Correspondence: Iftekhar A. Karimi, Department of Chemical and Biomolecular Engineering, National University of Singapore, 4 Engineering Drive 4, Singapore 117576. Singapore. Tel.: +65 6516 6359; fax: +65 6779 1936; e-mail: firstname.lastname@example.org
Rhodococcus erythropolis has been studied widely for potential applications in biodesulfurization. Previous works have been largely experimental with an emphasis on the characterization and genetic engineering of desulfurizing strains for improved biocatalysis. A systems modeling approach that can complement these experimental efforts by providing useful insights into the complex interactions of desulfurization reactions with various other metabolic activities is absent in the literature. In this work, we report the first attempt at reconstructing a flux-based model to analyze sulfur utilization by R. erythropolis. The model includes the 4S pathway for dibenzothiophene (DBT) desulfurization. It predicts closely the growth rates reported by two independent experimental studies, and gives a clear and comprehensive picture of the pathways that assimilate the sulfur from DBT into biomass. In addition, it successfully elucidates that sulfate promotes higher cell growth than DBT and its presence in the medium reduces DBT desulfurization rates. A study using eight carbon sources suggests that ethanol and lactate yield higher cell growth and desulfurization rates than citrate, fructose, glucose, gluconate, glutamate, and glycerol.
The increasingly stringent regulations for ultralow-sulfur fuels make desulfurization a crucial step in the processing of fossil fuels. The prevalent method for this is hydrodesulfurization, a chemical process. It is not only energy-intensive and expensive, but also incapable of removing sulfur from recalcitrant compounds such as benzothiophene and dibenzothiophene (DBT) (Song, 2003). Thus, there is a clear need for developing new, efficient, and more economical methods for deep desulfurization.
Biodesulfurization is considered an attractive technique, as it can proceed under ambient conditions without lowering the calorific value and is relatively economical (Soleimani et al., 2007). It involves the use of either whole cells or enzymes to remove sulfur from fuels. Several microbial strains belonging to various genera such as Pseudomonas, Rhodococcus, Gordonia, Paenibacillus, and Mycobacterium have been found to show desulfurization activity. These organisms use diverse biochemical mechanisms, such as Kodama and 4S pathways, to metabolize various polyaromatic sulfur heterocycles (PASHs). Of these, R. erythropolis IGTS8 was the first to be isolated for its ability to specifically cleave the C–S bond in PASHs without affecting the C–C bond (Kilbane & Jackowski, 1992). Since then, several Rhodococcus strains have been studied (Izumi et al., 1994; Li et al., 1996; Ohshiro et al., 1996; Honda et al., 1998; Davoodi-Dehaghani et al., 2010) for specifically desulfurizing DBT and its derivatives via the 4S pathway.
The desulfurization rates exhibited by the wild-type bacteria are too low for commercialization (Kilbane, 2006). Despite numerous experimental efforts including genetic manipulations, desirable desulfurization rates are yet to be attained. From our study, it seems that this may be due to the fact that most of these studies have solely targeted the 4S pathways and desulfurizing (dsz) genes. Because the cellular phenotypes are the manifestations of complex interactions among various gene products and environmental factors, a systems biology approach is critical for studying desulfurization. A comprehensive modeling approach can complement the existing and future experimental studies considerably. Such an approach would facilitate a more quantitative and insightful understanding of the interdependencies among the various pathways and associated reactions that largely determine the metabolic fluxes within a desulfurizing strain, and hence its desulfurization activity. The resulting knowledge can then guide the design of environment and re-engineering of strains for enhancing desulfurization via the 4S pathway.
This work represents the first attempt, to our knowledge, at reconstructing a stoichiometric model for the sulfur metabolism in R. erythropolis. It comprises a network of reaction pathways involved in sulfur and central metabolism, and quantitatively describes the assimilation of sulfur from different sources into various biomass precursors. It successfully predicts two independent cell growth data and several phenotypes reported in the literature such as the effects of sulfate and various carbon sources on biodesulfurization activity. We have successfully used the model to compare the effects of eight carbon sources (citrate, ethanol, fructose, gluconate, glucose, glutamate, glycerol, and lactate) on desulfurizing activity and cell growth.
Materials and methods
The flux-based models have been widely used to study the metabolic networks of various microorganisms in a holistic manner (Burgard & Maranas, 2003; Suthers et al., 2009; Orth et al., 2010; Thiele & Palsson, 2010). Such a model for an organism is built on the known and hypothesized reactions that may take place within the organism (Gonzalez et al., 2008) based on its genomic, biochemical, and physiological information (Park et al., 2009). In this work, our goal is to develop an in silico model that gives a quantitative description of the metabolic and biosynthetic functions of sulfur in R. erythropolis. Thus, we limited ourselves largely to the pathways dedicated to the syntheses of sulfur-containing metabolic precursors and their incorporation into biomass. However, we also added select pathways from the central metabolism to elucidate and examine the effects of carbon sources (Yan et al., 2000) on desulfurization activity and the key role of reducing equivalents (Oldfield et al., 1997) in the energy-intensive 4S pathway.
Our basis model used the information on pathways and reactions available in the Kyoto Encyclopedia of Genes and Genomes (Kanehisa & Goto, 2000) database. We curated the reactions manually and corrected them for carbon and sulfur balances. Further, we included some additional reactions from the literature (Oldfield et al., 1997, 1998; Beste et al., 2007; Jamshidi & Palsson, 2007) and MetaCyc (Caspi et al., 2008) to complete the pathways necessary for the biosynthesis and utilization of some key metabolites. For instance, we took the reactions for the 4S pathway from Oldfield et al. (1998), mycothiol biosynthesis from Rawat & Av-Gay (2007), and metabolism of glycerol and glutamate from MetaCyc (Caspi et al., 2008). Likewise, we adapted the pathways for the biosynthesis of thiamine and biotin from the existing reconstructed metabolic model of a related actinomycete, Mycobacterium tuberculosis (Beste et al., 2007; Jamshidi & Palsson, 2007). Table 1 shows the number of reactions taken from each of the above-mentioned sources. However, being limited in scope and pathways, the resulting model could still not synthesize (consume) some substrates (products) such as inositol, pantothenate, etc. that appear in the reactions. Therefore, we assumed an extracellular pool of such metabolites and added transport reactions with unlimited fluxes to simulate their necessary uptake (release).
Table 1. Number of reactions from different sources for the reduced model
A biomass equation represents cell growth in a flux-based in silico model. It is a synthetic reaction that consumes cell constituents in known constant proportions (derived from cell composition) to form a unit amount of cell biomass. However, as a quantitative analysis of the biomass constituents in R. erythropolis is unavailable in the literature, we adapted the biomass equation in our model from the known composition of a related actinomycete, M. tuberculosis (Beste et al., 2007; Jamshidi & Palsson, 2007). We kept only the precursors that contain sulfur or are involved in sulfur metabolism, and added other sulfur-containing cofactors such as biotin and thiamin to appropriately reflect the requirements of sulfur and its metabolism. However, we excluded sulfolipids, as they are known to confer pathogenic characteristics to M. tuberculosis.
For performing the flux balance analysis with the resulting model, we used metafluxnet (Lee et al., 2003).
Experimental data sources for model validation
Experimental data are indispensable for validating an in silico (computational) model. For this study, we used the experimental data of Izumi et al. (1994) and Davoodi-Dehaghani et al. (2010) on cell growth and metabolite concentration profiles. Izumi et al. (1994) reported that R. erythopolis D-1 desulfurized DBT to 2-hydroxybiphenyl (HBP) successfully. They used 500 mL of a glucose-based biosynthetic medium with 0.125 mM DBT as the sole sulfur source at 30 °C to examine the desulfurization activity of growing cells. They measured pH, cell growth, DBT concentration, and HBP concentration at various times during their experiment.
In another study, Davoodi-Dehaghani et al. (2010) isolated R. erythropolis SHT87. They used growing cells at 30 °C in a 50 mL solution of glycerol containing a synthetic medium with 0.25 mM of DBT as the sole sulfur source. They also measured cell growth, DBT concentration, and HBP concentration at different times over 120 h.
The experimental data from the above two independent studies provided a sound basis for validating our proposed model. We used their cell growth data and DBT/HBP concentration profiles from the exponential phase to compute specific cell growth rates (1 h−1) and DBT (HBP) uptake (secretion) rates (mmol g−1 dcw h−1).
Results and discussion
Model for sulfur metabolism in R. erythropolis
Our reconstructed model consists of 87 intracellular metabolic reactions, 66 transport reactions, and 196 metabolites related to either sulfur or central metabolism. The sulfur metabolism includes the 4S pathway; the CoA biosynthetic pathway; metabolism of inorganic sulfur, cysteine, and methionine; and biosynthesis of cysteine, methionine, mycothiol, biotin, and thiamine. The central metabolism includes gluconeogenesis, citric acid cycle, pentose phosphate pathway, and Embden Meyerhoff Paranas pathway for glycolysis. Figure 1 shows a complete picture of the pathways and reactions in our model, with full details in the Supporting information.
We simulated the experiments of Izumi et al. (1994) and Davoodi-Dehaghani et al. (2010) and compared our predicted cell growth rates with their measured data. As the 4S pathway is aerobic, we assumed unlimited oxygen flux in all of our validation studies and analyses. Sulfur was a limiting substrate in the experiments of Izumi et al. (1994) and Davoodi-Dehaghani et al. (2010). We inferred this from the fact that the stationary phase in their experiments was triggered, when DBT concentration went to zero and HBP concentration reached its maximum. Therefore, we allowed unlimited glucose flux for simulating the experiment of Izumi et al. (1994) and unlimited glycerol flux for Davoodi-Dehaghani et al. (2010). Then, we fixed the DBT uptake and HBP production rates (mmol g−1 dcw h−1) to be at some values computed from their data, and predicted specific cell growth rates at those values. Figure 2 shows that our growth predictions are in close agreement with the two experimental data. The accuracy of our predictions is confirmed by the argument that the limiting sulfur solely determines the growth.
Analysis of sulfur metabolism using alternate sources
We then studied the utilization of sulfate and DBT as sulfur sources to qualitatively demonstrate the consistency of our model with some literature observations (Omori et al., 1995; Honda et al., 1998) on cell growth and desulfurizing activity.
In silico growth on alternate sulfur sources
In a study on desulfurization by R. erythropolis IGTS8 in an acetate-based medium, Honda et al. (1998) observed that sulfate promoted higher cell growth than DBT. To study this phenotype, we performed flux balances for two scenarios (Table 2) with unlimited acetate uptake. In run 1, we fixed the DBT (sulfate) uptake at 20 (0.0) mg g−1 dcw h−1. In run 2, we fixed sulfate (DBT) at 20 (0.0) mg g−1 dcw h−1. Our model gave a higher cell growth rate (1.29 vs. 0.84 h−1) for sulfate (run 2) than DBT (run 1). Then, we fixed the acetate uptake at 20 mg g−1 dcw h−1 and studied two more scenarios (Table 2). In run 3, we allowed unlimited (zero) sulfate (DBT) uptake, and did the reverse in run 4. Again, we obtained a higher growth (1.4 vs. 1.06 h−1) for sulfate (run 3) than DBT (run 4).
Table 2. Flux analyses for the utilization of alternate sulfur sources with acetate as the carbon source and effect of sulfate on desulfurization activity with succinate as the carbon source
In silico growth of alternate sulfur sources
All uptakes in mg g−1 dcw h−1, growth in h−1, and desulfurizing activity in mmol g−1 dcw h−1.
Effect of sulfate on desulfurization activity
After studying sulfate and DBT separately, we also studied them together (run 5 in Table 2) for a fixed acetate uptake of 20 mg g−1 dcw h−1. We fixed the sulfate uptake at 2.16 mg g−1 dcw h−1 and allowed unlimited DBT. This sulfate uptake is 10% of its maximum (21.6 mg g−1 dcw h−1) observed in run 4. The model showed a higher growth rate of 1.12 h−1 compared with 1.06 h−1 obtained previously for run 3 (unlimited DBT, zero sulfate). The DBT uptake was also lower (22.08 vs. 25.76 mg g−1 dcw h−1). This suggests that the organism may grow faster when it fulfills a part of its sulfur needs via sulfate rather than DBT. In other words, the organism may prefer sulfate when both DBT and sulfate are present. Because sulfate yields a higher growth rate than DBT, the organism may use DBT only if sulfate is not present. This clearly confirms the results of Honda et al. (1998).
Honda et al. (1998) reasoned that the observed lower cell growth with DBT was due to the toxic effect of HBP (its desulfurized product). Because our model does not include such toxic effects, we cannot deny this as a probable explanation. However, we have the following alternate explanation from our study. Rhodococcus erythropolis needs sulfate and sulfide to synthesize its sulfur-containing biomass precursors. If it uses DBT as the sulfur source, then it must use the 4S pathway. 4S converts DBT to sulfite, which is converted to sulfate and sulfide by the sulfur metabolism and then incorporated into the biomass precursors. However, the organism needs 4 mol NADH mol−1 DBT to use DBT in the above manner. In contrast, the organism does not need this extra NADH for metabolizing sulfate. Thus, the organism prefers the energetically less expensive sulfate over DBT for its growth. Although our reduced model does not include all the reactions involving NADH, it is known that NADH is an essential component for growth. When the organism is forced to use DBT, NADH available for other growth-critical activities inside the cell reduces, and thus cell growth reduces.
Effect of sulfate on desulfurizing activity
The reduced DBT uptake in the presence of sulfate prompted us to study the effect of sulfate on desulfurization rates. Several literature studies have reported the effect of sulfate on desulfurization activity. Li et al. (1996) reported that although sulfate represses the dsz genes, it does not inhibit the activity of desulfurizing enzymes (Wang & Krawiec, 1996). They observed that the desulfurizing activity increased with decreasing amount of sulfate in the medium. Similarly, Omori et al. (1995) also observed enhanced desulfurizing rates arising from the removal of byproduct sulfate from a succinate-based medium. To understand this phenotype using our in silico model, we analyzed fluxes for three scenarios (Table 2) with a succinate uptake at 20 mg g−1 dcw h−1. In run 6, we allowed unlimited DBT as the sole sulfur source and obtained the maximum desulfurizing rate of 0.07 mmol g−1 dcw h−1. In run 7, we allowed unlimited sulfate as the sole sulfur source, and obtained the maximum sulfate uptake of 10.80 mg g−1 dcw h−1. Then, in subsequent runs, we allowed progressively increasing amounts of sulfate (from 0% to 100% of the maximum sulfate uptake of 10.80 mg g−1 dcw h−1 from run 7) with unlimited DBT. From Fig. 3, we see that the desulfurizing activity clearly decreases with increasing amount of sulfate. Thus, our model successfully explains the observations of Omori et al. (1995) and Li et al. (1996).
Our earlier comment on energy needs again readily explains this effect. When the desulfurizing enzymes are already present, then the organism is able to utilize (desulfurize) DBT. However, sulfate promotes higher growth at lower energy, and so the organism prefers sulfate consumption over DBT conversion. Only when sulfate is limited, it desulfurizes DBT. In other words, no desulfurization is possible even in the presence of desulfurizing enzymes if the medium has sufficiently high concentration of sulfate to meet the sulfur needs of R. erythropolis. To our knowledge, no previous experimental work has elucidated this phenotype, which our model made possible.
Effect of carbon source
Yan et al. (2000) studied the relative efficacy of ethanol, glucose, and glycerol as sole carbon sources for the growth and desulfurizing activity of R. erythropolis. They reported ethanol to yield the highest growth and desulfurizing rates, followed by glucose, and then glycerol. To simulate this phenotype, we considered three separate scenarios with unlimited DBT and one carbon source. In each scenario, we fixed the uptake of the respective sole carbon source at 20 mg g−1 dcw h−1 and used maximum biomass as the cellular objective. Our model gave the highest growth rate of 1.39 h−1 and the highest desulfurizing rate of 0.18 mmol HBP g−1 dcw h−1 for ethanol. In contrast, the rates were 0.60 h−1 and 0.08 mmol HBP g−1dcw h−1 for glucose, and 0.59 h−1 and 0.07 mmol HBP g−1 dcw h−1 for glycerol. Thus, our model qualitatively confirms the experimental results of Yan et al. (2000).
The extra NADH needed for the 4S pathway again successfully elucidates the above results. The major source of NADH in R. erythropolis is the carbon metabolism. Ethanol yields more NADH during this metabolism than glucose and glycerol. The additional NADH enables the cell to increase the flux (or desulfurizing rate) of the 4S pathway, which eventually helps it to increase growth. Extending this, we argue that a carbon source that provides more NADH is likely to enhance both the growth and the desulfurizing rates of R. erythropolis.
As our model predicted some experimental observations successfully, we examined the suitability of additional carbon sources for desulfurizing activity. We studied citrate, ethanol, fructose, gluconate, glucose, glycerol, glutamate, and lactate as possible sole carbon sources. We computed fluxes for each sole source separately with an uptake rate of 20 mg g−1 dcw h−1. Figure 4 shows the results of our eight simulation runs. The desulfurization and growth rates relative to those of ethanol decrease in the following order: ethanol (0.18 mmol HBP g−1 dcw h−1 as 100% and 1.39 h−1 as 100%)>lactate (67%)>citrate (48%)>glutamate (44%)>glucose=fructose (43%)>glycerol (42%)>gluconate (40%). However, as our model is reduced and has limited scope, this prediction is only qualitative in nature. An experimental verification of this prediction is clearly beyond the scope of this work. As a natural goal of any in silico model, our intention is simply to offer a new hypothesis that experimental researchers can verify.
Our reconstructed stoichiometric model for sulfur metabolism in R. erythropolis successfully predicted cell growth and several known/unknown phenotypes. Our analysis shows that NADH plays a critical role in desulfurization activity. Any changes in medium design or genetic manipulations that increase NADH regeneration and supply within the cellular metabolism are likely to enhance desulfurization activity. We are in the process of developing a full genome-scale model that can account for host functions other than just sulfur and central metabolism.