ProjecTILs murine reference atlas of virus-specific CD8 T cells, version 2

We have developed ProjecTILs, a computational approach to project  new data sets into a reference map of T cells, enabling their direct  comparison in a stable, annotated system of coordinates. Because new  cells are embedded in the same space of the reference, ProjecTILs  enables the classification of query cells into annotated, discrete  states, but also over a continuous space of intermediate states. By  comparing multiple samples over the same map, and across alternative  embeddings, the method allows exploring the effect of cellular  perturbations (e.g. as the result of therapy or genetic engineering) and  identifying genetic programs significantly altered in the query  compared to a control set or to the reference map.

We  illustrate the projection of several data sets from recent publications  over two cross-study murine T cell reference atlases: the first  describing tumor-infiltrating T lymphocytes (TILs), the second  characterizing acute and chronic viral infection.

Single-cell  data to build the virus-specific CD8 T cell reference map were  downloaded from GEO under the following entries: GSE131535, GSE134139  and GSE119943, selecting only samples in wild type conditions. Data for  the Ptpn2-KO, Tox-KO and CD4-depletion projections were obtained from  entries GSE134139, GSE119943, and GSE137007 and were not included in the  construction of the reference map.

To  construct the LCMV reference map, we split the dataset into five batches  that displayed strong batch effects, and applied STACAS  (https://github.com/carmonalab/STACAS) to mitigate its confounding  effects. We computed 800 variable genes per batch, excluding cell  cycling genes, ribosomal and mitochondrial genes, and computed pairwise  anchors using 200 integration genes, and otherwise default STACAS  parameters. Anchors were filtered at the default threshold 0.8  percentile, and integration was performed with the IntegrateData Seurat3  function with the guide tree suggested by STACAS.

Next,  we performed unsupervised clustering of the integrated cell embeddings  using the Shared Nearest Neighbor (SNN) clustering method implemented in  Seurat 3 with parameters {resolution=0.4, reduction=”pca”, k.param=20}.  We then manually annotated individual clusters (merging clusters when  necessary) based on several criteria: i) average expression of key  marker genes in individual clusters; ii) gradients of gene expression  over the UMAP representation of the reference map; iii) gene-set  enrichment analysis to determine over- and under- expressed genes per  cluster using MAST. In order to have access to predictive methods for  UMAP, we recomputed PCA and UMAP embeddings independently of Seurat3  using respectively the prcomp function from basic R package “stats”, and  the “umap” R package (https://github.com/tkonopka/umap).