Our team successfully developed a fat-reduction and muscle-enhancement system using Escherichia colias the chassis bacterium. Through codon optimization and genetic engineering, we designed engineered bacteria capable of secreting GLP-1 and Bimagrumab, aimed at reducing fat and increasing muscle mass. Additionally, we designed a DPP-4 inhibitor to extend the half-life of GLP-1, with its stability confirmed via molecular dynamics simulations. We also constructed arabinose operon-induced and cold-inducible suicide systems to prevent gene leakage from the engineered bacteria. The successful establishment of these systems lays the foundation for future applications of probiotics in fat reduction.

Design

Glucagon-like peptide-1 (GLP-1) is a hormone primarily produced by intestinal L-cells, which promotes insulin secretion and inhibits gastric emptying to achieve weight loss[1]. E. coli has several advantages as a chassis for genetic engineering, including rapid growth, ease of cultivation, straightforward genetic manipulation, low cost, and precise genome modification. This makes it an ideal choice for producing proteins and other biomolecules[2]. Therefore, we selected E. coli as the microbial chassis for GLP-1 production.

Build

First, we performed codon optimization of the GLP-1 sequence based on E. coli and commissioned Jinweizhi Company for fragment synthesis, including the synthesis of fragment primers. We chose BBa_J23100 as the promoter, BBa_B0034 as the ribosome binding site, and BBa_B0015 as the terminator codon (Fig 2). Since our goal is to develop a probiotic capable of continuously secreting GLP-1, we attached an extracellular signal peptide, PelB, to the N-terminus of GLP-1 to facilitate the successful secretion of GLP-1 by the engineered bacteria.

Next, we amplified the PelB-GLP-1 fragment using PCR (Fig 3.A) and ultimately integrated this fragment into the vector pET23b. The construct was then transformed into E. coli BL21 for protein expression verification. The transformed engineered bacteria were plated on LB agar containing ampicillin, resulting in the successful selection of monoclonal colonies (Fig 3.B).

Test

To determine whether the engineered bacteria could successfully express GLP-1 and secrete it extracellularly with the assistance of PelB, we centrifuged the culture containing the PelB-GLP-1 fragment to obtain the supernatant and bacterial pellet. The proteins were extracted from the bacterial pellet using a lysis method. We then used an ELISA kit (Shanghai Enzyme Linked, YJ022729) to verify the expression of GLP-1 in the engineered strain. The results indicated that in the absence of the PelB signal peptide, the intracellular concentration of GLP-1 reached 60.7 pg/mL, while the extracellular concentration was only 8 pg/mL. However, with the assistance of the PelB signal peptide, the extracellular level of GLP-1 increased to 62 pg/mL (Fig 4.B). Subsequently, we performed Western blotting on the extracted GLP-1 from both intracellular and extracellular sources. The results showed that without the PelB signal peptide, GLP-1 was predominantly localized intracellularly; whereas, after linking the PelB signal peptide to the N-terminus of GLP-1, the extracellular concentration significantly increased (Fig 4.A). Finally, we validated that GLP-1 and the PelB-GLP-1 fragment did not adversely affect the growth of the strains (Fig 4.C).

Learn

We successfully constructed engineered bacteria capable of secreting GLP-1 extracellularly. However, GLP-1 is prone to degradation by dipeptidyl peptidase-4 (DPP-4) in the bloodstream, resulting in a half-life (t₁/₂) of only a few minutes. Therefore, our team chose to design a protein inhibitor of DPP-4 to enhance the half-life of GLP-1 in vivo, allowing it to function more effectively.

Design

As there are no naturally occurring protein inhibitors that can competitively bind to DPP-4, we plan to utilize existing deep learning models for de novo design. We employed four deep learning models: AlphaFold3[3], RFdiffusion[4], ProteinMPNN[5], and ESMfold[6].

Build

To design a protein inhibitor for DPP-4, we established a comprehensive design and screening workflow: RFdiffusion generates the protein backbone, ProteinMPNN generates the side chains, and ESMfold filters for high-quality sequences. First, we modeled the protein complex of GLP-1 and DPP-4 using AlphaFold, then selected the key residues E167, E168, and Y624 in the DPP-4 pocket as hotspots using PyMOL. Next, we utilized the Practical Considerations for Binder Design module for the design. Subsequently, the selected backbone was used in the ProteinMPNN's FastRelax protocol for side chain generation. Finally, the complete amino acid sequences generated were modeled using ESMfold (Fig 5).

Test

For each step of the generated sequences, we established screening criteria:

Backbone Generation: We used structural observation methods to empirically select the most suitable structure for subsequent side chain generation.

Side Chain Generation: After generating the complete amino acid sequence, we modeled it using ESMfold and filtered for sequences with RMSD < 15 and pLDDT > 80 compared to the Cα of the protein backbone, ultimately selecting 18 sequences suitable for molecular dynamics simulation (Fig 6.A).

Molecular Dynamics Simulation: The selected sequences underwent long-term molecular dynamics simulations using the Amber program. This included a 100 ns equilibration followed by 100 ns of production dynamics, totaling 200 ns.

MM/PBSA Calculation: We performed MM/PBSA calculations on the molecular dynamics results, identifying one sequence with a binding energy to DPP-4 that was superior to that of GLP-1. The energy of the DPP-4 protein inhibitor (DPP-4PI) in solution was -6522.0487 kcal/mol, compared to -6474.9481 kcal/mol for GLP-1 with DPP-4 (Fig 6.B).

Trajectory Analysis: Finally, we analyzed the trajectories from the molecular dynamics simulations, confirming that both RMSD and RMSF trajectories remained within reasonable ranges (Fig 6.C and D).

Learn

We successfully established a workflow for artificially designed proteins and utilized existing deep learning models to effectively create a DPP-4 protein inhibitor, validating its rationality on a computational level. Although we have demonstrated the feasibility of this method computationally, we have not yet conducted experimental validation due to time constraints, which will be a core focus for our team in the future.

Design

Bimagrumab is a humanized monoclonal antibody that binds to type II hormone receptors on muscle cells. It is primarily used to increase skeletal muscle density for the treatment of muscle atrophy diseases and has been evaluated in several clinical trials. Therefore, we chose to utilize Bimagrumab to construct a muscle-building system for probiotic development.

Build

We first retrieved the sequence information for Bimagrumab from RCSB (PDB ID: 5NH3) and subsequently performed codon optimization. The optimized sequence and the designed primers were then synthesized by Genewiz. We selected BBa_J23100 as the promoter, BBa_B0034 as the ribosome binding site, and BBa_B0015 as the stop codon (Fig 7). Similar to the scenario with GLP-1, Bimagrumab also needs to be continuously secreted into the extracellular space, so we attached the PelB signal peptide to the N-terminus of Bimagrumab.

Next, we ligated the amplified sequence (Fig 8) into the vector pET23b and transformed it into E. coli BL21 for protein expression validation.

Test

We conducted Western blotting experiments to validate the expression of Bimagrumab and PelB-Bimagrumab both intracellularly and extracellularly. The results indicated that, in the absence of the PelB signal peptide, Bimagrumab was primarily located within the cells. However, after the addition of the PelB signal peptide to the N-terminus of Bimagrumab, its concentration in the extracellular space significantly increased(Fig 9).

Finally, we verified that the plasmids containing the Bimagrumab and PelB-Bimagrumab fragments did not affect the growth of the bacterial strain(Fig 10).

Learn

We successfully constructed the muscle-building system and verified that Bimagrumab can be effectively secreted into the extracellular space with the help of the PelB signal peptide. This indicates that our probiotic can reduce muscle loss while promoting fat reduction, helping patients maintain their health more effectively.

Design

To ensure that the genetic fragments do not leak into the environment and to allow for the timely removal of the engineered bacteria from the patient's body when needed, we used the arabinose operon as a switch to induce bacterial death. Our team first verified the function of the existing part BBa_I13453 in E. coli DH5α.

Build

To verify whether the arabinose operon can function properly in E. coli DH5α, we linked the mRFP fragment to the arabinose operon (Fig 11) and assessed the operon's functionality by measuring fluorescence values at different arabinose concentrations.

Test

We measured the expression of red fluorescent protein under the induction of the arabinose operon. As the concentration of arabinose increased, the Fluorescence/OD600 ratio also increased, demonstrating that the arabinose operon can induce the expression of the corresponding protein in response to arabinose (Fig 12).

Learn

We successfully tested the function of the arabinose operon in E. coli DH5α, which lays the foundation for the subsequent construction of the suicide system.

Design

Under typical circumstances, if the engineered bacteria leak into the environment, they cannot self-destruct via arabinose induction. Therefore, we chose a more widely occurring environmental factor—low temperature—as the condition to induce the engineered bacteria to self-destruct. In the parts library, we identified the low-temperature inducible promoter pCspA (BBa_K4987003) and began by verifying its functionality.

Build

Using a method similar to that used to validate the arabinose operon, we connected the mRFP segment to the low-temperature inducible promoter pCspA (Fig 13). We then validated the function of pCspA by measuring the fluorescence of E. coli at different cultivation temperatures.

Test

The results indicated that under low-temperature conditions, the OD600 value of BL21 was only 0.5 within 12 hours, while at 37°C, the OD600 value reached 2.3. However, under low-temperature conditions, the Fluorescence/OD600 value reached 278, significantly higher than the Fluorescence/OD600 value at 37°C (Fig 14). This demonstrates that the low-temperature inducible promoter CspA can effectively induce the expression of red fluorescent protein under low-temperature conditions.

Learn

We successfully validated the functionality of the low-temperature inducible promoter pCspA in E. coli DH5α, laying the groundwork for the subsequent construction of the suicide system.

Design

To successfully induce the suicide of engineered bacteria, we selected two lytic proteins derived from the T4 bacteriophage: T4 Holin and T4 Lysozyme. T4 Holin is a small membrane protein with two transmembrane domains, with the N-terminus and C-terminus located in the cytoplasm and extracellular space, respectively. T4 Lysozyme is a small globular protein that can break the chemical bonds between N-acetylglucosamine and N-acetylmuramic acid. The combined action of these two proteins can effectively kill the engineered bacteria.

Build

We first connected the T4 Holin and T4 Lysozyme fragments to the arabinose operon and low-temperature inducible promoter, respectively, and then linked the two systems using an 'OR' gate (Fig 15).

Test

We verified the survival status of bacteria under the induction of the two promoters. The results showed that in the presence of 0.5 mM/L arabinose, the OD600 of the culture significantly decreased after 5 hours of incubation, approaching 0 by the 20th hour (Fig 16.A). In contrast, under low-temperature conditions at 16°C, the engineered bacteria containing the lysozyme sequence remained at a low density, with the OD600 nearly constant at 0.3 (Fig 16.B).

Learn

Our team successfully constructed a suicide system using pBAD and pCspA as promoters and T4 Holin and T4 Lysozyme sequences as suicide genes, and we successfully verified the functionality in E. coli DH5α.

[1] Li Y, Hansotia T, Yusta B, et al. Glucagon-like peptide-1 receptor signaling modulates β cell apoptosis[J]. Journal of Biological Chemistry, 2003, 278(1): 471-478.

[2] Lee S Y. High cell-density culture of E. coli [J]. Trends in biotechnology, 1996, 14(3): 98-105.

[3] Abramson J, Adler J, Dunger J, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3[J]. Nature, 2024: 1-3.

[4] Watson J L, Juergens D, Bennett N R, et al. De novo design of protein structure and function with RFdiffusion[J]. Nature, 2023, 620(7976): 1089-1100.

[5] Dauparas J, Anishchenko I, Bennett N, et al. Robust deep learning–based protein sequence design using ProteinMPNN[J]. Science, 2022, 378(6615): 49-56.

[6] Song Y, Yuan Q, Chen S, et al. Accurately predicting enzyme functions through geometric graph learning on ESMFold-predicted structures[J]. Nature Communications, 2024, 15(1): 8180.