Scirpy: a Scanpy extension for analyzing single-cell T-cell receptor-sequencing data

1B and T lymphocytes are equipped with a vast repertoire of immune cell receptors that can recognize a wealth of different antigens. High-throughput sequencing technologies have enabled the study of these immune repertoires at unprecedented resolution and are advancing our understanding of adaptive immune responses in cancer, as well as in autoimmune and infectious diseases.

2Novel single-cell sequencing technologies now allow the joint profiling of transcriptomes and T-cell receptors (TCRs) in single cells. However, while the study of single-cell transcriptomes is facilitated by tools like Seurat and Scanpy, the bioinformatic analysis of paired α and β TCR chains is still in its infancy. Several methods to perform specific analytical tasks have been proposed (Supplementary Table S1), but the comprehensive characterization of TCR diversity from single cells is still hampered by the lack of ready-to-use computational pipelines.

3Here, we present Scirpy (single-cell immune repertoires in Python), a Python-based Scanpy extension that provides simplified access to various computational modules for the analysis and visualization of immune repertoires from single cells. Due to its tight integration with Scanpy, Scirpy allows the combination with scRNA-seq transcriptomic data to comprehensively characterize the phenotype and TCR of single T cells.

4Scirpy integrates different bioinformatic methods for importing, analyzing and visualizing single-cell TCR-sequencing data from human and mouse (Fig. 1). TCR data can be loaded from CellRanger (10× Genomics) csv or json files, TraCeR outputs generated from Smart-seq2 data or any delimited text file, including AIRR-compliant tsv files. The AnnData data structure provided by Scanpy is used to store TCR information together with matched transcriptomic profiles, when available.

5Scirpy uses a flexible TCR model supporting up to two α and β chains per cell, allowing the identification of dual-TCR T cells (Supplementary Note S1). It also flags cells with more than two chains, which potentially represent doublets (Supplementary Fig. S1) and may be discarded from downstream analyses. Scirpy defines clonotypes based on the nucleotide sequence of the TCR complementarity-determining region 3 (CDR3), but can further identify clonotype clusters based on CDR3 amino acid sequence identity or similarity. The latter approach, inspired by TCRdist, leverages the Parasail library to compute pairwise sequence alignments and identify clusters of T cells that likely recognize the same antigens. For building clonotype networks, Scirpy makes use of the sparse-matrix implementation from the scipy package, ensuring scalability to hundreds of thousands of cells (Supplementary Fig. S2).

6Scirpy offers a wide range of tools and visualization options that we demonstrate in Section 3. It allows inspecting TCR chain configurations (Supplementary Fig. S1), and exploring the abundance, diversity, expansion and overlap of clonotype repertoires across samples, patients or cell clusters derived from transcriptomics data (Supplementary Figs S3 and S4). Relationships between cells and clonotypes can be investigated with a graph-based approach (Supplementary Fig. S5), in addition to spectratype plots (which represent the distribution of CDR3 sequence lengths), and V(D)J-usage plots (Supplementary Fig. S6). Finally, TCR information can be integrated with transcriptomic data, for instance by overlaying Uniform Manifold Approximation and Projection (UMAP) plots (; Supplementary Fig. S4). Detailed tutorials on data loading and analysis with Scirpy are available at: https://icbi-lab.github.io/scirpy/tutorials.html.

7To demonstrate the applicability to a real-world scenario, we re-analyzed a recent single-cell dataset of ∼140k T cells. Single T cells were isolated from tumor, normal adjacent tissue, and peripheral blood of 14 patients with four different cancer types, and subjected to single-cell RNA and TCR sequencing with the 10× technology. Consistently with the original results, we found that the majority of clonotypes were singletons and only 9–18% of patients’ clonotypes were clonally expanded (Supplementary Fig. S3). Our results further confirm that effector, effector memory and tissue resident T cells comprised a large fraction of clonotypes that were expanded in both the tumor and normal tissue, while T cells consisted mostly of singletons (Supplementary Fig. S4). Moreover, leveraging Scirpy’s capability to group cells based on CDR3 sequence-similarity, we identified clonotype clusters indicating convergent TCR evolution (Supplementary Fig. S5). The analysis ran in 13 min on a single core of an Intel E5-2699A v4, 2.4 GHz CPU when defining clonotypes based on sequence identity, and in 42 min on 32 cores when using pairwise sequence alignment. A jupyter notebook to reproduce this case study is available at: https://icbi-lab.github.io/scirpy-paper/wu2020.html.

8Scirpy is a versatile tool to analyze single-cell TCR-sequencing data that enables seamless integration with the Scanpy toolkit, the de facto standard for analyzing single-cell data in Python. Scirpy is highly scalable to big scRNA-seq data and, thus, allows the joint characterization of phenotypes and immune cell receptors in hundreds of thousands of T cells. An extension of Scirpy to characterize γδ-TCR and B-cell receptor repertoires is planned for the next release.

9This work was supported by the Austrian Science Fund (FWF) [project number T 974-B30 to F.F. and I3978 to Z.T.] and by the European Research Council (ERC) [advanced grant agreement number 786295 to Z.T.]. Z.T. is a member of the German Research Foundation (DFG) [project number TRR 241(INF)].

10Conflict of Interest: none declared.