| 研究実績の概要 |
We have achieved the first important goal of developing our bioinformatics pipeline. Manuscripts describing the three core technologies used in this pipeline were published in 2019: Repertoire Buider (Schritt, D. et al. Mol Sys Des Eng), a method for high-throughput TCR or BCR 3D modeling; InterClone (Xu, Z. et al. Mol Sys Des Eng), a method for clustering TCRs or BCRs that target the same epitope (ImmuneScape; Li, S. et al. Meth Mol Biol), a method for constructing TCR-peptide-MHC 3D models. A fourth manuscript describing Adapt (Davila et al.), a method for estimating affinity lymphocyte-antigen complexes is in preparation.
We carried out two main experiments in our lab. First, we performed both bulk TCR and single-cell TCR+RNAseq sequencing of SKG mice, which are a model of Rheumatoid Arthritis, as described below. The second main experiment performed in our lab was to sequence 10,000 CD8+ T cells from a single (HLA-A*02) donor using 33 different barcoded peptide-MHC tetramers representing peptides from 10 viruses (CMV, EBV, HPV, Influenza, HCV, HBV). From this data, the epitope specificity of 153 clonotypes were determined. We have also made significant progress in analyzing clinical data from rhabdoid tumors, an aggressive pediatric sarcoma, provided by the Curie Institute in Paris. This dataset offered the opportunity to use all of the computational tools above to identify the TCRs of tumor infiltrating lymphocytes (TILs) and to predict their targeted epitopes.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
The software aspect of the project is going according to plan. The only serious bottleneck we encountered was that we are lacking sufficient experimental data to train our software for structure-based epitope prediction. To address this problem, we have taken two actions. The first is to improve the quality of our TCR-peptide-MHC binding affinity estimates. The result of this effort is the Adapt method, which was initially trained on BCR-antigen data, for which there is a large amount of experimental data. Our application of Adapt to ImmuneScape resulted in significant improvement in TCR-peptide-MHC modeling accuracy. The second action we have taken is to acquire more TCR-peptide-MHC experimental binding data. To this end, we utilized DNA barcoded peptide-MHC tetramers. With these tetramers, we first sort T cells, then sequence the T cells along with the barcoded peptide-MHC. One of the difficulties we encountered here was that 70% of the sequenced T cells consisted of a single CMV-positive clone. Therefore, we will follow up this experiment using pooled T cells from multiple donors in order to obtain a more diverse set of TCRs.
The experiments themselves have provided some surprises. For the SKG project, we unexpectedly found that the TCRs of CD4+ T cells in SKG mice are dominated by invariant chains (similar to those found in iNKT cells).
|
| 今後の研究の推進方策 |
We have four concrete plans for the current fiscal year. First, we aim to complete integration the Adapt-based scoring function with ImmuneScape in order to improve the accuracy of TCR-peptide-MHC modeling. This will proceed by the process of “cross-docking”: artificially pairing TCRs with both target and non-target peptide-MHCs and training the scoring function to recognize the “correct” pairs. Preliminary work along these lines suggest that dramatically different peptide-MHCs can be distinguished but that more subtle differences (i.e., where the MHC is the same but the peptide is only slightly different) are more difficult, as expected.
Second, we will incorporate new experimental data into the ImmuneScape-based TCR-epitope prediction. This data will include our own barcoded tetramer data and also publicly available data that has been deposited in VDJdb (currently 18,613 paired TCR sequences covering 177 unique epitopes). It is expected that well-studied epitopes, such as Yellowfever virus, will be rather easy to predict, but that less-studied epitopes will be more difficult to predict.
Third, we plan to carry out TCR-epitope predictions within rhabdoid tumors. We have already predicted candidate epitopes and now we are screening these by considering their interaction with TCRs.
Fourth, we plan to integrate InterClone-based clustering with ImmuneScape-based epitope prediction. We showed previously that such clustering was useful in reducing noise in our predictions, and we have been continuously improving InterClone using both paired and unpaired TCR sequencing data.
|
