Select sequencing of clonally expanded CD8⁺ T cells reveals limits to clonal expansion

Significance

T cells are a central component of our immune system. They express a T cell receptor (TCR) on their surface, which detects pathogens and stimulates the T cell to initiate an immune response. Upon exposure to a specific pathogen, some T cells are activated and clonally expand. These pathogen-specific clonally expanded T cells, however, are generally rare and difficult to isolate. We have developed a technology (SELECT-seq) to isolate this rare population and to analyze their transcriptome and TCR composition. Therefore, our work may be the key to achieving a better understanding of pathogen-specific T cell clonality and function.

Abstract

To permit the recognition of antigens, T cells generate a vast diversity of T cell receptor (TCR) sequences. Upon binding of the TCR to an antigen–MHC complex, T cells clonally expand to establish an immune response. To study antigen-specific T cell clonality, we have developed a method that allows selection of rare cells, based on RNA expression, before in-depth scRNA-seq (named SELECT-seq). We applied SELECT-seq to collect both TCR sequences and then transcriptomes from single cells of peripheral blood lymphocytes activated by a Mycobacterium tuberculosis (Mtb) lysate. TCR sequence analysis allowed us to preferentially select expanded conventional CD8⁺ T cells as well as invariant natural killer T (iNKT) cells and mucosal-associated invariant T (MAIT) cells. The iNKT and MAIT cells have a highly similar transcriptional pattern, indicating that they carry out similar immunological functions and differ considerably from conventional CD8⁺ T cells. While there is no relationship between expression profiles and clonal expansion in iNKT or MAIT cells, highly expanded conventional CD8⁺ T cells down-regulate the interleukin 2 (IL-2) receptor alpha (IL2RA, or CD25) protein and show signs of senescence. This suggests inherent limits to clonal expansion that act to diversify the T cell response repertoire.

The immune system of higher organisms is responsible for detecting and neutralizing a broad range of pathogens (1, 2). The collective immune response is complex and often requires the cooperation of many cell types. T cells play numerous essential roles in this system, from enabling antibody secretion by B lineage cells (3, 4), to detecting and eliminating pathogen-infected or cancerous cells in the organism. T cells can be highly specific, which they achieve by recombining and expressing heterodimeric T cell receptors (TCRs) on their cell surface, theoretically producing in excess of (∼10¹⁶) unique heterodimers (5). If a naïve αβ T cell’s TCR binds a specific antigen–MHC complex with sufficient affinity, the T cell will clonally expand while retaining its initial TCR sequence. This significantly increases the abundance of T cells that recognize a specific pathogen and consequently, enable an effective immune response. After clearance of the pathogen, most of these expanded T cells undergo apoptosis and the remainder mature to a memory stage. These cells can remain in circulation for decades in a quiescent but activatable state for a more immediate and robust response when recognizing the pathogen again (6, 7). Reactivation will lead to renewed T cell clonal expansion. During both phases of clonal expansion, however, the immune system must maintain a diverse T cell repertoire to fight any other invading insults. Recent data have demonstrated that repertoire richness in both CD4⁺ and CD8⁺ T cell compartments remains highly diverse even in aged people (8). To maintain such richness, T cell clonal expansion is well controlled by a balance of signals that impact the coordinately regulated processes of quiescence, proliferation, and cell death (9). However, how this balance is achieved to keep the T cell repertoire diversified is not well defined.

The main challenge in defining the mechanism of T cell repertoire diversity lies in the rarity of these cells. Memory T cells against a specific antigen are very rare (1–100 in 10⁵ T cells) (10) and only a limited number of cells (10∼1,000) may be sorted from one individual. Furthermore, to characterize clonal expansion and function of this rare cell population, both transcriptome and paired TCR sequences are needed from a single cell. For this purpose, gene expression profiling with high-throughput methods like 10× Chromium or DropSeq (11, 12) are not applicable because of their requirement for a large number of input cells (usually many thousands). An alternative is to use well-based sequencing strategies like Smart-Seq2, which is more attractive because it produces higher sequence resolution data relative to droplet-based methods (13). However, these are very costly. And, because for memory T cells there is no marker for those that are clonally expanded, much of the experimental costs will go to singletons that are often not of primary interest. This restriction on cell input numbers and the high costs are a limitation in single-cell immunology studies.

To circumvent these issues, we developed a direct approach that employs targeted PCR on Smart-Seq2–derived cDNA libraries, permitting a more precise selection of cells to analyze, avoiding the costly processing steps of tagmentation and in-depth sequencing in less relevant cells. We refer to this approach as SELECT-seq and applied it to the CDR3α and CDR3β TCR sequences, selecting duplicated TCR sequences, which indicates clonal expansion, to characterize the function of clonal expanded antigen-specific T cells in human.

We used SELECT-seq to characterize clonally expanded CD8⁺ memory T cells responding to a Mycobacterium tuberculosis (Mtb) lysate. Tuberculosis (TB) is one of the top 10 causes of death worldwide (14). While previous studies have shown an essential role of CD4⁺ T cells in protection against Mtb (15, 16), the role of Mtb-specific CD8⁺ T cells is unclear. Mtb-specific CD8⁺ T cells may be protective, since it is known that murine and human CD8⁺ T cells can recognize hundreds of Mtb epitopes (17–19). Besides conventional CD8⁺ T cells, unconventional, non–MHC-restricted T cells, including mucosal associated invariant T (MAIT) and invariant natural killer T (iNKT) cells, are known to be involved in Mtb and other bacterial responses as well (20–23). In contrast to conventional cells, which have highly diverse alpha and beta TCR sequences, MAIT and iNKT cells have limited TCR diversity, typically using a semiinvariant or invariant alpha chain, containing a signature amino acid motif (21).

Thus, to study Mtb-reactive CD8⁺ memory T cells, we stimulated peripheral blood mononuclear cells (PBMCs) from subjects with asymptomatic Mtb infection and sorted activated CD8⁺ T cells for SELECT-seq. In total, we analyzed over 3,000 Mtb-responsive peripheral CD8⁺ αβ T cells at the single-cell level. By analyzing TCR sequences, we found both conventional and unconventional (MAIT/iNKT) cells were activated by the Mtb lysate. We selected activated conventional and MAIT/iNKT cells for transcriptome sequencing. Analysis revealed a clear relationship between gene expression profiles and clone size in conventional CD8⁺ T cells. However, we did not find the same association within unconventional cells, nor did we observe differences in expression profiles between MAIT and iNKT cells. Highly expanded conventional clones decrease interleukin 2 (IL-2) receptor alpha (IL2RA, or CD25) expression and increase indicators of senescence, suggesting that this is a mechanism for limiting clonal expansion, thereby increasing TCR diversity.

Results

Development of SELECT-seq for Clonally Expanded T Cells.

The aim of SELECT-seq is to precisely select clonally expanded T cells for high-coverage sequencing within a single sequencing pipeline. This is achieved by screening single-cell cDNA libraries of large populations of cells with targeted PCR. We utilized a modified Smart-Seq2 protocol (24) to generate cDNA libraries (∼15 μL) from single T cells, from which a small aliquot (1 μL) was taken for nested PCR to amplify and sequence the CDR3 regions of both TCRα and TCRβ chains (25). We then applied tagmentation and transcriptome sequencing of selected cell libraries based on their TCR sequence to obtain high-coverage in-depth sequence data (Fig. 1A and SI Appendix, Fig. S1). Notably, the cost of cDNA library generation and TCR sequencing (∼$6) was much less than the combined cost of tagmentation and whole-transcriptome sequencing (∼$32), which at scale, and when the cell type of interest is rare, this provides substantial cost savings for high-resolution methods (SI Appendix, Fig. S1D).

Fig. 1.

Method for preferential selection of clonally expanded Mtb-reactive CD8⁺ T cells via TCR sequencing, for subsequent single-cell RNA sequencing. (A) Schematic view of the experimental workflow for determining TCR clonotypes and transcriptomes from single Mtb-reactive CD8⁺ T cells. (B) Table showing representative CDR3α and CDR3β sequences in different cell types derived from single T cells sorted based on CD137 and CD154 staining. Green text highlights “DSN” motif for MAIT cells, and red text highlights “DRG” motif for iNKT cells. (C) Circular phylogenetic trees representing the distribution of clone sizes for all activated TCR-sequenced cells, and the conventional and iNKT/MAIT cells selected for scRNA-seq. Each “leaf” (tip) represents one cell, and is grouped by identical TCR sequences (beta only for conventional, alpha and beta for iNKT/MAIT). The tree root is situated in the center of the circle, with branches for each clone, and colors distinguishing separate clones. The gray, rooted branches represent singlets. The inner ring color (blue and light blue) distinguishes neighboring clones but have no other significance. The outer ring color (rainbow) represents the size of the clone. (D) t-SNE analysis of all single cells (921 cells). Each point represents an individual cell. Colors indicate cell type and activation status; symbols indicate donor origin. (E) t-SNE analysis of all activated iNKT/MAIT cells (n = 169 and 276 for iNKT and MAIT, respectively). Each point represents an individual cell. Color indicates cell type. Symbol indicates donor origin. (F) Heatmap of the top 50 genes ranked by their contribution to PCA components 1–5 from all of our sequenced cells. The columns are individual cells ordered first by cell type (coloring as in D) then by donor, as shown in the color bars. The rows are genes clustered by expression similarity using hierarchical clustering. The z score is calculated by subtracting the gene’s expression in each cell by the mean expression across all cells. (G) t-SNE plot (from D) colored by the expression of the corresponding gene.

Applying this approach, we isolated PMBCs from donors who showed evidence of asymptomatic Mtb infection as gauged by a positive quantiferon-TB gold in-tube (QFT-GIT) test (26). PBMCs were activated with Mtb lysate ex vivo for 12 h, circumventing long-term culture after isolation from blood. This quick activation was sufficient to activate CD8⁺ T cells as measured by activation markers, permitting the enrichment of Mtb-reactive cells.

We then used fluorescence-activated cell sorting (FACS) to select both conventional and unconventional CD8⁺ T cells—MAIT and iNKT cells (SI Appendix, Fig. S2A). To define the surface markers to distinguish these cell types, we correlated CD8⁺ T cell surface marker expression to T cell types as defined by their TCR sequences. We found that in addition to CD137, an activated CD4⁺ T cell marker, CD154 (27–29) is up-regulated within CD8⁺ T cells after lysate stimulation (CD137⁺: 0.01∼0.3% and CD154⁺: 0.01∼0.7% of CD8⁺ T cells). While CD8⁺CD137⁺ T cells were enriched for conventional cells, CD8⁺CD154⁺ T cells were enriched for iNKT/MAIT cells (SI Appendix, Fig. S2 A and B). Further surface staining confirmed that only the CD8⁺CD154⁺ population is CD161⁺, a marker for iNKT/MAIT CD8⁺ T cells (SI Appendix, Fig. S2C) (30). In addition, both populations largely displayed memory T cell surface markers (CCR7⁻CD45RA⁺) and many contained identical TCR sequences, indicating clonal expansion (SI Appendix, Fig. S2C). Within the CD8⁺CD154⁺ population, invariant TCRα chains were ubiquitous, a property that defines both MAIT and iNKT CD8⁺ T cells (Dataset S1) (21). Thus, while it had been reported that activated iNKT/MAIT cells up-regulate CD154 (31, 32), CD154 appears to be a precise and sortable activation marker for iNKT/MAIT CD8⁺ T cells. We sorted, therefore, CD8⁺CD69⁺CD137⁺ and CD8⁺CD69⁺CD154⁺ populations for isolation of conventional and iNKT/MAIT cells, respectively. In addition, we sorted nonactivated cells for comparison (conventional: CD8⁺CD69⁻CD137⁻CD154⁻; iNKT/MAIT: CD8⁺CD161⁺CD69⁻CD137⁻CD154⁻).

We then assigned each cell a clone size based on the number of cells with an identical alpha and beta TCR sequence. For nonactivated conventional cells, we did not find any clonally expanded sequences, nor did any TCR sequence match those of the activated population, indicating that our clonally expanded and activated cells are indeed pathogen specific. We found some expanded nonactivated iNKT/MAIT TCR sequences, and one-third of nonactivated sequences overlapped with activated sequences (Dataset S1). This is in line with a reported low level of TCR sequence diversity in iNKT/MAIT cells (21, 30, 33).

Together, we sequenced 3,321 TCRs from five different donors from both conventional and iNKT/MAIT CD8⁺ T cells. Notably, only one in three activated T cells were clonally expanded. We detected 399 clonal (clone size >1) activated cells, with a wide range of clone sizes (Fig. 1C and see Materials and Methods). Many of these were doublets or very large clones (n = 51) so we deliberately picked out the most informative sequences. We selected over 1,000 cells for sequencing, 921 of which passed our quality-control thresholds and were used for downstream analysis (SI Appendix, Figs. S1C and S3 A–C and Dataset S2). Among them, the majority of the iNKT cells (152 of 169) are from donor 09/0769, while most of the MAIT cells are from four of five donors, and conventionally activated CD8⁺ T cells are from three of the five donors. In two donors (01/0468 and 09/0769), we found almost all iNKT/MAIT CD8⁺ T cells, with too few activated conventional CD8⁺ T cells to analyze.

Cell Type Classification and Gene Expression Signatures.

Each cell was assigned a cell type (conventional/iNKT/MAIT) based on its TCR sequence and activation marker, as well as its activation state (activated vs. nonactivated) (Fig. 1B). Applying t-distributed stochastic neighbor embedding (tSNE) to our transcriptome data, we detected distinct clusters that delineated conventional vs. unconventional and activated vs. nonactivated cell states (Fig. 1D).

Notably however, between CD8⁺ iNKT and MAIT cells, tSNE analysis found no observable differences in the overall gene expression profiles (Fig. 1E). In fact, we were only able to detect seven significantly differentially expressed genes between cells of these types (P < 0.05 and fold change >2, Dataset S3). Thus, we concluded that at least within CD8⁺ T cell lineages (a minority of iNKT cells), the iNKT/MAIT cells defined by TCR sequences, respond very similarly to Mtb lysate at a transcriptional level.

Additionally, we characterized the genes that were specific to each major cell type, which allowed for the first comprehensive comparison of genes specific to conventional and iNKT/MAIT cells in humans (Fig. 1 F and G and SI Appendix, Fig. S4). Both conventional and iNKT/MAIT subtypes showed up-regulation of cytotoxic genes GZMB, GZMH, and GNLY upon activation, consistent with their ability to kill Mtb-infected cells or Mtb directly (34, 35). While iNKT/MAIT cells showed more robust and ubiquitous IFN-gamma and TNF-alpha production, our data support only limited and sparse IFN-gamma gene expression from activated conventional cells. In terms of chemokine activity, conventional cells specifically expressed CCL3, CCL4, and CCL5, while iNKT/MAIT cells expressed mainly CCL20 (SI Appendix, Fig. S4). Elevated levels of CCL3, CCL4, and CCL5 in CD8⁺ T cells had been reported in viral infections (36) and also in liver-related infections (37). However, an elevated level of these cytokines in lung-related infections, where Mtb mainly resides, had not been characterized. Furthermore, we observe a set of NK cell receptors in conventional CD8⁺ T cells after activation, such as KLRK1, KIR3, ITM2A, KLRC, and HLA class II genes HLA-DRB1 and HLA-DRA (SI Appendix, Fig. S4). Furthermore, SLAMF1, SLC4A10, and DPP4 were expressed uniquely in iNKT/MAIT cells (Fig. 1 F and G). This was consistent with recent reports that MAIT cells preferentially express greater levels of SLAMF1 and DPP4 (38, 39). Interestingly, DPP4 is involved in the costimulatory signal essential for T cell receptor (TCR)-mediated T cell activation (Fig. 1 F and G).

In addition to well-defined clusters based on cell type and states, we find subclustering by donor (Fig. 1 D and F). This was driven by a set of genes differentially expressed between donors, including IFIT3, IFI44L, IFI6, and ISG15, which are part of the type I IFN signaling pathway. Although we could not associate a significance to these expression differences because we did not have the requisite patient phenotype information, we noted that it is consistent across multiple donors, including within the nonactivated cells. Therefore, these genes might reflect expected differences between environment and genetic backgrounds of donors, and our ability to detect them confirmed the sensitivity of our SELECT-seq approach.

Dynamic Gene Expression Profiles of Clonally Expanded T Cells.

To describe the dynamics of expression profiles based on clonal information, we performed principal component analysis (PCA) to determine if clustering was informed by clone size. Two observations stand out from this analysis: first, there is a striking dependence of conventional, but not iNKT/MAIT cell, gene expression profiles on clone size; second, the clone size distribution is the dominant layer of transcriptomic heterogeneity, i.e., it is associated with the first principal component (Fig. 2A). Thus, the transcriptional variation between large and small clones in conventional cells is greater than the difference between conventional and unconventional (iNKT/MAIT) cells. Indeed, we find that these two principal component dimensions contribute a much greater share of expression variance than the subsequent dimensions, demonstrating the importance of these features to their cell-state definitions (Fig. 2B). Together, this provides additional validation of our approach to specifically select clonally expanded cells because these cells are clearly unique relative to the singlet population.

Fig. 2.

Dynamic gene expression patterns among clonotypes and cell types. (A) PCA from all Mtb-reactive CD8⁺ T cells using highly variable genes. The size of each clonotype is represented by color; each dot represents one cell. Symbol represents cell type. (B) PCA, as in A, with the explained percent variances plotted for the top 10 PCA dimensions. (C) PCA of all Mtb-reactive CD8⁺ conventional cells using highly variable genes. The size of each clonotype is represented by color. (Lower) Cells are ranked according to their PC1 coordinate value and split into four equal-sized groups; a boxplot displays the distribution of clone sizes for these groups. Each boxplot represents the median (center line), the first quartiles and the third quartiles (box limits), and 1.5 times the interquartile range (whiskers). (D and E) As in C but for iNKT/MAIT cells, and includes PCA dimensions 1–4. Dotted-lined circles in D represent manually observed donor-dependent grouping (SI Appendix, Fig. S5B).

Although we do not observe an association between clone size and expression profiles in iNKT/MAIT cells, we explored these data further. We checked the first four principal components from PCA of activated iNKT/MAIT cells and found no association between clone size and expression profiles, but rather a primarily donor-specific contribution as mentioned above (Fig. 2 D and E and SI Appendix, Fig. S5B). In contrast, the first principal component of conventional cells revealed a clear correlation to clone size (Fig. 2C). We further analyzed this trend within each donor, and indeed found that conventional cells but not iNKT/MAIT cells cluster by clone size in aggregate, as well as within each donor (SI Appendix, Fig. S5 A and C).

To compare our results in additional contexts, we included similar analyses utilizing published datasets with single-cell RNA-seq (scRNA-seq) of clonally expanded CD8⁺ T cells. Surprisingly, none of these contexts yielded similar associations, or any discernible pattern, between clone size and gene expression profiles (SI Appendix, Fig. S6). These assays included T cells from blood, tumor, and adjacent tissue in cancer patients (40), as well as dextramer-isolated T cells from donors after receiving a yellow fever vaccine (41). Furthermore, a recently published study from patients hospitalized with severe influenza A virus also reported no difference in the transcriptomes of clonally expanded and nonclonally expanded T cells (42). A notable difference with our experimental design is our brief lysate activation. This may suggest that clonal-dependent gene expression profiles are only apparent immediately following activation, which in vivo studies may not capture. In Reinius et al. (41), the authors did sequence some T cells stimulated ex vivo; however, they allowed the cells to proliferate for 18 d, which increases the potential for artifacts. In addition, they included few clones, all of which were large (∼30–50), and thus the lack of an association may have been due to their limited clone size range (SI Appendix, Fig. S6).

Underlying Expression Patterns of Activated Conventional Cells.

Given the importance of clone size toward informing expression profiles in conventional cells only, we next focused on these cells by performing PCA with genes up-regulated upon T cell activation. We ranked all genes by their contribution to the first principal component, since PC1 reflects clone size (Fig. 3A). The highest ranked gene is IL2RA (CD25), which is induced by stimulation via the T cell receptor (43), and is critical in mediating T cell growth and proliferation (Fig. 3B and SI Appendix, Fig. S7 A and B). Two major cytotoxic cytokines LTA and IFNG (44) are also important contributors to PC1 and are generally expressed in small clones (Fig. 3 B and F).

Fig. 3.

Individual gene expression patterns associated with clone size in conventional cells. (A, Lower) Beeswarm plot of the PC1 coordinate values from PCA of activated conventional cells using the highly variable genes that are also up-regulated (P < 0.1 and fold change >0) in conventional cells (activated vs. nonactivated). The y axis has no significance. Color represents clone size. (A, Upper) Smoothed polynomial regression line of each cell’s clone size ordered by its PC1 coordinate. (B) Top 25 genes contributing to principal component 1, ranked by their percent contribution (x axis) calculated with “factoextra” (60). Negative values indicate a negative correlation between gene expression and PC1. Red means high expression in large clones, and purple reflects high expression in singlets and small clones. (C) As in A but for IL2RA expression. (D) Top 15 genes negatively correlated (Spearman’s rank) with PC1 coordinate values but positively correlated with clone size. (E) The highest scoring IPA network generated from genes that are highly correlated with clone size, as in B and D. (F and G) Boxplots showing individual gene expression among different clone sizes, and nonactivated conventional cells (CNTRL). Each boxplot represents the median (center line), the first quartiles, and the third quartiles (box limits), and 1.5 times the interquartile range (whiskers). Points are outliers.

Most of the genes that contribute the greatest level to PC1 variance are anticorrelated with clone size (Fig. 3B). Therefore, we next focused on genes specifically expressed in large clones (Fig. 3D). We found that these genes are part of the TCR signaling pathway (Fig. 3E), and are relevant to cytotoxic activity (PRF1 and GNLY), inflammatory chemokines (CCL4 and CCL5), in addition to, notably, senescence (TIGIT and KLRG1) (Fig. 3G).

We found similar indications in gene sets: chemokines are up-regulated in clonal cells, while cytokine and proliferation genes decrease expression with increasing clone size (SI Appendix, Fig. S7 E–G). Similarly, we looked at coreceptor genes, which colocalize with TCR molecules to inhibit or activate the TCR signaling pathway (45). We observed a decrease in expression of apoptotic coreceptor genes as clone size decreased, but a drop in costimulatory genes concordant with an increase of coinhibitory expression (SI Appendix, Fig. S7 H–J). Together, these data suggest that although large clones express genes important for cytotoxic and inflammatory activity, they also have lower expression levels of cytokines, costimulatory genes and proliferation related genes, suggesting that these large clones have a diminished cellular proliferation capacity and increased features of cellular senescence (46, 47).

Highly Expanded Clones Show Signatures of Senescence.

Next, we confirmed that the RNA-seq expression changes of key genes associated with a change in FACS profiling of their concomitant proteins. In addition to individuals with Mtb infection, we also assayed subjects with latent cytomegalovirus (CMV) infection using the same activation procedure, but with a CMV-pp65 derived peptide pool rather than a lysate. In two QFN+ (Quantiferon test positive) donors and one CMV+ donor, we observed a bimodal distribution of IL2RA (CD25) on antigen-specific CD8⁺ T cells (Fig. 4A). We compared these positive and negative IL2RA (CD25) populations by their expression of TIGIT and KLRG1 (Fig. 4B). We observed that the IL2RA (CD25) negative population is enriched in TIGIT and KLRG1 double positive cells while the IL2RA (CD25) positive population shows the opposite. Furthermore, we utilized index sorting to assay the expression of senescent markers, while retaining cell information to compare with TCR sequence data. Indeed, these data confirmed the scRNA-seq findings that IL2RA (CD25), and that costimulatory markers CD27 and CD28 (45), are expressed more frequently in singlets and small clones than in large clones. In contrast, the senescent markers, TIGIT and KLRG1, are expressed more frequently in large clones (Fig. 4C and SI Appendix, Fig. S8 A–E).

Fig. 4.

Functional characterization of TB and CMV-reactive CD8⁺ conventional T cell clones in relation to IL2RA (CD25). (A) Density plots showing the distribution of IL2RA (CD25) expression among Mtb-reactive (two donors) and CMV-reactive (one donor) CD8⁺CD137⁺ T cells after antigen-specific stimulation. Blue line indicates CD25⁻ population and the red line indicates CD25⁺. (B) Overlaid flow cytometry profiles of TIGIT and KLRG1 expression on CD25 negative (blue) and CD25 positive (red) cells, gated as in A. (C) Cells from A were single-cell index sorted and analyzed by single-cell TCR sequencing (25) six-parameter (Right of figure) phenotypic analysis of Mtb and CMV-specific CD8+ T cells. Individual T cells are grouped by TCR sequence; each color on the bar above the heat maps represents a distinct and clonal expanded TCR sequence. Red and blue represent positive and negative expression levels, respectively, of the protein acquired through index sorting. (D) Flow cytometry profiles showing combinations of CD25 expression with IFNγ-secreting CD8⁺CD137⁺ conventional T cells after stimulation with CMV-specific pp65 peptide pool. (E) Flow cytometry profiles showing CFSE dilution. CMV-specific CD25 positive and negative populations from donor CMV3405 were sorted, stained with CFSE, and stimulated for 7 d with pp65 peptide pool in different concentrations of IL-2 (Left and Middle) along with a positive control activated with phytohemagglutinin (PHA, Right).

To identify functional differences between large and small clones, we used an IFNγ capture assay, which showed decreased IFNγ secretion in the IL2RA (CD25) negative populations after pp65 peptide pool stimulation (Fig. 4D). Then, we assayed the proliferation potential of both IL2RA (CD25) positive and negative populations by a CFSE (carboxyfluorescein succinimidyl ester) assay (Fig. 4E and SI Appendix, Fig. S8F), by treating each population with the pp65 peptide pool and varying concentrations of IL-2. We found that the IL2RA (CD25) positive population proliferates robustly with both high and low IL-2 concentrations, while the negative population only proliferates well at high IL-2 concentrations. Together, these assays corroborate our hypothesis that large clones have a diminished capacity for cytokine secretion and proliferation, both features of cellular senescence.

Discussion

In this report, we develop a method, SELECT-seq, to provide high-quality scRNA-seq data of clonally expanded Mtb-responsive T cells. This approach enabled us to discover a strong but heterogenous activation response in CD8⁺ T cells upon M. tuberculosis lysate stimulation in infected individuals. In addition, we discovered an expression signature that limits further expansion within CD8⁺ clonally expanded conventional, but not in unconventional (iNKT/MAIT) T cells.

SELECT-seq is a straightforward approach to circumvent the problem of analyzing large numbers of cells, low sequencing resolution, high costs, and sorting by cell-surface proteins. It utilizes Smart-Seq2 to sequence rare cells and provide high-quality libraries. And, due to the targeted PCR step before expensive whole-transcriptome scRNA-seq, SELECT-seq significantly reduces the expense of untargeted single-cell sequencing. We applied this approach to select clonally expanded CD8⁺ T cells. This was achieved by identifying cells that had matching TCR sequences. Approximately two in five cells were clonally expanded, meaning the cost of tagmentation and sequencing all singlets would be $57,056, which can be avoided with SELECT-seq. Because it utilizes PCR to screen cDNA libraries, SELECT-seq may target any gene, or combination of genes, relevant to a specific cell population, like B cell receptors of B cells. Further, SELECT-seq is not limited to protein coding genes; it can also be applied to polyadenylated noncoding RNA or any expressed RNA if utilizing a total scRNA method (48, 49).

Applying SELECT-seq to major CD8⁺ subpopulations, we were able to distinguish MAIT and iNKT cells from conventional CD8⁺ T cells. Importantly, we found that gene expression profiles of CD8⁺ iNKT and MAIT cells are largely indistinguishable from each other in response to Mtb lysate. It is surprising that we see such a robust iNKT response to the Mtb lysate, since there are no known iNKT ligands from Mtb, versus MAIT cells which are known to recognize riboflavin derivatives from Mtb (23, 50). This indicates that even though iNKT and MAIT cells recognize different ligands, they carry out similar immunological functions, at least in our experimental context. In addition, it has been reported that both iNKT and MAIT cells could be activated by cytokines independent of TCR engagement (23, 50), thus an alternative explanation could be that iNKT and MAIT cells are activated by cytokines IL-12/IL-18/IL-15 in Mtb lysate stimulated PBMC culture medium. Thus, future studies comparing cognate antigen and cytokine activation of iNKT and MAIT cells with transcriptome analysis will be valuable. In this study, as we focused on the response of CD8⁺ T cells to Mtb, such responding iNKT and MAIT cells may only specify a particular subset of unconventional T cells, a similar comparison could be applied to iNKT/MAIT cells from additional lineages, such as CD4⁺ or CD4⁻CD8⁻ T cells, to determine if this pattern is consistent.

Another important finding was the significant gene expression changes along the continuum of clonally expanded conventional CD8⁺ T cells. Within the clonally expanded conventional cells we found an overall decrease in the expression of IL2RA (CD25), costimulatory receptors CD27 and CD28, and the number of genes expressed. Likewise, we found that the coinhibitory receptor TIGIT, senescent marker KLRG1, and cytotoxic and proinflammatory genes were up-regulated. Together, these findings suggested that these highly expanded T cells have become senescent. We confirmed our findings in multiple donors exposed to either a bacterial (Mtb) or viral infection (CMV) by showing that the most clonally expanded cells do not secrete IFNγ and are limited in their capacity for further proliferation.

The senescent phenotype observed in our study could be the outcome of low and consistent antigen dose stimulation, or homeostatic T cell proliferation considering the donors’ asymptomatic Mtb infection. In any case, an interesting consequence of this finding is that it would place an upper limit on the extent to which a given T cell could expand in response to antigens. This would allow “room” for other T cell clones to expand, diversifying the response considerably. This can be seen in the recent report by Stern and coworkers (51), who sequenced thousands of unique TCRs specific for one influenza peptide MHC and found an average of 516 unique TCRα clonotypes and 432 unique TCRβ clonotypes in each individual. In another report (52), the number of unique TCRs specific for NLV–HLA-A2 and GIL–HLA-A2 ranged from 21 to 784 for TCRα and from 13 to 1,030 for TCRβ per person. Similarly, in our studies (53) we found there are multiple TCR sequence solutions to particular viral or bacterial antigens in a given subject, even though they recognized the same ligand. An intrinsic limit to T cell clonal expansion could explain these TCR sequence results, generating a more diverse T cell response. This enforced diversity would limit the effectiveness of pathogen evolution to evade T cell recognition.

Given that we studied CD8⁺ T cells specific for Mtb antigens, the senescence signature we found is significant. While tuberculosis is highly prevalent and the immune systems response to Mtb infection is well studied (15), the role of CD8⁺ T cells in this disease is controversial. Their role is especially uncertain in human Mtb infection as there is no definitive evidence for a requirement of CD8⁺ T cells in mitigating this infection (54–56). Rather, CD4⁺ T cells have been implicated as the major functional responder to Mtb infection. However, we show that in asymptomatic Mtb-infected teenagers, not only is there extensive clonal expansion, but CD8⁺ T cells exhibit a senescent phenotype when reactivated, which suggests they are continually activated in vivo. Furthermore, these highly expanded clones show increased perforin and granulysin expression, which has been reported in terminally differentiated CD8⁺ T cells that are effective in killing Mtb (57). Unique to these expanded Mtb-specific CD8⁺ T cells, however, is that they express elevated levels of chemokines CCL4 and CCL5, ligands for CCR5, which attract white blood cells to sites of infection. Thus, while there are some unique characteristics of these senescent Mtb-specific CD8⁺ T cells, they are highly similar to CD8⁺ T cells in chronic diseases like HIV and CMV (58, 59), which are widely accepted as functional responders. Thus, the differences in CD8⁺ T cells responding to Mtb infection are more subtle than previously thought.

In conclusion, we developed SELECT-seq to permit economical, in-depth analysis of any specific cell type of interest. Without the limitation of cell-surface protein sorting, it provides significantly increased flexibility not associated with any current scRNA-seq method.

Materials and Methods

Detailed information on study cohort, antigen preparation, cell stimulation, flow cytometry, single-cell RNA-seq, TCR sequencing, and data analysis is provided in SI Appendix.