HPLC-based quantification of bacterial housekeeping nucleotides and alarmone messengers ppGpp and pppGpp

Here we describe an HPLC-based method to quantify bacterial housekeeping nucleotides and the signaling messengers ppGpp and pppGpp. We have replicated and tested several previously reported HPLC-based approaches and assembled a method that can process 50 samples in three days, thus making kinetically resolved experiments feasible. The method combines cell harvesting by rapid filtration, followed by acid extraction, freeze-drying with chromatographic separation. We use a combination of C18 IPRP-HPLC (GMP unresolved and co-migrating with IMP; GDP and GTP; AMP, ADP and ATP; CTP; UTP) and SAX-HPLC in isocratic mode (ppGpp and pppGpp) with UV detection. The approach is applicable to bacteria without the requirement of metabolic labelling with 32P-labelled radioactive precursors. We applied our method to quantify nucleotide pools in Escherichia coli BW25113 K12-strain both throughout the growth curve and during acute stringent response induced by mupirocin. While ppGpp and pppGpp levels vary drastically (40- and ≥8-fold, respectively) these changes are decoupled from the quotients of the housekeeping pool and guanosine and adenosine housekeeping nucleotides: NTP/NDP/NMP ratio remains stable at 6/1/0.3 during both normal batch culture growth and upon acute amino acid starvation.


SUPPLEMENTARY TEXT Introduction
We report here a workflow we have applied successfully, besides Escherichia coli to cultures of Bacillus subtilis, Acinetobacter baumannii, Pectobacterium wasabiae and We took into account the comprehensive knowledge from the literature and chose HPLC-UV analysis due to its relative simplicity, availability, ability to detect rich set of nucleotides and applicability to analysis of non-growing bacterial cultures or the ones grown in complex growth media.
The workflow from bacterial culture to quantitative estimates of nucleotide concentration in the cell can be divided into several steps (Figure 1): (i) sample acquisition, (ii) extraction, and (iii) quantification. Below we provide an overview of existing techniques and associated pitfalls.

Sample acquisition
It is essential that cellular metabolism must be rapidly quenched during the preparation of a nucleotide extract, thus 'freezing' the biologically relevant status quo. This can be achieved either in the course of sample acquisition and extraction or, alternatively, during a dedicated quenching step. Sample acquisition is performed either by separating cells from culture medium or by sampling whole culture broth. Separation of cells from culture medium, i.e. harvesting, can be performed either by filtration or centrifugation and results in a stronger signal during the detection step due to the reduced sample size leading to an increased concentration of metabolites. However, addition of a relatively slow harvesting step poses a challenge for the detection of high flux metabolites, such as nucleotides in a rapidly growing Escherichia coli culture (Cole, Wimpenny et al. 1967, Chapman, Fall et al. 1971, Lundin and Thore 1975, Payne and Ames 1982, Bolten, Kiefer et al. 2007, Ishii, Nakahigashi et al. 2007, Buckstein, He et al. 2008, Bennett, Kimball et al. 2009). To overcome this problem, the Rabinowitz lab has developed an approach utilizing cultivation of bacteria on a nitrocellulose filters on top of the agarose plates, so that the filters can be rapidly transferred to extraction solution (Brauer, Yuan et al. 2006).
Centrifugation is not applicable for bacterial nucleotide measurements since it dramatically alters the nucleotide levels, i.e. highly phosphorylated compounds are converted to less phosphorylated species (Cole, Wimpenny et al. 1967, Chapman, Fall et al. 1971, Lowry, Carter et al. 1971, Lundin and Thore 1975, Payne and Ames 1982, Buckstein, He et al. 2008. Rapid vacuum filtration, on the other hand, gives reliable results for B. subtilis (Ochi, Kandala et al. 1981) and E. coli (Franzen and Binkley 1961, Nazar, Lawford et al. 1970, Bagnara and Finch 1972, Walker-Simmons and Atkinson 1977, Payne and Ames 1982. Washing, if necessary, is a critical step during filtration since a difference in ionic strength of a washing solution and of the growth medium can dramatically perturb the results (Bolten, Kiefer et al. 2007). Addition of a dedicated quenching step prior to harvesting is commonly used to counter the abovementioned challenges. The most frequent quenchers are cold 60% methanol in an aqueous buffer (Buchholz, Takors et al. 2001, Buchholz, Hurlebaus et al. 2002, Bolten, Kiefer et al. 2007, Hiller, Franco-Lara et al. 2007, cold aqueous solution of 0.9% NaCl (Wittmann, Kromer et al. 2004), formaldehyde (Little and Bremer 1982) and cold glycerol (Wittmann, Kromer et al. 2004). However, some of these approaches can perturb the consequent nucleotide measurements: leakage of the cytoplasm from bacterial cells was observed while using methanol (Wittmann, Kromer et al. 2004, Bolten, Kiefer et al. 2007 or 0.9% NaCl (although it was suggested to be due to cold shock, the effect of centrifugation can not be ruled out) (Wittmann, Kromer et al. 2004); a combination of formaldehyde quenching with alkali extraction tends to give a weaker signal and introduces a lot of variation (for details and references, see next section on nucleotide extraction).
All of the above-mentioned problems with harvesting can be avoided altogether by opting for a whole-culture broth sampling. In this case, quenching can be done by snap-freezing the sample with liquid nitrogen (Dominguez, Rollin et al. 1998, Chassagnole, Noisommit-Rizzi et al. 2002 or rapidly boiling it (Schaub, Schiesling et al. 2006). A more usual approach is transferring the cell suspension into an extraction solute. There are, however, several disadvantages of the whole culture broth approach. First, it results in considerably more dilute solution of metabolites. This necessitates either a highly sensitive detection method, such as counting radioactivity after labeling with 32 P via phosphate or nucleotide precursors, or additional enrichment steps, such as separation, precipitation, evaporation and/or freeze-drying. Second, nucleotides in extracellular and intracellular material cannot be discriminated; performing an supplementary analysis of filtrate (Taymaz-Nikerel, de Mey et al. 2009) is a possible solution to this problem. An extreme example of this issue is quantification of nucleotide messenger cAMP which predominantly resides in the growth medium (Matin and Matin 1982). While 'housekeeping' nucleotides are not generally detected outside the cell (Lundin andThore 1975, Taymaz-Nikerel, de Mey et al. 2009), it has been reported that E. coli cultures might accumulate AMP together with enzymes that affect nucleotide stability in the growth medium during stationary phase (Chapman, Fall et al. 1971). Third, components of growth media can interfere with the nucleotide measurements: inorganic salts can interfere with nucleotide detection by liquid chromatography (LC) coupled to UV detector (Bhattacharya, Fuhrman et al. 1995) or to mass-spectrometer (Bolten, Kiefer et al. 2007); organic components can inhibit enzymatic assays (Lundin and Thore 1975).
At elevated temperatures, both enzymatic and chemical degradation of nucleotides are more likely. Cold extraction is therefore safer. It has its own drawbacks, however, a general and specific ones. Overarchingly, it does not necessarily completely stabilize the sample and nucleotides can still undergo a degradation even when kept at -10 to 4 °C (Holms, Hamilton et al. 1972, Lundin and Thore 1975, Rabinowitz 2007. This issue can be addressed by (1) chelation of Mg 2+ and other divalent ions-common cofactors for enzymes-by addition of EDTA (Lundin and Thore 1975) (however, this is likely to interfere with detection on HPLC, see our Supplemental Materials and Methods); (2) acidifying the solution (Rabinowitz and Kimball 2007); (3) acid precipitating the enzymes and removing precipitate, preferably before any neutralization (Lundin and Thore 1975); and (4) keeping the samples sufficiently cold i.e. working on ice and storing at -20 °C. Cold extraction is performed most commonly with chloroform (Coulier, Bas et al. 2006), ethanol (Cserjan-Puschmann, Kramer et al. 1999, Meyer, Liebeke et al. 2010, methanol (Meyer, Liebeke et al. 2010), acetonitrile (Au, Su et al. 1989), acidic acetonitrilemethanol-water (Rabinowitz and Kimball 2007), sodium formate at pH 3.4 (Cashel and Gallant 1968), and various acids (Franzen and Binkley 1961, Bagnara and Finch 1968, Nazar, Lawford et al. 1970, Fischer, Zimmerman et al. 1982. If organic extraction is a must, triphosphates are better extracted with acetonitrilemethanol-water than methanol-water mixtures (Rabinowitz and Kimball 2007). Most common is still the acidic extraction with perchloric (PCA) (Franzen and Binkley 1961), trichloroacetic (TCA) (Smith and Maaloe 1964), formic (Cashel andGallant 1969, Bochner and or acetic acid (Nazar, Lawford et al. 1970). The appeal of strong acids like TCA and PCA, compared to milder organic acids, lies in the fact that they are better at disruption of the cell envelope and quenching enzymatic activities: act fast and do not require freeze-thaw cycles to complete the extraction (Bagnara and Finch 1972). It is known, nevertheless, that some unidentified bacterial phosphatase activity can endure PCA (the activity can endure also boiling) and the best remedy is the addition of EDTA or removal of acid precipitate before neutralization (Lundin and Thore 1975). Yet strong acids, especially if not kept sufficiently cold, are more prone to break down highly phosphorylated species (Cashel and Gallant 1968, Nazar, Lawford et al. 1970, Au, Su et al. 1989) and accordingly, (p)ppGpp is not efficiently extracted from biological material with PCA and TCA (Cashel 1969). Cold formic acid extraction is therefore by far the most common solvent for (p)ppGpp extraction, even though it also has been claimed to introduce artifacts via ppGpp degradation to ppGp (Lagosky and Chang 1978) (This is part of the motivation why less used lysozyme (Lagosky and Chang 1978), and alkali extraction (for that, see discussion above) (Little and Bremer 1982) were devised for ppGpp quantification). Regardless of the acid of choice, prolonged incubation in acidic conditions during extraction or analysis (pH 4-6 will suffice if working at room temperature), results in degradation of NADPH and NADH (Lowry, Passonneau et al. 1961) and various resultant degradation products might interfere with detection and/or quantification (reduced forms, on the other hand, are stable in acid but not in alkaline conditions (Kaplan, Colowick et al. 1951)).
It is often desirable to get rid of the acid. Chemical lability is the major concern, phosphoanhydride bonds are stable in acidic conditions only if kept sufficiently cold.
There are three options: neutralization, extraction or evaporation. Perchloric acid has often been the acid of choice because of the ease of removal via neutralization: addition of KOH or K2CO3 will lead to precipitation of poorly water soluble KClO4. It has been estimated, however, that after neutralization and removal of insoluble KClO4, about 60 mM of salt remains in the solution (Pogolotti and Santi 1982). Accordingly, both PCA and TCA have been reported to interfere with downstream detection on HPLC (Au, Su et al. 1989, Gebelein, Merdes et al. 1992, Buchholz, Takors et al. 2001. It is likely that this interference can be surmounted using extraction of acid instead of neutralization (Khym 1975, Arezzo 1987. Amine-freon extraction separates the acid which forms a salt with amines and is partitioning into the freon phase, leaving the nucleotides in the aqueous phase (Khym 1975). TCA, soluble in ethyl ether, can be removed by repeated extraction with ether (Arezzo 1987). Instead of acid removal, nucleotides themselves could be extracted with either acid washed charcoal (Norit A) (Fiske 1934, Cabib, Leloir et al. 1953 or with ion exchange (Buckstein, He et al. 2008). Finally, volatile acids-such as TCA, formic, and acetic acid-can be removed by freeze-drying.
If the acid treatment was adequately strong-and/or acid precipitate removed before neutralization (see above)-to eliminate all enzymatic activities, neutralized aqueous nucleotide sample is often stable enough to be dried in centrifugal evaporator at room temperature (Khym 1975).

Nucleotide quantification
The last step is quantification of extracted nucleotides (Figure 1). This step is customarily based on liquid chromatographic separation. The alternatives to chromatography include enzymatic assays for adenosine nucleotides (Chapman, Fall et al. 1971, Schneider andGourse 2004) or dNTPs (Solter and Handschumacher 1969, Lindberg and Skoog 1970, Skoog 1970, and capillary electrophoresis (Soga, Ueno et al. 2002, Markuszewski, Britz-McKibbin et al. 2003. Enzymatic assay of dNTPs and capillary electrophoresis are not common for nucleotide detection. Luciferase assay of adenosine nucleotides, however, established the concept of adenylate energy charge (Atkinson 1968) and laid the foundations for all nucleotide quantifications that followed: energy charge is a crucial measure of sample acquisition, extraction and/or storage quality.
The most widely used chromatographic technique for nucleotide separation is thin layer chromatography (TLC) (Bochner and Ames 1982). TLC relies on isotope ( 32 P) labeling and is thus confined to actively growing cells and media where the concentration of phosphate can be readily manipulated to facilitate efficient uptake and labelling. Note that stringent response, for example, is known to interfere with the uptake of phosphate and thus incorporation of the radioactive label (Edlin and Neuhard 1967, Gallant and Cashel 1967, Cashel and Gallant 1968, Irr and Gallant 1969. High performance liquid chromatography (HPLC), however, is free of those limitations of labelling. HPLC can rely on photometric (Little andBremer 1982, Payne and or mass spectrometric detection (Buchholz, Takors et al. 2001, Soga, Ueno et al. 2002, Bennett, Yuan et al. 2008, Meyer, Liebeke et al. 2010). There are three well-established HPLC modes to separate nucleotide mixtures: ion exchange, reverse phase and ion-paired reverse phase. Ion exchange is the first LC approach developed for nucleotides (Cohn 1949), has gained a lot of popularity and works especially fine for highly charged nucleotides (tri-, tetra-and pentaphosphates). Reversed phase is suitable only for nucleosides and cyclic nucleotides (Payne and Ames 1982). To increase the retention of highly charged nucleotides on reversed phase column, a zwitterion, customarily tetrabutylammonium, is added to the mobile phase and the result is called ion-paired reverse phase (Hoffman and Liao 1977, Little and Bremer 1982, Payne and Ames 1982, Mack, Reed et al. 1985. Next, resolved nucleotides need to be quantified. UV absorbance is a feasible, less expensive and more often available, whereas MS excels in sensitivity and accurate identification. The highly charged nature of nucleotides, however, poses a challenge for analysis on LC-MS. First, high salt concentrations of strong anion exchange, necessary to elute charged nucleotides, are not easily compatible with electrospray ionization (ESI).
Second, nucleotides are not easily resolved on reverse phase, the most frequent and compatible LC mode for MS. Third, tetrabutylammonium salts of IPRP are not volatile and thus, again, not compatible with ESI. The best option therefore available is to use IPRP with either tributylamine (Luo, Groenke et al. 2007) or primary amine (Coulier, Bas et al. 2006) as a zwitterion. Hydrophilic interaction chromatography (HILIC) can also be attempted, preferably with an aminopropyl column (Bajad, Lu et al. 2006), but gives inferior separation as compared to IPRP.

Quality control
Relative levels of adenylate nucleotides  ATP, ADP and AMP  serve as a quality control for sampling, extraction and sample storage. ATP has an exceedingly high turnover rate with a half-life of around one-tenth of a second (Holms, Hamilton et al. 1972, Walsh andKoshland 1984) and therefore its levels report the efficiency of quenching enzymatic activity. ATP, same as other di-and triphosphates, can be further degraded chemically since phosphoanhydride bond is not stable in acid unless kept cold. One can readily spot the ATP degradation by assessing the 'adenylate energy charge' (AEC). AEC is calculated as = + 1 2 + + and it falls between 0.75-0.95 for rapidly growing cultures, be it bacterium E. coli (Chapman, Fall et al. 1971) or yeast Saccharomyces cerevisiae (Ball and Atkinson 1975).
A technical problem with AEC is that it requires concentrations of all the three adenylate species-ATP, ADP and AMP-and determining the relatively low levels of AMP can be challenging. Fortunately, a simpler parameter  ATP/ADP ratio  is nearly as informative as AEC; for values below 5 caution should be exercised and values above 10 are desirable as discussed by Pogolotti and colleagues (Pogolotti and Santi 1982). The biochemical and/or biological justification comes from the fact that the ATP/ADP ratio is a key physiological regulator of the glycolytic flux in E. coli (Koebmann, Westerhoff et al. 2002) and it was recently suggested to regulate protein synthesis via ribosomeassociated ABCF ATPase EttA/YjjK (Boel, Smith et al. 2014).

Discussion
The broadest scope of bacterial nucleotides from steady state exponential growth conditions are reported by Bochner and Ames (Bochner and Ames 1982), Bennett et al. (Bennett, Kimball et al. 2009) and Buckstein et al. (Buckstein, He et al. 2008)

(Supplementary Figure 8). Bochner and Ames have measured the nucleotide levels in
Salmonella typhimurium (rather than E. coli, the model organism characterized in the current report); however, due to conserved nature of central metabolism and exceptional breadth of the nucleotide species quantified, the seminal report has become the textbook reference of bacterial nucleotide levels. The work by Bennett and colleagues is remarkable in that they quantify not only nucleotides but also numerous other metabolites (Bennett, Kimball et al. 2009). Finally, the study by Buckstein and colleagues is apparently closest to ours in that only they have followed the rather complete set of nucleotides from exponential phase into stationary (Buckstein, He et al. 2008).
For abundance of guanosine nucleotides, the results of Buckstein et al. and are similar to ours in that GTP levels are gradually getting lower when cells go from exponential to stationary phase, and ppGpp is highest during stationary (Figure 4b). Yet we observe bigger elevation in levels of ppGpp during growth stop, and have some evidence for decrease in the level of all guanosine species. Moreover, the maximum levels of ppGpp in their work seem to occur right after cells have stopped growing whereas in our case, they occur right before the growth stop. Some of the differences in guanosine species are most likely because we have accomplished a better sensitivity whereas in their work, both GDP and ppGpp signal, given the scale, appear small and close to zero. For adenosine nucleotides, there are even less similarities: we do see an increase in ADP during the growth stop which is, however, very mild if it exists at all. In addition, we do not observe higher than stationary levels of ATP in exponential phase. Curiously, we get very similar overall shape of ATP and ADP levels except that, again, their results appear somewhat shifted along the growth curve towards later stationary. From that, in addition to the timing of ppGpp peak discussed above, it is also conceivable that we are over-interpreting some of the similarities by looking at the results which happened actually later in stationary phase in Buckstein and colleagues (Buckstein, He et al. 2008).
During the stringent response, accumulation of ppGpp precedes the curtailment of transcription (Cashel 1969, Gallant, Erlich et al. 1970). The initial drop in GTP must therefore result from allocation of guanosine pools in favor of (p)ppGpp. Next, when transcription-the major consumer of guanosines-stops, with all other things equal, GTP levels would be expected to expand whereas in fact they stay about half of the normal ( Figure 5 and references Harada 1969, Gallant, Erlich et al. 1970)). The fact that they do not expand suggests an inhibition of synthesis of GTP. Indeed, restriction of guanosine nucleotide biosynthesis by ppGpp is known from the literature (Gallant, Irr et al. 1971) and evident from our results (Figure 4b).
Besides drop in GTP and in ATP, the latter being often somewhat smaller, earlier reports of nucleotide levels during stringent response are contradictory. Edlin and Neuhard (Edlin and Neuhard 1967) report a gradual decline in all triphosphate pools to about a half within 30 min of stringent response. Both rapid shrinkage (Cashel and Gallant 1968) and increase (Edlin and Stent 1969) of UTP and CTP have been reported. Moreover, ATP and GTP stayed rather stable in according to Edlin and Stent (Edlin and Stent 1969). Also Edlin and Broda (Edlin and Broda 1968) record rather stable triphosphate pools except that GTP is clearly declining.
Most of the earlier work has been done with radioactive labeling-either with nucleosides or with 32 PO4. Possibly confounding the interpretation and/or explaining the contradictory results, in stringent strains at stress, phosphorylation of nucleotides is inhibited (Edlin and Neuhard 1967, Gallant and Cashel 1967, Cashel and Gallant 1968, Irr and Gallant 1969. In addition, stringent response can experimentally elicited in two different ways: (i) restraining the supply of amino acid or (ii) obstructing the aminoacylation of tRNA. The former, in principle, can be somewhat alleviated by the turnover of protein whereas the latter cannot be mitigated.

Nucleotide Samples for HPLC-A Whole Culture Approach Equipment and materials
(1) Manifold freeze-drier Some simple manifold system will do, equipment does not have to be resistant to aggressive chemicals. If manifold freeze-dryer is not available, skipping the freezedrying might be possible but we have not tried to do so.

Column concentration on Q Sepharose FF
(1) Equilibrate/wash column with water (we use ~6.5 mL/min) (2) Dilute sample 20x in mQ (deionized water) water (usually for 10 mL sample, fill up to 200 mL with ice cold water in ice-cold bottle) (3) Load 20x diluted sample onto Q-Sepharose FF column (~6.5 mL/min) (4) Wash with cold mQ (2-3 min, ~6.5 mL/min)  Use about 10-13 mL of E. coli culture of OD600 0.5, however, for fast exponentially growing unstressed cells that much of a starting material gives small peak, so larger culture volumes might be necessary (30-40 mL). In that case, use of larger filters or several 25 mm filters in parallel is appropriate.

After sample has thawed on ice, start the extraction with vortexing and continue with occasional vortexing
(3) Remove the filter:

General notes
(1) Column temp 26 °C, injection volume most often 100 µL (equivalent of 2.5 ODU of bacterial culture e.g. 5 mL of OD600 0.5. In unstressed conditions of fast growing bacteria, to detect ppGpp, be prepared to use 3-5 fold more material) (2) For column storage and wash, methanol does a better job than acetonitrile. Methanol is more efficient in cleansing of the column from charged compounds. Better cleansing results in faster outset and therefore shorter column uptime during the next session. We did not try phosphoric acid wash that is occasionally recommended (e.g. (Pogolotti and Santi 1982)).
(3) The main complication is the rapid deterioration of retention times (could be assigned to the complex, crude samples we usually have and very salty buffers) and short lifespan (only about 200-250 runs). Retention times decrease at different speed for nucleotides of interest and interfering peaks, making gradient program especially tricky to adjust and maintain. We have tried different column regeneration techniques but they seem to result in even faster wear off of the column. Retention decreases gradually during runs, however, expect to see more considerable decrease right after each storage of column, therefore, having several samples in one longer session is recommended.
(4) One has to remain careful when using high salt HPLC buffers. First, if the instrument is not cleaned properly, the remains of the salt can form crystals in pumps and seals; thus, using a pump seal wash system is recommended. Second, organic solvent for column storage or wash should be pumped through the system only after a thorough wash with water in order to avoid precipitation of buffer salts.

Notes on gradient program
(1) We have tried to improve the resolution (initial isocratic, stepwise gradients etc) but the linear gradient is the best, by a large margin.
(2) As column gets older, retention decreases. Therefore, decrease the final percentage of B (about 20% steps are appropriate: 100%, 80%, 60%, 40% and then the column is finished) and include short wash with 100% buffer B at the end of each run. If possible, consider mobile phase conditioning pre-columns as it is recommended for Partisil SAX (the one for Partisil is called Solvecon Guard Column), it helps to saturate the mobile phase with silica and helps to minimize column dissolution.
(3) Painstakingly, as the column gets older, retention decreases at different rate for different peaks-for peaks of interest and for interfering peaks-so that on top of adjusting buffer strength and/or gradient, one has to deal with certain ranges of runs in column lifespan during which one or the other nucleotide is nevertheless not resolved (4) We do realize the adjustments in buffer strength are tedious and obscure. We therefore describe now a typical gradient run session which consists of following steps.
First, the storage solution i.e. methanol is washed off with water. Second, the column is cleaned with high salt buffer by means of alternating 2-4 times between buffers A (low salt) and B (high salt) until the UV detection trace becomes reproducible. Third, the column is equilibrated with buffer A, followed by injection of the first sample. The first run is followed by an injection of the same sample spiked with nucleotide standards in in order to determine the identities of peaks. Fourth, sequence of samples is run and every 10 th -20 th sample is run twice, first without and then with the nucleotide standards spiked in. Finally, immediately after the session, salt is washed off the column with water and the column is stored in 100% methanol.
(5) Regarding the poor resolution of ATP in some early runs. What exactly seems to happen is that some peaks-with a retention time between ATP and GTP (where dATP and UTP are to be expected)-over the course of runs, due to change in retention time, we did not pursue to find out, however, the problem was more pronounced for E. coli samples than for B. subtilis (although the difference could have been due to somewhat different media used for the species, B. subtilis was grown as previously described (Kudrin, Varik et al. 2017)).
(6) As mentioned also in main text, increasing the resolution by using a longer column is likely to fail due to significant widening of the peaks (Supplementary Figure 1b). The widening, however, cannot be easily counteracted by increasing ionic strength of the buffer because of the already high salt concentrations necessary to elute the nucleotides from a shorter column. Notes on isocratic program.

Isocratic program
(1) Isocratic program is what we use nowadays almost exclusively as opposed to gradients. The routine is to measure ppGpp and ppGpp on SAX and quantify the rest of the nucleotides on Ion-Paired Reverse Phase (2) Most optimal for ppGpp measurements is to keep ppGpp retention time between 7.5 to 15 min. Be alert to point any interfering peaks, though, and if possible, verify the identity of a spectrum of a guanosine base.
(3) As the retention time decreases, come down with a flow rate-1.5 mL/min and 1 mL/min could be tried-and if that is not sufficient, come down with the buffer concentration, about 25% step is appropriate: 0.27M (29.5g/L), 0.19M (20.8g/L).
(4) We have not tested, but with a fresh column, instead of a 0.36M at 2 mL/min, one could try 0.5M at 1.5 mL/min or even at 1 mL/min.  Figures 2 and 3) 2.5 mL of OD600 0.5)

Preparation of buffers.
To make 1L of buffer A, dissolve 3.24 g 40% tbutylammoniumhydroxide (TBA-OH) in ~800 mL mQ. Add 4.08 g KH2PO4 and adjust pH to 6.0 with H3PO4. Finally, bring the volume to 1L with mQ and filtrate through 0.2 µm nylon filter. Store the buffers at 4 °C when not in use and re-filter every couple of days before use.

Wash with EDTA instead of including it in buffer or in sample.
Regardless if it is due to two-valent cations or something else, column needs periodic wash with EDTA (Supplementary Figure 2d, 2e and 2f), otherwise the shape of the peak of highly phosphorylated nucleotides deteriorates. Note that EDTA has no effect on fresh column (Supplementary Figure 2c). Every time we start HPLC, we thus perform a blank gradient run (in case of Waters Symmetry column, we performed a shortened 20 min gradient) in which we inject 100 µL 50 mM EDTA pH 8.0. Cserjan-Puschmann and colleagues recommend to include EDTA in buffer at 50 mM (Cserjan-Puschmann, Kramer et al. 1999). We find it to result in a baseline protrusion (a very wide and non-symmetric peak) in the middle of the gradient that (i) is only somewhat reproducible in size and location, therefore hampers the quantification of the nucleotides, (ii) does not disappear even at IPTG concentrations as low as 1 µM. We therefore resorted to EDTA injectionwash. Likewise, if contemplating to increase the stability of nucleotides via addition of EDTA during sample processing (Lundin and Thore 1975), retain that it might interfere with the HPLC (in fact, we had similar problems with EDTA on gradient SAX-HPLC).
Shutting down Kinetex Core-Shell C18 column. In our experience, one needs to perform shutdown of Kinetex Core-Shell C18 column with some caution, if higher than 65% ACN is to be used for storage. We did not investigate it thoroughly but it seems as if something is crashing out-probably ion-pair-and blocking the pre-column although the system had always been first washed with water. If such a blockage happens, changing the pre-column cartridge will restore the performance. We often stored the column in 65% ACN, however, favoured storage in 100% ACN and for the latter, we ran 5% ACN in water for 20 min at 0.8 mL/min (about 10 column volumes) to get rid of buffer salts, then brought flow rate to 0.4 mL min and switched to 65% ACN which we ran for 5 minutes. Finally, a linear gradient was ran from 65% ACN to 100% ACN in 15 minutes, after which, column was ready to be shut down.
General notes. As with SAX columns, we do get only 200-300 runs out of a C18 column in IPRP mode. That is, however frustrating, to be expected in case of complex biological material (Snyder, Kirkland et al. 1997). We tried some thorough wash regimes recommended by manufacturer but if anything, it resulted in faster wear-down of the column. We did not try wash with DMSO and with acetic acid (Gebelein, Merdes et al. 1992). Therefore, the only maintenance we do is regular replacement of pre-column, careful storage in solutions high in acetonitrile and EDTA injection wash.
We did attempt to use the IPRP-HPLC program as described by Buckstein and colleagues (Buckstein, He et al. 2008), however, we used HPLC-grade t-butyl ammonium phosphate instead of PIC A. For the majority of nucleotides, there was a massive increase in the baseline with the signal coming, most likely, from the eluting ion-pairing agent itself, or from impurities it contains. We therefore suggest either to use our protocol (which essentially adopts the one from Payne and Ames (Payne and Ames 1982)) or to follow the Buckstein protocol to the dot, including the use of PIC A.
Although dNTPs are probably not interfering with NTP measurements (dNTPs are 1-2 orders of magnitude less abundant (Nick McElhinny, Watts et al. 2010)) and quantification of dNTPs is out of the scope of this work, both isocratic and shallow gradient IPRP-HPLC hold the promise to skip the NTP removal step-which is usually done via periodate-methylamine oxidation (Garrett and Santi 1979, Tanaka, Yoshioka et al. 1984, Harmenberg, Cox et al. 1990) and/or boronate affinity chromatography (Payne and Ames 1982)-to resolve both dNTPs and NTPs in one run (Supplementary Figure   3b and 3c; see also (Arezzo 1987, Cross, Miller et al. 1993, Di Pierro, Tavazzi et al. 1995). ). Manual integration was avoided (used no more than for about one peak in 5-10% of the HPLC traces). Note that overlapping peaks, in case of a diode array detector, can be accurately quantified as discussed by Cross et al (Cross, Miller et al. 1993).  7a). Samples were drawn and stained with nigrosin (mixed with 1% nigrosin, spread onto microscope slide and air dried (Supplementary Figure 7b)) and microscopy images were analyzed by ImageJ (NIH), pixel-to-µm conversion was achieved by comparison with commercial beads. About 100-150 cells were measured for each time point. As described by Fry (Fry 1990), width (W) and length (L) were derived from perimeter (P) and area (A):

Initial Events For All Signals
Then, volume (V) was calculated approximating all bacteria to be a cylinder with two hemispherical caps: Finally, it was assumed that the periplasm accounts for a 0.21 fraction of cell volume (Stock, Rauch et al. 1977).
The cell volume decreases from 1.2±0.3 in exponential phase to 0.5±0.2 femtoliters in stationary phase (Supplementary Figure 7c and

Data Handling and Analysis
All the reported mean values and descriptors of spread are geometric unless indicated with '±' sign, in which case, arithmetic counterparts were used. Variation or uncertainty was expressed mostly with 95% confidence intervals; on graphs, however, we used standard error of the mean to avoid clutter in some cases and to keep it uniform in others.
Generally, outlier removal was avoided but when it was applied, Tukey fences (measurements deviating more than 1.5-times the interquartile range) were used on logarithmically transformed data (outliers did not account for more than about 2-3 % of the data). In an attempt to improve the resolution of gradient SAX-HPLC, we switched from a column length of 150 to 250 mm. As anticipated, this results in increased retention times and slightly improved resolution. All the gain in resolution, however, was negated by considerable widening of the peaks. What is shown here is our attempt to alleviate the widening, we increased salt concentration of the program to decrease retention times. Mack et al (Mack, Reed et al. 1985). (a) Based on ATP/ADP ratio of rapidly growing E. coli cells, filtration is the only form of harvesting that gets close to the ratios observed with least manipulative sampling i.e. no harvesting. Furthermore, centrifugation appears very inappropriate for harvesting even if preceded by quenching in cold glycerol or by aldehyde fixation. In general, ATP/ADP ratios way below 5 should be considered with caution as discussed by Pogolotti et al (Pogolotti and Santi 1982). (b) Similar pattern as for ATP/ADP, although less pronounced, was observed also for GTP/GDP ratios. (c) Acetic acid extraction of filtered cells gives stronger signal than formic acid. We speculate it is due to better compatibility with freeze-drying. Results are expressed as mean from two biological replicates with three technical replicates each. Error bars stand for s.e.m. (d) In case of acetic acid extraction of filtered cells, release of nucleotides from cells seems to become final during the freeze-drying step. Removal of acid precipitated material from sample prior freezedrying has resulted in weaker signal also elsewhere (Nazar, Lawford et al. 1970).   (Bennett, Kimball et al. 2009). The methodology implemented by Buckstein et al. (black circles) is the closest to our work and they quantify almost the very same set of nucleotides during growth from exponential into stationary phase (Buckstein, He et al. 2008). The results of 9 other reports (grey circles, aggregated set) were pooled together for descriptive statistics of median and interquartile range as shown by boxplots (Franzen and Binkley 1961, Neuhard and Munch-Petersen 1966, Edlin and Neuhard 1967, Bagnara and Finch 1968, Lowry, Carter et al. 1971, Holms, Hamilton et al. 1972, Poulsen and Jensen 1987, Vogel, Pedersen et al. 1991, Taymaz-Nikerel, de Mey et al. 2009). The whiskers stand for range still within 1.5-times the interquartile range of lower/upper quartile. Note that guanosine ratios reported by Bennett et al. were calculated without taking into account the (p)ppGpp levels since in this report (p)ppGpp was not quantified and, therefore, not reported (Bennett, Kimball et al. 2009). We believe that this did not significantly affect the final result since the (p)ppGpp levels in rapidly growing cells are low and constitute only a small fraction of the total guanosine pool. Exponentially growing cells were harvested by filtration and processed accordingly. B.

Supplementary
subtilis sample was subjected to further refinement by FPLC and ethanol precipitation.
Finally, samples were run on IPRP-HPLC-UV (black trace) and the identities of nucleotide peaks were revealed by a separate run with same samples but nucleotide standards spiked in. All cells were grown with vigorous aeration at 37 °C prior to harvesting. B.
subtilis was grown in defined medium supplemented with glucose and amino acids (Kudrin, Varik et al. 2017), P. wasabiae was grown in defined medium with glucose (Koiv, Andresen et al. 2013), whereas both A. baumannii and P. putida were grown in LB.