Photosynth Res DOI 10.1007/s11120-014-0061-1

REGULAR PAPER

Quantifying the effects of light intensity on bioproduction and maintenance energy during photosynthetic growth of Rhodobacter sphaeroides Saheed Imam • Colin M. Fitzgerald • Emily M. Cook Timothy J. Donohue • Daniel R. Noguera

•

Received: 1 April 2014 / Accepted: 19 November 2014 Ó Springer Science+Business Media Dordrecht 2014

Abstract Obtaining a better understanding of the physiology and bioenergetics of photosynthetic microbes is an important step toward optimizing these systems for light energy capture or production of valuable commodities. In this work, we analyzed the effect of light intensity on bioproduction, biomass formation, and maintenance energy during photoheterotrophic growth of Rhodobacter sphaeroides. Using data obtained from steady-state bioreactors operated at varying dilution rates and light intensities, we found that irradiance had a significant impact on biomass yield and composition, with significant changes in photopigment, phospholipid, and biopolymer storage contents. We also observed a linear relationship between incident light intensity and H2 production rate between 3 and 10 W m-2, with saturation observed at 100 W m-2. The light conversion efficiency to H2 was also higher at lower

Electronic supplementary material The online version of this article (doi:10.1007/s11120-014-0061-1) contains supplementary material, which is available to authorized users. S. Imam Program in Cellular and Molecular Biology, University of Wisconsin, Madison, WI, USA S. Imam  T. J. Donohue Department of Bacteriology, Wisconsin Energy Institute, University of Wisconsin – Madison, Suite 5166, 1552 University Avenue, Madison, WI 53726-4084, USA S. Imam  C. M. Fitzgerald  T. J. Donohue  D. R. Noguera DOE Great Lakes Bioenergy Research Center, University of Wisconsin, Madison, WI, USA C. M. Fitzgerald  E. M. Cook  D. R. Noguera (&) Department of Civil and Environmental Engineering, University of Wisconsin, Madison, WI, USA e-mail: [email protected]

light intensities. Photosynthetic maintenance energy requirements were also significantly affected by light intensity, with links to differences in biomass composition and the need to maintain redox homeostasis. Inclusion of the measured condition-dependent biomass and maintenance energy parameters and the measured photon uptake rate into a genome-scale metabolic model for R. sphaeroides (iRsp1140) significantly improved its predictive performance. We discuss how our analyses provide new insights into the light-dependent changes in bioenergetic requirements and physiology during photosynthetic growth of R. sphaeroides and potentially other photosynthetic organisms. Keywords Photosynthesis  Maintenance energy  Rhodobacter sphaeroides  Metabolic modeling  Bioenergetics  Hydrogen

Background Microbes hold great potential as industrial systems for economical production of high-value commodities (Gronenberg et al. 2013; Peralta-Yahya et al. 2012). To optimize these biological systems for efficient bioproduction, it is imperative we gain better knowledge of their physiological and bioenergetic processes, so improvements can be made in culture conditions or strain design. Photosynthetic microbes, with their ability to harness light energy, are the major contributor to global carbon cycling and can be instrumental in development of bioprocesses for sustainable energy and bioremediation (Atsumi et al. 2009; Gronenberg et al. 2013; Peralta-Yahya et al. 2012). Rhodobacter sphaeroides is perhaps the most well-studied photosynthetic bacterium and has been used as a model

123

Photosynth Res

system for investigating various aspects of photosynthesis (Imam et al. 2011; Mackenzie et al. 2007). In addition to its ability to grow under a variety of conditions (aerobic respiratory, anaerobic respiratory, and anoxygenic photosynthetic) and utilize a diverse array of growth substrates (Imam et al. 2013; Mackenzie et al. 2007), R. sphaeroides is also capable of producing significant quantities of valuable commodities including H2 (Kim et al. 2008; Kontur et al. 2011; Yilmaz et al. 2010), polyhydroxybutyrate (PHB) (Khatipov et al. 1999), and ubiquinone (Kien et al. 2010), as well as being able to fix CO2 and N2 (Atsumi et al. 2009; Connor and Atsumi 2010; Masepohl and Hallenbeck 2010; Wahlund et al. 1996). While several aspects of R. sphaeroides metabolic processes have been studied in detail, no detailed systems-level analysis of light absorption and its effects on bioenergetics and physiology has been conducted. In recent years, constraint-based modeling has proved to be a very useful tool for the systems-level analysis of cellular metabolism, yielding new insights into conditiondependent metabolic flux distributions and cellular physiology. Furthermore, constraint-based analysis can guide metabolic engineering efforts to optimize the production of cell biomass and metabolites of interest (Oberhardt et al. 2009). Previous work led to the construction and refinement of high-quality genome-scale metabolic models for R. sphaeroides, iRsp1095 (Imam et al. 2011) and iRsp1140 (Imam et al. 2013), which have been used for analysis of its metabolic processes (Carapezza et al. 2013; Imam et al. 2011, 2013). However, previous application of these and many other models of photosynthetic organisms have, with a few exceptions (Vu et al. 2012; Kliphuis et al. 2011), not considered measured light uptake as an important modeling constraint, generally relying on arbitrary thresholds and estimates (Montagud et al. 2010; Rex et al. 2013; Dal’Molin et al. 2011; Nogales et al. 2012). An important factor in determining the efficiency of a cell for bioprocessing is the amount of metabolic energy utilized for biomass production and maintenance of cellular homeostasis (Stouthamer and van Verseveld 1987; Russell and Cook 1995). In general, an ideal microbial cell factory would have low maintenance energy demands, allowing for the direction of more of its metabolic energy toward production of commodities of interest (Stouthamer and van Verseveld 1987; Tannler et al. 2008). There are several studies of how changes in light intensity impact growth of R. sphaeroides and other phototrophic microbes in batch or continuous culture (Aiking and Sojka 1979; Schumacher and Drews 1979; Golecki et al. 1980; Oelze 1988; Biel 1986; Campbell and Lueking 1983). However, the steadystate maintenance energy requirements for R. sphaeroides at different light intensities have not been determined. Knowledge of energy maintenance parameters is needed for the accurate prediction of biomass and metabolite

123

production using constraint-based models, and therefore, can be crucial for downstream strain design efforts. In this study, we aimed to (i) quantify the impact of light intensity on biomass formation, biomass composition, biomolecule production, and cellular maintenance energy requirements in R. sphaeroides; and (ii) use parameters estimated from these analyses to improve the calibration of iRsp1140 by updating biomass reactions and maintenanceassociated parameters. Using a series of chemostats run at varying dilution rates, we observed that biomass composition, as well as production of several target biomolecules (lipid, PHB, and H2) changed with variation in light intensity. We also found that the photosynthetic maintenance energy requirements were also significantly affected by changes in light intensity. The data were useful to understand the impact of light intensity on growth and metabolic flux distributions. Furthermore, the R. sphaeroides metabolic model, iRsp1140, updated with the experimentally determined biomass and maintenance energy parameters, was used to evaluate the impact of the model improvements in the prediction of biomolecule production. The incorporation of condition-dependent biomass and maintenance energy parameters significantly improved the predictive capability of iRsp1140. Overall, these analyses provide new insights into the light-dependent changes in bioenergetic requirements and cellular behavior during photosynthetic growth of R. sphaeroides.

Results and discussion The ability to efficiently convert available nutrient and energy sources into cell biomass or economically valuable end products is a desirable characteristic for microbes to be used in bioproduction. Thus, an evaluation of biomass and product yields is key to determining the suitability of candidate systems. Previous efforts at modeling R. sphaeroides metabolism resulted in the formulation of iRsp1140 (Imam et al. 2013). As with any model, its predictive ability depends on good calibration to experimental observations. Since light intensity can influence metabolic processes in photosynthetic organisms, we studied its effects on biomass yield and composition, biomolecule production, and ATP requirements for maintenance to better understand the impact of light on R. sphaeroides physiology. We also aimed to obtain lightdependent parameters from these analyses to better constrain iRsp1140 and thereby improve its predictions. Quantifying the effects of light and nutrient availability on biomass yield To assess the impact of light intensity on R. sphaeroides physiology and bioenergetics, we set up bioreactors

Photosynth Res

irradiated at low (3 and 5 W m-2), moderate (10 W m-2), and high (100 W m-2) light intensities (corresponding to *20, 33.5, 66.9, and 669 lmol photons m-2 s-1, respectively, at 800 nm). At each light intensity, multiple chemostats were run at retention times ranging between 8 and 39 h (i.e., dilution rates between 0.026 and 0.125 h-1). Cells were harvested to determine biomass and PHB compositions. The uptake rates for the components in the media with highest concentration, succinate, glutamate, aspartate, and sulfate were empirically determined, while H2 production was measured prior to harvesting cells. The specific light supply rate (SLSR), in mmol photons gDW-1 h-1, was also calculated from the measured biomass and light intensities (Table S1). Effect of light energy on biomass yield During photoheterotrophic growth, the cell’s ability to produce biomass is dependent on both available light and organic substrate(s). Under light-saturating conditions, biomass yield would be expected to be dependent on organic substrate availability. Conversely, under light limiting (and nutrient-replete conditions), biomass yield would be a function of light availability. To determine how varying light intensities affect growth yields, we assessed the relationship between SLSR and the dilution rate (D) (i.e., the specific growth rate). We observed a linear relationship for reactors illuminated at 3, 5, and 10 W m-2, with R2 of 0.9, 0.95, and 0.8 (p values of 0.0016, 0.024, and 0.0018), respectively (Fig. 1a), indicating that the supplied light was a limiting factor on steady-state growth. However, there was no correlation between the SLSR and D at 100 W m-2 (R2 = 0.007) (Fig. S1a), indicating that light is saturating at this intensity and suggesting that the

b

120 y = 838.65x + 16.708 R² = 0.8

100 80

y = 632.01x + 5.4468 R² = 0.9

60 40

y = 651.76x + 0.8287 R² = 0.95

3 W m-2 10 W m-2 5 W m-2

20

Cell dry weight (mgCOD/L)

Light supply rate (SLSR) (mmol photons gDW-1 h-1)

a

majority of the supplied light was not being used by the cells. The biomass yields on light energy (Yx,E) (i.e., the amount of biomass formed per mol of photons supplied) were calculated from the inverse of the slopes in Fig. 1a (Zijffers et al. 2010) to be 0.0016, 0.0015, and 0.0012 gDW mmol photons-1 at 3, 5, and 10 W m-2, respectively. These represent high biomass yields that are close to theoretical maximum biomass yields predicted for green algae (Zijffers et al. 2010), and are similar to those reported for other phototrophs (Kliphuis et al. 2011; Zijffers et al. 2010), though the growth mode used here is photoheterotrophic as opposed to photoautotrophic in the other studies. The lower biomass yield observed at 10 W m-2 could reflect the nitrogen source becoming limiting at this light intensity, which was not the case at the lower intensities (Table S1). Alternatively, it could indicate a drop in efficiency of light utilization. In addition, the average biomass yield on light at 100 W m-2 was 0.00013 gDW mmol photons-1, which is only *10 % of that observed at lower to moderate light intensity. This again is consistent with light saturation at this intensity. While most of the supplied light, under sub-saturating conditions, is apparently diverted toward biomass formation, cells also require additional energy for non-growthassociated processes including maintenance of cellular homeostasis. This additional energy requirement is obtained from the y-intercept in regression plots of biomass yield on light (Fig. 1a) (Zijffers et al. 2010). At sub-saturating light intensities, the light-derived maintenance energy requirements were 5.4, 0.83, and 16.7 mmol photons gDW-1 h-1 at 3, 5, and 10 W m-2, respectively. These values suggest that a greater amount of energy is required for cellular maintenance at higher light intensities.

2500 2000

R² = 0.92 1500

3 W m-2

R² = 0.92 1000

10 W m-2 5 W m-2

500

R² = 0.8

0

0 0

0.05

Dilution rate

0.1

0.15

(h-1)

Fig. 1 Relationship of dilution rate to SLSR and biomass production. Regression of a specific light supply rate (SLSR) against dilution rate D and b cell dry weight against D, at 3 (blue-filled diamond), 5

0

0.02

0.04

0.06

Dilution rate

0.08

0.1

0.12

(h-1)

(green-filled triangle), and 10 (red-filled rectangle) W m-2. There was no statistically significant correlation between these measurements at 100 W m-2 (Fig. S1)

123

Photosynth Res

This could be a reflection of increased turnover of cellular components induced by light-dependent stress or an increased amount of energy diverted toward energydependent maintenance of redox homeostasis via hydrogen production. We also observed a high correlation between the measured cell dry weight (DW) and D for reactors run at subsaturating light intensities, with R2 values ranging from 0.8 to 0.92 (Fig. 1b). The observed negative correlation between cell mass and D is consistent with cells growing under limiting conditions, as higher dilution rates result in less dense cultures. In these cultures, organic substrates in the media were abundant (Table S1), indicating that light intensity was the main limiting factor under these conditions. At the saturating light intensity of 100 W m-2, we did not observe a significant correlation between biomass and D (Fig. S1b), an observation that suggests light is not limiting under these conditions. These data indicate that light is the limiting factor on biomass yield at 3, 5, and 10 W m-2 and that R. sphaeroides is capable of achieving relatively high biomass yields under these conditions. These data also indicate that as the incident light increases, an increasing amount of light energy is devoted toward non-growth-associated processes, which could include maintenance of redox poise and coping with light-dependent stress. Alternatively, some supplied light might be unutilized or dissipated through mechanisms for non-photochemical quenching.

Effect of nutrient availability on biomass yield In contrast to the relationships between SLSR and D, the correlation between nutrient uptake and D was poor at subsaturating light intensities (Fig. S2). There was very little correlation between uptake of succinate (the major carbon source) or sulfate and D at these light intensities, indicating that neither of these was a limiting nutrient. However, glutamate (the main nitrogen source) uptake showed a better correlation to D (R2 [ 0.6) at these light intensities (Fig. S2). On the other hand, there were high correlations between uptake of all nutrients measured and D at an incident light intensity of 100 W m-2 (R2 [ 0.93) (Fig. S2). The observed biomass yields from carbon uptake (Yx,C) at 5, 10, and 100 W m-2 were 0.64, 0.61, and 0.57 gCODx gCOD-1 c , respectively (Fig. S3). However, the fit was much poorer at 3 W m-2, indicating organic substrates were not limiting. Overall, these observations are consistent with a model in which incident light intensity is the main constraint on growth at sub-saturating light intensities of 10 W m-2 and below. Under saturating light conditions (100 W m-2), the major limitation to growth becomes the rate of organic substrate utilization; hence, the

123

strong correlation between nutrient uptake and biomass observed under these conditions. Effect of light intensity on steady-state biomass composition Previous analyses of batch cultures have shown that light intensity significantly impacts R. sphaeroides physiology. In particular, the amount of light-capturing pigments produced increases significantly with decreases in incident light intensity (Aagaard and Sistrom 1972; Chory and Kaplan 1983). However, as these analyses were not conducted under steady-state conditions, we assessed the impact of changing light regimes on R. sphaeroides biomass composition in continuous culture. Incident light intensity had only a small effect on cellular protein content, which was slightly higher at low light intensities (3 and 5 W m-2) compared to moderate and high light intensities (10 and 100 W m-2) (Fig. 2a). On the other hand, light intensity had a significant, and previously unreported, impact on PHB production in R. sphaeroides. At low light intensities (3 and 5 W m-2), PHB made up only *3 % of the DW, which was more than fourfold less than that observed at moderate and high light intensities, where PHB accounted for 13–14.8 % of the total cell DW, respectively (Fig. 2b). PHB serves as a major route for dissipating excess electrons in R. sphaeroides (Yilmaz et al. 2010), as its production requires significant input of reducing power in the form of NADPH, which is mainly generated via the energy-dependent transhydrogenase reaction (Imam et al. 2013). Thus, the low levels of PHB formed at low light intensities may reflect the energy limitations under these conditions. Consistent with previous observations in batch cultures, the cellular bacteriochlorophyll (Bchl) content was dependent on incident light intensity. The cellular Bchl content was 0.71 and 0.91 % of DW at 3 and 5 W m-2, respectively. In contrast, a significant drop in Bchl content was observed at moderate (0.53 %) and high (0.17 %) light intensities, representing an *5 fold difference between high and low light conditions (Fig. 2c). These observations are consistent with the need for more photopigments at lower light intensities to absorb the limited amount of incident light. During photosynthetic growth, R. sphaeroides significantly increases its membrane surface area by forming intracytoplasmic membranes to house its photosynthetic apparatus (Kiley and Kaplan 1988). While varying light intensity resulted in significant changes in Bchl content, the relative magnitude of the change in cellular phospholipid levels was smaller. The highest amount of phospholipids was obtained for cells grown at 5 W m-2 (*6 % of DW), while the lowest was observed for cell grown at

Photosynth Res

a 80

b % Cell dry weight

% Cell dry weight

16 60

40

20

12

8

4

0

0

3W

m-2

5W

m-2

10 W

m-2

100 W

3 W m-2

m-2

5 W m-2

10 W m-2 100 W m-2

Light intensity

Light intensity

d

c 1.0

7

% Cell dry weight

% Cell dry weight

6 0.8 0.6 0.4

5 4 3 2

0.2 1 0.0

3 W m-2

5 W m-2

10 W m-2

100 W m-2

Light intensity

0

3 W m-2

5 W m-2

10 W m-2

100 W m-2

Light intensity

Fig. 2 Biomass composition is significantly affected by light intensity. Assessment of the contribution of a protein, b PHB, c bacteriochlorophyll, and d phospholipids to cell dry weight at 3, 5, 10, and 100 W m-2 light intensities. The values plotted in the bar charts

represent the average amounts of the various biomass components across all chemostats run at each light intensity. The error bars represent the standard error of mean of all chemostats at each light intensity

100 W m-2 (*4 % of DW), only a 1.5-fold difference (Fig. 2d). The light-dependent changes in biomass composition identified here are useful for generating conditiondependent biomass reactions (Table S2) for simulation of growth and flux distributions at different light intensities.

production helps maintain redox homeostasis by serving as a sink for excess reducing equivalents (Imam et al. 2011; Kontur et al. 2012; Yilmaz et al. 2010). Given the high energy demands of this process, the amount of H2 produced might be altered by changing light intensity. We observed a positive linear relationship between light intensity and the amount of H2 produced. At low (3–5 W m-2) to moderate (10 W m-2) light intensities, the increase in H2 production appeared to be directly proportional to incident light, increasing at a rate of *0.25 mmol gDW-1 h-1 of H2 production per unit of irradiance (*0.1 ml h-1 with the experimental conditions used; R2 [ 0.99; Fig. 3a). This observation indicates that light is a limiting factor for steady-state H2 production, consistent with previous observations in batch cultures (Miyake and Kawamur 1987), and it predicts that cells are using a proportional amount of the incident light for H2 production over this range. While there was an *1.7-fold increase in the

Effect of light intensity on metabolite secretion In addition to determining the effect of light intensity on R. sphaeroides biomass composition, we also analyzed its effects on selected metabolic end products at steady state. H2 production is dependent on light intensity Rhodobacter sphaeroides is capable of producing large amounts of H2 under photoheterotrophic conditions via ATP-dependent nitrogenase activity (Kontur et al. 2011; Masepohl and Hallenbeck 2010; Yilmaz et al. 2010). H2

123

Photosynth Res

b

4

H2 production (mmol gDW -1 h-1)

H2 production (mmol gDW -1 h-1)

a

3

2

y = 0.2454x + 0.4034 R² = 0.9994 1

6 5 4 3 2 1 0

0 0

5

Incident light (W

10

15

m-2)

0

50

Incident light (W

100

m-2)

Fig. 3 Relationship of H2 production rate to light intensity. a Regression of average H2 production rate against light intensity between 3 and 10 W m-2. b Plot of average H2 production against light intensity

extended to 100 W m-2, showing loss of linear relationship due to light saturation

amount of H2 produced when comparing moderate (10 W m-2) to high (100 W m-2) light, the increase was not proportional to light intensity, providing another indication of light saturation at 100 W m-2 (Fig. 3b), and suggesting that H2 production had reached its maximum rate [an average of *4.9 mmol gDW-1 h-1 (or *1.28 ml h-1)] under these conditions. The efficiency of light energy conversion is another useful parameter for assessing the effectiveness of photobiological H2 production (Koku et al. 2002; Miyake and Kawamur 1987). We found that the light energy conversion efficiency of steady-state cultures was relatively high at low and moderate light intensities with average efficiencies of 4.75, 6.35, and 5.77 % at 3, 5, and 10 W m-2, respectively. The energy conversion efficiencies obtained at these light intensities are in general agreement with reported values for cells grown in batch cultures (Koku et al. 2002; Miyake and Kawamur 1987). On the other hand, at 100 W m-2, the light conversion efficiency dropped down to just 0.77 %, consistent with light being in excess. In addition, to the light conversion efficiency, another important parameter is the substrate conversion efficiency (i.e., ratio of observed H2 produced to the theoretical maximum obtainable with the main growth substrate, succinate in our chemostats) (Koku et al. 2002; Sasikala et al. 1993). At 3 W m-2, the substrate conversion efficiency was *21 %, while the efficiencies significantly increased at higher light intensities—40, 34 and 34 % at 5, 10, and 100 W m-2, respectively. This suggests a significant energy limitation at 3 W m-2 that negatively impacts the cell’s ability to convert the growth substrate into H2.

Light intensity alters end product profiles

123

In addition to H2, R. sphaeroides excretes soluble microbial products (SMP), such as TCA cycle intermediates, lactate, and pyruvate, when growing in batch cultures (Yilmaz et al. 2010). To determine if the amount of SMPs produced during steady-state growth was affected by light intensity, we compared the measured chemical oxygen demand (COD) of the spent media to its calculated COD (based on measured extracellular concentrations of succinate, glutamate, and aspartate; see ‘‘Materials and methods’’ section). For chemostats operated at low light intensities (3 and 5 W m-2), the measured and estimated COD were within 10 % of each other. Conversely, at moderate to high light intensities (10 and 100 W m-2), the difference in measured and estimated COD was *20 %. When we analyzed the spent reactor media for organics, only small amounts of pyruvate, lactate, glycerol, xylitol, and xylose were observed. Thus, the chemical identity of most of the SMPs in steady-state cultures remains unresolved. In addition, the levels of these metabolites did not change significantly across the reactors. The increased extracellular COD observed at higher light intensities could reflect altered metabolic flexibility in these cultures, where conserving metabolic energy by nutrient oxidation is less of a constraint on the cells due to the increased availability of light energy. Alternatively, it could be an indication of increased cellular stress or photo-damage at higher light intensities, with the increased extracellular COD coming from secretion of cellular constituents into the media.

Photosynth Res

Effect of light intensity on maintenance energy requirements

our analysis, these differences did not appear to be significant.

The energy a cell expends for growth and for maintaining cellular homeostasis, is an important factor in determining its performance in the laboratory or in industrial settings (Stouthamer and van Verseveld 1987). Thus, we assessed the maintenance energy requirements of R. sphaeroides at varying light intensities to gain a better understanding of its bioenergetic requirements and limitations. In metabolic flux models, maintenance energy is divided into two major components: the growth-associated maintenance (GAM) and non-growth-associated maintenance (NGAM) energy requirements, used for biomass formation and maintenance of cellular homeostasis, respectively (Thiele and Palsson 2010). For photosynthetic organisms, the total amount of ATP predicted to be produced will depend on the efficiency with which the incident light is utilized by the cells. This efficiency of light utilization (U), which is dependent on the light absorbing capacity of the photosynthetic apparatus, can also be significantly impacted by cell shading, back scatter of light by cells, and reactor configuration (depth, etc.). Since U is an unknown parameter that influences the calculation of maintenance energy requirements (Table S3), we used iRsp1140 to computationally assess its impact on modeling predictions and to select an appropriate value. We constrained iRsp1140 using (i) the determined condition-dependent biomass compositions at varying light intensities; (ii) the measured uptake rates for succinate, glutamate, aspartate, and sulfate; and (iii) the observed steady-state growth rates to predict maximum ATP hydrolysis rates at candidate Us of 10, 25, 50, 75, and 100 %, and at incident light intensities of 3, 5, and 10 W m-2. Since the relationship between maximum ATP hydrolysis and observed growth rate is expected to result in a good linear fit, we selected U = 75 % as this produced the best fit for the sub-saturating light intensities (Fig. S4). Light utilization efficiencies of 10 and 25 % resulted in significantly poorer fits between the predicted ATP hydrolysis rates and l, whereas 75 % efficiency provided marginally better fits than 50 and 100 % (Fig. 4a, Fig. S4). Interestingly, the selected value of U is similar to the maximum light utilization efficiency of 80 % estimated for algal cells (Kliphuis et al. 2011). Nevertheless, given that the goodness of fit was relatively insensitive to U between 50 and 100 %, the selected U should not be interpreted as a unique descriptor of the physiological state of the cells. It should also be noted that U can vary with light intensity used, as efficiency might be expected to drop at light intensities that exceed the photosynthetic capacity of the cell. However, at the sub-saturating light intensities used in

Maintenance energy requirements vary with light intensity At light intensities of 3 and 5 W m-2, we estimated the GAM to be 282 and 297 mmol ATP gDW-1, respectively, while NGAM was estimated to be 2.6 and 0.5 mmol ATP gDW-1 h-1, respectively (Fig. 4b). Overall, there did not appear to be significant differences between these parameters at these two light intensities, which might be expected given the small difference in light intensities and the similarities in steady-state biomass composition under these conditions (Fig. 2). At a light intensity of 10 W m-2, the calculated GAM was 386 mmol ATP gDW-1, which is significantly higher than that obtained at lower light intensities (Fig. 4b). This likely reflects the significant differences in biomass composition between cells grown under these conditions. Furthermore, the estimated NGAM was 8.4 mmol ATP gDW-1 h-1, significantly higher than that observed at the two lower light intensities. This larger NGAM value could be the result of increased stress (heat, photochemical, or others) at higher light intensities. Alternatively, this difference could reflect the increased ATP required to maintain homeostasis, for instance, via ATP-dependent H2 production (Kontur et al. 2011), which is significantly higher at 10 W m-2. Incorporating ATP-dependent H2 production constraints into iRsp1140 with the calculated GAM and NGAM parameters showed that requiring additional ATP for this process significantly deteriorates the model’s predictions as light (and thus ATP) is limiting (Fig. S5). This suggests that ATP utilized for H2 production is already accounted for in the estimated maintenance energy parameters. As expected, the iRsp1140-derived maintenance energy values are very similar to the related, previously estimated, light-derived maintenance energy values obtained from regressing SLSR on D (Fig. 1a). Given that two photons lead to generation of *1 ATP during cyclic photophosphorylation (Allen 2003), the GAM (NGAM) calculated from that analysis is equivalent to 316 (2.7), 326 (0.4), and 419 (8.4) mmol ATP gDW-1 (mmol ATP gDW-1 h-1) at 3, 5, and 10 W m-2, respectively, very similar to that predicted by iRsp1140. The small differences between these values likely result from aspects of ATP utilization accounted for elsewhere in iRsp1140 (such as charging of tRNAs with amino acids), as well as the inclusion of nutrient-derived ATP in the model’s predictions and the rotational fold symmetry of R. sphaeroides ATP synthase. Overall, this indicates that the vast majority of energy used for growth and maintenance of cellular homeostasis during photoheterotrophic growth is derived from photon

123

Photosynth Res

a

b

1

60

0.8 0.7 0.6

3 W m-2 5 W m-2 10 W m-2

0.5 0.4 0.3 0.2 0.1 0 0

25

50

75

100

% Efficiency of light utilization

Max. ATP hydrolysis (mmol gDW-1 h-1)

Goodness of fit (R2)

0.9

y = 386.1x + 8.4 R² = 0.74

50 40 30

3 W m-2 10 W m-2 5 W m-2

y = 282.5x + 2.6 R² = 0.86

20 y = 297.6x + 0.5 R² = 0.93

10 0 0

0.05

0.1

Specific growth rate μ

0.15

(h-1)

Fig. 4 Assessing photosynthetic maintenance energy in R. sphaeroides. a Comparison of the goodness of fit (i.e., R2 values) of the regression of maximum ATP hydrolysis on specific growth rate l considering a range of efficiencies of light utilization (i.e., 10, 25, 50, 75, and 100 % of measured light intensity) in each simulation. The analysis was conducted with data obtained from chemostats operated

at 3 (blue-filled diamond), 5 (green-filled triangle), and 10 (red-filled rectangle) W m-2 light intensity. The best fit obtained across all data was at 75 % efficiency. b Regression of maximum ATP hydrolysis against specific growth rate l at 3 (blue-filled diamond), 5 (greenfilled triangle), and 10 (red-filled rectangle) W m-2 using a U of 75 %

absorption, leaving energy in organic substrates to be utilized for bioproduct formation. We could not use similar methods to accurately calculate maintenance energy requirements at 100 W m-2 as a plot of maximum ATP hydrolysis versus growth rate under these conditions did not show any significant correlation (Fig. S6). This observation might reflect the fact that light is saturating at this intensity, so only a small fraction of incident light is being used by cells under these conditions. However, utilizing more conservative estimates of light absorption efficiency did not improve these results (Fig. S6). An alternative explanation is that there is increased cellular damage from light, heat, photochemical stress, and other factors at high light intensities, which would complicate attempts to estimate maintenance energy under these conditions.

photons being supplied in excess under these conditions (Fig. 5a). The above analysis also suggests that LUR can have a significant impact on the observed and predicted behavior of photosynthetic cells. Thus, if LUR is ignored or inaccurately measured, it could have a negative effect on the validity of computational predictions. Indeed, plotting the predicted optimal growth rates against observed specific growth rates (i.e., dilution rate), when LUR and all measured uptake and production rates are taken into account (Fig. 5b), and comparing this to a plot of predictions generated by assuming that light is not limiting (i.e., omitting LUR), showed that inclusion of the LUR significantly improves the predictions of iRsp1140 (Fig. 5c). The fit between predicted and observed growth rates improves from an R2 of 0.63 (p = 2.9 9 10-5) to 0.85 (p = 7.2 9 10-9) with the inclusion of the measured LUR. Only cultures grown at low to moderate light intensities (3–10 W m-2) were considered for this analysis since at saturating light (100 W m-2), the SLSR is not an adequate estimation of optimal LUR (Fig. 5a). It should also be noted that inclusion of the experimentally determined parameters for GAM and NGAM also significantly improved the predictions of iRsp1140 compared to using the previously estimated maintenance energy requirements (Imam et al. 2011) (Fig. 5d). Since the determined GAM and NGAM parameters were estimated using all of the data generated in this study (at 3, 5, and 10 W m-2), a cross-validation of these parameters was carried out (Geisser 1975; Stone 1974). For each light intensity, we re-computed the GAM and NGAM values using all but one of the chemostat measurements for that light intensity. We then used the newly computed

Improvements in model predictions Our comprehensive analysis of light effects on biomass yield, composition, and maintenance energy requirements indicates that, at low and moderate light intensities, photons are a significant growth limiting factor, while this is not the case at high light intensities. Consistent with those observations, iRsp1140 predicts that at low to moderate light intensities (3–10 W m-2), the measured SLSR and the specific light uptake rate (LUR) required to achieve optimal growth (optimal LUR), given the substrate uptake rates, are comparable (Fig. 5a). In most cases, the measured SLSR is less than the predicted optimal LUR, consistent with photons being a growth limiting factor. Conversely, the measured SLSR was about 10 times the predicted optimal LUR at 100 W m-2, consistent with

123

5 W m-2

Light (mmol photons gDW-1 h -1)

a

10000 100 W m-2

1000

100

10

0.1

SLSR

R² = 0.851

0.08 0.06 0.04 0.02 0

0

Increasing growth rate

0.05

0.1

0.15

Observed specific growth rate μ (h-1)

d

0.12

Predicted growth rate (h-1)

R² = 0.6303

Predicted growth rate (h-1)

0.12

Optimal LUR

1

c

10 W m-2

3 W m-2

b Predicted growth rate (h-1)

Photosynth Res

0.1 0.08 0.06 0.04 0.02 0

0.12 R² = 0.6256

0.1 0.08 0.06 0.04 0.02 0

0

0.05

0.1

0.15

Observed specific growth rate μ (h-1)

0

0.05

0.1

0.15

Observed specific growth rate μ (h-1)

Fig. 5 Assessing the impact of SLSR, LUR, and maintenance energy on modeling predictions. a Plot comparing the measured SLSR and the specific LUR predicted by iRsp1140 to be required for optimal growth via FBA (optimal LUR), given the measured substrate uptake rates. Comparison of the observed and iRsp1140 predicted specific

growth rates using all measured parameters including LUR (b) or excluding LUR (c) as a constraint for FBA predictions. d Comparison of iRsp1140 growth rate predictions to the observed growth rates using previously estimated maintenance energy parameters (Imam et al. 2011)

GAM and NGAM values as input for the model, which was then constrained with uptake data from the chemostat that was left-out and used to predict the growth rate for that experiment. This rotation estimation analysis was conducted for all data points at each light intensity (Fig. S7). While we observed an R2 reduction from 0.85 to 0.80 (Fig. 5b and Fig. S7), the correlation was still significant (p = 8.7 9 10-8), indicating the ability of the trained model to reproduce the observation for experiments not included in the training dataset. In addition, the mean squared errors, calculated as the mean of the squared difference between predicted and observed growth rates across all light intensities, were smaller for both the crossvalidation and final model predictions (0.0001 and 0.00008, respectively) compared to predictions without the LUR constraint (i.e., Fig. 5c) or with the previously estimated maintenance parameters (i.e., Fig. 5d) (0.0002 and

0.00015 respectively). Thus, within the limits of reproducibility of experimental data generated in this study, the model updated with the newly estimated GAM and NGAM, and constrained with LUR, shows improved predictive capabilities. We also assessed iRsp1140’s ability to predict the production rate of metabolites such as H2 and PHB, which were measured as part of our analyses, but not used as input for calculating GAM and NGAM. We observed a good correlation between the predicted and observed values (R2 = 0.53, p = 0.005) for simulations conducted with LUR included as a constraint (Fig. 6a). Conversely, when LUR was excluded, no significant correlation was observed between the measured and predicted H2 production rates (R2 = 0.001) (Fig. 6b). Furthermore, simulations conducted using the previously estimated biomass and maintenance energy requirements (Imam et al. 2011) also

123

Photosynth Res

Predicted H2 (mmol gDW -1 h-1)

8

R² = 0.53

6

4

2

0 0

2

4

6

Predicted H2 (mmol gDW -1 h-1)

b

a

25

R² = 0.001

20 15 10 5 0 0

Observed H2 (mmol gDW -1 h-1)

4

Observed H2 (mmol

gDW -1

6

h-1)

25

R² = 0.0043

20 15 10 5 0 0

2

4

6

Observed H2 (mmol gDW -1 h-1)

Predicted H2 (mmol gDW -1 h-1)

d

c Predicted H2 (mmol gDW -1 h-1)

2

25

R² = 0.004

20 15 10 5 0 0

2

Observed H2 (mmol

4

gDW -1

6

h-1)

Fig. 6 Prediction of H2 production. Plots comparing the observed and iRsp1140 predicted maximum H2 production rate using all measured parameters including LUR (a) or excluding LUR (b) as a constraint for FBA predictions. c and d show the equivalent simulations using the previously estimated maintenance energy

parameters (Imam et al. 2011), with and without LUR as a constraint, respectively. For each simulation, the observed growth rate was used as an additional constraint, while H2 production was used as the objective function

showed poor correlation to observed H2 production rates, regardless of whether LUR was used as a constraint or not. These results point to an improved ability of the updated instance of iRsp1140 to predict H2 production in R. sphaeroides and highlight the importance of using experimentally determined LUR as a constraint. While inclusion of these newly estimated parameters and LUR also significantly improved iRsp1140’s ability to predict PHB production rate (Fig. S8), the agreement between maximum predicted and observed production rates was less than that for H2 production (R2 = 0.19). This suggests that iRsp1140 is not yet sufficiently well constrained to accurately predict the production rate of this particular metabolite.

(Mahadevan and Schilling 2003). However, with the application of additional relevant constraints, one can significantly reduce the size of the solution space (Reed 2012). We observed that LUR, in particular, has a significant impact on the solution space of iRsp1140 predictions. Using flux variability analysis (FVA) (Mahadevan and Schilling 2003), we assessed the range of predicted flux values for each reaction at optimal growth rate under lightlimiting and -saturating conditions (i.e., with and without use of the specific LUR as an additional constraint). Using all available nutrient uptake data from a representative chemostat operated at a light intensity of 10 W m-2, a total of 795 reactions were able to carry flux under light-saturating conditions, compared to 448 when the LUR constraint was imposed. Of these 448 reactions, which carried flux under both conditions, a total of 342 reactions (76.3 %) showed no variability in flux under light-limiting conditions, compared to only 123 reactions (27.5 %) with invariable fluxes under light-saturating conditions. These results indicate that there is a significant reduction in the solution space when the LUR constraint is included.

Impact of light limitation on the metabolic solution space While constraint-based analysis can be useful for studying metabolism on a genome-scale, one of its limitations lies in the large solution space of feasible flux distributions that can result from optimizing of a given objective function

123

Photosynth Res

Predicted flux range (mmol gDW-1 h-1)

3

A

B

C

D

E

F

G

H

I

J

K

L

M

N

O

P

Q

2

1

0

-1

-2 TCA cylce

Glycolysis

Fig. 7 Variability of predicted flux distributions under light-limiting and -saturating conditions. Box plots of flux distribution for reactions of the TCA cycle (A–J) and glycolytic pathway (K–Q) in R. sphaeroides under photoheterotrophic conditions. The first box plot in each group represents the flux distribution under light-constrained conditions, while the second box represents flux distributions with unconstrained light. Distributions were obtained by sampling 1,000 alternative optimal solutions. It should be noted that outliers were omitted from the box plots for the unconstrained light analysis to allow better visualization of the differences in predicted flux ranges. Data used in the simulations corresponded to a chemostat run at 10 W m-2.

In both sets of simulations, the model was constrained with the uptake rates for succinate, glutamate, aspartate, sulfate, as well as production rates of PHB and H2. For the light-constrained simulation, LUR was set to the measured SLSR. A citrate synthase; B aconitase; C isocitrate dehydrogenase; D and E a-ketoglutarate dehydrogenase; F succinylCoA synthetase; G succinate dehydrogenase; H fumarate reductase; I malate dehydrogenase; J phosphoenolpyruvate carboxykinase; K triosephosphate isomerase; L glyceraldehydes-3-phosphate dehydrogenase; M phosphoglycerate kinase; N phosphoglyceromutase; O phosphopyruvate hydratase; P fructose-bisphosphate aldolase; Q D-fructose 1,6-bisphosphatase

Examination of the fluxes through central metabolism helps illustrate this finding (Fig. 7), as the fluxes through most TCA cycle reactions show no change over the optimal solution space under light-constrained conditions, whereas significant variability in these fluxes is predicted with unconstrained light. This reduced solution space likely reflects the energy limitations under light-limiting conditions and predicts that flux through energy demanding reactions will be avoided or minimized when light is limiting.

potential as contributors to biotechnological innovation, as their ability to efficiently harness solar energy and produce value-added commodities can help develop a more sustainable society. Well-studied photosynthetic microbes, such as R. sphaeroides, can provide high-value products and guide future research given its well-studied photochemical cycle and the available genetic, genomic, and computational tools available to study this organism. To take advantage of photosynthetic systems, it is imperative we gain an improved understanding of their physiology and energetics, and the contribution of light intensity to these processes. In this study, we gained a better quantitative understanding of the impact of light on a photosynthetic organism by examining its effects on the physiology and bioenergetics of R. sphaeroides. We found significant light-

Conclusions Photosynthetic organisms play a key role in global carbon cycling and the planetary food chain. They also have

123

Photosynth Res Table 1 Properties of R. sphaeroides during photoheterotrophic growth 3 W m-2

5 W m-2

10 W m-2

100 W m-2

Carbon

Succinate

Succinate

Succinate

Succinate

Nitrogen

Glutamate

Glutamate

Glutamate

Glutamate

Biomass yield on light, Yx,E (gDW mmol photons-1)

0.0016 ± 0.0002

0.0015 ± 0.0002

0.0012 ± 0.0002

ND

Biomass yield on substrate, Yx,C (gCODx gCOD-1 c )

ND

0.64 ± 0.17

0.61 ± 0.07

0.57 ± 0.04

Substrates

Yields

Maintenance energy Light maintenance (mmol photons gDW-1 h-1)

5.4 ± 6.4

0.8 ± 5.4

16.7 ± 11.7

ND

GAM (mmol ATP gDW-1)

282 ± 51

298 ± 58

386 ± 86

ND

NGAM (mmol ATP gDW-1 h-1)

2.6 ± 3.2

0.5 ± 3

8.4 ± 5.9

ND

Hydrogen Average H2 production (mmol gDW-1 h-1)

1.1 ± 0.2

1.7 ± 0.1

2.9 ± 0.4

4.9 ± 0.3

Substrate conversion efficiency (%)

21 ± 5

40 ± 7

34 ± 3

34 ± 4

Light conversion efficiency (%)

4.8 ± 0.7

6.4 ± 0.4

5.8 ± 0.2

0.8 ± 0.05

Biomass (% DW) Protein

71.1 ± 2.0

73.3 ± 1.3

60.1 ± 4.6

63.8 ± 3.6

PHB

3.0 ± 1.0

3.3 ± 0.7

13.0 ± 2.0

14.8 ± 0.9

Pigment

0.7 ± 0.2

0.9 ± 0.06

0.5 ± 0.05

0.2 ± 0.05

Lipid

4.5 ± 0.6

6.1 ± 0.2

5.5 ± 0.3

4.0 ± 0.3

dependent changes in biomass composition as well as PHB and H2 production in steady-state cells. Furthermore, our analysis showed that the metabolic energy required for cellular maintenance under photosynthetic conditions is significantly affected by light intensity, but it is also linked to resultant differences in biomass composition and the need to maintain redox homeostasis. At low light intensities, R. sphaeroides maintenance energy requirements were relatively low, suggesting that it could serve as an excellent microbial factory for bioproduction under these conditions. Maintenance energy increased significantly at higher light intensities, but with a tradeoff of producing significantly larger amounts of potentially useful end products like PHB and H2. We also showed that use of the condition-dependent parameters for biomass and maintenance energy can significantly improve predictions of iRsp1140. By incorporating measured photon uptake for the first time in modeling photosynthetic growth of R. sphaeroides, we also significantly improved the predictive ability of iRsp1140. Furthermore, we showed that the solution space of modeling predictions is significantly reduced when light-limiting conditions are considered, with only a handful of reactions in the network having variable fluxes under these conditions. Overall, these analyses have improved our understanding of the effects of light on R. sphaeroides metabolism, bioenergetics, and physiology. In addition, it has provided several new parameters (Table 1) that can be

123

used to improve the modeling and performance of this and related photosynthetic organisms for industrial uses.

Materials and methods Chemostat operation Rhodobacter sphaeroides 2.4.1 cells were grown on Sistrom’s Minimal Medium (SMM) (Sistrom 1960) with 8.1 mM glutamate used to replace ammonia as the main nitrogen source. Under these conditions, H2 production via nitrogenase activity is promoted (Kontur et al. 2011; Yilmaz et al. 2010). Cultures (16.5 ml) were maintained in continuously stirred, sealed 20 ml chemostats at 30 °C, which were illuminated by an incandescent light box at 3, 5, or 10 W m-2, or with an incandescent flood light source at 100 W m-2. Reported light intensities were measured with a Yellow–Springs–Kettering model 65-A radiometer through a Corning 7-69 filter [transmission 750–900 nm, i.e., the photosynthetically active radiation (PAR) for R. sphaeroides]. Culture turbidity was monitored using a Klett– Summerson photoelectric colorimeter (Klett MFG Co., NY). The chemostats were operated in batch-fed mode, in which a portion of spent media was wasted for 1 min immediately followed by a 3-min feeding cycle of an equivalent volume of new media. This cycle was repeated every 20 min. The feeding and wasting rates were varied to

Photosynth Res

attain a range of retention times varying from 8 to 39 h. Reactors were started in batch mode until cells reached an optical density of [100 Klett units (1 Klett unit equals *107 cells ml-1) (Imam et al. 2011). Cells were continuously fed with medium using Watson Marlow model 120U peristaltic pumps (Watson Marlow Inc., Wilmington, MA), and media were wasted using Masterflex model C/L peristaltic pumps (Cole-Palmer Instrument Co., Vernon Hills, IL). Reactors were checked approximately every 12 h, and when necessary, pumping was manually adjusted to correct for any changes in reactor volume due to imbalances of liquid inflow and outflow. Cultures were grown for at least five retention times and analyzed when steady state was established (verified from constant turbidity measures and/or constant gas production rates). Light supply rates The measured incident light intensity at the PAR measured in W m-2 was converted into standard units of photon flux (PF) of lmol photon m-2 s-1 at a wavelength of 800 nm using the equation: PF ¼

I  k  109 ; h  c  NA

where I is the irradiance in W m-2; k is the wavelength of light in nm; h is Planck’s constant (6.626068 9 10-34 m2 kg s-1); c is the speed of light (2.997925 9 108 m s-1); and NA is Avogadro’s number (6.02 x 1017 lmol-1). The PF was then converted into a SLSR in mmol gDW-1 h-1 for model simulations with the equation: SLSR ¼

PF  3600  Areactor  U ; DW  Vreactor

where PF is the photon flux in lmol photon m-2 s-1; Areactor is the surface area of the bioreactor in m2; DW is the cell DW in g l-1; Vreactor is the volume of the bioreactor in ml; and U is the efficiency of light utilization taken to be 0.75 based on best fits from regression of max. ATP production on l (see ‘‘Effect of light intensity on maintenance energy requirements’’ section). The biomass yield on light at 100 W m-2, which could not be determined from the regression of SLSR and D, was calculated as the average biomass yield as previously described (Zijffers et al. 2010).   Biomass yield gDW mmol photons1 DW  l  Vreactor ¼ PF  Areactor  3600 Light conversion to H2 efficiency was calculated (Koku et al. 2002) with the equation: % Efficiency ¼

33:61  qH2  VH2 ; IA

where qH2 is the density of H2 (0.08988 g/l); VH2 is the volume of H2 produced per hour in l hr-1; I is the irradiance in W m-2; and A is the total surface area of the bioreactor in m2. Substrate conversion to H2 efficiency was calculated using succinate as the substrate (Koku et al. 2002; Sasikala et al. 1993). Quantification of biomass and cell dry weight Samples from individual chemostats were centrifuged at 5,500 rcf at 4 °C to obtain cell pellets for biomass analysis. Total cellular protein was quantified via the Lowry assay (Hartree 1972; Lowry et al. 1951), while total cellular Bchl was determined spectrophotometrically (Cohen-Bazire et al. 1957). The phospholipid component was determined by total phosphorus assay on lipids extracted via standard chloroform/methanol extraction (Rouser et al. 1970). The PHB content of cells was determined by GC–MS (GC2010 gas chromatograph coupled to a QP-2010S mass spectrometer detector; Shimadzu Scientific) (Yilmaz et al. 2010). Other biomass components were assumed to remain constant based on previous measurements (Imam et al. 2011). Cell DW was calculated using the COD mass balance approach (Imam et al. 2011; Yilmaz et al. 2010). Quantification of nutrient uptake and biomolecule secretion To analyze the presence of media nutrients and products, culture supernatants were filtered through a 0.22-lm Nylon Filter (Fisher Scientific, Pittsburg, PA) and frozen at -80 °C until analysis. The uptake rates for succinate, glutamate, and aspartate were measured as previously described (Yilmaz et al. 2010) by comparing their concentrations in the filtered supernatant from the chemostat cultures to that in the initial media. Sulfate was quantified using Dionex 2100 Series Ion Chromatogram (Thermo Scientific) using EPA method 300.1. Ammonia concentrations were analyzed using the salicylate method (Method 10031, Hach Company, Loveland, CO). COD was analyzed using the reactor digestion method (Method 8000, Hach Company, Loveland, CO). Lactate, pyruvate, and malate were measured using GC– MS (Yilmaz et al. 2010). Glucose, xylose, pyruvate, xylitol, glycerol, formate, acetate, and ethanol were quantifying using high-performance liquid chromatography (HPLC) with a C18 column and a refractive index detector (RID-10A) on an Agilent system (Agilent Technologies, Santa Clara, CA), in collaboration with the Metabolomics Facility of the Great Lakes Bioenergy Research Center. Total gas production and the gas composition of the headspace were determined as previously described (Yilmaz et al. 2010).

123

Photosynth Res

Assessment of media for secretion of untargeted SMPs To test if other SMPs not targeted in our analysis might be present in the media, the measured COD of the culture supernatants was compared to the estimated COD of the supernatant. The estimated COD of the supernatant was calculated based upon the measured concentrations of succinate, glutamate, and aspartate in the culture supernatant. The estimated COD contribution for each compound was calculated using the following formula: X X ThCODi ¼ Ci þ 0:25  Hi þ 0:5 X X   Ni  0:75  Oi  ð32 mg O2 =mmolÞ  ½Conci ; where RCi, RHi, RNi and ROi are the number of carbon, hydrogen, nitrogen, and oxygen molecules, respectively, present in compound i; [Conci] is the concentration of compound i in units of mmol l-1. ThCODi is the estimated CODi in units of mgCOD l-1 (Rittmann and McCarty 2001). Constraint-based analysis The genome-scale metabolic model for R. sphaeroides, iRsp1140 (Imam et al. 2013), was used for all simulations. Flux balance analysis (FBA) (Varma and Palsson 1994) was used to simulate in silico growth by solving the linear programming problem: max vbiomass s:t: S  v¼0 vmin  v  vmax ; where mbiomass is the flux through biomass objective function; m is the vector of steady-state reaction fluxes; and mmin and mmax are the minimum and maximum allowable fluxes set to -1,000 and 1,000 mmol gDW-1 h-1 for reversible reactions and 0 and 1,000 mmol gDW-1 h-1 for forward only reactions. Measured uptake rates for light, succinate, glutamate, aspartate, and sulfate, as well as the measured growth rates and the production rates of PHB and H2 were used as initial input constraints for modeling when required. Other media components were allowed to be freely exchanged with the extracellular space. FVA (Mahadevan and Schilling 2003) and alternative optima analysis (Lee et al. 2000; Reed and Palsson 2004) were conducted as previously described (Imam et al. 2011). For assessing differences in flux distribution between lightlimiting and -saturating conditions, 1,000 alternative solutions were sampled from the optimal solution space, and

123

summary statistics obtained from these were used generate the box plots in Fig. 7. All simulations were conducted under the GAMS programming environment (GAMS Development Corporation, Cologne, Germany) using the CPLEX solver. Determination of maintenance energy requirements To determine the cellular maintenance energy requirements at a given light intensity, the maximum ATP hydrolysis rate for each reactor was calculated by constraining the model with uptake rates for light, succinate, glutamate, aspartate, and sulfate and the experimentally determined growth rate. At each light intensity, the photosynthetic biomass reaction of iRsp1140 (RXN1306) was replaced with a new biomass reaction corrected for condition-specific differences in biomass composition and removing the previously estimated GAM value of 53.65 mmol ATP gDW-1 from the reaction. Simulations were conducted by maximizing the ATP hydrolysis reaction (RXN0765) under these constraints. For each light intensity, the maximum ATP hydrolysis rate (qATP-total) for each reactor was plotted against the specific growth rate (l). The ATP balance for anaerobic photoheterotrophic growth can be considered as qATPcarbon þ qATPlight ¼ qATPtotal ¼ GAMATP  l þ NGAMATP where qATP-carbon is the ATP production rate from carbon sources taken up from the media (succinate, glutamate, and aspartate), qATP-light is the ATP production rate from light, and qATP-total is the ATP production/hydrolysis rate all in mmol gDW-1 h-1. GAMATP is the GAM energy requirement (in mmol gDW-1), and NGAMATP is the NGAM energy requirement (in mmol gDW-1 h-1). GAMATP and NGAMATP were determined from the slope and y-intercept, respectively, of the regression of qATP-total on l. Acknowledgments This work was funded in part by the Department of Energy, Office of Science, Great Lakes Bioenergy Research Center (DE-FC02-07ER64494), and the Genomics:GTL and SciDAC Programs (DE-FG02-04ER25627). SI was supported during part of this work by a William H. Peterson Predoctoral Fellowship from the University of Wisconsin-Madison Bacteriology Department.

References Aagaard J, Sistrom WR (1972) Control of synthesis of reaction centre bacteriochlorophyll in photosynthetic bacteria. Photochem Photobiol 15(2):209–225 Aiking H, Sojka G (1979) Response of Rhodopseudomonas capsulata to illumination and growth rate in a light-limited continuous culture. J Bacteriol 139(2):530–536 Allen JF (2003) Cyclic, pseudocyclic and noncyclic photophosphorylation: new links in the chain. Trends Plant Sci 8(1):15–19

Photosynth Res Atsumi S, Higashide W, Liao JC (2009) Direct photosynthetic recycling of carbon dioxide to isobutyraldehyde. Nat Biotechnol 27(12):1177–1180 Biel AJ (1986) Control of bacteriochlorophyll accumulation by light in Rhodobacter capsulatus. J Bacteriol 168(2):655–659 Campbell TB, Lueking DR (1983) Light-mediated regulation of phospholipid synthesis in Rhodopseudomonas sphaeroides. J Bacteriol 155(2):806–816 Carapezza G, Umeton R, Costanza J, Angione C, Stracquadanio G, Papini A, Lio P, Nicosia G (2013) Efficient behavior of photosynthetic organelles via Pareto optimality, identifiability, and sensitivity analysis. ACS Synth Biol 2(5):274–288 Chory J, Kaplan S (1983) Light-dependent regulation of the synthesis of soluble and intracytoplasmic membrane proteins of Rhodopseudomonas sphaeroides. J Bacteriol 153(1):465–474 Cohen-Bazire G, Sistrom WR, Stanier RY (1957) Kinetic studies of pigment synthesis by non-sulfur purple bacteria. J Cell Physiol 49(1):25–68 Connor MR, Atsumi S (2010) Synthetic biology guides biofuel production. J Biomed Biotechnol. doi:10.1155/2010/541698 Dal’Molin CG, Quek LE, Palfreyman RW, Nielsen LK (2011) AlgaGEM—a genome-scale metabolic reconstruction of algae based on the Chlamydomonas reinhardtii genome. BMC Genom 12(suppl 4):S5 Geisser S (1975) The predictive sample reuse method with applications. J Am Stat Assoc 70(350):320–328 Golecki JR, Schumacher A, Drews G (1980) The differentiation of the photosynthetic apparatus and the intracytoplasmic membrane in cells of Rhodopseudomonas capsulata upon variation of light intensity. Eur J Cell Biol 23(1):1–5 Gronenberg LS, Marcheschi RJ, Liao JC (2013) Next generation biofuel engineering in prokaryotes. Curr Opin Chem Biol 17(3):462–471 Hartree EF (1972) Determination of protein: a modification of the Lowry method that gives a linear photometric response. Anal Biochem 48(2):422–427 Imam S, Yilmaz S, Sohmen U, Gorzalski AS, Reed JL, Noguera DR, Donohue TJ (2011) iRsp1095: a genome-scale reconstruction of the Rhodobacter sphaeroides metabolic network. BMC Syst Biol 5:116 Imam S, Noguera DR, Donohue TJ (2013) Global insights into energetic and metabolic networks in Rhodobacter sphaeroides. BMC Syst Biol 7(1):89 Khatipov E, Miyake M, Miyake J, Asada Y (1998) Polyhydroxybutyrate accumulation and hydrogen evolution by Rhodobacter sphaeroides as a function of nitrogen availability. In: Zaborsky O, Benemann J, Matsunaga T, Miyake J, San Pietro A (eds) BioHydrogen. Springer, US, pp 157–161 Kien NB, Kong IS, Lee MG, Kim JK (2010) Coenzyme Q10 production in a 150-l reactor by a mutant strain of Rhodobacter sphaeroides. J Ind Microbiol Biotechnol 37(5):521–529 Kiley PJ, Kaplan S (1988) Molecular genetics of photosynthetic membrane biosynthesis in Rhodobacter sphaeroides. Microbiol Rev 52(1):50–69 Kim E, Lee M, Kim M, Lee JK (2008) Molecular hydrogen production by nitrogenase of Rhodobacter sphaeroides and by Fe-only hydrogenase of Rhodospirillum rubrum. Int J Hydrog Energy 33(5):1516–1521 Kliphuis AM, Klok AJ, Martens DE, Lamers PP, Janssen M, Wijffels RH (2011) Metabolic modeling of Chlamydomonas reinhardtii: energy requirements for photoautotrophic growth and maintenance. J Appl Phycol 24(2):253–266 Koku H, Eroglu I, Gunduz U, Yucel M, Turker L (2002) Aspects of the metabolism of hydrogen production by Rhodobacter sphaeroides. Int J Hydrogen Energy 27(11–12):1315–1329 Kontur WS, Ziegelhoffer EC, Spero MA, Imam S, Noguera DR, Donohue TJ (2011) Pathways involved in reductant distribution

during photobiological H2 production by Rhodobacter sphaeroides. Appl Environ Microbiol 77(20):7425–7429 Kontur WS, Schackwitz WS, Ivanova N, Martin J, Labutti K, Deshpande S, Tice HN, Pennacchio C, Sodergren E, Weinstock GM, Noguera DR, Donohue TJ (2012) Revised sequence and annotation of the Rhodobacter sphaeroides 2.4.1 genome. J Bacteriol 194(24):7016–7017 Lee S, Phalakornkule C, Domach MM, Grossmann IE (2000) Recursive MILP model for finding all the alternate optima in LP models for metabolic networks. Comput Chem Eng 24(2–7):711–716 Lowry OH, Rosebrough NJ, Farr AL, Randall RJ (1951) Protein measurement with the Folin phenol reagent. J Biol Chem 193(1):265–275 Mackenzie C, Eraso JM, Choudhary M, Roh JH, Zeng X, Bruscella P, Puskas A, Kaplan S (2007) Postgenomic adventures with Rhodobacter sphaeroides. Annu Rev Microbiol 61:283–307 Mahadevan R, Schilling CH (2003) The effects of alternate optimal solutions in constraint-based genome-scale metabolic models. Metab Eng 5(4):264–276 Masepohl B, Hallenbeck PC (2010) Nitrogen and molybdenum control of nitrogen fixation in the phototrophic bacterium Rhodobacter capsulatus. Adv Exp Med Biol 675:49–70 Miyake J, Kawamur S (1987) Efficiency of light energy conversion to hydrogen by the photosynthetic bacterium Rhodobacter sphaeroides. Int J Hydrogen Energy 12(3):147–149 Montagud A, Navarro E, de Cordoba PF, Urchueguia JF, Patil KR (2010) Reconstruction and analysis of genome-scale metabolic model of a photosynthetic bacterium. BMC Syst Biol 4:156 Nogales J, Gudmundsson S, Knight EM, Palsson BO, Thiele I (2012) Detailing the optimality of photosynthesis in cyanobacteria through systems biology analysis. Proc Nat Acad Sci USA 109(7):2678–2683 Oberhardt MA, Palsson BO, Papin JA (2009) Applications of genome-scale metabolic reconstructions. Mol Syst Biol 5:320 Oelze J (1988) Regulation of tetrapyrrole synthesis by light in chemostat cultures of Rhodobacter sphaeroides. J Bacteriol 170(10):4652–4657 Peralta-Yahya PP, Zhang F, del Cardayre SB, Keasling JD (2012) Microbial engineering for the production of advanced biofuels. Nature 488(7411):320–328 Reed JL (2012) Shrinking the metabolic solution space using experimental datasets. PLoS Comput Biol 8(8):e1002662 Reed JL, Palsson BO (2004) Genome-scale in silico models of E. coli have multiple equivalent phenotypic states: assessment of correlated reaction subsets that comprise network states. Genome Res 14(9):1797–1805 Rex R, Bill N, Schmidt-Hohagen K, Schomburg D (2013) Swimming in light: a large-scale computational analysis of the metabolism of Dinoroseobacter shibae. PLoS Comput Biol 9(10):e1003224 Rittmann B, McCarty PL (2001) Environmental biotechnology: principles and applications. McGraw-Hill Science Engineering, New York Rouser G, Fkeischer S, Yamamoto A (1970) Two dimensional then layer chromatographic separation of polar lipids and determination of phospholipids by phosphorus analysis of spots. Lipids 5(5):494–496 Russell JB, Cook GM (1995) Energetics of bacterial growth: balance of anabolic and catabolic reactions. Microbiol Rev 59(1):48–62 Sasikala K, Ramana VC, Rao RP, Kovacs KL (1993) Anoxygenic phototrophic bacteria: physiology and advances in hydrogen production technology. Adv Appl Microbiol 38:211–295 Schumacher A, Drews G (1979) Effects of light intensity on membrane differentiation in Rhodopseudomonas capsulata. Biochim Biophys Acta 547(3):417–428 Sistrom WR (1960) A requirement for sodium in the growth of Rhodopseudomonas spheroides. J Gen Microbiol 22:778–785

123

Photosynth Res Stone M (1974) Cross-validatory choice and assessment of statistical predictions. J Royal Stat Soc 36(2):111–147 Stouthamer AH, van Verseveld HW (1987) Microbial energetics should be considered in manipulating metabolism for biotechnological purposes. Trends Biotechnol 5(5):149–155 Tannler S, Decasper S, Sauer U (2008) Maintenance metabolism and carbon fluxes in Bacillus species. Microb Cell Fact 7:19 Thiele I, Palsson BO (2010) A protocol for generating a high-quality genome-scale metabolic reconstruction. Nat Protoc 5(1):93–121 Varma A, Palsson BO (1994) Metabolic flux balancing: basic concepts, scientific and practical use. Nat Biotechnol 12:994–998 Vu TT, Stolyar SM, Pinchuk GE, Hill EA, Kucek LA, Brown RN, Lipton MS, Osterman A, Fredrickson JK, Konopka AE, Beliaev

123

AS, Reed JL (2012) Genome-scale modeling of light-driven reductant partitioning and carbon fluxes in diazotrophic unicellular cyanobacterium Cyanothece sp. ATCC 51142. PLoS Comput Biol 8(4):e1002460 Wahlund TM, Conway T, Tabita FR (1996) Bioconversion of CO2 to ethanol and other compounds. American Chemical Society Division of Fuel Chemistry 3:1403–1405 Yilmaz LS, Kontur WS, Sanders AP, Sohmen U, Donohue TJ, Noguera DR (2010) Electron partitioning during light- and nutrient-powered hydrogen production by Rhodobacter sphaeroides. Bioenerg Res 1:55–66 Zijffers JW, Schippers KJ, Zheng K, Janssen M, Tramper J, Wijffels RH (2010) Maximum photosynthetic yield of green microalgae in photobioreactors. Mar Biotechnol (NY) 12(6):708–718