1. Experiment: Determinate Binding Sites
2. Time: 2024.08.03-2024.08.07
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Method: Through reference research, we identified the native binding sites for IL-2 receptor subunits β and γ as the active sites.
5. Attachment: Determinate Binding Sites
1. Experiment: RFDiffusion-Based Protein Design
2. Time: 2024.08.29-2024.09.28
3. Member: Xudong Tang, Yang Jin, Binxuan, Zhang, Kaiqing Zhang, Xuantong Liu
4. Method:
RFDiffusion is a cutting-edge method for de novo protein structure and function design, leveraging the capabilities of deep learning and diffusion models. It is based on the RoseTTAFold (RF) structure prediction network, fine-tuned on protein structure denoising tasks. The principle behind RFDiffusion is to generate protein backbones by iteratively refining noise-corrupted structures, eventually producing functional and structurally accurate proteins from simple molecular specifications.
(1) Core Components:
① Denoising Diffusion Probabilistic Models (DDPMs): These models are trained to reverse the
process of adding Gaussian noise to protein structures, thereby generating new, realistic
protein structures.
② Rotational Equivariance: RFDiffusion maintains rotational equivariance, allowing it to model
three-dimensional (3D) structures in a global representation frame independent manner.
③ Conditioning Information: The model can be guided towards specific design objectives by
providing conditioning information at each step of the generation process.
(2) The procedure of RFDiffusion:
① Model Training
- Data Preparation: Sample protein structures from the Protein Data Bank (PDB) and introduce noise to create training inputs.
- Noise Application: Perturb Cα coordinates with 3D Gaussian noise and apply Brownian motion to residue orientations.
- Model Training: Train the RFDiffusion model by minimizing the mean-squared error (MSE) loss between frame predictions and the true protein structure.
② Protein Design
- Initialization: Start with random residue frames.
- Denoising Iterations: Iteratively refine the protein structure by denoising the noisy input, adding noise at each step to generate the input for the next iteration.
- Sequence Design: Use the ProteinMPNN network to design sequences encoding the generated protein structures.
③ Conditioning for Specific Designs
- Unconditional Design: Generate diverse protein structures without additional input.
- Topology-Constrained Design: Provide secondary structure and/or fold information to guide the design towards specific topologies.
- Symmetric Oligomer Design: Specify point group symmetry to create symmetric oligomeric structures.
④ Experimental Characterization
- Expression and Purification: Express the designed proteins in a suitable host and purify them for further analysis.
- Structural Verification: Use techniques such as circular dichroism (CD) and cryo-electron microscopy (cryo-EM) to verify the structure and stability of the designed proteins.
- Functional Validation: Assess the functionality of the designed proteins through binding assays, enzymatic activity tests, or other relevant functional assays.
5. Attachment: RFDiffusion Based Protein Design
1. Experiment: Using HDOCK to evaluating the binding energy
2. Time: 2024.09.10-2024.09.12
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Method:
HDock is an efficient protein-protein interaction docking tool that combines template matching and free docking algorithms to accurately predict intermolecular interactions. During the HDock precision screening phase, binding score calculations consider structural alignment between bait proteins and target proteins, enabling more effective identification of protein interactions with enhanced stability and higher binding affinity. This serves as a crucial screening metric that improves result accuracy.
(1) The procedure of HDock:
① Data input: The HDOCK web server is available at http://hdock.phys.hust.edu.cn/. Both sequence and structure information of ligand and receptor should be uploaded through the website separately. The docking process will be carried out in the server. Click submit to wait for the results. Generally, the waiting time is approximately 30 minutes. The submitted sequences should be in FASTA format and contain only standard amino acids. The structure should be in PDB format. In addition, residue information of the binding site is also possible to be submitted as a filter in the docking process.
② Docking preparation: First, the homologous sequences of molecules will be searched by sequence similarity search. The template with the highest sequence coverage, highest sequence similarity and highest resolution will be selected to build the template-based model. ClustalW is used for sequence alignment, and MODELLER is used for model construction. This step will be executed on the server.
③ Global docking: HDOCKlite, a layered docking program based on FFT, is used to globally sample the presumed binding directions. We subjected the screened proteins to HDock refined screening, where adjusting the Spacing and Angle parameters to 1.2 and 15 provides highly accurate docking results, offering data references for protein-protein interaction strength. Docking scores are generally negative values, with more negative scores indicating a higher probability of valid binding models. The ranked binding modes are clustered with an RMSD cutoff of 5 Å. if two binding modes have a ligand RMSD of ≤ 5 Å, the one with the better score is kept. This step will be executed on the server. The docking model of HDOCKlite and the template-based docking model of MODELLER will be provided for download through web interaction.
④ Results analysis: The server will send the top 10 docking models to the website. The results page contains an interactive visualization window and a summary table. The summary table usually contains docking energy score, confidence score and ligand RMSD. If the SAXS data file is provided, the SAXS chi-square of the model can also be displayed. Results can also be downloaded as a PDB file, using the molecular visualization software PyMOL to open and view the predicted model. The docking results of HDock are generally referenced by the docking score and input into the formula: Confidence_score = 1.0/[1.0 + e^(0.02*(Docking_Score+150))]. Taking a docking score of -200 as a reference, the confidence score at this point is approximately 0.7. In the official documentation, a confidence score > 0.7 (closer to 1) indicates a high likelihood that the two protein molecules will bind; a confidence score between 0.5 and 0.7 indicates that the two proteins may bind; a confidence score < 0.5 indicates that it is unlikely that the two molecules will bind. Therefore, a docking score lower than -200 corresponds to a very high probability of binding.
5. Attachment: Using HDOCK to evaluating the binding energy
1. Experiment: Hydrogen bond depiction
2. Time: 2024.09.30-2024.10.05
3. Member: Fan Yang, Qiwen Jiang, Men Sun, Xinxin Zhang, Ziyan Yu
4. Method:
(1) The PDB file for the protein is prepared.
(2) In PyMol, chains should be renamed. Then, hydrogen bonds are found by clicking "find-polar-contacts-to other atoms in object". The stick structures of the protein molecules are displayed. After the amino acids connected by hydrogen bonds are clicked, we show the cartoon figure of the molecules and show the stick structure of the amino acids clicked. Then, the sequence numbers and names of the amino acids can be annotated by clicking on "label-residue".
(3) Click "draw/ray-save image to file".
5. Attachment: Hydrogen bond depiction
1. Experiment: The optimization of sequence
2. Time: 2024.10.04-2024.10.05
3. Member: Xudong Tang, Qiwen Jiang, Binxuan Zhang, Xuantong Liu
4. Method:
(1) Notably, although the candidate sequences, specifically B0, B34, and B51, along with G9, G16, G35, G42, G57, and G71, engaged the binding interface via fewer than four amino acids, a substantial portion of their sequences remained uninvolved in IL-2 binding. Consequently, these sequences were truncated to retain only the 20 amino acids implicated in binding interface contacts, with the exception of B34, which was truncated to 39 amino acids for further investigation (Table.1, 2).
(2) In PyMOL, based on the pre-existing file, we clicked "Display" → "Sequence", selected the corresponding amino acid residues, and executed "Remove Atoms" to generate the result shown in the figure below.
5. Attachment: The optimization of sequence
1. Experiment: Sequence design
2. Time: 2024.10.05-2024.10.13
3. Member: Fan Yang, Qiwen Jiang, Xinxin Zhang, Kaiqing Zhang, Meng Sun
4. Method:
(1) In this study, a homologous recombination strategy was employed for the experimental design. The selected peptide sequences, namelyB0, B34, B51 and G9, G16, G35, G42, G57, G71, along with their corresponding reverse sequences, were conjugated via a flexible linker. Reverse sequences were incorporated due to the flexibility of peptide linker and the inherent directionality of protein-receptor binding. Furthermore, glycosylphosphatidylinositol a (GPI) membrane-anchoring sequence was fused to the construct to retain the IL-2 mimetics on the cell surface rather than their secretion into the extracellular space, facilitating activation of NK cells. Additionally, Flag tag was coupled with the recombinant sequences for subsequent detection and isolation purposes.
(2) The specific sequences of the genetic elements are listed in the table below.
5. Attachment: Sequence design
1. Experiment: Lentivirus Packaging and Infection of CAR on NK-92 cell line
2. Time: 2024.10.26-2024.12.01
3. Member: Fan Yang, Qiwen Jiang, Xinxin Zhang, Kaiqing Zhang, Meng Sun
4. Principle: Lentiviral packaging is a technique to generate replication-incompetent recombinant lentiviral particles for efficient delivery of exogenous genes into target cells. Its core principle relies on the split of viral functional components into separate plasmids, which are co-expressed in packaging cells to assemble infectious but non-replicating viral particles.
5. Materials: Lenti-293 T cells, 5% CO2 incubator, 5% FBS, CPT Transfection Kit (Viraltherapy), Centrifuge 5810 R (Eppendorf, 5810R), Eclipse TS100 (Nikon, TS100), 96-well plates, 5% CO2 incubator, complete medium (RPMI-1640 with 10% FBS), polybrene.
6. Method:
(1) One day before transfection, Lenti-293 T cells was passaged and evenly plated on 100 mm culture dishes (at 8-10×106 cells per dish). Cells were cultured at 37 °C in a 5% CO2 incubator. The medium was replaced with 10.5 mL of pre-warmed, resistance-free medium (5% FBS) 2-4 hours before transfection.
(2) A mixture of twenty-three plasmids in equimolar amounts was prepared, and transfection was performed using the calcium phosphate transfection method with the CPT Transfection Kit (Viraltherapy). The transfection mixture composition was as follows:
7. Attachment: Lentivirus Packaging and Infection of CAR on NK 92 cell line
1. Experiment: Single-Cell Sorting and Planting
2. Time: 2024.12.21-2024.12.28
3. Member: Fan Yang, Qiwen Jiang, Xinxin Zhang, Kaiqing Zhang, Meng Sun
4. Principle: Fluorescence-activated cell sorting (FACS) uses target cells' specific fluorescent signals to isolate individual cells into 96-well plates, ensuring clonal growth from a single cell. Second, in vitro culture mimics physiological conditions to sustain cell viability and proliferation, while inverted microscopy tracks clonal expansion. For sequencing, gentle centrifugation preserves cell integrity during collection; cell lysis releases nucleic acids, enabling high-quality sequencing of clonally pure single-cell samples.
5. Materials: NK cell media (Gibco, supplemented with 5% FBS and 500 IU/mL IL-2), IL-2 (Gibco), 5% hAB serum (Gibco), Trypan Blue
6. Attachment: Single-Cell Sorting and Planting
1. Experiment: BLI measurement of cytokine-receptor binding affinity
2. Time: 2025.05.26-2025.05.31
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Principle: Bio-Layer Interferometry (BLI) is a label-free detection technology that converts optical interference signals generated on the surface of BLI biosensors into real-time response signals for detection. Its fundamental principle is wave interference. The biosensor consists of a glass optical fiber layer and an optical layer. At the interface between the optical fiber layer and the optical layer, the first reflected light is generated, while the second reflected light is produced at the interface between the optical layer and the external solution. The superposition of these two waves forms an interference spectrum curve. When the bait binds to the ligand immobilized on the sensor surface, the thickness of the optical layer increases, causing the path length of the second reflected light to become longer than before. This results in a change in the interference spectrum curve, shifting it to the right. When the bait and ligand dissociate, the bait detaches from the biosensor surface into the solution, causing the interference spectrum curve to shift to the left. A plot of the shift distance of the interference spectrum curve against the reaction time is called a sensorgram. Based on various binding models, the association constant (Ka or Kon), dissociation constant (Kd or Koff), and affinity (KD) can be fitted. Using the sensorgram, fitted curves, and affinity kinetic parameters, the affinity of designed IL-2R agonists for IL-2Rβ and IL-2Rγ can be evaluated and ranked.
Biosensors offer diverse methods for immobilizing bait molecules, including amine coupling, biotin/streptavidin, antibody/anti-Fc, His-tag/Ni-NTA, GST/anti-GST, or FLAG/anti-FLAG interactions, to form a biomolecular layer on the tip of the fiber optic biosensor. In this experiment, due to the widespread use, high affinity, and effectiveness of SA sensors, we selected SA sensors to immobilize the bait IL-2Rβ or IL-2Rγ for analyzing the affinity of IL-2R agonists. SA sensors utilize the interaction between streptavidin and biotin. Therefore, we purchased IL-2Rβ and IL-2Rγ with a His-tag (for purification) from Novoprotein, with the sequence number C33A. In subsequent operations, these were used as bait and pre-biotinylated
5. Materials: SA Sensors, Greiner 96-well Black Plate, PBST Buffer (0.02% Tween 20), 1.5 mL EP Tubes, 15 mL Centrifuge Tube, Empty Sensor Tray, Octet® RED96 BLI instrument, IL-2Rβ, IL-2Rγ, IL-2 mimics
6. Attachment: BLI measurement of cytokine-receptor binding affinity
1. Experiment: Molecular dynamics simulation
2. Time: 2025.05.10-2025.05.29
3. Member: Xudong Tang, Yang Jin, Kaiqing Zhang, Binxuan Zhang, Xuantong Liu
4. Materials:
(1) Software: GROMACS 2022.5, Alphafold3, HADDOCK, Visual Molecular Dynamics (VMD)
(2) Database: Protein Data Bank (PDB)
(3) Force Fields: CHARMM36m, TIP3P, CHARMM
(4) Simulation Environment: Periodic Cubic Water Box, NaCl
5. Attachment: Molecular dynamics simulation
1. Experiment: Cell proliferation assessment by cell counting method
2. Time: 2025.02.10-2025.02.29
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Cell culture flasks, automated cell counter, pipettes and pipette tips, cell culture medium, incubator, centrifuge, cell culture dishes, sterile tubes or containers, growth curve plotting software, IL-2
5. Attachment: Cell proliferation assessment by cell counting method under acidic condition
1. Experiment: LDH assay for measuring cytotoxic activity
2. Time: 2024.05.26-2024.05.31
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: K562 cell line, Microplate reader, CO2 incubator, LDH assay kit, 96-well plate, Pipette
5. Attachment: LDH assay for measuring cytotoxic activity
1. Experiment: Flow Cytometric Analysis of NK Cell Activity
2. Time: 2025.05.10-2025.05.29
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Flow cytometer, CD107a-APC, CD56-PE, CD69-FITC, Ice-cold PBS Staining buffer, Brefeldin A, 4% paraformaldehyde (PFA), 0.1% Triton X-100
5. Attachment: Flow Cytometric Analysis of NK Cell Activity
1. Experiment: Flow Cytometric Analysis of NK Cell PFN & GrB
2. Time: 2025.05.10-2025.05.29
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Flow cytometer, Anti-human Perforin antibody (BioLegend, 308104), Anti-Human Granzyme B Antibody (BioLegend, 515404),Ice-cold PBS Staining buffer, Brefeldin A, 4% paraformaldehyde (PFA), 0.1% Triton X-100
5. Attachment: Flow Cytometric Analysis of NK Cell PFN & GrB
1. Experiment: Software structure prediction and visual inspection
2. Time: 2024.06.07-2024.06.15
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Xuantong Liu, Kaiqing Zhang
4. Method:
(1) Alphafold3 structure prediction
① Open the website, find an interface for inputting the protein sequence. Select "protein" for all options, and then input data.
② Initially, input IL-2Rβ sequence and IL-2Rγ sequence. These two sequences are always consistent and no need to be changed in subsequent steps.
③ For the third entry, input our predicted protein structure: β + linker + γ. Through sequencing, we selected three complexes B34G35R-G, B51G35R-G, and B51G9-G, and chose GALA peptide which repeat four to six times as linker to connect protein fragments. Through arranging and combining them, we got 9 combinations.
④ Evaluate the predicted protein structures through scores and predicted 3D model.
(2) Visual inspection
① Notably, we use the candidate sequences, specifically B34, and 51, along with G9 and 35. These sequences were truncated to retain the 20 amino acids implicated in binding interface contacts, with the exception of β34, which was truncated to 39 amino acids for further investigation.
② In PyMOL chains should be renamed. Then, hydrogen bonds are found by clicking "find-polar-contacts-to other atoms in object". The stick structures of the protein molecules are displayed. After the amino acids connected by hydrogen bonds are clicked, we show the cartoon figure of the molecules and show the stick structure of the amino acids clicked. Then, the sequence numbers and names of the amino acids can be annotated by clicking on "label-residue".
③ Click " draw/ray-save image to file "
5. Attachment: Software structure prediction and visual inspection
1. Experiment: BLI-based affinity assessment of IL-2 Mimics under acidic conditions
2. Time: 2025.06.26-2025.07.10
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: SA Sensors, Greiner 96-well Black Plate, 0.5M MES buffer (pH=6.0, J62574.AK 2-Morpholinoethanesulfonic acid sodium salt: 71119-23-8), 1.5 mL EP Tubes, 15 mL Centrifuge Tube, Empty Sensor Tray, Octet RED96 BLI instrument, IL-2Rβγ, IL-2 mimics(GALA peptide conjugation)
5. Attachment: BLI-based affinity assessment of IL-2 Mimics under acidic conditions
1. Experiment: Detect the activating signal pathway by Western Blotting under acidic condition
2. Time: 2025.07.08-2025.07.14
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Lysis solution, PBS, protein loading buffer, electrophoresis buffer, transfer buffer, blocking buffer, centrifugal tube, electrophoresis tank, transfer membrane device, NC membrane, β-actin antibody (CST, 4970) , Phospho-Jak1 antibody (CST,74129), Jak1 antibody (CST, 29261), Phospho-Jak3 antibody (CST, 5031), JAK3 antibody (CST,80331), Phospho-ERK antibody (CST), ERK antibody (CST), Phospho-STAT1 (CST,9167), STAT1 antibody (CST,14994), Phospho-STAT3 (CST, 9145), STAT3 (CST, 9139), Phospho-AKT (CST, 9271), T-AKT (CST,4691)
5. Attachment: Detect the activating signal pathway by Western Blotting under acidic condition
1. Experiment: Cell proliferation assessment by cell counting method under acidic condition
2. Time: 2025.07.10-2025.07.29
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Cell culture flasks, automated cell counter, pipettes and pipette tips, cell culture medium, incubator, centrifuge, cell culture dishes, sterile tubes or containers, growth curve plotting software, IL-2 , lactic acid
5. Attachment: Cell proliferation assessment by cell counting method under acidic condition
1. Experiment: LDH assay for measuring cytotoxic activity under acidic condition
2. Time: 2024.08.26-2024.08.31
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: K562 cell line, Microplate reader, CO2 incubator, LDH assay kit, 96-well plate, Pipette
5. Attachment: LDH assay for measuring cytotoxic activity under acidic condition
1. Experiment: Flow Cytometric Analysis of NK Cell Activity under acidic condition
2. Time: 2025.09.10-2025.09.20
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Principle: This experiment uses flow cytometry for multi-parameter cell analysis to explore how acidic conditions affect NK cell activity. NK cells are identified via CD56-PEas CD3⁻CD56⁺ populations. CD107serves as an activation marker: surface expression indicates cytotoxicity (CD56⁺CD107a⁺ = active NK cells). Flow cytometry excludes doublets/dead cells, then quantifies the ratio of active NK cells to total NK cells to assess functional impacts.
5. Materials: Flow cytometer, CD107a-APC, CD56-PE, Ice-cold PBS Staining buffer, Brefeldin A, 4% paraformaldehyde (PFA), 0.1% Triton X-100, 15 mM lactic acid, K562 cell line
6. Attachment: Flow Cytometric Analysis of NK Cell Activity under acidic condition
1. Experiment: Flow cytometric analysis of NK cell PFN & GrB under acidic condition
2. Time: 2025.09.10-2025.09.15
3. Member: Xudong Tang, Yang Jin, Binxuan Zhang, Kaiqing Zhang, Xuantong Liu
4. Materials: Flow cytometer, Anti-human Perforin antibody (BioLegend 308104), anti-Human Granzyme B Antibody (BioLegend 515404), Ice-cold PBS Staining buffer, Brefeldin A, 4% paraformaldehyde (PFA), 0.1% Triton X-100
5. Attachment: Flow cytometric analysis of NK cell PFN & GrB under acidic condition
