Structural and mechanistic divergence of the small (p)ppGpp synthetases RelP and RelQ

The nutritional alarmones ppGpp and pppGpp (collectively: (p)ppGpp) are nucleotide-based second messengers enabling bacteria to respond to environmental and stress conditions. Several bacterial species contain two highly homologous (p)ppGpp synthetases named RelP (SAS2, YwaC) and RelQ (SAS1, YjbM). It is established that RelQ forms homotetramers that are subject to positive allosteric regulation by pppGpp, but structural and mechanistic insights into RelP lack behind. Here we present a structural and mechanistic characterization of RelP. In stark contrast to RelQ, RelP is not allosterically regulated by pppGpp and displays a different enzyme kinetic behavior. This discrepancy is evoked by different conformational properties of the guanosine-substrate binding site (G-Loop) of both proteins. Our study shows how minor structural divergences between close homologues result in new functional features during the course of molecular evolution.


Results
RelP and RelQ share an equal architecture. To better understand RelP at the molecular level, we determined the crystal structures of RelP homologues from S. aureus (Sa) and B. subtilis (Bs) at 2.25 and 3.3 Å resolution, respectively (Table S1). Both, SaRelP and BsRelP form highly symmetrical and oval-shaped homotetramers with a prominent cleft in their centers highly reminiscent of BsRelQ (Figs 1b and S1a). Helix α1 at the N-terminus of each monomer stabilizes the medial sides of the homotetramer interface via hydrogen bonds and salt bridges (buried surface area of ~1200 Å 2 ). Helices α5 and α6 at the C-terminus of each monomer establish the lateral sides of the homotetramer interface mainly due to polar contacts (buried surface area of ~1200 Å 2 ). The (p) ppGpp synthetase monomers of SaRelP and BsRelP are highly identical and consist of a mixed β-sheet build by five β-strands (β1-β5) that is surrounded by alpha helices (α1-α6, Figs 1c and S1b).
Structural comparison of RelP and RelQ reveals the architecture of the homotetramer as well as each of the monomers is highly similar (r.m.s.d. of 1.292 over 138 Cα atoms for the RelP and RelQ monomers). However, RelP and RelQ differ in the orientation of helix α2, which appears to be shifted approximately 3 Å towards the active site center in RelQ when compared to RelP ( Fig. 1c; right panel). Another interesting observation is that the loop connecting β3 and β4, which is disordered in the structure of BsRelQ, could be resolved in both structures of RelP (Fig. 1c). Taken together, RelP and RelQ share highly conserved ternary and quaternary structures, but also reveal subtle differences that might be of functional relevance (see below).

RelP and RelQ differ in their (p)ppGpp synthetase activity. The most distinguished features of
BsRelQ lie in the apparent positive cooperativity of (p)ppGpp synthesis and its susceptibility to allosteric stimulation of by pppGpp but not ppGpp 33 . To test whether both features would also be present in RelP, we performed an in-depth kinetic analysis. We used the same buffer composition for characterization of SaRelP as previously for BsRelQ to ensure maximal comparability. SaRelP was incubated together with 5 mM ATP and varying concentrations of GDP or GTP (Of note: BsRelP exhibited no (p)ppGpp synthetase activity under our assay conditions for unclear reasons). SaRelP synthesized ppGpp more efficiently than pppGpp as evidenced from an approximately 4-fold higher V max value ( Fig. 2a and b). A similar preference for the product ppGpp was previously observed for BsRelQ 33 and RelQ from other organisms 25,30 . However, the K m values for (p)ppGpp synthesis drastically differ between both enzymes in that they are significantly lower for SaRelP (i.e. 0.3 ± 0.2 for GDP and 0.1 ± 0.1 for GTP) than for BsRelQ (i.e. 1.7 ± 0.1 for GDP and 1.2 ± 0.1 for GTP; Fig. 2b). It also seemed to us that SaRelP monomers displays less cooperativity within the tetramer than BsRelQ indicated by Hill coefficients closer to 1 (Fig. 2b).
Amino acid sequence analysis of RelP shows that the amino acid residues required for allosteric binding of pppGpp to RelQ are replaced in RelP proteins (Fig. 2c). Indeed, this different set of amino acids found in SaRelP seems incapable to coordinate pppGpp in similar fashion as BsRelQ (Fig. S2) strongly suggesting to us that SaRelP cannot be allosterically stimulated by the alarmone. In agreement with our structural analysis, no change in the enzymatic activity of SaRelP was observed in the absence and the presence of ppGpp or pppGpp (Fig. 2d). Taken together, RelQ and RelP do not differ much in their V max values of (p)ppGpp synthesis, while significantly differing in the in K m values. Moreover, RelP is not subject to allosteric stimulation by pppGpp.

ATP-binding to RelP and RelQ is identical.
To gain further insights into the disparate enzymatic activities of RelQ and RelP, we attempted to solve the structure of SaRelP in presence of the non-hydrolysable ATP analogue AMPCPP (α,β-methyleneadenosine 5′-triphosphate) and GDP or GTP. However, we could only obtain crystals and solve the structure of SaRelP in presence of AMPCPP ( Fig. S3a and Table S1). Coordination of AMPCPP within all the four active sites of SaRelP is guided by π-stacking interactions of the adenine base with the arginine residues 78 and 112 of SaRelP (Fig. S3b). The ribose moiety of the adenosine is coordinated by hydrogen bonding via His190. Interactions with the phosphate moieties of AMPCPP are mainly established by lysine and arginine residues residing in β1 and α2 (i.e. Lys80, Lys88 and Arg91) and Ser84 contacting the 5′ α-phosphate. AMPCPP adopts a kinked conformation that is enforced by a magnesium ion coordinated by Asp107 and Glu174 (Fig. S3b). An identical conformation of AMPCPP is observed in the active site of BsRelP (Fig. S3c). As all ATP-coordinating and catalytic amino acid residues are strictly conserved among RelP/RelQ proteins (Fig. S3d), we suspect a common ATP-binding mode and mode of catalysis.
G-Loop rigidity governs the activity of RelP and RelQ. If binding of ATP to RelP and RelQ is identical (see above), then the different enzymatic properties of both enzymes should originate from differences in binding of GDP/GTP and/or a different susceptibility to allosteric stimulation by pppGpp. As mentioned above, our structural analysis of RelP and RelQ indicated a different conformational flexibility of the loop connecting strands β3 and β4 (Fig. 1c). This loop contains a conserved tyrosine residue (i.e. Tyr151 in SaRelP and Tyr116 in BsRelQ, Fig. 3a) critical to guanosine nucleotide binding in all (p)ppGpp synthetases. Therefore, we decided to term the loop connecting β3 and β4 'G-Loop' . To our surprise, the different configurations of the G-Loop seem to be a common theme among RelP/RelQ proteins. In the apo-and ATP-bound states of BsRelQ, the G-Loop is disordered, and could therefore not be modeled in these structures (Fig. 3b). In stark contrast, the G-Loop of SaRelP was well-ordered and could be unambiguously modeled in its apo-and ATP-bound structures (Fig. 3c). We speculated that the difference in enzymatic activity between RelQ and RelP is founded in the different conformational properties of the G-Loop.
Inspection of the amino acids of the G-Loop reveals the presence of proline in RelP proteins with no correspondent in RelQ (Fig. 3a). We hypothesized that the absence of this proline in RelQ renders the G-Loop less rigid, while its presence in RelP results in a well-ordered G-Loop that might easily facilitate GDP/GTP coordination ( Fig. 3d). We challenged this notion by introducing proline into the disordered G-Loop of RelQ (i.e. BsRelQ-H111P). BsRelQ-H111P produces (p)ppGpp as efficient as SaRelP and the V max (i.e. 243 ± 9 and 194 ± 8 nmol min −1 nmol −1 for ppGpp and pppGpp, respectively), K m (i.e. 0.4 ± 0.2 for GDP and 1.9 ± 0.2 for GTP) and Hill-coefficient (i.e. 1.6 ± 0.2 for GDP and 1.0 ± 0.1 for GTP) of BsRelQ-H111P more resemble SaRelP than BsRelQ ( Fig. 3e and compare to Fig. 2a and b). Moreover and unlike BsRelQ, BsRelQ-H111P is not amenable to allosteric stimulation by pppGpp (Fig. 3f). These results demonstrate a strong dependence of RelP/RelQ activity on the rigidity of the G-Loop.

Allosteric stimulation of RelQ by pppGpp acts via the G-Loop. Our results indicated that RelP
proteins synthesize (p)ppGpp more efficiently than RelQ, because RelP can more readily bind the GDP/GTP substrate through increased rigidity of the G-Loop. Moreover, pppGpp stimulates the activity of RelQ, while it does not for RelP (Figs 2d and 3f). Therefore, we hypothesized that binding of pppGpp to the central cleft of RelQ might be translated into an increased (p)ppGpp synthesis via the G-Loop. Superimposition of the crystal structures of apo-BsRelQ and pppGpp-bound BsRelQ (PDB: 5DEC and 5DED 33 , respectively) allowed tracing a structurally possible path, which would connect the presence of pppGpp within the allosteric cleft of RelQ with the G-loop (Fig. 4). In short, two opposing subunits of the BsRelQ tetramer are involved in coordination of one allosteric pppGpp in the central cleft 1,33 . Coordination of pppGpp leads to a displacement of Phe42, Thr44 and Asn148 by ~1-2 Å towards the cleft (Figs 4b; S2). Helix α4 comprising Asn148 follows this movement and rotates by approximately 15° in a counterclockwise manner. This movement is relayed onto helix α5 through the hydrophobic core between both helices constituted by Phe149 (α4), Leu183 and Met187 (both α5, Fig. 4c). Rotation of α5 turns Glu178 towards the G-Loop and enables formation of a salt bridge between Glu178 and Arg117 (Fig. 4c). Further contacts between α5 and the G-Loop are established between His111/Glu178 and Glu113/ Gln174 (Fig. 4d).
To probe the participation of these amino acids, we replaced them by alanine and measured the (p)ppGpp synthesis of the resulting BsRelQ variants in pppGpp-dependent manner (Figs 4e and S4). Variation of His111 and Glu113 does not affect stimulation of BsRelQ. However, upon replacement of Gln174 and Glu178 the pppGpp-stimulatory effect is decreased and completely abolished when Arg117 is replaced (Figs 4e and S4).
Finally, we tested how the allosteric pppGpp affects the enzyme kinetic behaviour of BsRelQ by determining the (p)ppGpp synthesis BsRelQ in presence of different concentrations of pppGpp (i.e. 0, 2.5, 10, 25, 100, 250 µM). While addition of increasing amounts of pppGpp to BsRelQ does only slightly elevate V max of (p)ppGpp synthesis, the K m values for the substrates GDP and GTP decrease dramatically (Figs 4f and S5). Also, BsRelQ displays a less cooperative behaviour indicated by a loss of the sigmoidal shape of the v/S characteristic when pppGpp is present. It therefor appears to us that the apparent cooperativity of BsRelQ rather originates from pppGpp produced during the enzymatic reaction rather than from a positive cooperativity between the four active sites of BsRelQ (compare to Fig. 2a and ref. 29 ). Noteworthy, at the highest concentration of pppGpp tested (i.e. 250 µM), the enzyme kinetic behavior of BsRelQ is highly similar to BsRelQ-H111P and SaRelP.
These results show that allosteric binding of pppGpp causes structural rearrangements of BsRelQ that are translated into an increased (p)ppGpp synthetase activity via an induced structural rigidity of the G-Loop.

Discussion
Two small alarmone synthetases (i.e. RelP/SAS2 and RelQ/SAS1) are typically found together in members of the Firmicutes phylum e.g. B. subtilis, S. aureus or L. monocytogenes 20 . RelP and RelQ share similarities of ~50 percent on the amino acid sequence level. Our structural analysis shows that RelP and RelQ possess a highly similar (p)ppGpp synthetase domain and both establish highly similar homotetrameric complexes (Fig. 1b and c). Nevertheless, both enzymes decisively differ in their ability to produce (p)ppGpp in that RelP is much more active than RelQ (Fig. 2a). Why is that the case? Our analysis demonstrates that binding of ATP proceeds in identical fashion in RelP/RelQ proteins, because both proteins harbor an identical architecture of their ATP-coordination site (Fig. S3). However, RelP and RelQ inherently differ in their ability to coordinate the GDP and GTP substrates. This is caused by a different structural flexibility of their G-Loops. While the G-loop of RelQ is highly disordered, the equivalent region of RelP is highly ordered and can therefore readily coordinate GDP/GTP (Fig. 3). However, the activity of RelQ can be enhanced by coordination of pppGpp within the central cleft 33 . This pppGpp results in a rearrangement of helices α4 and α5 at the lateral sides of the RelQ homotetramer and, by establishing a salt bridge between Glu178 (α5) and Arg117 (G-Loop) (Fig. 4), results in a more ordered (and active) conformation of the G-Loop. The (p)ppGpp synthetase activity of the so-stimulated RelQ resembles RelP. Notably, the K m values obtained for SaRelP ( Fig. 2a and b) and allosterically stimulated BsRelQ (Figs S4 and S5) accord with the intracellular concentrations of GDP and GTP, estimated as 200-500 µM and 1-5 mM, respectively 34,35 . Under these conditions, both enzymes are highly sensitive to small changes in GDP/GTP levels. Non-stimulated BsRelQ, in contrast, appears rather insensitive to changes in GDP/GTP levels because of its high K m values for both substrates (Fig. 2a and b). In summary, RelP always appears as a highly active alarmone synthetase, while  RelQ can switch between a passive state with low and an active (i.e. pppGpp-stimulated) state with high (p)ppGpp synthetase activity.
Having elucidated the different properties of RelP and RelQ, we wondered how this divergence might be relevant for the bacterial cell. In our current understanding, RelQ can appear in two passive states. In the apo-state, RelQ's central cleft is unoccupied while in the RNA-bound state a so far uncharacterized RNA 29,36 might reside in the central cleft (Fig. 5). We suspect that RelQ is predominantly found in either of those passive states in nutrient-rich conditions, because the (p)ppGpp hydrolytic activity of Rel should keep (p)ppGpp levels below the limit of RelQ stimulation. When the microorganism is suddenly confronted with nutrient limitation, Rel will recognize and bind to stalled ribosomes. When doing so, Rel could provide the pppGpp needed to bring RelQ into its active (i.e. pppGpp-bound) state by the intricate mechanism involving helical rearrangements and loop stabilization (Fig. 5). RelQ would then simply serves as an amplifier of the stress signal given by Rel. Additionally, the RNA bound to RelQ would be outcompeted by pppGpp and might result in the transcription of stress genes. Unfortunately, it is unclear so far, which genes might be differentially regulated, as the 'real' RNA bound by RelQ in vivo still remains to be identified 29,36 . Seemingly, RelQ's activity is intensively coupled to Rel (Fig. 5). Although experimental data for this functional link of Rel and RelQ are missing so far, the outlined scenario would provide an elegant way for an instant rise of (p)ppGpp levels dominated by the activity Rel and aided by RelQ.
RelP, in contrast to RelQ, is always a highly active enzyme that possesses all features enabling efficient (p) ppGpp synthesis, mainly an ordered G-Loop (Fig. 5). RelP should therefore not rely on the signal provided by Rel but might rather work independently. The presence of a central cleft within the tetramer of RelP nevertheless allows hypothesizing that an unknown factor might regulate the activity of RelP (Fig. 5). Noteworthy, the different activities of RelP and RelQ seem to be perfectly matched with their disparate transcriptional profiles. The switchable RelQ, predominantly transcribed during logarithmic growth 27 , can counteract a sudden nutrient limitation with the help of the Rel protein. The presence of RelP during logarithmic growth, however, might be detrimental for the microorganism. Consequently, RelP transcripts appear only during early stationary phase and in response to treatment with antibiotics, ethanol, high salt and acidic or alkalic pH stress conditions 25,37,38 . Also, RelP has been implicated in mediating inactivation of ribosomes by forming translation-inactive ribosome dimers thereby providing an elegant and fast shutdown mechanism for the bacterial metabolism 32,39 . In conclusion, our study strengthens the understanding of disparate roles of RelP/RelQ proteins and sets the stage for future investigations on this class of (p)ppGpp synthetases.

Materials and Methods
Cloning and mutagenesis. Genes encoding for RelP (ywaC and SA2297, respectively) were amplified from B. subtilis PY79 and S. aureus strain Newman genomic DNA by polymerase chain reaction using Phusion High-Fidelity DNA polymerase (NEB) according to the manufacturer's manual. The forward primer for SA2297 encoded a hexahistidine-tag in frame with the DNA sequence of relP. The forward primer for ywaC encoded a strep-tag in frame with the DNA sequence. The resulting PCR fragments were cloned into the pET24d(+) vector (Novagen) at the NcoI/XhoI restriction sites. Mutations within RelP were generated by overlapping PCR. BsRelP was purified by a similar procedure using a 1-ml StrepTrap column (GE Healthcare). Lysis buffer without imidazole was employed for cell lysis, column equilibration and washing and elution from the column was conducted with 5 CV of SEC buffer containing 2.5 mM desthiobiotin.

Protein Production and
Preparation of ppGpp and pppGpp. (p)ppGpp was produced essentially as described previously 33 . In brief, 5 µM SAS1 were incubated in SEC buffer together with 10 mM ATP and 10 mM GDP for 30 min at 37 °C to produce ppGpp or together with 10 mM ATP and 10 mM GTP for 2 h at 37 °C to produce pppGpp. Afterwards, the reaction was mixed with the same volume of chloroform and centrifuged (17300 × g, 5 min, 4 °C). The aqueous phase was removed and the organic phase mixed with one volume of double-destilled water and centrifuged (17300 × g, 5 min, 4 °C). The combined aqueous phases were subjected to anion-exchange chromatography using a ResourceQ. 6-ml column (GE Healthcare) at a flow rate of 6 ml/min and the nucleotides eluted with a gradient of NaCl. Fractions containing ppGpp or pppGpp were pooled followed by addition of lithium chloride with a concentration of 1 M and four volumes of ethanol. The suspension was then incubated at −20 °C for 20 min and centrifuged (5000 × g, 20 min, 4 °C). The resulting pellets were washed with absolute ethanol, dried and stored at −20 °C. Quality of the so-prepared alarmones was controlled by HPLC and yielded ppGpp and pppGpp in purities of 98% and 95%, respectively.
Kinetic analysis of RelP/RelQ. The enzyme kinetic behavior of RelP and RelQ (compare to Figs 2a, 3e, 4f and S5), were monitored by HPLC. Reactions were prepared in SEC buffer supplemented with 100 mM HEPES-Na pH 7.5 by incubating 0.2 µM protein together with 5 mM ATP and varying concentrations of GDP or GTP (i.e. 0.05, 0.1, 0.2, 0.3, 0.5, 1, 3 and 5 mM; 2 and 4 mM were included where necessary). For the analysis of pppGpp affecting the kinetic behavior of BsRelQ, pppGpp was also added to the reaction in concentrations of 0/2.5/10/25/100/250 µM. Samples were taken after different time points (i.e. 2, 4, 6, 8 and 10 minutes) and stopped as follows: two volume parts of chloroform were added to the sample, thoroughly mixed for 15 seconds, kept at 95 °C for 15 seconds and flash-frozen in liquid nitrogen. While thawing, the samples were centrifuged (17300 × g, 30 min, 4 °C) and the aqueous phase used for analysis. HPLC measurements were conducted on an Agilent 1100 Series system (Agilent technologies) equipped with a C18 column (EC 250/4.6 Nucleodur HTec 3 µM; Macherey-Nagel). Nucleotides were eluted isocratically with a buffer containing 50 mM KH 2 PO 4 , 50 mM K 2 HPO 4 , 10 mM TPAB (tetrapentylammonium bromide) and 20% (v/v) acetonitrile and detected at 260 nm wavelength in agreement with standards. Analysis of enzymatic measurements was performed with GraphPad Prism version 6.04 for Windows, (GraphPad Software, San Diego, California, USA). The velocity of (p)ppGpp synthesis was obtained by linear regression of the amount of AMP quantified after different incubation times. Kinetic parameters (K m , V max and the Hill coefficient (h) ± standard deviation) were obtained from the fit of the v/S characteristic according to the equation v = V max S h /(K m h + S h ).

Stimulation of RelP/RelQ by (p)ppGpp.
In experiments probing the stimulatory effect of (p)ppGpp Crystallization and structure determination. Crystallization was carried out at room temperature by sitting drop vapor diffusion in SWISSCI MRC 2-well plates (Jena Bioscience) with a reservoir volume of 50 µl and the drop containing 0.5 µl of protein and crystallization solution each. Crystals of BsRelP were obtained from a 10 mg/ml solution after 1 week from 0.1 M CHES pH 9.5 and 30% (w/v) PEG 3000. Crystals of SaRelP were obtained from a 15 mg/ml solution after 1 week in 0.1 M CHES pH 9.5 and 40% (v/v) PEG600. For crystallization of SaRelP-AMPCPP, a 15 mg/ml concentrated protein solution was incubated together with 5 mM AMPCPP for 30 minutes on ice. Crystals of SaRelP-AMPCPP were obtained after 2 days from 0.1 M Tris pH 8.5, 0.2 M lithium sulfate and 30% (w/v) PEG4000.
To harvest crystals, 0.5 µl of a cryo-protecting solution containing mother liquor supplemented with 20% (v/v) glycerol was added to the drop, crystals looped and flash-frozen in liquid nitrogen. Diffraction data were collected at the European Synchrotron Radiation Facility (ESRF) Grenoble, France, at beamlines ID23-1 and ID29 under laminar nitrogen flow at 100 K (Oxford Cryostream 700 Series) with a DECTRIS PILATUS 6M detector. Data were processed with XDS 40 and CCP4-implemented SCALA 41 . Crystal structures were determined by molecular replacement (MR) employing BsRelQ (PDB: 5DEC 33 ) as search model using the CCP4-implemented PHASER 41 . Structures were manually built in COOT 42 and refined with PHENIX 43 . Figures were prepared with PYMOL (www.pymol.org).