SI6 REPORT: Next Generation Lateral Flow Tests
AI structural modelling and peptides - A Proof-of-Concept step to prepare for novel lateral flow tools that detect myrtle rust or kauri dieback with specificity by design
May 2024
Sun X, Rikkerink E. 2024. AI structural modelling and peptides - A Proof-of-Concept step to prepare for novel lateral flow tools that detect myrtle rust or kauri dieback with specificity by design. A Plant & Food Research report prepared for: Biological Heritage, Ngā Rākau Taketake Theme 5 Control, Protect, Cure. Milestone No.102185. Contract No. 39454 Var 2. Job code: P/313069/13. PFR SPTS No. 25595. 10 p.
NOTE
This report is embargoed until 31/05/2025. For further information please contact Xiaolin Sun at Plant and Food Research: Xiaolin.Sun@plantandfoodresearch.co.nz
ABSTRACT
Background
Lateral flow devices to detect pathogens are becoming increasingly common. They are easy to deploy, stable, rapid, and relatively cheap and they have economies of scale as well. One of the biggest problems with lateral flow devices is that they are typically based on monoclonal antibodies that act as “binding agents” to the “target” organism, but also often bind other closely related organisms. As antibodies are not really designed, but created by random processes and then selected based on their ability to bind a target, finding an antibody that both binds well AND is specific is often a difficult, time-consuming and costly task; it involves screening of a large panel of potential candidates before an applicable one is found. The crux of our idea is to avoid the need for generating antibodies and move to trial “designed peptides” as the “binding agents” on lateral flows instead of antibodies. One remaining advantage of antibodies, however, is that they have multiple binding sites and when more than one site binds to the target, this creates a cooperative binding scenario that leads to the highest possible binding strengths. Single peptides may not be able to match these binding strengths. Our idea is to design a pair of peptides that bind to distinct parts of the target, and see if this can lead to similar cooperative binding advantages for peptides. Antibodies have been widely used in lateral flow tests to detect particular target analyte proteins or protein fragments. A stable and strongly binding antibody with high specificity is needed to obtain a high-quality test strip. However, this is the bottleneck of the production in terms of the excessive cost as well as time of making a suitable antibody. One of the properties of antibodies that makes them such a good binding reagent in immune systems is that they are made up of several variable sites. The natural process of antibody generation creates a large number of variations at these sites that are then combined to create a large pool of differential binding affinities. From this pool, antibodies that bind well to the foreign protein targets detected are ultimately selected and then amplified to protect the host. When more than one variable region binds the target, the strength and selectivity of the binding are both increased through co-operative binding effects at two or more of these variable sites.
Project Origin and Scope
We have been working on constructing peptides that bind to known targets. In one case we have used the recent boon in predicted protein structures that flow from structural prediction Artificial Intelligence (AI) programmes, such as AlphaFold2 (AF2). We propose to use these tools as a proof-of-principle to test the idea of whether we can significantly improve binding strength to the target by using two binding peptides together. We have a test bed machine that we use to assess peptide binding kinetics (the BiacoreX100) and also have existing antibodies that bind to the same target that we can compare with the two-peptide approach. We put forward here an innovative idea to replace binding antibodies with peptides produced from scratch based on AI-generated AF2 structural protein models of relevant protein target(s) such as surface proteins in the kauri dieback disease agent Phytophthora agathidicida. AF2 models have now been generated for many proteins in public databases and we expect more to be generated over time. Our AF2 modelling research with a particular pair of proteins that interact in planta as well as in the AI-predicted models has suggested that we may be able to design peptides from these models that will bind their protein binding target. In our model, a series of alpha helices from one protein bound our target plant protein RIN4. We sought to first demonstrate that peptides that produce two of these helices could be used to bind to this protein target, and then secondly to test if we could combine (or “multiplex”) the two peptides together. This two-peptide approach would be tested to determine if it could mimic improved co-operative binding, analogous to the way many useful antibodies show this effect. This proof-of-concept project used the information, tools, protein extracts and data we already had for the interacting protein pair, but was extended to include the idea of a two-peptide system and the use of AI models to assess their binding potential. This concept could potentially be extended to include single proteins for which AF2 models exist, using internal, interacting secondary structures such as alpha helices and beta sheets in the AF2 structure predictions.
Project Aims
If our approach proves successful, we will have a way to computationally create multiplexed-binding reagents for any protein with predicted structural data to replace the “trial and error prone”, wet laboratory approach of antibody generation. Such a “design-based approach” would potentially revolutionise the creation of detection systems for particular organisms or proteins such as lateral flow devices that currently rely on having a high-quality antibody binding reagent made from costly procedures including animal facilities. Our first proposed target for this novel technology would be Phytophthora agathidicida, as soon as sufficient high-quality genome and gene curation data are available to identify likely surface-exposed proteins and to generate the required AF2 models.
Results Summary
Our initial results demonstrated that the two peptides we generated could bind the target protein if that protein were bound to a chip used in our BiacoreX100 test system. Initial tests to repeat the assay with the peptides bound to such chips were not successful, and we now have a probable explanation for this result. We surmise the linkers we used to attach these peptides were not long enough and thus far enough away from the surface of the chip to allow the protein target to fold into its recognising protein shape and bind. This inability of the peptides to bind their target thus prevented us from assaying co-operative binding, which we were intending to demonstrate by increased binding affinity to mixtures of the two peptides bound to chips (when compared with the peptides by themselves).
This research was funded by the Ministry of Business, Innovation and Employment (Ngā Rākau Taketake – Myrtle Rust and Kauri Dieback Research, C09X1817).
There are no views created for this resource yet.
Additional Information
| Field | Value |
|---|---|
| Data last updated | unknown |
| Metadata last updated | 24 May 2024 |
| Created | unknown |
| Format | unknown |
| License | CC-BY-SA-4.0 |
| Created | 2 years ago |
| Datastore active | False |
| Has views | False |
| Id | 0ddbbd60-296a-4ff9-b480-b26de5969c83 |
| Package id | 73cdc842-c015-4f9d-99cd-5edf58c54834 |
| Position | 19 |
| State | active |