Dual regulation of activity and intracellular localization of the PASTA kinase PrkC during Bacillus subtilis growth

The activity of the PrkC protein kinase is regulated in a sophisticated manner in Bacillus subtilis cells. In spores, in the presence of muropeptides, PrkC stimulates dormancy exit. The extracellular region containing PASTA domains binds peptidoglycan fragments to probably enhance the intracellular kinase activity. During exponential growth, the cell division protein GpsB interacts with the intracellular domain of PrkC to stimulate its activity. In this paper, we have reinvestigated the regulation of PrkC during exponential and stationary phases. We observed that, during exponential growth, neither its septal localization nor its activity are influenced by the addition of peptidoglycan fragments or by the deletion of one or all PASTA domains. However, Dynamic Light Scattering experiments suggest that peptidoglycan fragments bind specifically to PrkC and induce its oligomerization. In addition, during stationary phase, PrkC appeared evenly distributed in the cell wall and the deletion of one or all PASTA domains led to a non-activated kinase. We conclude that PrkC activation is not as straightforward as previously suggested and that regulation of its kinase activity via the PASTA domains and peptidoglycan fragments binding occurs when PrkC is not concentrated to the bacterial septum, but all over the cell wall in non-dividing bacillus cells.


Results and Discussion
PASTA domains are not necessary for septal localization of PrkC. The extracellular domain has been shown to be required for PknB proper localization at mid-cell and at the poles in M. tuberculosis 25 , and for StkP localization at the septum in S. pneumoniae 2,27 . To test if the PASTA repeats are also required for PrkC localization at mid-cell and at the poles in B. subtilis growing cells 20 , we constructed several PrkC truncated proteins fused to GFP (Fig. 1). These proteins were deleted of one, two or three PASTA repeats as well as other deletions of the cytoplasmic domain or of the entire external domain. We analyzed their localization by fluorescence microscopy and noticed that all the truncated forms of PrkC were properly situated at the septum or at the poles of the cell. The only exception was the catalytic domain alone (GFP-PrkCc) that showed fluorescence throughout the cytoplasm of the bacteria (Fig. 1E). The deletion of one or two PASTA repeats or of all the external domain of PrkC had no effect on its localization (Fig. 1B,C,D). Even the deletion of the catalytic domain, alone or coupled with the deletion of the external domain, had no effect ( Fig. 1F and G), as long as the TM domain is present. Since one of the PASTA domain properties is its ability to bind muropeptide fragments of the cell wall, we prepared some PG fragments from an exponential growing cell culture of B. subtilis and tested their addition to the cell culture producing GFP-PrkC at the concentration range known to stimulate spore germination as previously described 19 . In agreement with the former observations, this supplement had no effect on the localization of the protein ( Fig. 1A and A'). The deletions of PASTA domains or their interaction with PG fragments have therefore no consequences on the localization of PrkC. All these data suggest that its septal and polar localization in B. subtilis cells is only mediated by the TM domain. This could be due to the TM domain itself or to its interactions with membrane proteins located at the division sites or at the poles like proteins of the divisome. Some of them may be involved in the localization of PrkC via an interaction of their transmembrane domains. We can already exclude the GpsB, EzrA and DivIVA cell division proteins whose deletion has no effect on PrkC localization 20 . Some proteins recycling the cell wall like hydrolases or PBPs 28,29 may be potential candidates.
The deletion of the PASTA domains has no effect on the kinase activity of PrkC in vitro. The catalytic domain of PrkC 30 alone possesses an efficient kinase activity in vitro 17,23,31 . We thus wanted to test if the deletion of the PASTA repeats could have an effect on the enzymatic activity of PrkC in vitro. For this purpose, we constructed and then purified some truncated forms of PrkC deleted of one or two PASTAs or of the entire extracellular domain (Fig. 2A). The production and purification yield was low depending on the extracellular region  (Fig. 2B); however, we used these preparations to test the kinase activity of the corresponding proteins in vitro with radiolabeled ATP. Since GpsB has been shown to stimulate PrkC 20 , the experiments were also carried out in the presence of GpsB. We observed that the autophosphorylation of PrkC as well as its kinase activity were not significantly modified by the deletions of the PASTA domains (Fig. 2C, odd numbered lanes and Fig. S1). In all cases, the activity was stimulated by GpsB (Fig. 2C, even numbered lanes). These data clearly show that the The binding of muropeptides to PrkC doesn't modify its kinase activity in vitro. It has been shown by different biochemical approaches that PrkC PASTA repeats and homologues from other species, Stk1, PknB or StkP, are all able to bind muropeptides in vitro 10,11,15,19,25,32 . However, these studies were carried out with the extracellular domain alone, in the absence of the catalytic domain, and the effect of this binding on the kinase activity has never been tested so far in vitro. In addition, for PrkC from B. subtilis, a specific interaction of DAP-type-tetrapeptide was proposed with the PASTA3 by NMR 11 . To test the effect of muropeptide binding on the kinase activity of PrkC, we first checked by limited proteolysis the ability of our purified full-length PrkC protein to bind a purified PG fragments. The analysis of constituent muropeptides of the vegetative cell wall peptidoglycan of B. subtilis show that they are mostly composed of glycan strands ending with MurNAc units in the anhydro form and of peptide chains containing three or four amino acids with a clear preference for DAP-amidated-type 33 . We then decided to use the disaccharide tetrapeptide GlcNAc-MurNAc-L-Ala-γ-D-Glu-meso-DAP-D-Ala (or TCT) 34 in our tests (Fig. 3A). As purified muropeptides are usually used at the micromolar range in the previously cited biochemical studies, we decided to test TCT in a scale from 50 to 250 µM. We observed that the digestion profile of PrkC was modified by the addition of TCT suggesting a conformational change due to TCT binding to the protein. Indeed, we observed a decrease in the intensity of a specific protein band (see arrow) with the addition of increasing amounts of TCT ( Fig. 3A left gel). This observation could not be made with the PrkCc protein used as negative control (Fig. 3A right gel). These data confirmed that the binding of muropeptides to PrkC protein is dependent of its PASTA domains. This interaction was shown here for the first time by analyzing the entire protein and not only its extracellular domain. The amount of TCT necessary to visualize its binding to PrkC is in agreement with an affinity at the micromolar range for this muropeptide. Furthermore, we tested the effect of this binding on the kinase activity of the same preparation of PrkC using the same amounts of TCT. The catalytic domain PrkCc that is unable to bind the muropeptide was used as negative control. As shown in the autoradiogram, the binding of TCT to PrkC has no effect on its kinase activity in vitro (Fig. 3B). These observations were made in vitro but clearly run counter to the current activation model of PrkC and we decided to test it in vivo.
Addition of PG fragments to full-length PrkC induces its oligomerization. The current activation model of PrkC is based on the observation that B. subtilis spores of a prkC mutant are unable to germinate in response to PG fragments compared to a WT strain 19 , combined to biochemical data showing the interaction between the PASTA3 domain and muropeptides in vitro 11 . However, a clear evidence of a direct muropeptide activation of PrkC in vivo is still missing. We therefore decided to measure the autophosphorylation of PrkC full-length or of truncated forms of PrkC in vivo from cells cultivated in the absence and then in the presence of increasing concentrations of PG fragments. Being limited in the amount of TCT available, we decided to use PG fragments prepared from a B. subtilis cell culture stopped at the exponential growth phase and at the concentration range known to stimulate spore germination as previously described 19 . In order to make sure that our PG preparation was containing enough muropeptides able to interact with PrkC, we checked this binding by Dynamic Light Scattering (DLS). We also included the mutant protein PrkC(R500A) as a negative control in the experiment since this mutation, located in the PASTA3, was shown to abolish PG fragments binding. The effect of the increasing amount of PG fragments on the hydrodynamic size of PrkC, PrkC(R500A) and PrkCc respectively was determined by measuring the intensity of the light diffused by the molecules and their translational speed in solution as presented in the correlograms (  PG. However, the analysis of correlograms, Z-averages ( Fig. 4B) and diameters by intensity distribution median (Di50) (Fig. S2) suggested a difference between the proteins. All three showed differences in their correlograms in the exponential decay and fluctuating times (Fig. 4A). When PG fragments were added to PrkC, we observed a concentration-dependent increase of the fluctuation time and the broadening of the decay was slow showing a progressive increase in polydispersity suggesting a specific binding of PG fragments to PrkC leading to the oligomerization of the protein. When PG fragments were added to PrkC(R500A), the concentration-dependent increase of the fluctuation time was slower and the broadening of the decay was narrower showing that specific binding of PG fragments to PrkC(R500A) was less efficient and consequently the oligomerization too. Conversely, for the catalytic domain, the addition of PG fragments increased the time at which the correlation begins to decay with a broadening of the fluctuation time suggesting an aggregation of PrkCc by PG fragments and high polydispersity thus a nonspecific binding. These results were confirmed by the Di50 (Fig. S2) and Z-average values (Fig. 4B) that increased constantly and to less than 100 nm with increasing concentrations of PG fragments for PrkC. However, for PrkCc, the Z-average increased beyond 100 nm in the presence of 0.15 µg/ml of PG fragments and continued to increase at 1.5 µg/ml. Small variations of the Z-average values (Fig. 4B) and no variation of Di50 values (Fig. S2) for PrkC(R500A) confirmed that PG fragments binding was impaired on this protein. We also measured the denaturation temperatures for both proteins with or without PG fragments (data not shown). The denaturation temperature of PrkC was identical with or without PG fragments (50 °C) suggesting a good stability of the protein that was not modified by muropeptides binding. On the contrary, the denaturation temperature of PrkCc dropped from 70 °C to 50 °C with the addition of the PG fragments confirming a destabilizing effect of muropeptides on the catalytic domain. Altogether, these results suggest that PG fragments interact specifically with PrkC to induce its oligomerization whereas it interacts non-specifically with PrkCc to induce its aggregation. Deletion of PASTA domains or addition of PG fragments has no effect on PrkC activity during exponential phase. Since PG fragments bind to PASTA domains, we first analyzed the effect of PASTA deletions on PrkC activity in B. subtilis growing cells. For this purpose, strains expressing the truncated-PrkC-GFP proteins were grown in a rich medium until the mid-exponential phase (OD 600 = 0.8) and crude extracts were prepared and analyzed by western blots. Using antibodies against the GFP protein, we first checked that all the forms of PrkC were produced at the same level in the cell (Fig. 5A). We then measured the autophosphorylation of PrkC in vivo using antibodies against phospho-threonine (P-Thr). We observed that all the forms of PrkC were phosphorylated except the extracellular domain of PrkC (GFP-TM-P1P2-Ig-like) lacking the catalytic intracellular kinase domain that served as the negative control. These data clearly show that all the forms of PrkC containing the catalytic domain are active. However, the absence of the PASTA domains has no effect on PrkC autophophorylation and therefore on its kinase activity in vivo. This observation has also been made recently for the homologous protein IreK from E. faecalis but to a lesser extent 24 . Indeed, a truncated form of IreK lacking its five PASTA repeats was still active indicating that the extracellular domain of IreK is not required for an answer to a signal associated with growth and/or cell division. However, the specific stimulation of IreK kinase activity in response to cell-wall antimicrobials seemed to be dependent of the PASTA module, which suggests multiple parameters for sensory input in vivo 24 .
Since our preparation of PG fragments contains muropeptides able to bind to PrkC, we decided to test it on PrkC autophosphorylation in vivo. For this, strains producing GFP-PrkC or GFP-PrkCc as negative control were grown in a rich medium in the presence of increasing amount of PG fragments until the mid-exponential phase (OD 600 = 0.8). Then crude extracts were prepared and analyzed by western blots. Using antibodies against PrkC, we observed that both proteins were produced at the same level whatever the amount of PG fragments added (Fig. 5B). In addition, the phosphorylation signal detected in all lanes with antibodies against P-Thr showed that the level of PrkC autophosphorylation was similar (Fig. 5B)  PASTA domains are necessary for PrkC activation during stationary phase growth. Since we did not detect any effect of PASTAs deletions or addition of PG fragments during exponential phase, we decided to investigate the localization and the activity of the protein in stationary phase cells. We first analyzed the localization by fluorescence microscopy of GFP-PrkC as well as of the truncated forms of PrkC deleted of one or several PASTA repeats. Unexpectedly, we noticed that all the forms of PrkC were distributed all over the cell wall (Fig. 6A). These observations revealed, for the first time, that the localization of PrkC varies according to the growth phase (Fig. S3). However, the PASTA domains are not necessary for this new positioning of the kinase. Thus, PrkC is concentrated at the septum and at the poles during cell division and delocalizes throughout the cell wall when the cells enter into a non-dividing state, but in both cases, the external domain is not required for PrkC cellular localization. We then wanted to test if a role of the PASTA domains would be to stimulate the kinase activity by autophosphorylation in non-dividing cells. We thus prepared crude extracts from stationary phase cultures producing the PASTA-truncated forms of GFP-PrkC or the mutant protein GFP-PrkC(R500A) unable to bind PG fragments and to oligomerize. These crude extracts were then analyzed by western blots using antibodies against the GFP protein to check that all the forms of PrkC were produced at the same level in the cell and using antibodies against P-Thr to measure the autophosphorylation of PrkC in vivo ( Fig. 6B and C). Whereas all forms of PrkC were expressed at the same level, only the entire protein was still detected by the antibodies against P-Thr. The removal of the PASTA3 or the mutation of the Arg500 to Ala, known to interact with PG fragments, was were grown on LB medium with 0.5% xylose at 37 °C until OD 600 = 0.8. After centrifugation, the pellets were resuspended in 1/100 th volume of lysis buffer and treated as described in Materials and Methods. For each strain, 16 μl of crude extract were separated by SDS-PAGE. After blotting, phosphorylated PrkC was detected using antibodies directed against P-Thr residues. To estimate the relative quantity of PrkC in crude extracts, we used anti-GFP antibodies. (B) Strains SG278 (ΔprkC, amyE::gfp-prkC) and SG465 (ΔprkC, amyE::gfp-prkCc-TM) were grown on LB medium with 0.5% xylose and increasing amounts of PG fragments (0, 0.003, 0.015, 0.03, 0.3 and 3 µg/ml) at 37 °C until OD 600 = 0.8. After centrifugation, the pellets were resuspended in 1/100 th volume of lysis buffer and treated as described in Materials and Methods. For each strain, 16 μl of crude extract were separated by SDS-PAGE. After blotting, phosphorylated PrkC was detected using antibodies directed against P-Thr residues. To estimate the relative quantity of PrkC in crude extracts, we used a specific antibody against PrkC. Full-length blots are presented in the supplemental data. sufficient to lose the phosphorylation signal. These results clearly show that, during stationary phase growth, the binding of PG fragments to the extracellular domain of PrkC is required to allow the autophosphorylation of the protein and therefore to stimulate its kinase activity.
A sophisticated model for kinase activation of PrkC. Our data suggest that PrkC regulation is a complex mechanism that may vary depending on the physiological state of the cell, i.e. growing cells versus non-dividing cells (stationary phase cells or spores) and according to the ligand and the partners of the protein when it is involved in one cellular process or another. We therefore propose a new model for the modulation of PrkC kinase activity depending on its cellular localization, the growth phase and the cellular process regulated (Fig. 7). The current model described for germination conditions and which can also be proposed for stationary phase conditions is summarized in (Fig. 7 top panel) and the designed regulatory model for PrkC kinase activity during exponential growth is presented in (Fig. 7 lower panel). During stationary phase (or germination), PrkC is located all around the cell wall (or the spore membrane), PG fragments are available, and the cells (or spores) probably need to sense an environmental signal to ensure the best rate of cell survival (or induce the exit from dormancy). Thus, stimulation by PG fragments through the PASTA domains of the protein may be required to induce PrkC dimerization and thus kinase activation via its autophosphorylation. This layout could be similar in other sporulating bacteria in which a PASTA-containing protein kinase is involved in spore germination. During growth, PrkC is concentrated at mid-cell or at the poles and, in such conditions, it likely has no access to the extracellular medium and thus to free PG fragments. Actually, no stimulation of its kinase activity by addition of PG fragments has been detected in vivo but it is stimulated by interaction with the division protein GpsB 20 . Moreover, we did not detect any stimulation of the kinase activity by TCT in vitro, although it binds to PrkC. This could be explained by the strong probability for two protein molecules to be in contact in solution before the addition of TCT and to phosphorylate each other. A high percentage of the PrkC molecules are therefore already active. In vivo, ligands like lipid II or other periplasmic molecules may play a role in PrkC dimerization that can also be mediated by its recruitment, via its TM domain, by partners located at the septum and/or at the poles like proteins of the division machinery, PBPs and hydrolases or other membrane associated proteins. During cell division, PrkC molecules are thus focused at the poles and septa where they can phosphorylate each other to become active. Furthermore, GpsB and to a lesser extent EzrA and DivIVA, increase PrkC kinase activity via a yet unknown mechanism. In these conditions, the stimulation by PG fragments may not be necessary. Our results are consistent with the regulation observed in E. faecalis cells, where IreK can respond to cell wall stress by enhancing its kinase activity via its PASTA repeats and can also respond to signal associated with growth and/ or division independently of the presence of its extracellular PASTA-containing domain 24 . We can conclude that PASTA-kinases are subtly stimulated according to their cellular localization and the cellular processes they regulate. This fine-tune regulation may also differ between species.

Methods
Plasmids and strains constructions. Standard procedures for molecular cloning and cell transformation of B. subtilis and E. coli were used. All the strains and plasmids used in this study are listed in Table 1. Primers The scale bar for microscopy images represents 2 µm. (B) PrkC-phosphorylation analysis by western blots. The same strains SG278, SG467, SG466 and SG465 were grown on LB medium with 0.5% xylose at 37 °C for 23 hours (final OD 600~ 6). After centrifugation, the pellets were resuspended in 1/30 th volume of lysis buffer and treated as described in Materials and Methods. For each strain, 16 μl of crude extract were separated by SDS-PAGE. After blotting, phosphorylated PrkC was detected using antibodies directed against P-Thr residues. To estimate the relative quantity of PrkC in crude extracts, we used anti-GFP antibodies. (C) PrkC(R500A)-phosphorylation analysis by western blots. The strains SG278 and SG622 were grown on LB medium with 0.5% xylose at 37 °C for 23 hours (final OD 600~6 ) and the culture pellets were treated as mentioned in (B). Full-length blots are presented in the supplemental data. used in this study are available upon request. Sequencing of PCR-derived DNA fragments in the plasmid constructs was carried out by GATC-Biotech to ensure error-free amplification.
The truncated or mutated versions of PrkC were overexpressed in E. coli, with the corresponding genes amplified by PCR using B. subtilis 168 genomic DNA as a template and a primer pair containing SacI and PstI restriction sites. The amplified products were digested with SacI and PstI and then ligated to the pETDuet vector. The resulting plasmids are listed in Table 1.
PG preparation and purification of TCT muropeptide. B. subtilis PG fragments from growing cells were prepared as previously described in 19 . The pellet containing the cell wall PG freed of proteins and lipoteichoic acids was quantified by weighing then resuspended and digested with mutanolysin (10 µg/ml) overnight at 37 °C prior to inactivation of the enzyme at 80 °C for 20 min. The concentration of PG fragments was determined by quantitative aminoacid (diaminopimelic acid) and aminosugar (muramic acid, glucosamine) analysis with a Hitachi L8800 analyzer (ScienceTec, Les Ulis, France) after hydrolysis of samples in 6 M HCl at 95 °C for 16 h. PG fragments were then used in the several tests. TCT muropeptide was produced and purified as described in 36 .  mean size of the sample, smaller samples fluctuate quicker than larger samples is solution. The steeper the exponential decay, the more monodisperse (single population) the sample, and if the decay is extended the greater the sample polydispersity (several populations). Size distribution assays were performed at 25 °C. For each assay three measurements were performed; each one consisting in 10-15 runs of 10 seconds. The scattering angle was 173°. Temperature trend assay to calculate aggregation points contained a sequence from 20 to 70 °C with 10 °C intervals. We displayed our results using the correlograms, Z-average (mean intensity size of sample) and Di50 (diameter by intensity of 50% of the molecules in solution) using the Zetaziser software 7.12. PrkC and PrkC(R500A) were prepared at 0.2 mg/ml in 20 mM Tris-HCl pH 7.5, 200 mM NaCl, 7% glycerol following centrifugation for 15 min at 14000 rpm at 6 °C. PrkCc was prepared at 0.1 mg/ml in 20 mM Tris-HCl pH 7.5, 200 mM NaCl, 7% glycerol following centrifugation for 15 min at 14000 rpm at 6 °C. The proteins were analyzed in the absence or presence of increasing concentrations of PG fragments from 0.15-1.6 µg/ml.

Microscopy.
Strains were grown on LB medium at 37 °C. The gfp-prkC gene fusion and all truncated versions of prkC were expressed from the inducible P xyl promoter in the presence of 0.5% xylose. The PrkC localization was analyzed by fluorescent microscopy on a Zeiss Upright Axio Imager M2 microscope as described previously 3 .
Western Blot. The cells were grown at 37 °C in 30 ml of LB medium to OD 600 = 0.8 for exponential phase extracts and to OD 600~6 for stationary phase extracts (23-hour cultures) then centrifuged for 10 min at 8000 rpm at 4 °C. When needed, the exponential phase cultures were realized in the presence of increasing amounts of PG fragments (0, 0.003, 0.015, 0.03, 0.3 and 3 µg/ml). Cell pellets were resuspended in 1/100 th and 1/30 th volumes of lysis buffer for exponential and stationary phase cultures, respectively, containing 10 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.1% NP40, 1 mM PMSF, 1 mM DTT, 25 U.ml −1 benzonase and 10 mg.ml −1 lysozyme, and incubated for 30 min at 37 °C. 1/10 th volume of 10% SDS and 1/2 volume of (2×) Laemmli buffer were added to the extracts that were heated at 100 °C for 10 min. Samples were run on a 10% or 12.5% SDS-PAGE and transferred to hybond-ECL membrane by electroblotting. The membrane was blocked with PBS solution containing 5% milk powder (w/v), for 3 hours at room temperature with shaking, then incubated with anti-GFP or anti-PrkC antibodies diluted to 1/10000 th or 1/1000 th , respectively, overnight at 4 °C. After three washes, the secondary antibody, a peroxidase-conjugated Goat anti-Rabbit (DAKO) antibody, was used at 1/10000 th dilution for one hour. After three washes, the membrane was incubated with ECL reagents (Perkinelmer) and scanned for chimioluminescence with an ImageQuant LAS4000 (GE Healthcare). A second membrane was used for anti-P-Thr antibodies as previously described in 20 . Protein purification. Plasmids overproducing 6His-tagged PrkC truncated proteins were used to transform E. coli C41(DE3). Purification of 6His-tagged recombinant proteins was performed with Ni 2+ -NTA resin (Qiagen) as previously described in 18 for PrkCc, the PASTA-truncated forms of PrkC, PrkC and PrkC(R500A). Before purification, in order to solubilize the proteins containing a TM domain, 0.4% Triton X100 was added to the crude extracts, shacked for 1 h at 4 °C then centrifuged at 35000 rpm for 1 h at 4 °C to removed membrane fragments. In addition, the buffer used for the purification steps of these proteins contained 0.2% Triton X100.
Protein phosphorylation. 2 μg of GpsB were incubated for 15 min at 37 °C with 2.5 μg of the several forms of PrkC protein in a 15 μl reaction mixture containing 10 mM HEPES, pH 8.0, 1 mM MgCl 2 , 2 μg of MBP (Myelin Basic Protein) that was shown to be phosphorylated by PrkC 18 and 1 mM [γ-33 P] ATP (1 μCi). The phosphorylation reaction was stopped by adding 5× SDS-sample buffer to the reaction mixtures before SDS-PAGE analysis. Gels were then dried and exposed to autoradiography. When muropeptides were tested, increasing amounts (from 0 to 250 µM) of TCT were added in the reaction mixture containing 2.5 μg of the PrkC or PrkCc protein, 10 mM HEPES, pH 8.0, 1 mM MgCl 2 , 2 μg of MBP and 1 mM [γ-33 P] ATP (1 μCi).

Limited proteolysis.
For each 20 μl sample, 3 μg of PrkC or PrkCc proteins were pre-incubated for 10 min at 37 °C with 40 mM NaCl, 1 mM MgCl 2 , 10 mM Tris-HCl, pH 8.0 in the absence or presence of increasing amounts of TCT muropeptide (50, 100, 150 and 250 µM). After addition of 0.01 μg of trypsin (Promega), the reaction mixture was incubated for 10 min at 37 °C. The digestion was stopped by adding an equal volume of electrophoresis loading buffer to the assay mixtures and by heating 5 min at 100 °C before applying the samples onto a 12.5% or 15% SDS-PAGE. The gels were colored with Coomassie blue then scanned.
Data and materials availability. Data and materials will be made available upon request.