Introduction
Background
Autism Spectrum Disorder
Autism Spectrum Disorder (ASD) is one of the most prevalent neurodevelopmental disorders, affecting an estimated 1 in 36 children. ASD is highly genetically heterogeneous and may be caused by interaction of environment and genes. ASD is characterized by repetitive behaviors and impairments in social communication and interaction1.
However, autism is now seen as a spectrum that can range from very mild to severe. Moreover, many (but not all) individuals with ASD require lifelong support of some kind.
Etiology
To sum up, autism spectrum disorder is autism in a broad sense. Since people began to study in 1943, people still haven't found out the cause of autism. " What is the etiology of autism?" is also listed by Science as one of the 125 new most challenging scientific questions(https://www.science.org/content/resource/125-questions-exploration-and-discovery).
At first, people thought that due to external reasons such as parent-child interaction disorders, parental personality traits, family structure, etc., psychoanalysis was adopted in treatment, which often had no curative effect2.At present, the existing research is still inconclusive and undecided. Basically, there is a consensus that the extensive developmental disorders of autistic patients are mainly caused by factors of brain biology.
One of the causes of the disease may be the presence of harmful autoantibodies in our body. We believe that the effective treatment is to block autoantibodies. In order to determine the antibodies to be blocked, we searched the relevant literature and found some relevant autoantibodies.
Why do we use metabolites of the gut microbiota?
Over the past decade, a surge of research has emerged that investigates the bidirectional relationship between the brain and the human gut microbiome, known as the brain-gut-microbiome (BGM) system. The BGM system includes the central and enteric nervous systems (CNS; ENS) and various neural, metabolic, endocrine, and immune mediators.
On one hand, the gut microbiota plays a crucial role in modulating crosstalk between the intestinal and nervous systems, as well as promoting gastrointestinal (GI) homeostasis, and they also impact higher cognitive functions. Within the intestinal epithelial barrier, enteroendocrine cells can sense the luminal contents, such as nutrients and bacterial metabolites. These cells are directly connected to the gut environment and the brainstem through incoming vagal nerve fibers.
On the other hand, certain microbiota, along with their molecular by-products, neuroactive metabolites, and related inflammatory mediators, can traverse both the intestinal and blood-brain barriers, allowing transmission along the brain-gut-microbiome (BGM) system3. Some of these metabolites derived from gut microbiota have therapeutic potential, which is one of the reasons why we chose to study gut microbiota metabolites.
Overview
Method
Homology modeling
A. Why we use homology modeling?
Homology modeling is one of the computational structure prediction methods used to determine protein 3D structure from its amino acid sequence. It is considered to be the most accurate of the modern computational structure prediction methods, which consists of multiple steps that are straightforward and easy to apply4.
There are many tools and servers that are used for homology modeling. Model building approaches can be classified as rigid-body assembly methods, segment matching methods, spatial restraint methods, and artificial evolution methods.
B. First choice of Swiss model on Wemol
SWISS-MODEL workspace is a web-based integrated service dedicated to protein structure homology modelling, using rigid-body assembly methods5. It pioneered the field of automated modelling 25 years ago and been continuously further developed, which aims at extending the scope of automated homology modelling to address the modelling of protein assemblies by efficiently using the information on quaternary structures available in the PDB6.
In order to provide objective assessments of modelling performance, SWISS-MODEL participates in the CAMEO project (https://cameo3d.org). Taking some inspiration from CASP, CAMEO aims to provide a continuous, fully automated, assessment of predictions produced by various modelling servers using a common benchmark dataset of targets. Based on the CAMEO results in the ‘3D Structure Prediction’ category, SWISS-MODEL is consistently ranked among the top-modelling servers for several crucial modelling aspects.
To use Swiss-model, we choose Wemol as our Homology modeling platform because it simplifies and automates complex computing processes and supports low-code custom development and flexible extension. Even non-computing professionals are able to use. It’s easy to control and free to use, and embedded Swiss-model as its homology modeling algorithm.
C. Turning point: AlphaFold2
AlphaFold2 is an AI system developed by DeepMind that predicts a protein’s 3D structure from its amino acid sequence. It regularly achieves accuracy competitive with experiment.
In 2020, Alphafold2 won the CASP14 for achieving the most accurate prediction of protein-folding with less than the diameter of one carbon atom confidence error in average (< 1Å in average)7.
AlphaFold2 incorporated novel neural network architectures and training procedures based on the evolutionary, physical and geometric constraints of protein structures. It directly predicts the 3D coordinates of all heavy atoms for a given protein using the primary amino acid sequence and aligned sequences of homologues as inputs.
Molecular docking: Autodock vina
A. Why we do Molecular docking?
Since its first appearance in the mid-1970s, docking has proved to be an important tool to help understanding how chemical compounds interact with their molecular targets, and for drug discovery and development.
Till now, molecular docking is an established in silico structure-based method widely used in drug discovery. Docking enables the identification of novel compounds of therapeutic interest, predicting ligand-target interactions at a molecular level, or delineating structure-activity relationships (SAR), without knowing a priori the chemical structure of other target modulators8.
In particular, Reverse Docking (RD), which allows predicting the biological targets of a molecule of interest9, represents a valuable approach for computational target fishing and profiling10.
As the set-up of reverse docking screening workflows requires more efforts and longer preparation with respect to standard virtual screening, various tools and web platforms have also been recently developed to facilitate RD. Most of them entrust in already compiled libraries of disease-relevant targets and implement standard programs (e.g., DOCK11 , AutoDock12, and AutoDock Vina13) for performing reverse docking calculations.
B. Why we choose Autodock vina?
AutoDock Vina is an open-source program for doing molecular docking. It was originally designed and implemented by Dr. Oleg Trott in the Molecular Graphics Lab (now CCSB) at The Scripps Research Institute14.
Additionally and independently, AutoDock Vina has been tested against a virtual screening benchmark called the Directory of Useful Decoys by the Watowich group, and was found to be “a strong competitor against the other programs, and at the top of the pack in many cases”.
For its input and output, Vina uses the same PDBQT molecular structure file format used by AutoDock. PDBQT files can be generated (interactively or in batch mode) and viewed using MGLTools. Other files, such as the AutoDock and AutoGrid parameter files (GPF, DPF) and grid map files are not needed.
Results
I. Basis for project establishment
We try to find Ways to reduce and remove autoantibodies in the brain:
Retrieval method 1:
Understand the treatment for autoimmune encephalitis 15
Retrieval method 2:
Step1: Find the all types of autoantibodies
Step2: Find out whether each autoantibody exists in cerebrospinal fluid and how to eliminate it
The following autoantibodies are highly-related:
1.anti-GM1 16
| anti_GM1_125.fasta |
>:H
QVQLKESGPGLVAPSQSLSITCTVSGFSLTSYGVHWVRQPPGKGLEWLGTIWAGGSTNYNSALM
SRLSINRDTSKSQVFLKLNSLHTDDTAMYYCARDWRTGPYFDYWGQGTTLTVSS
>:L
DVQITQSPSYLAASPGETITINCRASKSISRYLAWYQEKPGKTNKLLVYSGSTLQSGVPSRFSG
SGSGTDFTLTISSLEPEDFAMYYCQQHNEYPYTFGGGTKLEIK
|
2.anti-MBP 17
| anti_MBP_F23C6.fasta |
>:H
QVQLKESGPGLVAPSQSLSITCTVSGFSLTGYGVHWVRQPPRKGLEWLGMIWGDGSTDYNSALK
SRLSISKDKSKRQVFLKMNSLQTDDTARYYCARDYDYGAMDYW
>:L
DIVLTQSPATLSVTPGDSVSLSCRASQSISNSLHWYQQKSHESPRLLIKYASQSISGIPSRFSG
SGSGTDFTLSINSVEAEDFGMYFCQQTNSWPHTFGGG
|
3.anti-GAD 18
| anti_GAD_MICA7.fasta |
>:H
QVQLVQSGAEVKKPGSSVKVSCKASGGTFNIYSFSWVRQAPGQGLEWMGGIIPIYRPANY
AQNFQGRVTITADESTSTVYMDLSSLRSDDTAMYYCARGTGGYYNSWGQGTLVTVSS
>:L
DIQMTQSPSTLSASVGDRVTITCRASQNINSWLAWYQQKPGKAPNLLISKASTLESGVPS
RFSGSGSGTEFTLTISSLQPDDFASYYCQQYKNYSWTFGQQTKVEMK
|
II. Design: Homologous modeling and Molecular docking
To identify how to block these autoantibodies, the first step we need to take is obtaining structural files for molecular docking. Since there are no ready-made structural files available for the autoantibodies for subsequent procedures, our plan is to first homology model several highly relevant autoantibodies and identify a few small molecule metabolites to block them.
1.Data Preparation: Set up our microbe metabolites list
As for the molecular docking, we search the Human Virtual Metabolism Database - Human and Intestinal Bacteria and Nutrition and Diseases to set up a microbe metabolites list.
The VMH database(https://www.vmh.life/#microbes/metabolites) integrates five types of resources: human metabolism, intestinal flora, disease, nutrition and ReconMaps, including 5,180 metabolites, 17,730 chemical reactions, 3,695 human genes, 255 genetic diseases, 818 microorganisms, 632,685 Microbial genes and 8790 foods.
Because the majority of these metabolites lack structural files, we collaborated with all team members to draw 2D structures together. It is really a bunch of work and everyone get allocated one page and a half. With the 2D structures, we create the 3D structure with Chem3D 20.0 to build our 3D micro-molecules database.
2.Structure Prediction
We executed our antibody structure prediction using the AlphaFold-Multimer model with 20 cycles on Google's Colab platform(https://colab.research.google.com/github/deepmind/alphafold/blob/ main/notebooks/AlphaFold.ipynb.)
We provided the Variable Light chain (VL) and Variable Heavy chain (VH) sequences of the antibody as input.
We provided the Variable Light chain (VL) and Variable Heavy chain (VH) sequences of the antibody as input. Subsequently, we utilized AlphaFold to generate the three-dimensional structures of the variable regions of anti-GM1，anti-MBP and anti-GAD antibodies, as depicted in Figure 1A, 1B and 1C.
Figure 1: Three-Dimensional Structures of the Variable Regions of Antibodies. A. Schematic representation of the three-dimensional structure of anti-GM1 antibody (Blue represents the three CDR regions of the heavy chain, pink represents those of the light chain, and the same color scheme applies below).
B. Schematic representation of the three-dimensional structure of anti-MBP antibody.
C. Schematic representation of the three-dimensional structure of anti-GAD antibody.
In our study, we employed similar batch docking procedures to assess the interactions between anti-GM1, anti-MBP and anti-GAD antibodies and a diverse set of 1190 small-molecule metabolites sourced from the VMH Metabolite Database.
After docking analysis, we selected two promising molecules: Ferrichrome and Protoporphyrin IX (Figure 2A and 2B). They exhibit highly advantageous structural characteristics, which result in numerous interactions with three different antibodies, indicating a higher degree of affinity. (Figure 3A to 3F)
Figure 2: The structures of Ferrichrome and Protoporphyrin IX A. Ferrichrome B. Protoporphyrin IX
Figure 3: 2D diagrams of interactions A. Ferrichrome and anti-GM1 autoantibody (A represents the heavy chain and B repesents the light chain)
B. Ferrichrome and anti-MBP autoantibody
C. Ferrichrome and anti-GAD autoantibody
D. Protoporphyrin IX and anti-GM1 autoantibody
E. Protoporphyrin IX and anti-MBP autoantibody
F. Protoporphyrin IX and anti-GAD autoantibody
The two molecules bind to the surface of the complementarity-determining regions (CDRs) of each antibody, interacting with the amino acids in the CDR regions, resulting in an inhibitory effect on antigen-antibody binding (Figure 4A to 4F and Figure 5A to 5D).
Figure 4: Interaction of Ferrichrome with Various Antibodies A, C, and E. Binding sites of Ferrichrome on the antibody surfaces. A, C, and E respectively refer to anti-GM1, anti-MBP, and anti-GAD autoantibodies.
B, D, and F. Conventional hydrogen bonds formed between Ferrichrome and amino acid residues of antibody protein. B, D, and F respectively refer to anti-GM1, anti-MBP, and anti-GAD autoantibodies. Orange: Fe (Iron), Red: O (Oxygen), Blue: N (Nitrogen), White: H (Hydrogen). Bond length unit: Å (Angstroms). A represents the heavy chain and B repesents the light chain.
Figure 5: Interaction of Protoporphyrin IX with Various Antibodies A, C, and E. Binding sites of Protoporphyrin IX on the antibody surfaces. A, C, and E respectively refer to anti-GM1, anti-MBP, and anti-GAD autoantibodies.
B, D, and F. Conventional hydrogen bonds formed between Protoporphyrin IX and amino acid residues of antibody protein. B, D, and F respectively refer to anti-GM1, anti-MBP, and anti-GAD autoantibodies. Red: O (Oxygen), Blue: N (Nitrogen). Bond length unit: Å (Angstroms). A represents the heavy chain and B repesents the light chain.
Overview
How we do it?
Our molecular docking process requires substantial computational resources. Therefore, we utilize our own server equipped with an Intel Xeon E5-2696 v4 @2.20GHz processor and 512GB of memory. Our server app is ‘Termius’, which also has SFTP functionality. If this is your new server, or if you have newly acquired server account, we recommend that you first install Anaconda. You can refer to the official tutorial or Google search for the installation method of Anaconda on the server. Our Conda version is 23.3.1. If you have already installed Anaconda, please skip this step. See 'Guide' for details.
Executive Summary
Conclusion & Discussion
In summary, we have successfully identified promising small molecule drug candidates from the VMH database targeting anti-GM1, anti-MBP, and anti-GAD through homology modeling and molecular docking. We have also examined their effectiveness and handed over the results to the wet lab team for subsequent synthetic biology design.
During this process, we encountered several challenges. Firstly, due to ethical constraints, the availability of cerebrospinal fluid from autistic individuals was limited, and we had to rely on older articles with less resolved antibody structures of interest. Consequently, we resorted to homology modeling for our subsequent work. Additionally, when attempting to validate the docking results through molecular dynamics simulations, we unfortunately discovered that the CHARMM force field does not include iron atoms, which led to the interruption of our simulations when using GROMACS.
It's important to note that many of our steps were simulated, and this may reduce consistency with real-world conditions, potentially affecting the success rate of subsequent wet lab experiments. However, this remains a rational and efficient approach to save significant time in the screening of ideal drug molecules.
Importance
To our knowledge, there is currently no team worldwide focused on developing small molecule drugs targeting antibodies. We have undertaken pioneering research in this field, providing a foundation for future studies and improvements.
Notebook
reference
1 Lord C, Elsabbagh M, Baird G, Veenstra-Vanderweele J. Autism spectrum disorder. Lancet. 2018 Aug 11;392(10146):508-520
2 Temelturk RD, Yurumez E, Cıkılı Uytun M, Oztop DB. Parent-child interaction, parental attachment styles and parental alexithymia levels of children with ASD. Res Dev Disabil. 2021 May;112:103922
3Kang DW, Adams JB, Gregory AC, Borody T, Chittick L, Fasano A, Khoruts A, Geis E, Maldonado J, McDonough-Means S, Pollard EL, Roux S, Sadowsky MJ, Lipson KS, Sullivan MB, Caporaso JG, Krajmalnik-Brown R. Microbiota Transfer Therapy alters gut ecosystem and improves gastrointestinal and autism symptoms: an open-label study. Microbiome. 2017 Jan 23;5(1):10
4 Muhammed MT, Aki-Yalcin E. Homology modeling in drug discovery: Overview, current applications, and future perspectives. Chem Biol Drug Des. 2019;93(1):12-20
5 Arnold K, Bordoli L, Kopp J, Schwede T. The SWISS-MODEL workspace: a web-based environment for protein structure homology modelling. Bioinformatics. 2006;22(2):195-201
6 Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R, Heer FT, de Beer TAP, Rempfer C, Bordoli L, Lepore R, Schwede T. SWISS-MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res. 2018 Jul 2;46(W1):W296-W303
7 Jumper J, Evans R, Pritzel A, et al. AlphaFold 2[J]. Fourteenth Critical Assessment of Techniques for Protein Structure Prediction; DeepMind: London, UK, 2020.
8 Pinzi L, Rastelli G. Molecular Docking: Shifting Paradigms in Drug Discovery. Int J Mol Sci. 2019 Sep 4;20(18):4331
9 Xu X, Huang M, Zou X. Docking-based inverse virtual screening: methods, applications, and challenges. Biophys Rep. 2018;4(1):1-16
10 Gloriam DE. Bigger is better in virtual drug screens. Nature. 2019 Feb;566(7743):193-194
11 Lang PT, Brozell SR, Mukherjee S, Pettersen EF, Meng EC, Thomas V, Rizzo RC, Case DA, James TL, Kuntz ID. DOCK 6: combining techniques to model RNA-small molecule complexes. RNA. 2009 Jun;15(6):1219-30
12 Morris GM, Huey R, Lindstrom W, Sanner MF, Belew RK, Goodsell DS, Olson AJ. AutoDock4 and AutoDockTools4: Automated docking with selective receptor flexibility. J Comput Chem. 2009 Dec;30(16):2785-91
13 Trott O, Olson AJ. AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. J Comput Chem. 2010 Jan 30;31(2):455-61
14 Eberhardt J, Santos-Martins D, Tillack AF, Forli S. AutoDock Vina 1.2.0: New Docking Methods, Expanded Force Field, and Python Bindings. J Chem Inf Model. 2021 Aug 23;61(8):3891-3898
15M Mané-Damas, C Hoffmann, S Zong, et al. Autoimmunity in psychotic disorders. Where we stand, challenges and opportunities[J]. Autoimmun Rev, Sep, 2019, 18(9): 102348
16 Weng NP, Ritter E, Yucel E, Zhang D, Ritter G, Marcus DM. Specificity and structure of murine monoclonal antibodies against GM1 ganglioside. J Neuroimmunol. 1994 Nov;55(1):61-8
17 Maier CC, LeBoeuf RD, Zhou SR, Whitaker JN, Jarpe MA, Blalock JE. The structure of a myelin basic protein-associated idiotope. J Neuroimmunol. 1993;46(1-2):235-243
18 Richter W, Jury KM, Loeffler D, Manfras BJ, Eiermann TH, Boehm BO. Immunoglobulin variable gene analysis of human autoantibodies reveals antigen-driven immune response to glutamate decarboxylase in type 1 diabetes mellitus. Eur J Immunol. 1995;25(6):1703-1712
