Sulfur Isotope Enrichment during Maintenance Metabolism in the Thermophilic Sulfate-Reducing Bacterium Desulfotomaculum putei

ABSTRACT Values of Δ34S (\batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \({=}{\delta}^{34}S_{HS}{-}{\delta}^{34}S_{SO_{4}}\) \end{document}, where δ34SHS and \batchmode \documentclass[fleqn,10pt,legalpaper]{article} \usepackage{amssymb} \usepackage{amsfonts} \usepackage{amsmath} \pagestyle{empty} \begin{document} \({\delta}^{34}S_{SO_{4}}\) \end{document} indicate the differences in the isotopic compositions of the HS− and SO42− in the eluent, respectively) for many modern marine sediments are in the range of −55 to −75‰, much greater than the −2 to −46‰ ε34S (kinetic isotope enrichment) values commonly observed for microbial sulfate reduction in laboratory batch culture and chemostat experiments. It has been proposed that at extremely low sulfate reduction rates under hypersulfidic conditions with a nonlimited supply of sulfate, isotopic enrichment in laboratory culture experiments should increase to the levels recorded in nature. We examined the effect of extremely low sulfate reduction rates and electron donor limitation on S isotope fractionation by culturing a thermophilic, sulfate-reducing bacterium, Desulfotomaculum putei, in a biomass-recycling culture vessel, or “retentostat.” The cell-specific rate of sulfate reduction and the specific growth rate decreased progressively from the exponential phase to the maintenance phase, yielding average maintenance coefficients of 10−16 to 10−18 mol of SO4 cell−1 h−1 toward the end of the experiments. Overall S mass and isotopic balance were conserved during the experiment. The differences in the δ34S values of the sulfate and sulfide eluting from the retentostat were significantly larger, attaining a maximum Δ34S of −20.9‰, than the −9.7‰ observed during the batch culture experiment, but differences did not attain the values observed in marine sediments.


Dissimilatory SO 4
2Ϫ reduction is a geologically ancient, anaerobic, energy-yielding metabolic process during which SO 4 2Ϫ -reducing bacteria (SRB) reduce SO 4 2Ϫ to H 2 S while oxidizing organic molecules or H 2 . SO 4 2Ϫ reduction is a dominant pathway for organic degradation in marine sediments (23) and in terrestrial subsurface settings where sulfur-bearing minerals dominate over Fe 3ϩ -bearing minerals. For example, at depths greater than 1.5 km below land surface in the fractured sedimentary and igneous rocks of the Witwatersrand Basin of South Africa, SO 4 2Ϫ reduction is the dominant electron-accepting process (3,26,46,48,61).
The enrichment of 32 S in biogenic sulfides, with respect to the parent SO 4 2Ϫ , imparted by SRB, is traceable through the geologic record (10,54). The magnitude of the ⌬ 34 S (ϭ ␦ 34 S pyrite Ϫ ␦ 34 S barite/gypsum , where ␦ 34 S pyrite and ␦ 34 S barite/gypsum are the isotopic compositions of pyrite and barite or gypsum) increases from Ϫ10‰ in the 3.47-billion-year-old North Pole deposits to Ϫ30‰ in late-Archaean deposits (55), to Ϫ75‰ in Neoproterozoic to modern sulfide-bearing marine sediments (13).
Experiments performed during the 1960s found that the magnitude of ε 34 S was inversely proportional to the csSRR for organic electron donors (16,31,38,39) when SO 4 2Ϫ was not limiting. More-recent batch culture experiments on 3 psychrophilic (optimum growth temperature, Ͻ20°C) and mesophilic (optimum growth temperature, between 20°C and 45°C) SRB strains (7) and on 32 psychrophilic to thermophilic SRB strains (22), however, have failed to reproduce such a relationship. In 2001, Canfield (11) reported an inverse correlation between ε 34 S and reduction rate using a flowthrough sediment column and demonstrated that ε 34 S values of approximately Ϫ35 to Ϫ40‰ were produced when organic substrates added by way of amendment were limited with respect to SO 4 2Ϫ . Because it was not possible to readily evaluate changes in biomass in the sediment column with changes in temperature or substrate provision rate, it was inferred that changes in ε 34 S were related to changes in the csSRRs. More recently, Canfield et al. (12) observed a 6‰ variation in ε 34 S values related to the temperature of the batch culture experiments relative to the optimum growth temperature. The few early experiments that were performed using H 2 as the electron donor yielded ε 34 S values ranging from Ϫ3 to Ϫ19‰ (22,38,39), which appear to correlate with the csSRR (39). Hoek et al. (32) also found that the ε 34 S values for the thermophilic SO 4 2Ϫ reducer Thermodesulfatator indicus increased from between Ϫ1.5‰ and Ϫ10‰ in batch cultures with high H 2 concentrations to between Ϫ24‰ and Ϫ37‰ in batch cultures grown under H 2 limitation with respect to SO 4 2Ϫ . Detmers et al. (22) found that the average ε 34 S of SRB that oxidize their organic carbon electron donor completely to CO 2 averaged Ϫ25‰, versus Ϫ9.5‰ for SRB that release acetate during their oxidation of their organic carbon electron donor. Detmers et al. (22) (14) but is not consistent with the lack of significant fractionation observed during oxidative reactions (29). To explain the ⌬ 34 S values of Ϫ55 to Ϫ77‰ reported to occur in interstitial pore waters from 100-to 300-m-deep, hypersulfidic ocean sediments (51,64,67), where the presence of a S-oxidative cycle is unlikely, an alternative, elaborate model of the SO 4 2Ϫ -reducing pathway has been proposed by Brunner and Bernasconi (9 (20,25,41,42,52,66). The conditions under which the maximum ε 34 S values might be expressed are a combination of elevated HS Ϫ concentrations, electron donor limitations, nonlimiting SO 4 2Ϫ concentrations, and a very low csSRR. The csSRR for subsurface environments has been estimated from biogeochemical-flux modeling to be 10 Ϫ6 to 10 Ϫ7 fmol cell Ϫ1 h Ϫ1 (23), with a corresponding cell turnover rate greater than 1,000 years (37).
Batch and chemostat culture systems, despite low growth rates, cannot completely attain a state of zero growth with constant substrate provision and therefore do not accurately reflect the in situ nutritional states of microbes in many natural settings. Retentostats, or recycling fermentor vessels, recycle 100% of biomass to the culturing vessel, allowing experimenters to culture microbial cells to a large biomass with a constant nutrient supply rate until the substrate supply rate itself becomes the growth-limiting factor and cells enter a resting state in which their specific growth rate approaches zero and they carry on maintenance metabolism (1,47,53,58,59,62,63). Utilizing this approach, Colwell et al. (18) were able to obtain a cell-specific respiration rate of 7 ϫ 10 Ϫ4 fmol of CH 4 cell Ϫ1 h Ϫ1 for a mesophilic marine methanogen, a rate that is comparable to that estimated for methanogenic communities in deep marine sediments off the coast of Peru (49).
In this study, the conditions that Brunner and Bernasconi (9) hypothesized would lead to the large ⌬ 34 S values seen in na-ture were recreated in the laboratory by limiting the electron donor supply rate with respect to the SO 4 2Ϫ supply rate in a retentostat vessel. The S isotopic enrichment by a resting culture of Desulfotomaculum putei at an extremely low csSRR was compared to that of a batch culture experiment to determine whether the ε 34 S values produced under the former conditions approach the ⌬ 34 S seen in nature.

MATERIALS AND METHODS
Cultivation of D. putei. D. putei strain TH15 represents an appropriate model subsurface SRB because it is the closest culturable phylogenetic relative to the 16S rRNA clone sequences belonging to the SRB that are prevalent in South African subsurface sediments (26,44,46), and it was isolated from an ϳ2.7-km depth in the Taylorsville Triassic Basin, VA (45). D. putei is a strictly anaerobic, Batch culture cultivation. D. putei was grown in batch culture for 72 h at 65°C and sampled for changes in biomass, aqueous geochemistry, and sulfur isotope enrichment. The lactate and SO 4 2Ϫ concentrations were 4 and 20 mM, respectively, consistent with that of the retentostat experiments. Changes in cell concentration and morphology were determined by staining 100-l aliquots of cells with 5 nM Sytox-13 green fluorescent dye and analyzing them using flow cytometry (FACScan; Becton-Dickinson). The cell concentration and size were calibrated to 1-m fluorescent microspheres at known concentrations (Molecular Probes, Eugene, OR). The uncertainty for enumeration was determined by triplicate analyses, and the 95% error of confidence was typically Ϯ6%.
Aqueous . Sulfide samples were collected in 1.5-ml microcentrifuge tubes, and the sulfide was immediately quantified by UV spectroscopy by following the method of Cord-Ruwisch (19), which has a detection limit of 0.01 mM and a 95% probability envelope of Ϯ5%.
Cell-specific rates of SO 4 2Ϫ reduction (csSRRs in mol of SO 4 cell Ϫ1 h Ϫ1 ) were calculated for the exponential phase using the changes in the SO 4 2Ϫ levels and cell concentrations between time points 1 and 2 (T 1 and T 2 , respectively) [SO 4 2Ϫ(1) and SO 4 2Ϫ(2) and X 1 and X 2 , respectively] according to the following equation (22): The standard deviation of the csSRR was determined from duplicate measurements and was equal to ϳ25%. Retentostat cultivation. All retentostat experiments were performed within an incubator that was housed in an anaerobic glove bag (Coy Laboratory Products, Inc., Grass Lake, MI) filled with N 2 -CO 2 (70:30). The retentostat experiments were performed using a 1-liter culture vessel with a constant rate of fresh medium supplied at a rate of 22 ml h Ϫ1 for runs 1 and 2 and of 10 ml h Ϫ1 for run 3. The medium contained 4 mM lactate and 26 mM SO 4 2Ϫ for run 1 and the same lactate concentration and 20 mM SO 4 2Ϫ for runs 2 and 3. The other medium constituents were the same as described above. The vessel was inoculated with 10 ml of batch culture.
A Pellican XVG50 tangential-flow filter (Millipore Inc., CA) with a maximum pore size of 0.2 m was used to recycle cells back into the culture vessel while allowing effluent medium to be removed at the same rate that fresh medium was being added (Fig. 1). The medium was constantly stirred with a magnetic stir bar. Samples were collected simultaneously with inoculation for time zero measurements, and the retentostat was then sampled three times every 24 h. The retentostat vessel fluid volume, which fluctuated due to slight variations in the pump 5622 DAVIDSON ET AL. APPL. ENVIRON. MICROBIOL. rates ( Fig. 1), was recorded to correct for the calculation of substrate supply rates. The maximum vessel volume fluctuation was never more than 5% of the total volume. Biomass samples were collected in 1.5-ml microcentrifuge tubes from an in-line sampling port on the recycling biomass loop and were fixed immediately in a 2.2% glutaraldehyde solution. Cell concentration and morphology for each time point was determined by flow cytometry as described above.
Aqueous geochemistry (SO 4 , acetate, and lactate) samples were collected in 1.5-ml microcentrifuge tubes from the effluent port ( Fig. 1) and immediately frozen at Ϫ80°C until analysis by ion chromatography-mass spectrometry, as described above. Sulfide samples were collected in 1.5-ml microcentrifuge tubes and were immediately quantified as described above. The pH of the effluent was measured at each time point using pH paper in two ranges, pHs 5 to 10 (Colorphast intermediate range; accuracy, 0.3 to 0.5; EMD Chemicals) and pHs 6.5 to 10 (Colorphast narrow range; accuracy, 0.2 to 0.3; EMD Chemicals). The dissolved inorganic carbon was measured using a LI6252 apparatus (LICOR Biosciences, Lincoln, NE). The H 2 concentration in the glove bag headspace was measured by gas chromatographic analyses (Kappa5; Trace Analytical, Menlo Park, CA) on gas samples collected in serum vials before the exponential growth phase and during the maintenance metabolism phase of run 3.
The csSRR for each time point (T i ) of the experiment (csSSR i ) was calculated using the following formula: where VI is the volume flow rate for the incoming fresh medium (liters h Ϫ1 ), VO i is the volume flow rate of the outgoing medium for sample concentration at time zero, and X i is the cell concentration (cells liter Ϫ1 ) for sample i. In order to test the effects of various csSRRs on the ε 34 S, the initial cell concentration was increased from 4.4 ϫ 10 10 cells liter Ϫ1 for run 1 to 1.5 ϫ 10 11 cells liter Ϫ1 for run 2 and to 4.7 ϫ 10 12 cells liter Ϫ1 for run 3. To determine if temperature affected the ε 34 S, the temperature of run 3 was reduced from 65 to 60°C at 640 h and then from 60 to 55°C at 692 h. The temperature was reduced from 55 to 50°C at 840 h, but the fresh medium had become contaminated at this point and the experiment was terminated.
Sulfur isotope methods. Sulfur isotope samples were collected by adding 1 ml of 2 M CdCl 2 solution to the 10 ml of medium in each tube during the course of the batch culture experiment. An uninoculated medium tube served as the negative control and was processed in the same manner to obtain the initial SO 4 2Ϫ isotopic composition. Sulfur isotope samples were collected for the retentostat experiments by collecting 45 ml of eluent in sterile 120-ml, O 2 -free serum vials preloaded with 5 ml of a 2 M CdCl 2 solution, capped with butyl rubber stoppers, and frozen at Ϫ20°C until processing, as previously described (8,22). In the presence of H 2 S, CdS was precipitated and recovered on quartz filters, and the filters were dried in an oven overnight at 60°C. The filtrate solution was acidified to about pH 3 and heated to 80°C. Drops of a saturated BaCl 2 solution were added to the filtrate until there was no visible whitening due to suspended BaSO 4 crystals. The BaSO 4 precipitate was washed and centrifuged three times using distilled, deionized water from a Millipore gradient Mili-Q water filtration system with a QuantumEX and Q-Gard-2 dual-filtration system.
Sulfur isotope ratios for BaSO 4 and CdS were determined using an EA1110 elemental analyzer coupled to a Finnigan Mat 252 isotope ratio mass spectrometer via a ConFlo II split interface (57). Isotope data are reported as the ␦ 34 S per mille deviation from the Vienna Cañon Diablo Troilite (VCDT) calculated by using equation 3: where r sample is equal to 34 S sample / 32 S sample i and r VCDT is equal to 0.0441638 (24). Each set of 12 samples was bracketed by three working standards ranging in isotopic composition from Ϫ32.5‰ to ϩ38.7‰ based on calibration to the International Atomic Energy Agency (IAEA) S-1 (Ϫ0.30 ‰), IAEA S-3 (Ϫ32.5‰), and NBS-127 (ϩ20.3 ‰) external standards. Corrections were assumed to be linear over the entire range of isotopic compositions, and analytical reproducibility was better than Ϯ0.2‰ for standards and analytical duplicates.
The ε 34 S during exponential-phase growth in batch culture was calculated using the following standard Rayleigh model for closed systems: where ␦ 34 S product is the isotopic composition in the integrated HS Ϫ pool, ␦ 34 S init is the isotopic composition in the initial SO 4 2Ϫ pool, and f is the fraction of SO 4 2Ϫ remaining at a given time. The ␦ 34 S init was 9.97‰. The ε 34 S during retentostat operation can be modeled as a steady-state system (50) using the following equation: where ␦ 34 S SO4 and ␦ 34 S HS are the isotopic compositions of the SO 4 2Ϫ and HS Ϫ in the eluent (Fig. 1).
The overall S isotope mass balance was verified by comparing the total S isotopic composition of the eluent to that of the SO 4 2Ϫ in the fresh medium supply by using the following relationship: The ␦ 34 S value of the SO 4 2Ϫ in the fresh medium supply was determined at the initiation of each experiment and was Ϫ0.41‰ for run 1, 0.40‰ for run 2, and 5.44‰ for run 3. These calculations also take into account the HS Ϫ present in the inoculum (ϳ10 mM in 10 ml) and in the 1-liter culture vessel at the beginning of each experiment (ϳ0.256 mM HS Ϫ ) and in the 3 liters of fresh medium supply in the case of run 2. The ␦ 34 S value of the HS Ϫ in the medium supply was determined at the initiation of each experiment and was 8.74‰ for run 1, 4.94‰ for run 2, and 5.94‰ for run 3. Geochemical modeling. The retentostat experimental data were simulated using the React module of Geochemist's Workbench standard version 7.04 (5) by using the "flush" command and by following the dual-Monod reaction model of Jin and Bethke (34,35): where the reaction rate, k, is in mol kg of H 2 O Ϫ1 s Ϫ1 , V max is the maximum cellular rate for SO 4 2Ϫ reduction in mol mg of biomass Ϫ1 s Ϫ1 , and X is the cell concentration in mg of biomass kg of H 2 O Ϫ1 . F A is the factor controlled by the electron-accepting reaction, F T is the thermodynamic-potential factor, and F D is the kinetic parameter controlled by the electron-donating reaction and is defined by the equation where m D PD and m Dϩ PDϩ are the concentrations of the electron donor reactant and product, respectively. Similarly, the electron-accepting reaction is controlled by the kinetic parameter F A , defined by where m A PA and m AϪ PA are the concentrations of the reactants of the electronaccepting reactant and product, respectively. PA and PD are assumed to be 1 for all reactions and PAϪ and PDϩ are assumed to be 0 for all reactions. K D and K A are in molar units and are equivalent to the half-saturation constants K m and K s . The K D value was constrained by matching the initial rate of lactate depletion. The K A value of 10 Ϫ5 M reported for SRB by Ingvorsen and Jorgensen (33) was assumed, but because of the high SO 4 2Ϫ concentrations used in these experiments, the K A had no effect on the outcome of the simulations.
F T is defined as where R is the real gas constant 8.314472 J K Ϫ1 mol Ϫ1 , T is the temperature in kelvins, x is the average stoichiometric number (34) or the ratio of the freeenergy change of the overall reaction to the sum of the free energy changes for each elementary step, and ⌫ is the net thermodynamic driving force of the reaction defined by equation 11: where ⌬G is the free-energy change of the redox reaction and ⌬G p is the free energy for the phosphorylation reaction ADP ϩ P 3 ATP (35). ⌬G p can vary from about Ϫ40 to Ϫ70 kJ mol Ϫ1 , depending upon the temperature, pH, and concentrations of ADP and ATP in the cell (43), and more likely ranges from Ϫ50 to Ϫ88 kJ mol Ϫ1 if a thermodynamic efficiency of 80% is assumed. n is the number of moles of ATP generated per mole of reaction mixture. Values of four-thirds and 0.5 were utilized for n and x, respectively, according to the method of Jin and Bethke (35), but the ⌬G p was constrained by matching the lactate concentration in the maintenance phase of the retentostat experiments, i.e., the threshold concentration of the lactate. The cell concentration is given by the following relationship: where dX/dt is the change in cell concentration with time, i.e., the growth rate, Y is the yield in mg of biomass mol of reaction mixture Ϫ1 , k is the reaction rate in mol of reaction mixture kg of H 2 O Ϫ1 s Ϫ1 , and D is the death rate in s Ϫ1 . Cell counts were converted to mg of biomass by assuming a cellular mass of 1.56 ϫ 10 Ϫ13 g cell Ϫ1 (30). Y and V max were determined by simultaneously fitting the rate of SO 4 Ϫ2 reduction and the cell concentrations during the growth phase. D was determined by the trend in cell concentrations during the maintenance phase. Kinetic enrichment of S isotopes was simulated by introducing 34 S species into the thermodynamic database, defining a 34 S microbial reaction, and then adjusting the V max , D, and K A for the 34 S microbial reaction to fit the data.
TEM protocol. Transmission electron microscope (TEM) observations of the samples were performed subsequent to pelleting of cell suspensions and fixation with 2.5% glutaraldehyde in 200 mM Na cacodylate (pH 7.2) for 2 h on ice. The pellets were then washed three times with 100 mM Na cacodylate buffer (pH 7.2) and left in the buffer overnight at 4°C. The pellets were then fixed with 2% Os tetroxide in Na Veronal buffer (pH 7.2) for 1 h on ice, followed by three rinses with the Na Veronal buffer and three rinses with 50 mM Na maleate buffer (pH 5.2). The pellets were stained with 0.5% uranyl acetate in the Na maleate buffer overnight at room temperature and then rinsed three times with 50 mM Na maleate buffer. The pellets were then sequentially dehydrated with 30%, 50%, 70%, and 95% ethanol and twice with 100% ethanol, followed by immersion in propylene oxide for 30 min, a propylene oxide-resin (1:1) mixture overnight, a propylene oxide-resin (1:2) mixture for 24 h, and straight resin overnight. The pellets were then placed in Beem capsules and polymerized at 60°C overnight. Unstained 70-nm sections were obtained using a diamond knife on a Leica UC6 ultramicrotome and observed at 80 kV on a Zeiss 912AB TEM equipped with an Omega energy filter (at Princeton University) and on a Philips 420 TEM operating at 80 kV (at the University of Western Ontario). Micrographs were captured using a digital camera from Advanced Microscopy Techniques and saved as TIFF files onto a Dell personal computer (at Princeton University) and using a Hamamatsu digital camera and HP computer (at the University of Western Ontario).
Lipid analyses. Biomass samples for membrane lipid analyses were shipped to the University of Tennessee on dry ice and stored at Ϫ80°C until analyzed. The total lipids were extracted using methanol-chloroform-buffer (2:1:0.8 [vol/vol/ vol]) and subsequently fractionated on a silicic acid column with only the polar lipids then transesterified into phospholipid fatty acid (PLFA) methyl esters (65). The PLFA methyl esters were separated, quantified, and identified by gas chromatography-mass spectrometry (65).

RESULTS
Batch culture experiment. During the batch culture experiments, 6.6 mM SO 4 2Ϫ and 5.4 mM lactate were consumed over the 72-h period, with the production of 5.8 mM acetate. S 2 O 3 2Ϫ was not detected. The csSRR for exponential-phase growth of D. putei was (0.9 Ϯ 0.4) ϫ 10 Ϫ15 mol cell Ϫ1 h Ϫ1 , with a Y of 5,180 mg mol of lactate Ϫ1 , and an ε 34 S of Ϫ9.7‰ Ϯ 0.4‰ was measured.
Retentostat experiment. The pH and temperature remained at 7.8 Ϯ 0.5 and 65°C, respectively, with no fluctuations throughout the duration of the experiments, except in run 3, where the temperature was purposely decreased in two steps to 55°C toward the end. During all three runs, the medium in the retentostat became progressively darker. During run 1, the SO 4 2Ϫ concentrations decreased from 25 mM to ϳ19 mM and the lactate concentrations decreased from 4.0 mM to below the detection level (Fig. 2) by the end of exponential-phase growth, when an inflection occurs in the model growth curve (Fig. 3). SO 4 2Ϫ concentrations decreased from 20 mM to ϳ14 mM and HS Ϫ concentrations increased from 0.1 mM to a maximum of 6 mM in runs 2 and 3 (Fig. 2) by the end of exponential-phase growth (Fig. 3). S 2 O 3 2Ϫ concentrations increased from Ͻ6 M to 0.3 Ϯ 0.1 mM in a manner synchronistic with HS Ϫ production (Fig. 2). Lactate concentrations decreased from 4 mM to 4.8 Ϯ 1.7 M (Fig. 2) by the end of the exponential-phase growth (ϳ30 h in Fig. 3) of run 2 and from 4 mM to 77 Ϯ 44 M (Fig. 2) by the end of the exponential-phase growth (ϳ92 h in Fig. 3) of run 3 and remained at these low concentrations for the remainder of these experiments. Acetate increased from below 0.2 M to 9 Ϯ 2 mM by the end of the exponentialphase growth of run 2 and increased to 8 Ϯ 2 mM by the end of the exponential-phase growth of run 3; it remained at these concentrations for the remainder of these experiments. No formate was detected. The HCO 3 Ϫ varied from 8 mM to 13 mM. The decrease in SO 4 2Ϫ and lactate concentrations and the increase in HS Ϫ concentrations during the course of each run were consistent with the following reaction stoichiometry: The declines in SO 4 2Ϫ concentrations for runs 1, 2, and 3 ( Fig.  2) were best fitted with specific V max reaction rates of 8.7 ϫ 10 Ϫ10 , 1.7 ϫ 10 Ϫ10 , and 5.3 ϫ 10 Ϫ12 mol, respectively, of SO 4 2Ϫ mg of biomass Ϫ1 s Ϫ1 . Geochemical modeling indicated that the pH should remain at 7.8, as it is buffered by the precipitation of dolomite and the 30% CO 2 in the headspace, which is consistent with the observed pH and explains the darkening of the medium. The average threshold lactate concentration was fit with a ⌬G p value of Ϫ63 Ϯ 1 kJ (mol of SO 4 2Ϫ ) Ϫ1 for run 2 and of Ϫ75 Ϯ 1 kJ (mol of SO 4 2Ϫ ) Ϫ1 for run 3. As has been observed in other retentostat experiments, cell concentrations increased from the inoculum concentration through an exponential phase of growth until the substrate feed became limiting and the growth rate declined and, in the case of run 1, approached zero (Fig. 3). The geochemically modeled curves simulate the overall trend in declining growth rates (Fig. 3). A maximum cell concentration of 2 ϫ 10 9 cells ml Ϫ1 was observed for run 1 by 408 h, 8 ϫ 10 9 cells ml Ϫ1 was observed for run 2 by 171 h, and 3 ϫ 10 10 cells ml Ϫ1 was observed for run 3 by 450 h (Fig. 3). Run 2 may have been terminated prior to attaining maximum cell concentration and zero growth rate. Toward the end of run 3 when the temperature was reduced by a total of 10°C in two 5°C increments, the cell concentration attained 4 ϫ 10 10 cells ml Ϫ1 . The lowest csSRR values attained at the end of the runs were 0.10 Ϯ 0.02, 0.016 Ϯ 0.012, and 0.0017 Ϯ 0.0003 fmol of SO 4 2Ϫ cell Ϫ1 h Ϫ1 . In runs 1 and 2, the cell concentrations declined briefly at the end of exponential growth (Fig. 3), indicating that some cell death process was occurring in these experiments, and this was seen by TEM. TEM observations also confirmed that dividing cells were more abundant in the samples collected prior to the onset of maintenance, were detected by the presence of lysed cells, indicated a reduction in cell volume from 4.9 to 0.5 m 3 , and revealed the formation of endospores (Fig. 4). Spores comprised ϳ10% of the total cell counts, and the csSRR values were corrected for this nonactive portion of the population.
Lipid analyses. PLFA results were obtained for samples from the end of the exponential phase through the maintenance phase of runs 2 and 3 (Fig. 5). The total PLFA ranged from 286 to 2,470 pmol ml Ϫ1 . These concentrations compared to the cell counts correspond to a cell number (pmol of PLFA Ϫ1 ) ratio of 2.7 ϫ 10 7 with a Ϯ1 S.D. of ϩ3.3 ϫ 10 7 and Ϫ1.6 ϫ 10 7 . This value is ϳ1,000 times the 2.5 ϫ 10 4 normally used for conversion of PLFA concentrations into cell concentrations (4). The PLFA composition was dominated by normal saturates (33 to 57%) and terminally branched saturates (29 to 52%), with the monoenoics comprising 3 to 11%, the branched monoenoics comprising 4 to 17%, and the one cyclic fatty acid present, 17:0 cyc, comprising only 0 to 3%. The relative proportions of the major fatty acid groups present in the retentostat experiment match those reported for D. putei (45), but the normal saturate abundance was somewhat higher and the monoenoics lower than those reported by Liu et al. (45). At the beginnings of runs 2 and 3, the proportions of normal saturates were greater than those of the terminally branched saturates, but further into the maintenance phase the terminally branch saturates became proportionally greater than the normal saturates, with the crossover occurring at ϳ150 h for run 2 and ϳ400 h for run 3 (Fig. 5). Toward the end of run 3, the proportions of terminally branched saturates decreased and of normal saturates increased as the temperature decreased from 65 to 55°C. S isotope analyses. The S mass balances for all aqueous sulfur species (SO 4 2Ϫ , HS Ϫ , and S 2 O 3 2Ϫ ) in runs 2 and 3 remained within 7% and 10%, respectively, of the initial 20 mM SO 4 2Ϫ pool. The S mass balance for run 1 could not be confirmed due to inaccurate HS Ϫ measurements. The S isotope mass balance for runs 2 and 3 as determined using equation 6 were ϳ1‰ Ϯ 1‰ greater than the S isotopic composition of the supply medium.  Assuming that the isotopic composition of S 2 O 3

2Ϫ
, which was not determined, accounts for this small isotope imbalance, its ␦ 34 S values ranged from Ϫ55 to Ϫ100‰ and from Ϫ20 and Ϫ60‰ for runs 2 and 3, respectively.
The ε 34 S values ranged from 12 to Ϫ15‰ during the course of run 1, with the most negative ε 34 S recorded at 408 h (Fig. 6). The ε 34 S values ranged from 5 to Ϫ21‰ during the course of run 2, with the most negative ε 34 S recorded at 95 h (Fig. 6). The ε 34 S values ranged from 1 to Ϫ18‰ during the course of run 3, with the most negative ε 34 S recorded at 192 h (Fig. 6). In the cases of runs 2 and 3, the most negative ε 34 S occurred just after the exponential phase and before the maintenance phase, and as the cells entered the state of maintenance metabolism, the ε 34 S steadily declined. This was most readily apparent in run 2, but run 3 exhibited an extension of the same trend. As the temperature dropped from 65 to 60°C, the ε 34 S changed from Ϫ15 to Ϫ5‰. As the temperature dropped from 60 to 55°C, the ε 34 S changed from Ϫ5 to Ϫ10‰.

DISCUSSION
The retentostat cellular concentration profiles and substrate consumption curves correspond well with those found in previously published retentostat studies (47,53,58,59,62,63), with the exception that distinct linear regions in the cellular concentration profiles were not readily discernible (Fig. 3). The average calculated maintenance coefficients during the final stages of runs 1, 2, and 3 were 4.7 ϫ 10 Ϫ17 , 1.2 ϫ 10 Ϫ17 , and 1.
, thereby adding to the measured acetate pool and ultimately affecting the interpretation of the ε 34 S data due to mixed electron donor availability. This hypothesis seems unlikely, however, because of the following: (i) analyses of the glove bag headspace before exponential phase and during maintenance metabolism phase yielded 20 parts per million by volume H 2 at both time points, indicating that H 2 existed in the glove bag but had remained unconsumed throughout the course of the experiment; (ii) even if all of this H 2 were consumed to produce acetate, given that the volume of the glove bag is ϳ30 mol of gas, less than 1 mmol of acetate could be produced compared to the ϳ52 mmol of acetate produced during run 3; (iii) at 65°C only submicromolar concentrations of the H 2 would be dissolved in the medium, and this would be insufficient to account for the millimolar concentrations of acetate recorded; and (iv) excess acetate was observed in the batch culture where no H 2 was added or detected. Second, peptones in the fresh medium supply (see Materials and Methods) may have provided an alternate source of organic C to the system. Assuming that the 0.2% (by weight) peptones in the supply medium contained 40% organic C by weight (2) and that 100% of it was bioavailable and could subsequently contribute to acetate production via fermentation, as has been reported for other bacterial strains (50), then ϳ33 mM could potentially result from this source compared to the 8 to 10 mM acetate concentrations observed. If true, this would mean that our calculated maintenance coefficients are minimum estimates.
Third, as cells are nutritionally deprived, they tend to shrink in order to maximize their surface area-to-volume ratio for increased transmembrane transport of nutrients, as well as for conservation of energy by limiting production of potentially superfluous macromolecules required for growth and cell division. Self-cannibalization could occur at this time, depleting intracellular carbon stores and once again producing acetate by direct substrate level phosphorylation. TEM observations revealed that cell size reduction did occur, with an increasing proportion of cells (Fig. 4B) being smaller than the cells of more normal size seen in the growth phase (Fig. 4A). Results of PLFA analyses are compatible with a change in membrane composition reflecting the change from vegetative cells (normal saturates dominate) to resting cells in maintenance metabolism (terminally branched saturates dominate) (Fig. 5).
Whether the last two processes have an effect on the ε 34 S results is uncertain, as the electron balance indicated that SO 4 2Ϫ reduction occurred primarily by the complete oxidation of lactate.
S isotopic enrichment during maintenance metabolism. The S mass balance results indicate that the S species were conserved for the duration of the retentostat run, and the ϳ1‰ Ϯ 1‰ S isotope mass imbalance that was based only on the ␦ 34 S of SO 4 2Ϫ and HS Ϫ suggests that the ␦ 34 S of the S 2 O 3 2Ϫ was more negative than that of the HS Ϫ . Theoretically, H 2 S comprises only 4.5% of the total sulfide at pH 7.8 and 65°C, and the theoretical equilibrium ⌬ 34 S H2 S Ϫ HS Ϫ is Ϫ1.1‰ and Ϫ1.2‰ for 65°and 55°C, respectively. The ␦ 34 S of the total sulfide, HS Ϫ ϩ H 2 S, would be only 0.05 to 0.06‰ more negative than the values measured for HS Ϫ , and this falls within the error of measurement.
The lowest csSRRs attained in our retentostat experiments are substantially lower than the lowest csSRRs reported in the literature for batch culture experiments. The lowest reported csSRR for a batch culture experiment was 2 ϫ 10 Ϫ17 mol cell Ϫ1 h Ϫ1 from a psychrophilic SRB strain incubated at Ϫ2°C (7), and this rate is 10 times higher than that of run 3. The lowest reported csSRR for a thermophilic batch culture experiment was 3 ϫ 10 Ϫ16 mol cell Ϫ1 h Ϫ1 for Desulfotomaculum geothermicum (22), which is 100 times higher than that of run 3. The lowest csSRR attained in our experiments, 1.7 ϫ 10 Ϫ8 mol of SO 4 cell Ϫ1 h Ϫ1 , is still 3 orders of magnitude higher than the in situ planktonic csSRR inferred from geochemical estimation (ϳ4 ϫ 10 Ϫ21 mol of SO 4 cell Ϫ1 h Ϫ1 ) for the deep terrestrial subsurface of the Witwatersrand Basin (44) but is of the same magnitude as the csSRR measured by 35 S microautoradiography for core-bound SRB (Ն8 ϫ 10 Ϫ17 mol of SO 4 cell Ϫ1 h Ϫ1 ) from the same environment (21). The maximum ε 34 S attained during retentostat cultivation, Ϫ20.9‰ in run 2, is much greater than that for the batch culture performed with the same temperature, pH, and lactate/ SO 4 2Ϫ ratio, Ϫ9.7‰. The ε 34 S trends for all three experiments are similar, becoming increasingly negative during the exponential phase, attaining a peak negative value, which is then followed by a slow retreat from this peak negative value during the maintenance stage (Fig. 6). The decrease in ε 34 S during the exponential phase represents the combined effect of flushing and replacement of the initial HS Ϫ by microbially respired HS Ϫ . The decline in ε 34 S values during the maintenance stage ( Fig. 6) cannot be attributed to changes in the membrane transport rates or HS Ϫ production rates as a function of temperature as outlined in the Canfield et al. (12) model for SO 4 2Ϫ reduction, because the retentostat runs were isothermal in this region. The TEM data indicate that the cell surface-to-volume ratio increases from the end of the exponential stage to the maintenance stage. If vegetative cells express an inherently more negative ε 34 S than their resting cell counterparts, then one possible explanation for the slow decline in the magnitude of ε 34 S during the maintenance stage is the decrease in the number of vegetative cells and the increase in the number of resting cells. In order for the resting cells to yield a lesser ε 34 S than their vegetative progenitors, they would have to be more efficient in the conversion of intracellular SO 4 2Ϫ to HS Ϫ rather than less efficient. The observed decrease and subsequent increase in the ε 34 S as the temperature was decreased at the end of run 3, on the other hand, could conceivably be related to temperature-dependent changes in the membrane transport and SO 4 2Ϫ reduction rates as espoused by Canfield et al. (12) for their model.
The ε 34 S values during the maintenance stage did not exhibit any correlation the with S 2 O 3 2Ϫ or HS Ϫ concentrations. Comparing the results from all three experiments, the ε 34 S values during the maintenance stage also did not exhibit any correlation with the csSRR across 3 orders of magnitude (Fig. 6). This was surprising given that the SO 4 2Ϫ reduction pathway is the cumulative expression of multiple reductive and potentially reversible steps, each with its own ε 34 S (9, 12). By performing the retentostat experiments under electron donor-limited conditions and at a very low csSRR, we appear to have enhanced the reverse flow of S species in the cells as witnessed by the presence of S 2 O 3 2Ϫ in the medium, and the inferred ␦ 34 S of the S 2 O 3 2Ϫ is enriched in 32 S by 20 to 100‰. These observations are compatible with the hypothesis of Brunner and Bernasconi (9) that the final reduction of SO 3 2Ϫ to HS Ϫ is indeed a multistep process for D. putei. Although the retentostat experiments yielded greater ε 34 S values than did the batch culture, the final HS Ϫ product was still 40 to 50‰ less enriched in 32 S than the 70‰ enhancement predicted by Brunner and Bernasconi (9). Further retentostat studies of SRB should be performed at lower temperatures to attain rates comparable to that in nature to fully test this theory.