Reconstruction and analysis of a genome-scale metabolic model for Eriocheir sinensis eyestalks

Abstract

The eyestalk of Eriocheir sinensis has significant biological functions with many nerve peptide hormones expressed in the X-organ which exists in the eyestalk. A metabolic network model is an effective tool for the systematic study of E. sinensis eyestalks. In this work, we reconstructed a metabolic network model for E. sinensis eyestalks based on transcriptome sequencing. The model contains 1304 reactions, 1381 unigenes and 1243 metabolites distributing in 98 pathways. The reconstructed metabolic network model was used for the functional module and block metabolite analysis of eyestalks, which reveals that the function of the eyestalk network agrees with its function as the centre of the E. sinensis endocrine system. The difference expression analysis of reactions in the model indicates that the eyestalk mainly functions in the regulation of amino acids, carbohydrate and nucleotide metabolism.

---

Introduction

E. sinensis, also known as crabs, is one of the most important aquatic economic animals with a high commercial value as a food source. Many studies have been performed focusing on several or single genes,¹ proteins² or a specific pathway³ to accelerate the growth or further investigate the metabolic system of E.sinensis. In 2014, Sun Yan et al. conducted transcriptome sequencing on the hepatopancreas, the eyestalk and the Y-organ of E. sinensis.⁴ Subsequently, Hao Tong et al. constructed a protein interaction network of these organs.⁵ Eyestalks which contain the X-organ sinus gland (XO-SG) are known as the endocrine control center of E. sinensis. It secretes and produces a variety of neuropeptide hormones which regulates important physiological activities of the crustacean animals, such as growth, metabolism and reproduction.

Many research studies have focused on the profiles or genes of eyestalks.^6,7 However, a systematic study of the eyestalk metabolic system is still missing. In recent decades, a forward trend of the network biology is to use genomic data and molecular mechanisms to construct predictive models.⁸ A genome-scale metabolic network model is an important part in this area and is a useful tool for a systematic study of specific organisms or organs.⁹ In recent years, with the development of high throughput data analysis, the reconstruction of a genome-scale metabolic network model has developed rapidly^10,11 with interaction of knowledge on biochemistry and physiological metabolism.¹² Since Edwards and Palsson et al. constructed the first genome-scale metabolic network model for Haemophilus influenzae in 1999,¹³ plenty of models have been constructed and applied in the analysis of various biological phenotypes, which reflects the important role of the genome-scale metabolic network model in the research of metabolism in a wide range of organisms. Currently, more than one hundred metabolic network models have been reconstructed for bacteria,¹⁴ plants,^15,16 animals¹⁷ and human.^18,19 The models have a wide use in areas such as metabolic engineering, antibiotic design, and organismal and enzyme evolution.²⁰

For the systematic study of the metabolic function of E. sinensis eyestalks, we reconstructed a genome-scale metabolic model for eyestalks based on transcriptome sequencing. The model was revised with the simulation on the synthesis capability of non-essential amino acids. Functional analysis of the reconstructed model was used to reveal the fundamental function of the eyestalk metabolism system and the vital function as the center of the crustaceans endocrine system. The differentially expressed unigenes were analyzed for further understanding the influence of 5-HT and glucose on the metabolism of eyestalks. The model provides a useful tool for the in-depth study on the function of E. sinensis eyestalks, as well as provides an important reference for the future study of aquatic crustacean species.

Methods

Preliminary reconstruction of the genome-scale metabolic network model

The unigene information was obtained from the transcriptional sequencing of E. sinensis eyestalks.⁴ The KO annotate of the unigenes was supplied in the work of Yan Sun et al.⁴ The related biochemical information of proteins, metabolic reactions, pathways and metabolic subsystems was matched from the KEGG database with KO as a bridge. The draft reconstruction is obtained with the relationship of unigene–protein-reaction is established according to the KEGG database.

The draft reconstructed metabolic network on the genome-scale is far from perfect. Further curation is needed. Firstly, the chiral molecule of glucose is revised. In the KEGG database, glucose was shown in three forms: D-glucose, alpha-D-glucose and beta-D-glucose. D-glucose is the general name of glucose, which includes both alpha-D-glucose and beta-D-glucose. In order to avoid the confusion of glucose in the following curation and simulation, all the glucose in the network was unified to be alpha-D-glucose as the alpha type is the main existing form of most sugar in biological organisms.

Gap filling

The gaps in the network were filled with the method based on the connectivity of weak connected components (WCCs).^21,22 Firstly, the reactions that exist in the whole reaction list from the KEGG database but not the preliminary reconstructed network were considered as the candidate dataset for gap filling. Secondly, as graph theory has been proved to be a useful train of thought for biological analysis,^23–26 the network was converted into a reaction map with reactions as nodes and metabolites as edges according to the metabolic flux direction. In this way, the metabolic network was divided into different WCCs. Finally, the reactions that could be used to connect different WCCs were searched in the candidate dataset. These reactions were identified as the gap filling reactions.

The gaps were filled both on the pathway²¹ and global scales.²² These two methods were used sequentially to fill the gaps in the network to a maximum extent. Gap filling on the pathway scale prefers to decrease the number of WCCs in the same pathway, whereas that on the global scale inclines to reduce the WCCs in the whole network.

Simulative curation of the model

In order to determine whether the reconstruction can correctly reflect the metabolic processes of the living organisms, we firstly convert the reaction list to standard SBML document which can be read by COBRA²⁷ and analyzed with flux balance analysis (FBA) which is the commonly used optimization simulation method^28–30 for metabolic networks.

The flux ranges of reactions in the network were limited for simulations. The upper and lower flux of the reversible reaction, the irreversible reaction and the transport reaction were set to be (−1000, 1000)¹² (0, 1000), and (−1000, 1000) mmol g DW⁻¹ h⁻¹, respectively. The limits of exchange reactions for essential amino acids, glucose, H₂O, Fe³⁺, alpha-D-glucose, potassium, magnesium, calcium, which are needed to be exchanged with the environment, were set to be (−5, 1000) mmol g DW⁻¹ h⁻¹, while those for non-essential amino acids which can be synthesized by E. sinensis were set to (0, 1000) mmol g DW⁻¹ h⁻¹.

In order to investigate the biosynthesis capability of non-essential amino acids (alanine, asparagine, aspartic acid, cysteine, glutamine, glutamate, glycine, proline, serine, and tyrosine), 10 non-essential amino acids were set to be objective functions, respectively. If an amino acid cannot be synthesized, there must be the absence of reaction(s) in the network. The missing reaction(s) are found by backtracking along the synthesis pathway of the amino acid. With the addition of missing reaction(s), the production of the amino acid was calculated again to confirm the synthesis capability of the model.

Network decomposition

Model decomposition is performed by applying the method based on the combination of dendrogram and modularity.^31,32 The network was decomposed with four steps. Firstly, the network was converted into a reaction map. In order to achieve the conversion, current metabolites (such as ATP, ADP, H₂O, NAD+, NADH, Pi³³) are firstly removed. Reactions exist as nodes and the metabolites act as edges in the map. Secondly, the reaction map was preliminary decomposed according to the dendrogram tree with no more than 50 nodes in a cluster. Thirdly, the preliminary decomposition was optimized by the elimination of the Very Small Cluster (VSC) with no more than 10 nodes as the first optimization step. Fourthly, the modularity of the network was maximization in the second optimization step with the final number of cluster set to be 10. With the decrease of module numbers in the fourth step, the partition with highest modularity was considered as the final decomposition result.

Detection of differentially expressed protein in eyestalks

The healthy E. sinensis (body weight 5–6 g), with good vitality, was bought from Huada, and then acclimatized in freshwater at 18–20 °C for 3–5 days.

A total of 180 E. sinensis that were selected randomly from the rearing animals in the laboratory were equally divided into three groups. The group that was injected with 5-HT was the experimental group (A1). The second group that was injected with glucose was the experimental group (A2). The third group, which was injected with distilled water, was the control group (A0) (The weight of each E. sinensis was measured before injection. The volume injected (ul) was the same as the weight (g) in terms of number.) One hour later, eyestalks of E. sinensis from all these groups were removed and frozen in liquid nitrogen immediately storing at −80 °C until the extraction of RNA.

To extract RNA, the following steps were taken. Firstly, the total RNA from eyestalks in groups A0, A1 and A2 was extracted using the TRIzol method (Invitrogen, USA). Secondly, the cDNA was synthesized using the superScript III first-strand synthesis system (Invitrogen, USA). After extracting the total RNA from the three samples, the generated cDNA library was sequenced using Illumina HiSeqTM2000.

Results and discussion

Preliminary reconstruction of the network

Totally 48 835 unigenes were obtained in the transcriptional sequencing of eyestalks, in which 12 864 were found to be matched with KO from the KEGG database.⁴ The preliminarily reconstructed metabolic network was obtained through matching unigenes to related reactions and pathways from the KEGG database through KO as a bridge. The network included 1208 reactions and 1381 unigenes, which distributed in 98 pathways and 11 subsystems (additional file 1). 6 reactions with D-glucose were curated to be alpha-D-glucose for the standardization of chiral metabolites.

Gap filling

A ‘gap’ is defined as a reaction that takes place in a pathway/network different to that of its two neighbor reactions. Network gaps will lead to unbalanced flux distribution. Therefore, it is necessary to fill them before simulation. Filling gaps can improve the connectivity of the network and enable a pathway to perform its normal function.³⁴

Gaps were filled on the pathway and global scale sequentially. With gap filling on the pathway scale, the number of reactions in the network increased from 1028 to 1186. 158 new reactions were added, most of which were involved in metabolism of arginine, proline, caffeine, cysteine, methionine, sphingolipid, cytochrome P450 and biosynthesis of phenylpropanoid and steroid. With gap filling on the global scale, 60 more reactions were added. The number of WCCs decreased from 218 to 79 in the gap filling step, in which 108 decreases on the pathway scale and 31 decreases on the global scale. After gap filling, the network contains 1246 reactions.

Addition of transport and exchange reactions

In order to distinguish the intracellular and extracellular parts of the cell, the metabolites and reactions in the network were divided into two compartments: intracellular and extracellular. The metabolites were discriminated as intracellular or extracellular according to their locations. The reactions with all the related metabolites located in intracellular are identified as intracellular reactions. The reactions with all the related metabolites in extracellular are identified as extracellular reactions. The reactions with related metabolites distribute in both intracellular and extracellular are identified as transport reactions.

Actually most reactions in the metabolic network are intracellular reactions, whereas the organisms cannot grow without exchanging substances with the extracellular environment by absorbing nutrients or discharging metabolic products. In order to enable the model the capability of exchanging substances with the outside world, transport and exchange reactions for the basic nutrients that the eyestalk cells may exchange with external environment were added. We totally added transport and exchange reactions for 26 substances including 20 kinds of amino acids, H₂O, Fe³⁺, potassium, alpha-D-glucose, magnesium and calcium. With addition of transport and exchange reactions, 1299 reactions were included in the network. Both intracellular and extracellular metabolites are involved in the transport reactions, and all the metabolites are extracellular in the exchange reactions.

Simulation curation

In order to investigate the quality of the model, the capability of synthesizing non-essential amino acids was verified. The result showed that two of non-essential amino acids, asparagine and cysteine, cannot be synthesized. According to the map of Alanine, aspartate and glutamate metabolism in KEEG, L-asparagine (KEGG NO.: C00152) is the product of two reactions (R00483 and R00578). When R00483 (ATP + L-aspartate + ammonia ⇔ AMP + diphosphate + L-asparagine) is added to the network, asparagine can be successfully synthesized. Therefore, it was finally found that the break of the asparagines synthesis route is due to the absence of reaction R00483. With the same way, the synthesis route of cysteine (C00097) was remedied by the addition of four reactions (R00528, R00858, T00027, E00027). These reactions enable the network to absorb sulfate from the environment³⁵ and convert it into sulfide which is necessary in the synthesis of cysteine. With these curations, all the ten non-essential amino acids can be successfully synthesized by the reconstructed network model. The curated model contains 1340 reactions, 1243 metabolites, 1381 unigenes, which distributes in 98 pathways and 11 subsystems (ESI,† file S1). The overview characteristics of the reconstructed model are shown in Table 1.

Table 1 The characteristics of a genome-scale metabolic network model of eyestalks

Items	Count
Unigenes	1381
Reactions	1304
Metabolic reactions	1250
Transport reactions	27
Exchange reactions	27
Metabolites	1243
Intracellular metabolites	1214
Extracellular metabolites	29
Pathways	98
Subsystems	11

---

Topology and functional analysis of the reconstructed metabolic network model

Topology analysis. The genome-scale metabolic model for E. sinensis eyestalks was composed of 1304 reactions. Its topological features, which were calculated with Pajek software,³⁶ are shown in Table 2. Besides the exchange reactions, there were 1277 reactions in the network. After removing of currency metabolites, 1246 main reactions are extracted, in which 1128 are non-isolated reactions. Therefore, the reaction map of the network, which was composed of the non-isolated reactions, contains 1128 reactions. The bow-tie structure of the network is clearly reflected by the GSC, S and P part in the topological analysis. The existence of bow-tie structure in this network is in consistent with the conclusion that the bow-tie structure is widespread in the metabolic network in different levels, sizes, chemical and spatial units, which is proved by Zhao Jing et al.³⁷ The metabolic network was divided into 71 WCCs which include nine bigger WCCs that contain more than 10 reactions. The biggest one included 797 reactions, which takes up about 70.66% of the total number of reactions in the metabolic network (as shown in Fig. 1). In order to further analysis the function of the metabolic network, the biggest weak component was decomposed with the method described in the Method section. After two times of optimization in the decomposed process, the connection within modules increased gradually, while that between modules gradually reduced. The network modularity increased (as shown in Table 3), which indicates that the modularity of the metabolic network was gradually strengthened. When the number of the modules decreased to 16 we got the highest modularity value 0.827. Therefore, the 16 modules were taken as the final decomposition result (ESI,† file S2).

Table 2 The basic topological features of the reconstructed metabolic network

Graph matrics		Value
Nodes		1128
	Arcs	1970
	Edges	769
Density		0.002757
Average degree		4.856383
Average path length		8.95101
Diameter		21
Biggest cluster	Nodes	797
	Arcs	1702
	Edges	717
Bowtie of biggest cluster	GSC	394(49.4%)
	S	75(9.4%)
	P	253(31.7%)
	IS	75(9.4%)

---

Fig. 1 The WCCs (weak connected components) in the eyestalk metabolic network. Totally 71 components are shown which are distinguished by different colors.

Table 3 Decomposition results of the eyestalk metabolic network

	Threshold	Modules	Modularity
Preliminary	50	57	0.622
First	10	22	0.812
Second	10	16	0.827

---

Functional analysis. In order to further understand the metabolic function of eyestalks, the main function of each module in the biggest weak component was further analyzed. Its main functions include basic amino acid metabolism, carbohydrate metabolism, nucleic acid metabolism and lipid metabolism (Fig. 2). Most of the modules were dominated by one or two kinds of subsystems suggesting that the function of each module was relatively independent. The amino acid metabolism dominated in the metabolic network of eyestalks (Table 4). Amino acid is the basic substance in the synthesis of neuropeptide hormones. Therefore, the domination of amino acid metabolism in the modules is in consistent with the fact that X-organ sinus gland (XO-SG) secretes neuropeptide hormones, which is the main function of eyestalks.³⁸ In addition, carbohydrate metabolism also takes up a large proportion in the network because of its close relationship with other metabolisms, indicating that eyestalks may also play a role in the energy offering and converting process. The fact that different metabolic pathways are contained in one module shows that one metabolic function may be accomplished through several pathways. For example, aspartic acid in module 14 can be converted to arginine; module 9 includes the metabolism of folic acid and purine, which coincides with one of the functions of folic acid: involvement in the biosynthesis of purine ring and deoxynucleotide,³⁹ module 14 includes the metabolism of folic acid and glutamic acid. Folic acid, also known as pteroylglutamic acid, is involved in the biosynthesis of glutamic acid and the metabolism of arginine⁴⁰ (some scholars regard that folic acid in the body exists in two forms: transport folic acid in the form of mono-glutamic acid, storage folic acid in the form of poly-glutamic acid⁴¹).

Fig. 2 Functional analysis of the reconstructed metabolic network. Each pie chart represents a single module divided into several different subsystems. The percentage of reactions in a certain subsystem are shown by the area in a pie chart. The thickness of the line between modules represents the strength of the connection.

Table 4 Distribution of amino acid metabolism in modules

Module number	Metabolic process
Module 2	Lysine biosynthesis and degradation
Module 3	Leucine and isoleucine degradation
Module 4	Glycine metabolism
Module 5	Tryptophan metabolism
Module 6	Isoleucine degradation
Module 7	Serine and glycine metabolism
Module 8	Cysteine metabolism
Module 10	Tyrosine and phenylalanine metabolism
Module 14	Arginine, proline, glutamate metabolism
Module 15	Cysteine metabolism
Module 16	Lysine, isoleucine degradation; glycine metabolism

---

To further confirm the core function of the modules, we considered the reactions with the highest degree in the module as the core reactions and analyzed the subsystem of the core reactions, which may inflect the core function of the related module. The first three reactions with the highest degrees were checked. We found that the function of the core reactions in each module agreed well with the main functions in Fig. 2 (ESI,† file S3).

In the common metabolism process of a cell, exogenous nutrients are firstly decomposed into primary metabolites and then synthesized into 12 kinds of common precursors, such as glucose-6-phosphate, fructose-6-phosphate, phosphoenolpyruvic acid, pyruvic acid (as shown in Fig. 3). The common precursors are subsequently used for the synthesis of the key components of the cell, such as amino acids, nucleotides, fatty acid and sugar.⁴² In order to verify the basic metabolic functions of the genome-scale metabolic network model for E. sinensis eyestalks, the synthesis process of the 12 common precursors was investigated. As shown in Fig. 3, all the 12 precursors can be found in the network. They distribute in three modules, and half of them were in module 12 where reactions were mainly associated with carbohydrate metabolism. The TCA cycle scattered in two modules, which agreed with the diversity of the TCA pathway.⁴³ In addition to the common precursors, module 16 contained a number of building blocks from these precursors, such as aspartate and asparagine. The above analysis showed that the eyestalk has the basic capability of synthesizing common precursors and building blocks.

Fig. 3 Distribution of 12 precursors in the modules of the eyestalk metabolic network. All precursors were distributed in three modules, and half of them centered in module 12.

Analysis of differentially expressed unigenes

In the comparison of the mRNA expression from A0 with that from A1 and A2, we got 51 810 and 53 833 differentially expressed unigenes respectively. In these unigenes, both 135 can be found in the metabolic network respectively (ESI,† file S4). The result of the expression profile is shown in Table 5.

Table 5 The results of the experiment

Experimental	Differentially expressed unigenes			Differentially expressed unigenes in module			Differentially expressed unigenes in the biggest WCC
A0	51 810	Up	12 716	135	Up	11	89	Up	5
A1		Down	39 094		Down	124		Down	84
A0	53 833	Up	49 048	135	Up	118	89	Up	77
A2		Down	4785		Down	17		Down	12

---

5-HT injection results in the expression changes of 90 unigenes which are related to 90 reactions. These reactions were mainly in modules 7, 11, 12, 13, 14, and 15. To be specific, the metabolic difference reaction of amino acid mostly distributed in module 5 (in which metabolism of cofactors and vitamins dominates), modules 7, 10, 14 (in which amino acid metabolism dominates) and module 15 (in which nucleotide metabolism and carbohydrate metabolism dominate); the metabolic difference reaction of carbohydrate was in module 15 and module 12 (in which carbohydrate metabolism dominates); the metabolic difference reaction of lipid existed in modules 11 and 13 (in which nucleotide metabolism dominates); the metabolic difference reaction of cofactors and vitamin centered in modules 3 and 14 (in which amino acid metabolism dominates); the metabolic difference reaction of the nucleotide was in modules 13 and 15 (as shown in Fig. 4). 5-HT is a neurotransmitter in the nervous system of crustaceans.^44,45 Its function is to help the X organ-sinus gland complex (XO-SG) in the eyestalk of crustaceans to release some neurohormones, such as high blood sugar hormone (CHH),^46,47 molt inhibiting hormone (MIH).⁴⁸ The differential expression results show that the function of 5-HT in eyestalks is accomplished by various metabolic subsystems. The differences between the differentially expressed reactions and the dominated reactions in a module indicate that the injection of 5-HT may influence the dominated metabolisms in the module mainly through the differentially expressed reactions and thereby influence the function of eyestalks.

Fig. 4 Functional distribution of differentially expressed reactions in the modules of the eyestalk network.

Differentially expressed genes found after the injection of glucose were the same as those after the injection of 5-HT. It is because that the eyestalk secretes neuropeptide hormone whose chemical nature is protein which was composed of amino acids. Precursors synthesized into amino acid are in the pathways of carbohydrate metabolism. Nucleotide metabolism and amino acid metabolism are closely linked because they are all related to nitrogen. Lots of unigenes have different change trends after the injection of glucose and 5-HT. The main metabolic processes influenced by these unigenes are shown in Table 6. That may be because the injection of 5-HT motivates the release of CHH which improves the increase of blood sugar, whereas the injection of glucose may inspire the regulation for decreasing the blood sugar. 5-HT/glucose exerts an effect on the metabolism of several substances in eyestalks, such as ammonia acid, proline, amino sugars and uracil, thus affecting the metabolic process of E. sinensis.

Table 6 The different influences on the metabolic process between the injection of 5-HT and glucose

Pathway		Process	General trend
			5-HT	Glucose
Amino acid metabolism	Arginine and proline metabolism	L-Arginine → L-arginine phosphate; L-proline → hydroxyproline	Down	Up
	Glycine, serine and threonine metabolism	Sarcosine ↔ glycine ↔ L-threonine; serine ↔ L-cysteine	Down	Up
	Cysteine and methionine metabolism	AdoMet → S-adenosylhomocysteine → homocysteine	Down	Up
Carbohydrate metabolism	Pentose phosphate pathway	Beta-D-fructose 6-phosphate → alpha-D-glucose 6-phosphate	Down	Up
	Amino sugar, nucleotide sugar metabolism	GlcNAc → GlcNAc-6P → GlcNAc-1P → UDP-GlcNAc → chitin	Down	Up
Nucleotide metabolism	Uracil metabolism	UDP ↔ UMP ↔ Uracil ↔ pseudouridine 5′-phosphate	Down	Up
	Adenine metabolism	GMP → IMP ← AMP ↔ ADP ↕ hypoxanthine	Down	Up
Lipid metabolism	Sphingolipid metabolism	Dihydroceramide → ceramide ↔ sphingomyelin	Down	Up
	Glycerophospholipid metabolism	Ethanolamine → ethanolamine phosphate → CDP-ethanolamine	Down	Up

---

Conclusions

A genome-scale metabolic network model has been widely used in the analysis of the metabolic function and processes for many kinds of organisms. In this work, we reconstructed a genome-scale metabolic network for E. sinensis eyestalks on the basis of transcriptome sequencing. The biological feature of the model was evaluated by the topology analysis. The analysis of the biggest WCC shows that amino acid metabolism and carbohydrate metabolism constitute the largest proportions in the eyestalk which indicates the function of eyestalks in peptide hormone secretion and energy metabolism. 5-HT and glucose injections lead to the changes in amino acid metabolism and carbohydrate metabolism, which illustrates that 5-HT and glucose may influence the neuropeptide hormone secretion of the eyestalk, and further certified that the eyestalk is the control center of the endocrine system in E. sinensis.

Acknowledgements

This work was supported by National High-Tech Research and Development Program of China (863 programs, 2012AA10A401), Grants of the Major State Basic Research Development Program of China (973 programs, 2012CB114405), National Key Technology R&D Program (2011BAD13B07 and 2011BAD13B04), National Natural Science Foundation of China (21106095, 61100124), Natural Science Foundation of Tianjin (15JCYBJC30700), Foundation of Introducing Talents to Tianjin Normal University (5KQM110003), Academic innovation promotion project of Tianjin Normal University for young teachers (52XC1403), and “131” Innovative Talents cultivation of Tianjin.

Notes and references

1. A. Q. Yu, X. K. Jin, X. N. Guo, L. Shuang, M. H. Wu, W. W. Li and Q. Wang, Fish Shellfish Immunol., 2013, 35, 1282–1292.
2. Y. Wang, Y. Zhang, Y. Sun, Y. Liu, X. Geng and J. Sun, J. Fish. China, 2013, 37, 987–993.
3. X. Li, Z. Cui, Y. Liu, C. Song and G. Shi, PLoS One, 2013, 8, 374.
4. Y. Sun, Y. Zhang, Y. Liu, S. Xue, X. Geng, T. Hao and J. Sun, PLoS One, 2014, 9, e95827.
5. H. Tong, Z. Zheng, B. Wang, Y. Zhang, Y. Liu, X. Geng and J. Sun, BMC Syst. Biol., 2014, 8, 417–422.
6. G. Zhengbing, Y. Shui, X. Zhou, X. Zenghong and Z. Zhaoyang, Jiang Su Agric. Sci., 2012, 40, 28–30.
7. P. S. Sun, Mol. Mar. Biol. Biotechnol., 1995, 4, 262–268.
8. E. Wang, N. Zaman, S. Mcgee, J. S. Milanese, A. Masoudi-Nejad and M. O’Connor-Mccourt, Semin. Cancer Biol., 2015, 30, 4–12.
9. J. D. Orth, T. M. Conrad, J. Na, J. A. Lerman, H. Nam, A. M. Feist and B. Ø. Palsson, Mol. Syst. Biol., 2011, 7, 535.
10. R. A. Notebaart, F. H. V. Enckevort, C. Francke, R. J. Siezen and B. Teusink, BMC Bioinf., 2006, 7, 296.
11. C. Francke, R. J. Siezen and B. Teusink, Trends Microbiol., 2005, 13, 550–558.
12. I. Thiele and B. Ø. Palsson, Nat. Protoc., 2010, 5, 93–121.
13. J. S. Edwards and B. O. Palsson, J. Biol. Chem., 1999, 274, 17410–17416.
14. T. Hao, B. Han, H. Ma, J. Fu, H. Wang, Z. Wang, B. Tang, T. Chen and X. Zhao, Mol. BioSyst., 2013, 9, 2034–2044.
15. R. Saha, P. F. Suthers and C. D. Maranas, PLoS One, 2011, 6, e21784.
16. S. Mintz-Oron, S. Meir, S. Malitsky, E. Ruppin, A. Aharoni and T. Shlomi, Proc. Natl. Acad. Sci. U. S. A., 2012, 109, 339–344.
17. M. Sigurdsson, N. Jamshidi, E. Steingrimsson, I. Thiele and B. Palsson, BMC Syst. Biol., 2010, 4, 140.
18. H. Ma, A. Sorokin, A. Mazein, A. Selkov, E. Selkov, O. Demin and I. Goryanin, Mol. Syst. Biol., 2007, 3, 135.
19. N. Duarte, S. Becker, N. Jamshidi, I. Thiele, M. Mo, T. Vo, R. Srivas and B. Palsson, Proc. Natl. Acad. Sci. U. S. A., 2007, 104, 1777–1782.
20. E. O’Brien, J. Monk and B. Palsson, Cell, 2015, 161, 971–987.
21. T. Hao, H.-W. Ma, X.-M. Zhao and I. Goryanin, BMC Bioinf., 2010, 11, 393.
22. T. Hao, H.-W. Ma, X.-M. Zhao and I. Goryanin, Mol. BioSyst., 2012, 8, 663–670.
23. A. Liu, Z. Wang, W. Nie and Y. Su, Inf. Sci., 2015, 320, 429–442.
24. A. A. Liu, K. Li and T. Kanade, IEEE Trans. Med. Imaging, 2012, 31, 359–369.
25. A. A. Liu, Y. T. Su, P. P. Jia, Z. Gao, T. Hao and Z. X. Yang, IEEE Trans. Cybern., 2015, 45, 1194–1208.
26. A. A. Liu, Y. T. Su, W. Z. Nie and Z. X. Yang, PLoS One, 2015, 10, e0130884.
27. S. A. Becker, A. M. Feist, M. L. Mo, G. Hannum, B. Ø. Palsson and M. J. Herrgard, Nat. Protoc., 2007, 2, 727–738.
28. J. M. Lee, E. P. Gianchandani and J. A. Papin, Briefings Bioinf., 2006, 7, 140–150.
29. J. M. Park, T. Y. Kim and S. Y. Lee, Biotechnol. Adv., 2009, 27, 979–988.
30. D. Koschützki, B. H. Junker, J. Schwender and F. Schreiber, J. Theor. Biol., 2010, 265, 261–269.
31. H. W. Ma, X. M. Zhao, Y. J. Yuan and A. P. Zeng, Bioinformatics, 2004, 20, 1870–1876.
32. T. Hao, B. Wang, A. Yu, A. Liu and J. Sun, MNetDec: A Flexible Algorithm for Metabolic Network Decomposition, 2015.
33. H. Ma and A.-P. Zeng, Bioinformatics, 2003, 19, 270–277.
34. R. L. Chang, L. Ghamsari, A. Manichaikul, E. F. Hom, S. Balaji, W. Fu, Y. Shen, T. Hao, B. Ø. Palsson and K. Salehi-Ashtiani, Mol. Syst. Biol., 2011, 7, 518.
35. J. M. Dreyfuss, J. D. Zucker, H. M. Hood, L. R. Ocasio, M. S. Sachs and J. E. Galagan, PLoS Comput. Biol., 2013, 9, e1003126.
36. V. Batagelj and A. Mrvar, Lecture Notes in Computer Science, 2002, vol. 2265, pp. 77–103.
37. J. Zhao, H. Yu, J.-H. Luo, Z.-W. Cao and Y.-X. Li, BMC Bioinf., 2006, 7, 386.
38. W. Zaizhao and X. Jianhai, J. Fish. China, 2001, 175–180.
39. J. Dongcai, J. Han, J. Xiaojun and Y. Haiming, China Feed, 2014, 29, 31–36.
40. K. r. C. Williams, N. E. Renthal, R. D. Gerard and C. R. Mendelson, Mol. Endocrinol., 2012, 26, 1857–1867.
41. J. KAN, L. LI and J.-y. XU, Chin. J. Biochem. Pharm., 2009, 4, 020.
42. M. Csete and J. Doyle, Trends Biotechnol., 2004, 22, 446–450.
43. L. J. Sweetlove, K. F. Beard, A. Nunes-Nesi, A. R. Fernie and R. G. Ratcliffe, Trends Plant Sci., 2010, 15, 462–470.
44. R. R. Eitenmiller, J. H. Orr, W. W. Wallis, Chemistry and biochemistry of marine food products, 1982, pp. 39–50.
45. L. Laxmyr, Comp. Biochem. Physiol., C: Comp. Pharmacol., 1984, 77, 139–143.
46. R. Keller, Cell. Mol. Life Sci., 1992, 48, 439–448.
47. W. Lüschen, A. Willig and P. P. Jaros, Comp. Biochem. Physiol., C: Comp. Pharmacol., 1993, 105, 291–296.
48. M. P. Mattson and E. Spaziani, Biol. Bull., 1985, 169, 246–255.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5mb00571j

---