Metabolic Network

Global Metabolic Network (GMN)

Goals

We will optimize β-Glucan production by reconstructing the metabolic network of Aureobasidium melanogenum BZ-11. Due to the lack of existing genome-scale metabolic models (GEMs) for BZ-11, we decided to start with the metabolic networks of related species and build our own model. We will use flux balance analysis (FBA) and the OptForce optimization algorithm to identify pathways affecting β-Glucan production, providing essential support for experimental design.

Reconstruct Metabolic Network

The A. melanogenum BZ-11 strain was chosen for its ability to efficiently produce β-Glucan. We aim to regulate its metabolic pathways at the genetic level through gene editing to increase β-Glucan yield. However, due to the complexity of such a vast metabolic system, it is impractical to rely on kinetic calculations alone. After discussions with the SDU-China team, we were inspired to use GEMs to regulate metabolic pathways. As a systems biology tool, GEMs are based on genome-wide annotations and experimental data, utilizing gene-protein-reaction associations. They encompass system-wide information, including metabolites, reactions, genes, and compartments to describe the metabolic network of an organism. By using optimization techniques, GEMs allow us to predict the metabolic flux of system metabolites. ^[1]
Our initial plan was to construct GEMs for A. melanogenum, but we find official databases such as Biomodels and BiGG Models do not list any information for A. melanogenum, let alone the BZ-11 subspecies. We spent a long time addressing this problem and explored various solutions.

Phase one: We decided to look for metabolic networks in closely related species instead.
We reviewed literature and found a phylogenetic tree of 19 Ascomycota species constructed based on the internal transcribed spacer (ITS) region. ^[2]
By comparing the phylogenetic tree, we identified the closest species with a metabolic network available in the database, which is the GEMs network of Aspergillus niger. However, we observed that many pathways in the GEMs of Aspergillus niger differ from what we know about A. melanogenum[3], such as the synthesis of Melanin. A. melanogenum utilizes the dihydroxynaphthalene (DHN) pathway, while Aspergillus niger utilizes the L-3,4-dihydroxyphenylalanine (L-DOPA) pathway. This discrepancy significantly impacts our analysis, and this piecemeal approach is also inefficient and undesirable.

Phase two: We decided to construct the metabolic network ourselves.
After gathering information, we found that constructing GEMs requires whole-genome sequencing of the strain, which posed another challenge for us. Given the current situation and the issues we faced, we consulted our instructor, who supported our solution and generously provided us with the whole-genome sequencing information for the A. melanogenum P16 strain. This was of great help to us. (A. melanogenum P16 is a widely used strain that is highly similar to our A. melanogenum BZ-11 strain.)

Phase three: We will look for a suitable tool to construct the GEM metabolic network.
After reviewing literature and gathering information [4], we ultimately chose the CarveMe reconstruction tool. This is a Python-based automated tool for GEMs reconstruction that uses the Diamond sequence aligner (an efficient alternative to BLAST in bioinformatics) to compare sequences in the database and build the metabolic network. To maintain high efficiency, we ran it on Linux Ubuntu, successfully obtaining the GEMs. This is a critical step for us. However, this was not enough; we manually modified the network, correcting some mass balance issues and adding additional metabolic pathways. While this helps us, it further assists other iGEM teams in need of GEM metabolic network analysis as a guidance.

Search Target

FBA Simulation
GEMs networks analysis is flux balance analysis (FBA), which is a widely used linear programming (LP) approach for studying the flow in large biochemical networks. It is based on the steady-state assumption and seeks to find feasible solutions that maximize the objective function while satisfying constraints, resulting in the optimal solution. FBA forms a matrix of reactions and metabolites, applying flux to each reaction, thus creating constraints. Under the defined final objective function, it selects the optimal solution.^[5].

Fig. 1 Principles of FBA Simulation, adapted from previous work[5]

Assumptions:
1. Steady-State Assumption: The system can reach a stable state, after stabilization, the concentration of metabolites does not change over time, and the net production rate of the system is zero.
2. Enzyme Kinetics: We ignore the kinetic properties of the metabolic network and focus solely on the distribution of metabolic flux rather than specific enzyme kinetic parameters.

To perform flux analysis, we will use COBRApy for simulation. This is a Python-based tool that enables simple flux balance analysis and reaction reconstruction. To observe the maximization of growth rate and β-Glucan production, we optimized the model twice, using β-Glucan and growth as objective functions, respectively (Fig. 2). Additionally, to prevent cell death caused by the overproduction of secondary metabolites, we must consider the impact of β-Glucan on cell growth and have established the relationship between cell growth and the flux of β-Glucan (Fig. 3).

Fig. 2 FBA Simulation Results for Wild-type

Fig. 3 Flux Between β-Glucan and Growth

From the analysis results, we find that β-Glucan can reach a maximum of 667 mmol/gDW/h, while growth can reach 87 mmol/gDW/h. Additionally, we observe that the flux of β-Glucan does not affect growth until it reaches 175 mmol/gDW/h. After reaching the maximum production, growth decreases by approximately 87%. Thus, while aiming to maximize β-Glucan production, we need to minimize its impact on cell growth and reduce the system's burden.

OptForce Simulation
Inspired by the 2023 Leide team, we use the OptForce algorithm based on the COBRA Toolbox from MathWorks. This algorithm, which utilizes flux variability analysis (FVA), identifies the range of each flux while satisfying flux constraints to implement optimization strategies. OptForce compares the wild-type and mutant MUST sets, which encompass the pathways that need to be altered to achieve the desired goals.

1. First-order MUST sets consist of individual reactions, where MUSTL indicates reactions that must be downregulated and MUSTU indicates reactions that must be upregulated.
2. Second-order MUST sets are more complex and involve greater computational effort, as they logically combine reactions that collectively affect metabolic flux distribution. MUSTLL represents logical combinations of reactions that must be downregulated, while MUSTUU indicates those that must be upregulated. If certain flux ranges in the wild-type cannot overlap with those of the mutant, it indicates these reactions cannot achieve the excessive production of the target product. OptForce will use this information to modify the reactions accordingly. ^[6]

If certain flux ranges in the wild-type cannot overlap with those of the mutant, it indicates that these reactions cannot achieve the excessive production of the target product. OptForce will use this information to modify the reactions accordingly ^[6].

Fig. 4 Wild-Type and Target Overproduction MUST Set Model, adapted from previous work[6]

Assumptions:
1. Under the steady-state assumption, we simulate by glucose uptake while considering
2. biological constraints such as β-Glucan production.

Through reasonable constraints and calculations, we obtained the first-order and second-order MUST sets and conducted simulations using the OptForce algorithm, which returned the top 10 single reaction force sets. (Fig. 5)

Fig. 5 OptForce Simulation Results

Combining the reactions, we analyzed the following: Reaction 2 and Reaction 3 (these are transport reactions); Reaction 4, Reaction 5, and Reaction 10 (these are involved in energy metabolism); Reaction 1 (which is a key step in carbohydrate metabolism); Reaction 6 (which is involved in the synthesis of Pullulan); Reaction 7 (which participates in cell wall synthesis); Reaction 8 (which is related to amino acid metabolism); and Reaction 9 (which is involved in fatty acid β-oxidation).
Considering the premise of not affecting cellular metabolism, we ultimately chose the Pullulan synthesis reaction Pullulan_01 as our final target for OptForce. At the same time, we evaluated the preceding reaction for Pullulan synthesis, Pullulan_00, and conducted FBA simulations for both. (Fig. 6)

Fig. 6 Flux Between Pullulan_00 and β-Glucan Flux at Different Pullulan_01

Under the same Pullulan_00 flux, an increase in Pullulan_01 flux leads to a decrease for β-Glucan. And as Pullulan_00 flux increases, β-Glucan flux also decrease. It is evident that both Pullulan_00 and Pullulan_01 have a significant impact on β-Glucan production. While we focus on Pullulan_01, we should also pay attention to the presence of Pullulan_00.
At the same time, we have also achieved some results in wet experiments. By analyzing the extracellular polysaccharide components, we discovered the presence of Pullulan, which was a pleasant surprise. Combining this with our model predictions, we found targeting Pullulan can not only enhance the intrinsic yield of β-Glucan but also help avoid impurity issues during the purification of extracellular polysaccharides! This also indicates that the direction of our dry experiments is correct.
However, we also observed Melanin deposition in the extracellular space, which adversely affects the color quality of our product. After reviewing the literature and discussing with our instructor and PI, we decided to target the synthesis pathways of Pullulan and Melanin for modification. This is still not sufficient. Although both are secondary metabolites, we cannot completely knock out the pathways for these two reactions, as that is not feasible. We decided to conduct further FBA simulations and gene knockout simulations. By observing the flux and reduced costs returned from FBA, as well as the target product quantities from the knockout simulations, we performed a more detailed analysis. However, we found that controlling these sub-reactions was not sensitive, our target product showed only minor changes in yield. We were unable to draw direct conclusions from the simulation results, and further research is needed.

Summary

Ultimately, through the analysis of the GEMs metabolic network, we identified the pathways that influence β-Glucan production, specifically in Pullulan synthesis, leading to a preliminary optimization plan for the target product's metabolic pathway. We also provided a modified GEMs for A. melanogenum P16 to other iGEM teams that may need it.
However, our modeling has its limitations, modifications to the model require continuous iterations and experimental validation, which is a time-consuming process. We cannot complete such a large workload in a short period, we will continue to optimize the model and contribute to the development of GEMs in the future.

Local Metabolic Network (LMN)

Goals

After analyzing the GMN, we narrowed our "scope of focus" to concentrate solely on the metabolic pathways of β-Glucan, Melanin, and Pullulan. Based on this, we decided to adopt traditional kinetics for modeling to conduct a more detailed analysis and further identify the key reactions we aim to knock out.

Biochemical Network Simulation

We used MathWorks' SimBiology for kinetic modeling, which is widely used for pharmacokinetic simulations and is also an efficient tool for simulating small metabolic systems. It allows us to observe the dynamic behavior of the network and provides more precise control and optimization strategies. Through literature searches and readings ^[7][8][9], we constructed the metabolic network. (Fig. 7)

Fig. 7: local metabolic network

Biochemical network simulations allow us to dynamically observe the changes of various substances over a certain time scale, which helps us understand the dynamic behavior of the network better.

Assumptions:
1. We assume that the amount of enzymes in the system is sufficient and has not reached saturation. We also neglect enzyme inhibition and activation effects, focusing primarily on which reactions the enzymes participate in, thereby further simplifying the model load.
2. The simulation focuses on the dynamic changes of low concentrations substances for preliminary predictions and simulations.

We describe the kinetic equations for each substance by linear equations. The X is the amount of the target substance, and K_fi and K_fj are the kinetic constants for the reactions R_i (synthesis of X) and R_j (decomposition of X), respectively.

Fig. 8: Metabolic network simulation results

As shown in Fig. 8, we visualized the concentration changes of Glucose, Pullulan, β-Glucan, and Melanin. β-Glucan is produced at a high yield, while the contributions of Melanin and Pullulan to the system should not be overlooked.

Local Sensitivity Analysis

Sensitivity analysis is used to measure the response of the network to small changes in different parameters. By observing the response results, we can identify the parts of the system that are highly sensitive and combine this with the results of the dynamic simulations to gain a deeper understanding of our system's behavior.
Considering that there are not many synergistic effects in the network, we adopted a lightweight local sensitivity analysis (LSA). This approach uses partial derivatives for calculations, assuming that all other parameters remain constant and only considering changes in a single parameter, focusing on small perturbations around a specific point in the model, making it suitable for simpler model analyses.

The sensitivity of the target substance X is analysed in terms of how much X is affected by a change in the reaction rate constant K_f. If the value of S is large, then X is very sensitive to changes in K_f.
We will focus on downregulating the metabolic pathways of Melanin and Pullulan and perform sensitivity analyses on the parameters of each reaction.

Fig. 9 LSA analysis of the Melanin synthesis pathway

Fig. 10 LSA analysis of the Pullulan synthesis pathway

Fig. 9 and Fig. 10 show the Melanin and Pullulan's sensitivity versus time and integral plots for different parameter conditions. We can find the sensitivity values for the Melanin production reaction regulated by Kf_25 and the Pullulan production reaction regulated by Kf_26 are the highest, indicating that these two metabolic pathways have the greatest impact on the variations of their respective final products. The results of these LSA analyses provide valuable insights for the system, and we will subsequently focus on the regulation of these two genes.

Parameter Scan

In order to further validate the results of the sensitivity analysis and the degree of fluctuation of the product effects, we observed the yield of the target products under different parameters and iterated the parameters of the top three reactions that are highly sensitive to the synthesis of Pullulan and Melanin.

Fig. 11: Parameter scan of the three reactions involved in Melanin synthesis

Fig. 12: Parameter scan of the three reactions involved in Pullulan synthesis

From the analysis results (Fig. 11 & Fig. 12), it is clear that the fluctuations in the curves for Kf_26 involved in Pullulan synthesis and Kf_25 involved in Melanin synthesis are significantly greater than those of the other groups, highlighting their impact on the corresponding products. This not only further validates the correctness of our LSA but also allows us to focus on these two reactions. This provides substantial assistance in understanding the system dynamics and predicting future experiments.

Summary

We constructed a LMN and further explained and analyzed the system. Through system simulations, LSA, and parameter scans, we conducted multi-dimensional analyses and identified key pathways affecting β-Glucan production, specifically those related to Pullulan and Melanin synthesis. This provides critical insights for our project. Our experiments will focus on the Melanin_01 and Pullulan_01 reactions. However, due to the compact nature of our experimental section, some parameters have not been fitted, making complete quantitative analysis currently unattainable. We recognize this as one of the directions for future model optimization.

GMN Re-analysis

Goals

Through further analysis of the LMN, we identified the key reactions for gene knockout. We compared the results of the LMN with the GEMs, and through additional visualization and analysis, we aimed to identify the shortcomings of the GMN analysis.

Visualization

We recursively analyzed the model using our custom scripts and identified the pathways related to Melanin, β-Glucan, Pullulan, and those discovered by OptForce from thousands of reactions. We then visualized these pathways using the SAMMI tool.

Fig. 13: SAMMI Visualization Network (Red marked are the reactions we're concerned)

Verification

We re-evaluated the global metabolic network to predict β-Glucan production and investigated the effect of Pullulan_01 on β-Glucan under different Melanin_01 conditions (Fig. 14).

Fig. 14 Flux Between Pullulan_01 and β-Glucan Flux Under Different Melanin_01

By maximizing the flux of Melanin_01, we found that under the same conditions of Pullulan_01 flux, the yield of β-Glucan further decreased. The area in between represents the impact of Melanin_01 on β-Glucan production. Our analysis indicates that Melanin_01 does indeed affect the synthesis of β-Glucan to some extent, revealing pathways that were not predicted by OptForce. This also suggests that there are many unknown pathways in the fungus that influence β-Glucan production. This necessitates integrating theoretical insights into practical fermentation for deeper exploration. Our understanding of the metabolic pathways not only provides suggestions for other iGEM teams but also offers numerous ideas for optimizing and improving our future projects.

Summary

We used our own scripts to identify the key pathways we are focusing on and conducted visualizations. We compared the results from the LMN back to the GMN to observe the effects of Melanin and Pullulan on β-Glucan. This also demonstrated that the GMN has many undiscovered key pathways, highlighting the need to optimize the model and add constraints for further exploration. This work also provides valuable suggestions for other iGEM teams.