Targeted reconstruction of T cell receptor sequence from single cell RNA-seq links CDR3 length to T cell differentiation state

Abstract

The T cell compartment must contain diversity in both T cell receptor (TCR) repertoire and cell state to provide effective immunity against pathogens. However, it remains unclear how differences in the TCR contribute to heterogeneity in T cell state. Single cell RNA-sequencing (scRNA-seq) can allow simultaneous measurement of TCR sequence and global transcriptional profile from single cells. However, current methods for TCR inference from scRNA-seq are limited in their sensitivity and require long sequencing reads, thus increasing the cost and decreasing the number of cells that can be feasibly analyzed. Here we present TRAPeS, a publicly available tool that can efficiently extract TCR sequence information from short-read scRNA-seq libraries. We apply it to investigate heterogeneity in the CD8+ T cell response in humans and mice, and show that it is accurate and more sensitive than existing approaches. Coupling TRAPeS with transcriptome analysis of CD8+ T cells specific for a single epitope from Yellow Fever Virus (YFV), we show that the recently described 'naive-like' memory population have significantly longer CDR3 regions and greater divergence from germline sequence than do effector-memory phenotype cells. This suggests that TCR usage is associated with the differentiation state of the CD8+ T cell response to YFV.

Figure 1.

TRAPeS—an algorithm for TCR reconstruction in single cell RNA-seq Illustration of the TRAPeS algorithm. First, the V and J segment are identified by searching for paired reads with one read mapping to the V segment and its mate mapping to the J segment. Then, a set of putative CDR3-originating reads is identified as the set of unmapped reads whose mates map to the V, J and C segments. Finally, an iterative dynamic programming algorithm is used to reconstruct the CDR3 region.

Figure 2.

Validation of TRAPeS and comparison to other methods (A) Success rates for reconstruction of productive CDR3 in various CD8⁺ T cell data sets. Each line depicts the fraction of cells with a productive alpha or beta chain in a given data set with each one of the following methods—150 bp sequencing (black line), short paired-end data reconstructed using TRAPeS (red), TraCeR (turquoise), scTCRseq (gray), VDJPuzzle (dark blue) or Trinity (light blue). (B) Specificity of TRAPeS. Fraction of cells with identical CDR3 sequence between 150 bp data and the 25–30 bp data reconstructed either by TRAPeS, TraCeR, scTCRseq, VDJPuzzle or Trinity. This was calculated as the fraction out of cells with a productive chain in both 150 and 25–30 bp data. (C) Sensitivity of TRAPeS. Same as b, except the fraction of cells is calculated out of the total number of cells that had a successful reconstruction using 150 bp sequencing only. (D) Single cell RNA-sequencing captures a variety of clonal responses. Bars represent the Gini coefficient of each human CD8⁺ T cell data set. The Gini coefficient can range from zero (a complete heterogeneous population) to one (a complete homogenous population). Pie charts represent the distribution of clones in each population, n represents the number of cells with a successful reconstruction of both alpha and beta chains.

Figure 3.

Transcriptome analysis reveals distinct subpopulation of YFV-specific cells exhibiting a naive-like profile. (A) t-SNE projection of 353 CMV-specific, effector memory, YFV-specific, and Naïve cells, using normalized transcripts per million (TPM) values of 10,827 transcripts. Ellipses indicate three distinct spatial clusters. A discrete subset of YFV-specific cells cluster with Naive. (B) Genes differentially expressed between relevant phenotypic groups. YFV-specific cells were classified as effector memory-like or naive-like using SC3, a non-spatial consensus clustering approach (Figure S7). (C) t-SNE projections, each cell colored by relative signature score. Shown are two signatures from the ImmuneSigDB distinguishing CMV-specific from YFV-specific cells, and two signatures distinguishing Naïve or YFV-specific naive-like cells from effector memory, CMV-specific and YFV-specific effector memory-like populations. (D) FACS protein expression of CCR7 and CD45RA surface molecules from index sort of Effector Memory, YFV-specific effector memory-like, YFV-specific naive-like, and Naïve cells.

Figure 4.

YFV-specific subpopulations display different TCR structure. (A) YFV-specific naive-like cells tend to have longer CDR3. Distribution of the YFV-specific effector memory-like and naive-like CDR3 lengths in both alpha (left) and beta (right) chains. P-values were calculated with K–S test. (B) Differences between naive-like and effector memory-like CDR3 lengths are due to added nucleotides. Distribution of the YFV-specific effector memory-like and naive-like CDR3 germline scores, defined as the number of nucleotides in the CDR3 encoded by the V, D or J segments divided by the total number of nucleotides in the CDR3, for both alpha (left) and beta (right) chains. P-values were calculated with K–S test. (C) Signature analysis reveals significant correlation between CDR3 length and cell state. The plot depicts the rescaled normalized mutual information score between CDR3 length and transcriptional signatures of CD8⁺ T cells from ImmuneSigDB. Signatures identified as statistically significant using a permutation test (FDR-adjusted P-value < 0.1) are highlighted in red. (D) YFV-specific cells with long CDR3 tend to have a higher transcriptomic naive signature than cells with short CDR3. Plot represents the score of each cell for a transcriptional signature of a naive versus effector CD8⁺ T cell state. A high signature score means that a cell has higher expression of naive signature genes compared to effector signature genes.