRT tetramer, which exists in apo- or pppGpp-bound states, into dimers. By comparing sequences, structures, and oligomerization properties of HPRT homologs sensitive or refractory to (p)ppGpp inhibition, the authors concluded that tetramerization of HPRT triggers structural rearrangements to enhance (p)ppGpp binding, and that a dimer-dimer interface supporting tetramerization is the defining feature of (p)ppGpp sensitivity among HPRT homologs. They call this interplay between oligomeric state and ligand affinity "oligomeric allostery". Using ancestral protein reconstruction, the authors also suggested that the emergence of the dimer-dimer interface is likely coupled to the evolution of (p)ppGpp sensitivity.

Although there was enthusiasm for the work, the reviewers collectively raised a number of concerns, some about the work itself and some about the interpretation or presentation of the data. Some of these concerns should be straightforward to address experimentally. Others may be more involved, particularly points 4-5 and the broader issue of whether this mechanism is relevant in vivo given the physiological concentrations of the molecules in question. If the authors feel that all of the issues could be addressed, the reviewers would welcome a revision.

Essential revisions:

1) The authors cited others' work and their own ITC data and claimed that HPRT is not an allosteric enzyme. In fact, apo-HPRT is likely a tetramer at the working concentration for ITC (45 μM) so that the experiment is not expected to reveal cooperativity. Conversely, it is possible that apo-HPRT falls apart into dimers when diluted to ~100 nM for biochemical assays so that the cooperativity for pRpp was not observed. Nonetheless, if pRpp binding dissociates HPRT tetramers into dimers, ITC of HPRT with pRpp should reveal cooperativity, and the [pRpp]-Vi relationship should deviate from the Michaelis-Menten model when HPRT concentration is sufficient to maintain a tetrameric configuration in the apo-form. The authors should present these data as important supporting evidence for their oligomeric allostery model.

2) Throughout the second half of the manuscript, the authors emphasize the significance of tetramerization on tight (p)ppGpp binding, but touched very little on the other side of the allostery, namely, how tetrameric vs dimeric HPRT affects the enzyme activity? For instance, is HPRT made constitutively tetrameric (e.g., crosslinked with disulfides) less active? And can the phylogenetic studies be used not to break a tetrameric HPRT (as is already done) but to make a normally dimeric HPRT tetrameric and thus subject to ppGpp-based regulation?

3) Subsection “(p)ppGpp regulation of HPRT is conserved across bacteria and beyond”: "these results suggest significant inhibition of HPRT under physiological, basal levels of (p)ppGpp" Provided that inhibition by (p)ppGpp is competitive with respect to pRpp, the magnitude of inhibition also depends on the physiological level and Km for pRpp. In other words, it's difficult to say in vivo what the effect of ppGpp will be (especially at low concentrations) unless one also knows the concentration of pRpp and the enzyme's Km for pRpp. Can these values be measured? If not, it may significantly impact what can be concluded about the level of inhibition by ppGpp in vivo.

4) In the end, it's not clear how important the ppGpp-driven stabilization of a tetramer is for inhibition. In the chimera experiment in Figure 4, the authors disrupt tetramerization and the affinity for ppGpp decreases 20-fold. But is this chimera no longer inhibited by ppGpp or does it just take higher levels? As the authors point out, ppGpp levels can get into the mM regime, so how much does a change in affinity from 1 to 20 μm matter in vivo? Related to this, the authors only examine enzyme inhibition in Figure 1 at one concentration: 25 uM. Maybe some of the enzymes, like the L. pneumophila enzyme used in creating the chimera in Figure 4, are more fully inhibited at even slightly higher concentrations. We wouldn't expect the authors to examine enzyme inhibition in detail for every organism's HPRT, every mutant, and the chimera in Figure 4, but in a few select cases (especially the chimera and a couple of the orthologs) it seems essential to do so.

5) What are the in vivo consequences of not forming a tetramer in B. subtilis, with respect to GTP homeostasis?

6) Ancestral reconstruction

The ancestral reconstruction method used was not correct and needs to be redone. This is a critical problem, as they authors do not establish that the branching pattern of their phylogeny is correct for the deepest nodes (Anc 1-7). If the branching pattern was different, these ancestors simply would not have existed. As a result, it is not clear whether the ancestor was a dimer that became a tetramer (as they claim) or was instead a tetramer that became a dimer.

The solution to these issues would be to do the reconstruction using established methods (see below) using a much larger set of sequences than the 141 chosen. It may be that their conclusions are robust to a better reconstruction, but they must establish this before publication.

A) None of the nodes they reconstructed (Anc 1-7) have bootstrap values above 90. They do not report the support values for these nodes, but I suspect that the support for these nodes is very low (see D below). Even though the posterior probabilities on the reconstruction of ancestral amino acids are high for the sites they identify as important, the authors have not convinced me that these nodes actually existed historically. Maybe the lineage lacking the dimer-dimer motif evolved later, representing a loss of ancestral tetramerization. This possibility cannot be distinguished from the hypothesis they present on the basis of their tree.

B) They constructed their tree using the WAG+G+I substitution model, but then reconstructed their ancestors using the JTT model. This mismatch is not correct and likely changed their reconstructed ancestral amino acid states. This should be re-done using consistent models.

C) They never justified the substitution models they chose (usually done using something like ProtTest or a manual AIC calculation). This should be done.

D) If the branch length unit shown on their tree (Figure 5A) is substitutions per site (as it usually is), their placement of the S. cerevisiae protein is highly suspect. By eye, that branch accumulated ~1.5 subs/site, indicating it likely aligned very poorly. All one can see from the tree is that the bootstrap value for this placement is <90 – but likely it is much lower. There are several strategies they could/should employ to get around this. First, build a much larger alignment that includes multiple representative sequences from the outgroup clade. This should better resolve the phylogeny. Second, report support values on all nodes-particularly the nodes that they use for reconstruction studies. If the support for these nodes remains low, strong claims about the order of branching and order of evolution of dimerization and tetramerization cannot be made.

E) They show a branch length label of "0.2" on their tree (Figure 5A), but do not define it in the legend. This should be done.

F) They did not describe what method they used for the reconstruction. They cite Jones, 1992, but this is the JTT substitution model, not the reconstruction method. Presumably they are using a marginal probability reconstruction (e.g. Yang, 1995). This should be stated.

G) To ensure reproducibility, the authors should publish their alignment in the supplement.

7) "Oligomeric Allostery"

The authors spend significant time in the discussion arguing that their work reveals a new phenomenon they call "oligomeric allostery." It does not seem it was ever defined precisely in the text; however, as we understood it, it is a change in oligomeric state that controls ligand specificity. They point out in numerous places that they think is a new and important observation (Discussion section).

We do not think it is new. The authors note that there are many instances in which ligand binding controls oligomeric state (e.g. subsection “Protein oligomerization allosterically alters ligand specificity”). But, from the perspective of linkage thermodynamics, this is equivalent to saying ligand binding controls oligomeric state. The earliest thinking regarding allostery and cooperativity was informed by studies of oligomeric states (Perutz, MWC, etc.).

In our view, this framing hurts the manuscript, but not the underlying study. We think they could recast the work without relying on this dubious premise. The authors would be better served to argue that ligand binding and oligomeric states are often tightly linked to one another. They could then describe their work as a cool way that this interplay enables evolutionary change (in their case, by allowing mutations to accumulate away from a conserved binding pocket to promote a change in specificity).

[Editors' note: further revisions were requested prior to acceptance, as described below.]

Thank you for submitting your article "Evolution of (p)ppGpp-HPRT regulation through diversification of an allosteric oligomeric interaction" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Michael Marletta as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

The reviewers were generally satisfied with your responses to the initial set of concerns raised. There are, however, a couple of remaining issues that you should address in a revision.

1) The first is that the reviewers were not entirely convinced by the proposed mechanism. If we understood it correctly, you are still proposing a cooperative mechanism but have no data supporting cooperativity. This came out most directly in the revision cover letter, where you suggest that cooperativity would not arise if (1) a single (p)ppGpp is sufficient to lock the tetramer in place and (2) four PRPP molecules are necessary to cause dissociation into a dimer. But this did not make sense to the reviewers – one of them even pieced together a set of linkage equations compatible with your verbal model and explored them on a computer and could not find any parameters that did not display cooperativity in PRPP. Your revision should clarify your proposed mechanism and whether cooperativity is a key feature and if so, what results support that interpretation.

2) It was difficult to assess your revisions as you did not always state whether the individual points had given rise to textual changes or not. Some were obvious, but others less so. In your next revision, please address the following:

a) How is the oligomeric state of each structure assessed? While it is clear that the PRPP + 9- deazaguanine bound form is dimeric from the packing in the supplied PDB file, the tetramer observed in the other two structures could in theory be a result of crystal packing. Usually, a tool like PISA is used to assess the validity of proposed multimers in crystal structures, but the authors make no reference to this. Please confirm that PISA agrees with the conclusions about oligomeric state.

Thank you for including the PISA analysis in the response to reviewers, this is quite clear. I also notice that you now mention PISA in the Materials and methods section. However, I believe it will be of value to the readers to have this information also at the relevant place in the Results section, i.e. whenever you conclude on oligomeric state, I suggest you say that this is supported by PISA analysis.

b) Subsection “(p)ppGpp binds the conserved active site of HPRT and closely mimics substrate binding”. Have the authors considered that the pppGpp preparation could contain a ppGpp contamination that crystallised with the protein? Would it be possible to check for this?

Could the authors please indicate whether this has resulted in any changes to the manuscript?

c) Subsection “(p)ppGpp prevents PRPP-induced dissociation of HPRT dimer-of-dimers” and Figure 3. What is the reason for excluding 9-deazaguanine during gel filtration and cross- linking? The comparison to the crystal structure would be more direct if both substrates were present. Is it possible that the oligomeric state of the protein only in the presence of PRPP is different?

Could the authors please indicate whether this has resulted in any changes to the manuscript? In this case I would suggest that you include this additional experiment as supplementary data in case other readers wonder.

d) Subsection “Dimer-dimer interaction allosterically positions loop II for potentiated (p)ppGpp binding”. Can the authors exclude that the structural variations observed in loop II to some extent are induced by crystal contacts?

Did this give rise to textual changes?

https://doi.org/10.7554/eLife.47534.055

Author response

Essential revisions:

1) The authors cited others' work and their own ITC data and claimed that HPRT is not an allosteric enzyme. In fact, apo-HPRT is likely a tetramer at the working concentration for ITC (45 μM) so that the experiment is not expected to reveal cooperativity. Conversely, it is possible that apo-HPRT falls apart into dimers when diluted to ~100 nM for biochemical assays so that the cooperativity for pRpp was not observed. Nonetheless, if pRpp binding dissociates HPRT tetramers into dimers, ITC of HPRT with pRpp should reveal cooperativity, and the [pRpp]-Vi relationship should deviate from the Michaelis-Menten model when HPRT concentration is sufficient to maintain a tetrameric configuration in the apo-form. The authors should present these data as important supporting evidence for their oligomeric allostery model.

This is a very good point. To address this question, we conducted ITC between PRPP and HPRT (~45 μM). The isotherm fit a single site binding model without observed cooperativity for PRPP binding (with a Kd ~ 7 μM, new data added as Figure 4 —figure supplement 5). Because we have already observed with dynamic light scattering at 175 μM of HPRT that PRPP still induces tetramer to dimer formation (Table 3), we expect at the 45 μM concentration used for ITC, PRPP would also induce tetramer to dimer transition. Thus, at high HPRT concentrations sufficient to maintain a tetrameric configuration in the apo form, we have not observed cooperativity for PRPP binding despite its ability to induce the tetramer to dimer transition. We assume that the [PRPP]-Vi relationship will not deviate from the Michaelis-Menten model at HPRT concentrations sufficient to maintain a tetrameric configuration in the apo-form.

We propose that the lack of cooperativity is best explained by PRPP having to associate with all four binding sites before it induces the tetramer to dimer transition. This would also explain why (p)ppGpp binding can lock the HPRT as a tetramer despite addition of high concentrations of PRPP – one (p)ppGpp binding may be sufficient to prevent dimeric HPRT formation.

Finally, we reasoned that our original ‘oligomeric allostery’ model where the enzyme is not allosteric is confusing. To remove this confusion, we have removed (as suggested by reviewers in comment 6) the term ‘oligomeric allostery’.

2) Throughout the second half of the manuscript, the authors emphasize the significance of tetramerization on tight (p)ppGpp binding, but touched very little on the other side of the allostery, namely, how tetrameric vs dimeric HPRT affects the enzyme activity?

We found that naturally occurring dimeric HPRT homologs have similar activity compared to the tetrameric B. anthracis Hpt-1 and B. subtilis HPRT (Table 4). Between the naturally occurring homologs, the biggest difference as far as we can perceive, is their ability to be regulated by (p)ppGpp.

In addition, we found that the majority of the dimeric homologs also tend to have lower Km values for PRPP. This suggests that the evolution of the oligomer is not just affecting the affinity for (p)ppGpp but also for PRPP, and the dimer favors PRPP binding but the tetramer favors (p)ppGpp binding, which should also play a role in their reduced inhibition by (p)ppGpp. This point has been elaborated in the Discussion section.

For instance, is HPRT made constitutively tetrameric (e.g., crosslinked with disulfides) less active? And can the phylogenetic studies be used not to break a tetrameric HPRT (as is already done) but to make a normally dimeric HPRT tetrameric and thus subject to ppGpp-based regulation?

To address this comment, we constructed a presumably constitutively tetrameric HPRT with a Y117C variant. This variant had significantly lower activity than WT (~6% of the Vmax). Also, while this variant bound pppGpp as well as the WT HPRT, it was nearly resistant to high concentrations of competing PRPP. This suggests that PRPP is unable to bind and dissociate a locked tetramer. These data are now shown as Figure 4—figure supplement 4.

We have also attempted to make a normally dimeric HPRT a tetramer. We replaced interface residues of the dimeric L. pneumophila HPRT with corresponding residues from B. subtilis HPRT. However, this variant was insoluble, perhaps due to improper folding. Despite the strength of the phylogenetic analyses, the HPRT homologs are too sequence diverse to pinpoint all compensatory changes.

3) Subsection “(p)ppGpp regulation of HPRT is conserved across bacteria and beyond”: "these results suggest significant inhibition of HPRT under physiological, basal levels of (p)ppGpp" Provided that inhibition by (p)ppGpp is competitive with respect to pRpp, the magnitude of inhibition also depends on the physiological level and Km for pRpp. In other words, it's difficult to say in vivo what the effect of ppGpp will be (especially at low concentrations) unless one also knows the concentration of pRpp and the enzyme's Km for pRpp. Can these values be measured? If not, it may significantly impact what can be concluded about the level of inhibition by ppGpp in vivo.

We thank the reviewer to remind us to clarify this point. The regulation by basal level is highly relevant in vivo.

1) Physiological PRPP levels in bacteria are roughly 0.1 – 1 mM (Bennett et al., (2009); Jensen, (1983)). In B. subtilis, the PRPP level is also in this range (0.2 – 1 mM) (Berlin and Stadtman, (1966)). All inhibition assays were performed with 1 mM PRPP, the high end of PRPP concentration. At this concentration, (p)ppGpp at lower concentrations (25 μM) can still strongly inhibit HPRT activity (Figure 2E). On the other hand, for the constitutive dimeric homologs, at 1 mM PRPP, (p)ppGpp can no longer inhibit HPRT (Figure 2E). We now add this information in the Discussion section.

2) We measured Km for PRPP for B. subtilis HPRT and B. anthracis Hpt-1 (166 μm and 112 uM, data shown in Table 4). Most bacterial HPRTs have a Km for PRPP in the range of 50 – 200 μM (Guddat et al., (2003); Zhang et al., (2016)). On the other hand, Kd values for pppGpp typically range from 0.1-10 μM. Both of these parameters are within the physiological range of PRPP and (p)ppGpp.

3) We performed enzymatic assays showing that (p)ppGpp is a competitive inhibitor of HPRT with its substrate PRPP.

4) In the end, it's not clear how important the ppGpp-driven stabilization of a tetramer is for inhibition. In the chimera experiment in Figure 4, the authors disrupt tetramerization and the affinity for ppGpp decreases 20-fold. But is this chimera no longer inhibited by ppGpp or does it just take higher levels? As the authors point out, ppGpp levels can get into the mM regime, so how much does a change in affinity from 1 to 20 μm matter in vivo?

This is a valid point that we are now able to clarify during revision. Here, it important to point out that regulation by (p)ppGpp is both important at basal level and at the induced level (e.g. 20-fold higher). In fact, basal level of (p)ppGpp has been shown to be very important for physiology, however, most of the (p)ppGpp biochemistry only showed that induced levels of (p)ppGpp regulate its enzymatic targets (i.e. upon starvation or other stress). The key to HPRT regulation is basal level (p)ppGpp. Our previous experiment (Kriel et al., 2012) showed that basal levels of (p)ppGpp are important for GTP homeostasis in vivo, protecting cells by inhibiting the purine salvage pathway at a basal (p)ppGpp concentration that is in strong contrast to the induced sub-millimolar or millimolar concentrations. In the context studied in Kriel et al., guanine is added as a source of purine in cells supplemented with both glucose and amino acids. In wild type cells grown with guanine, (p)ppGpp remained at its basal level (estimated to be only ~30 μM), yet this level of (p)ppGpp prevents excess GTP production through purine salvage and maintains homeostasis. Therefore, our analysis is highly relevant at least in the case of Bacillus species.

To translate the importance of a change in affinity from 1 to 20 μM, at a physiological concentration of PRPP (1 mM, assuming PRPP Km = 0.1 mM), a Ki of 1 μM for pppGpp will result in ~75% inhibition at 30 μM pppGpp. On the other hand, a Ki of 20 μM will result in ~12% inhibition at 30 μM pppGpp. Even upon (p)ppGpp inducing to 0.5 mM, a Ki of 20 μM will result in ~70% inhibition of HPRT activity. Thus, lowering the affinity for pppGpp by 20-fold would mean that it is no longer an effective competitor of PRPP, especially at basal (p)ppGpp, and would likely keep the enzyme active.

Related to this, the authors only examine enzyme inhibition in Figure 1 at one concentration: 25 uM. Maybe some of the enzymes, like the L. pneumophila enzyme used in creating the chimera in Figure 4, are more fully inhibited at even slightly higher concentrations. We wouldn't expect the authors to examine enzyme inhibition in detail for every organism's HPRT, every mutant, and the chimera in Figure 4, but in a few select cases (especially the chimera and a couple of the orthologs) it seems essential to do so.

In the revised version, IC50 curves of multiple orthologs, including C. burnetii, are shown in Figure 2E. For C. burnetii, IC50 > 100 μM, much stronger than the 10 μM IC50 for B. subtilis.

5) What are the in vivo consequences of not forming a tetramer in B. subtilis, with respect to GTP homeostasis?

We performed several experiments to examine the in vivo consequences of weakening the affinity of HPRT for (p)ppGpp with respect to GTP homeostasis. We first attempted to introduce the (p)ppGpp-refractory C. burnetii hprT into B. subtilishprT + amyEcbuhprT that expresses C. burnetii HPRT from an IPTG inducible promoter). However, the homolog has a weak expression in B. subtilis making it unable to utilize guanine efficiently (shown by a lack of growth rate increase upon guanine addition). We next constructed a codon-optimized C. burnetii hprT, but this allele is unable to transform into B. subtilis, presumably because of an unidentified toxicity.

Eventually, we found a way to demonstrate the physiological importance of basal (p)ppGpp regulation of guanine salvage by introducing a E. coli guanine salvage enzyme XGPRT into B. subtilis. We biochemically characterized this enzyme to show that its ability to catalyze PRPP and guanine conversion to GMP is inhibited by (p)ppGpp at IC50 of 45 μM. In other words, this enzyme is inhibited by induced, but not basal level of (p)ppGpp (Figure 1B). When we introduced this construct into an ecgmk background (a background where B. subtilis GMK which is strongly inhibited by (p)ppGpp is replaced by E. coli GMK that is resistant to (p)ppGpp), we are able to observe a much higher concentration of GTP (4-fold higher) upon addition of guanosine (Figure 1C). Therefore, we are able to verify in vivo consequences of an HPRT that is inhibited only by induced (p)ppGpp, with respect to GTP homeostasis.

6) Ancestral reconstruction

The ancestral reconstruction method used was not correct and needs to be redone. This is a critical problem, as they authors do not establish that the branching pattern of their phylogeny is correct for the deepest nodes (Anc 1-7). If the branching pattern was different, these ancestors simply would not have existed. As a result, it is not clear whether the ancestor was a dimer that became a tetramer (as they claim) or was instead a tetramer that became a dimer.

The solution to these issues would be to do the reconstruction using established methods (see below) using a much larger set of sequences than the 141 chosen. It may be that their conclusions are robust to a better reconstruction, but they must establish this before publication.

We are thankful to the reviewers for their helpful criticism of the phylogenetic methods. We re-performed the phylogenetic analysis with an expanded 430 HPRT sequences with the specific issues raised below in mind. With the new analysis, we still obtain a phylogenetic tree that branches into two broad lineages: one sensitive to basal (p)ppGpp and one insensitive to basal (p)ppGpp. The ancestor to the (p)ppGpp-sensitive clade has a relatively low bootstrap value (55-60), indicating low confidence in this ancestor, as pointed out in issue A below. The key is that the residues we had previously identified as being important within each branch are still found to be important with the new analysis – one branch contains residues associated with dimeric HPRTs and the other does not. The Anc1 does not have residues associated with dimeric HPRTs, and it likely contains the Lys81 associated with the (p)ppGpp-regulated clade, suggesting that it is a tetramer.

The lower confidence in these branches inspired us to analyze another enzyme differentially regulated by (p)ppGpp: GMK, which has a much higher sequence identity across organisms than HPRT. Even with fewer sequences, this resulted in a more robust tree that also differentiates (p)ppGpp-sensitive and insensitive GMKs into two broad clades and suggests that the ancestral GMK was also regulated by (p)ppGpp.

A) None of the nodes they reconstructed (Anc 1-7) have bootstrap values above 90. They do not report the support values for these nodes, but I suspect that the support for these nodes is very low (see D below). Even though the posterior probabilities on the reconstruction of ancestral amino acids are high for the sites they identify as important, the authors have not convinced me that these nodes actually existed historically. Maybe the lineage lacking the dimer-dimer motif evolved later, representing a loss of ancestral tetramerization. This possibility cannot be distinguished from the hypothesis they present on the basis of their tree.

This is a very good point and is addressed above.

B) They constructed their tree using the WAG+G+I substitution model, but then reconstructed their ancestors using the JTT model. This mismatch is not correct and likely changed their reconstructed ancestral amino acid states. This should be re-done using consistent models.

We used ProtTest to identify the best substitution model with the lowest AIC score as LG+G+I. Ancestral reconstruction was performed with the same model.

C) They never justified the substitution models they chose (usually done using something like ProtTest or a manual AIC calculation). This should be done.

See (B) above.

D) If the branch length unit shown on their tree (Figure 5A) is substitutions per site (as it usually is), their placement of the S. cerevisiae protein is highly suspect. By eye, that branch accumulated ~1.5 subs/site, indicating it likely aligned very poorly. All one can see from the tree is that the bootstrap value for this placement is <90 – but likely it is much lower. There are several strategies they could/should employ to get around this. First, build a much larger alignment that includes multiple representative sequences from the outgroup clade. This should better resolve the phylogeny. Second, report support values on all nodes-particularly the nodes that they use for reconstruction studies. If the support for these nodes remains low, strong claims about the order of branching and order of evolution of dimerization and tetramerization cannot be made.

This branch length unit is the substitutions per site. We included more sequences, particularly in the outgroup clade, to get a better alignment. All bootstrap values are now reported, especially for the ancestral states.

E) They show a branch length label of "0.2" on their tree (Figure 5A), but do not define it in the legend. This should be done.

The branch length label has been defined in the legend.

F) They did not describe what method they used for the reconstruction. They cite Jones, 1992, but this is the JTT substitution model, not the reconstruction method. Presumably they are using a marginal probability reconstruction (e.g. Yang, 1995). This should be stated.

Correct, the reconstruction method was according to Yang, 1995. This has been cited in the Results section and described in the Materials and methods section.

G) To ensure reproducibility, the authors should publish their alignment in the supplement.

The alignments have been included as Source Data files (Figure 6—source data 1 and Figure 7—source data 1).

7) "Oligomeric Allostery"

The authors spend significant time in the discussion arguing that their work reveals a new phenomenon they call "oligomeric allostery." It does not seem it was ever defined precisely in the text; however, as we understood it, it is a change in oligomeric state that controls ligand specificity. They point out in numerous places that they think is a new and important observation (Discussion section).

We do not think it is new. The authors note that there are many instances in which ligand binding controls oligomeric state (e.g. subsection “Protein oligomerization allosterically alters ligand specificity”). But, from the perspective of linkage thermodynamics, this is equivalent to saying ligand binding controls oligomeric state. The earliest thinking regarding allostery and cooperativity was informed by studies of oligomeric states (Perutz, MWC, etc.).

In our view, this framing hurts the manuscript, but not the underlying study. We think they could recast the work without relying on this dubious premise. The authors would be better served to argue that ligand binding and oligomeric states are often tightly linked to one another. They could then describe their work as a cool way that this interplay enables evolutionary change (in their case, by allowing mutations to accumulate away from a conserved binding pocket to promote a change in specificity).

We thank the reviewers for their feedback on this point and for the proposed reinterpretation. We have altered our model and modified the text at multiple points to incorporate these ideas.

[Editors' note: further revisions were requested prior to acceptance, as described below.]

The reviewers were generally satisfied with your responses to the initial set of concerns raised. There are, however, a couple of remaining issues that you should address in a revision.

1) The first is that the reviewers were not entirely convinced by the proposed mechanism. If we understood it correctly, you are still proposing a cooperative mechanism but have no data supporting cooperativity. This came out most directly in the revision cover letter, where you suggest that cooperativity would not arise if (1) a single (p)ppGpp is sufficient to lock the tetramer in place and (2) four PRPP molecules are necessary to cause dissociation into a dimer. But this did not make sense to the reviewers – one of them even pieced together a set of linkage equations compatible with your verbal model and explored them on a computer and could not find any parameters that did not display cooperativity in PRPP. Your revision should clarify your proposed mechanism and whether cooperativity is a key feature and if so, what results support that interpretation.

We thank the reviewers for the opportunity to clarify this point. We do not think that cooperativity is a key feature of our model. We can confidently say that variation in the dimer-dimer interface influences the selectivity for substrate versus inhibitor. The potential cooperativity of PRPP binding (or lack thereof) does not change our findings that the dimer prefers PRPP over (p)ppGpp and the tetramer prefers (p)ppGpp over PRPP.

We thank the reviewers for pointing out that cooperativity would still arise if single (p)ppGpp is sufficient to lock the tetramer in place or four PRPP molecules are necessary to cause dissociation into a dimer. We agree that cooperativity is likely detectable under certain conditions. However, in the conditions we tested, within physiological ranges of (p)ppGpp and PRPP, we did not detect cooperativity. Our literature search of this extensively characterized enzyme did not find observation of cooperativity of native HPRT. On the other hand, kinetic cooperativity between PRPP and HPRT has been reported in mutant human HPRTs (Lightfoot et al., 1994; Balendiran et al., 1999). Thus, observing this cooperativity requires altering the native enzyme.

In light of these investigations, we revised the argument about the physiological relevance of cooperativity in the main text. We feel that the rest of the manuscript would be clear and impactful without the argument of cooperativity related to PRPP.

2) It was difficult to assess your revisions as you did not always state whether the individual points had given rise to textual changes or not. Some were obvious, but others less so. In your next revision, please address the following:

a) How is the oligomeric state of each structure assessed? While it is clear that the PRPP + 9- deazaguanine bound form is dimeric from the packing in the supplied PDB file, the tetramer observed in the other two structures could in theory be a result of crystal packing. Usually, a tool like PISA is used to assess the validity of proposed multimers in crystal structures, but the authors make no reference to this. Please confirm that PISA agrees with the conclusions about oligomeric state.

Thank you for including the PISA analysis in the response to reviewers, this is quite clear. I also notice that you now mention PISA in the Materials and methods section. However, I believe it will be of value to the readers to have this information also at the relevant place in the Results section, i.e. whenever you conclude on oligomeric state, I suggest you say that this is supported by PISA analysis.

Thanks! We added these statements in subsection “p)ppGpp prevents PRPP-induced dissociation of HPRT dimer-of-dimers”.

b) Subsection “(p)ppGpp binds the conserved active site of HPRT and closely mimics substrate binding”. Have the authors considered that the pppGpp preparation could contain a ppGpp contamination that crystallised with the protein? Would it be possible to check for this?

Could the authors please indicate whether this has resulted in any changes to the manuscript?

We have added this observation in subsection “(p)ppGpp binds the conserved active site of HPRT and closely mimics substrate binding”.

c) Subsection “(p)ppGpp prevents PRPP-induced dissociation of HPRT dimer-of-dimers” and Figure 3. What is the reason for excluding 9-deazaguanine during gel filtration and cross- linking? The comparison to the crystal structure would be more direct if both substrates were present. Is it possible that the oligomeric state of the protein only in the presence of PRPP is different?

Could the authors please indicate whether this has resulted in any changes to the manuscript? In this case I would suggest that you include this additional experiment as supplementary data in case other readers wonder.

The data have been added as Figure 4—figure supplement 4, and changes have been made to the Figure 4 legend.

d) Subsection “Dimer-dimer interaction allosterically positions loop II for potentiated (p)ppGpp binding”. Can the authors exclude that the structural variations observed in loop II to some extent are induced by crystal contacts?

Did this give rise to textual changes?

This has been clarified in subsection “p)ppGpp prevents PRPP-induced dissociation of HPRT dimer-of-dimers” and the figure has been added as Figure 4—figure supplement 1.

https://doi.org/10.7554/eLife.47534.056

Article and author information

Author details

  1. Brent W Anderson

    Department of Bacteriology, University of Wisconsin, Madison, United States
    Contribution
    Conceptualization, Data curation, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-2785-5343
  2. Kuanqing Liu

    Department of Bacteriology, University of Wisconsin, Madison, United States
    Contribution
    Conceptualization, Formal analysis, Investigation, Methodology
    Competing interests
    No competing interests declared
  3. Christine Wolak

    Department of Biomolecular Chemistry, University of Wisconsin, Madison, United States
    Contribution
    Formal analysis, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  4. Katarzyna Dubiel

    Department of Biomolecular Chemistry, University of Wisconsin, Madison, United States
    Contribution
    Formal analysis, Investigation, Writing—review and editing
    Competing interests
    No competing interests declared
  5. Fukang She

    Department of Bacteriology, University of Wisconsin, Madison, United States
    Contribution
    Investigation
    Competing interests
    No competing interests declared
  6. Kenneth A Satyshur

    Department of Biomolecular Chemistry, University of Wisconsin, Madison, United States
    Contribution
    Data curation, Formal analysis, Investigation
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-9371-2493
  7. James L Keck

    Department of Biomolecular Chemistry, University of Wisconsin, Madison, United States
    Contribution
    Resources, Data curation, Supervision, Visualization, Writing—review and editing
    Competing interests
    No competing interests declared
  8. Jue D Wang

    Department of Bacteriology, University of Wisconsin, Madison, United States
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Project administration, Writing—review and editing
    For correspondence
    wang@bact.wisc.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-1503-170X

Funding

National Institute of General Medical Sciences (R35 GM127088)

  • Jue D Wang

Howard Hughes Medical Institute (Faculty Scholar)

  • Jue D Wang

National Science Foundation (GRFP DGE-1256259)

  • Brent W Anderson

National Institutes of Health (R01 GM084003)

  • Jue D Wang

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Esra Gumuser and Matthew Macgilvray for technical help, and members of the Wang lab for feedback on this manuscript. We thank Federico Rey, Bob Kerby, JD Sauer, Dan Pensinger, Helen Blackwell, and Kayleigh Nyffeler for assistance in obtaining bacterial genomic DNA. We thank Katrina Forest for assistance with dynamic light scattering. This research used resources of the Advanced Photon Source, a U.S. Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. Use of Life Sciences Collaborative Access Team beamline 21-ID-D was supported by the Michigan Economic Development Corporation and the Michigan Technology Tri-Corridor (Grant 085P1000817). This work was funded by NIH R35 GM127088 and R01 GM084003 to JDW, and BWA was supported by NSF GRFP DGE-1256259.

Senior Editor

  1. Michael A Marletta, University of California, Berkeley, United States

Reviewing Editor

  1. Michael T Laub, Massachusetts Institute of Technology, United States

Publication history

  1. Received: April 9, 2019
  2. Accepted: September 24, 2019
  3. Accepted Manuscript published: September 25, 2019 (version 1)
  4. Version of Record published: October 8, 2019 (version 2)

Copyright

© 2019, Anderson et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 507
    Page views
  • 102
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Biochemistry and Chemical Biology
    2. Cell Biology
    Xiaocheng Zhao et al.
    Research Article Updated
    1. Biochemistry and Chemical Biology
    Shannon N Mostyn et al.
    Research Article