Two distinct subpopulations of human stem-like memory T cells exhibit complementary roles in self-renewal and clonal longevity

Danai Koftori; Charandeep Kaur; Laura Mora Bitria; Yan Zhang; Linda Hadcocks; Ada W. C. Yan; Piotr F. Burzyński; Kristin Ladell; Daniel E. Speiser; Katrina M. Pollock; Derek Macallan; Becca Asquith

doi:10.1371/journal.pbio.3003179

Abstract

T stem cell-like memory cells (T_SCM cells) are considered to be essential for the maintenance of immune memory. The T_SCM population has been shown to have the key properties of a stem cell population: multipotency, self-renewal and clonal longevity. Here we show that no single population has all these stem cell properties, instead the properties are distributed. We show that the human T_SCM population consists of two distinct cell subpopulations which can be distinguished by the level of their CD95 expression (CD95int and CD95hi). Crucially, using long-term in vivo labelling of human volunteers, we establish that these are distinct populations rather than transient states of the same population. These two subpopulations have different functional profiles ex vivo, different transcriptional patterns, and different tissue distributions. They also have significantly different TREC content indicating different division histories and we find that the frequency of CD95hi T_SCM increases with age. Most importantly, CD95hi and CD95int T_SCM cells also have very different dynamics in vivo with CD95hi cells showing considerably higher proliferation but significantly reduced clonal longevity compared with CD95int T_SCM. While both T_SCM subpopulations exhibit considerable multipotency, no single population of T_SCM cells has both the properties of self-renewal and clonal longevity. Instead, the “stemness” of the T_SCM population is generated by the complementary dynamic properties of the two subpopulations: CD95int T_SCM which have the property of clonal longevity and CD95hi T_SCM which have the properties of expansion and self-renewal. We suggest that together, these two populations function as a stem cell population.

Citation: Koftori D, Kaur C, Mora Bitria L, Zhang Y, Hadcocks L, Yan AWC, et al. (2025) Two distinct subpopulations of human stem-like memory T cells exhibit complementary roles in self-renewal and clonal longevity. PLoS Biol 23(6): e3003179. https://doi.org/10.1371/journal.pbio.3003179

Academic Editor: Avinash Bhandoola, National Cancer Institute, UNITED STATES OF AMERICA

Received: August 2, 2024; Accepted: April 23, 2025; Published: June 20, 2025

Copyright: © 2025 Koftori et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: All relevant data are within the paper and its Supporting Information files.

Funding: BA is funded by the Wellcome Trust (103865Z/14/Z, 220794/Z/20/Z), the European Union Seventh Framework Programme (FP7/2007–2013) under grant agreement 317040 (QuanTI), the European Union H2020 programme under grant agreement 764698 (QUANTII) and Leukemia and Lymphoma Research (15012). AWCY was supported by an Imperial College Research Fellowship. KMP was supported by a St Mary’s Development Trust fellowship. The EAVI programme was funded by the European Union’s Horizon 2020 research and innovation programme grant 681137. DCM was partially funded by the Jefferiss Trust. CK and LMB received salaries from the European Union H2020 programme under grant agreement 764698 (QUANTII). Infrastructure support was provided by the NIHR Imperial Biomedical Research Centre and the NIHR Imperial Clinical Research Facility. The funders did not play any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Abbreviations: CT, cycle threshold; CTV, cell trace violet; MFI, median fluorescence intensity; IQR, interquartile range; T_CM, T central memory; T_EM, T effector memory; T_N, T naïve; TEMRA, T effector memory expressing RA; TREC, T cell receptor excision circle; T_SCM, T stem-cell memory

Introduction

T cell memory is central to protective immunity yet how memory is maintained is poorly understood. In 2011, in a seminal paper, Gattinoni and colleagues identified, in humans, a population of antigen-experienced T cells with stem cell-like properties that included a high capacity for self-renewal in vitro [1]. It was observed that these T stem cell-like memory cells (T_SCM cells) were long-lived and multipotent, i.e., they could differentiate into all other T cell memory subsets including central memory (T_CM), effector memory (T_EM) and effector memory expressing RA (T_EMRA) [1,2]. It was suggested that T_SCM cells are a dedicated population of cells whose role is to maintain T cell memory; i.e., they serve as a long-lived reservoir that can maintain themselves for years and be called upon to differentiate into effector populations in the event of re-exposure to cognate antigen. Since then, there has been considerable interest in T_SCM cells as preventive and therapeutic tools and in the role of T_SCM in the longevity of the vaccine-induced T cell response [3,4].

T_SCM cells are not the only candidate for the stem cell-like precursor population. Others have argued that T_CM rather than T_SCM are the main population responsible for maintaining memory [5]. This position is particularly supported in murine models. Although T_SCM cells were first described in mouse models of graft v host disease [6], can be generated with murine cells in vitro and are detectable in mice in vivo [7], it has proved difficult to identify their antigen-specificity or to generate them following in vivo challenge. Furthermore, cell fate–tracking experiments in mice (using genetic barcoding and single-cell transfer), showed that T_CM cells have the capacity to self-renew and that a single murine T_CM cell could reconstitute immune protection against an otherwise lethal pathogen [8,9].

Outside of murine models, T_SCM cells are less controversial and evidence for the stem-cell like behavior of T_SCM cells has continued to accumulate. T_SCM cells were identified in non-human primates (rhesus macaques and pigtail macaques) and importantly these T_SCM populations could be readily generated following acute infection [10,11]. In humans, following hematopoietic stem cell transfer, the number of persisting donor cells was strongly correlated with the number of T_SCM cells infused. Furthermore, stable isotope labelling studies of human volunteers established that, even in healthy, lymphocyte-replete individuals, T_SCM cells were long-lived and had the dynamic properties of self-renewal and clonal longevity consistent with the hypothesis that T_SCM cells are stem-like cells responsible for maintaining immune memory in humans [12,13].

The stable isotope labelling studies also made an unexpected finding: instead of being a single homogenous population as previously assumed, there was strong evidence that both CD4⁺ and CD8⁺ T_SCM populations were in fact heterogeneous and consisted of subpopulations with different dynamics [12]. Stable isotope labelling is a destructive assay in that once cells are sorted, DNA is isolated, and the identity of the cell is lost. Therefore, it was not possible to go back and obtain the phenotype of the subpopulations with differential kinetics. Consequently, although there was strong indirect evidence for heterogeneity the subpopulations could not be identified.

Here, we identify these T_SCM subpopulations in CD8⁺ T cells. We first use Ki67 expression to identify cell subpopulations with different proliferation rates and establish that CD95 expression distinguishes between the subpopulations. We show that T_SCM cells with high levels of CD95 expression (henceforth CD95hi) have significantly higher levels of Ki67 expression than T_SCM cells with intermediate levels of CD95 expression (henceforth CD95int). We then investigate the functional properties of the CD95hi and CD95int T_SCM subpopulations. We show that CD95hi T_SCM cells have significantly higher self-renewal and effector functionality compared to CD95int T_SCM cells. Transcriptionally, we show that CD95hi T_SCM cells are closer to central memory cells while CD95int cells are closer to naïve T (T_N) cells. We also show that CD95int and CD95hi T_SCM cells have different physical locations with CD95int T_SCM enriched in lymph nodes while CD95hi T_SCM are enriched in the blood. Returning to their dynamics, we show that CD95hi T_SCM cells have a significantly lower TREC content than CD95int T_SCM cells and that their frequency increases significantly with donor age. Finally, we go back in vivo and use stable isotope labelling, this time sorting on CD95 expression, and show that CD95hi and CD95int T_SCM populations have different dynamics in vivo. Importantly, the stable isotope traces for the two populations remain distinct for the duration of the labelling experiment (140 days), ruling out the possibility that there is rapid interconversion between the two subpopulations and establishing them as bona fide distinct subsets. We fitted different models to investigate the lineage relationship between the cell subpopulations. We find that, of the models we consider, a linear model in which upon meeting cognate antigen, T_N undergo clonal expansion and differentiate into CD95int T_SCM which in turn differentiate into CD95hi T_SCM, is most consistent with the data. Furthermore, we find that the CD95int subpopulation, but not the CD95hi subpopulation, harbors long-lived clones, with half lives of more than 10 yrs. This finding is independently verified by the quantification of tetramer-positive cells post-yellow fever vaccination.

While we identify many differences between the CD95int and CD95hi T_SCM subpopulations, it is the difference in their dynamics that is critical. CD95int T_SCM have the longest life span of any memory subpopulation identified to date [14]. This long life span will be central to their ability to maintain immune memory.

Results

CD95 expression identifies T_SCM subpopulations with differential ex vivo proliferation

Our aim was to identify subpopulations of CD8⁺ T_SCM cells (CD8⁺CD3⁺CD45RA⁺CCR7⁺CD95⁺) with different kinetics. We therefore stained human peripheral blood mononuclear cells (PBMC) with a panel of markers (“Materials and methods”) and, by backgating on the Ki67 positive cells established that the median fluorescence intensity (MFI) of CD95 expression was significantly higher in Ki67⁺ T_SCM cells than in Ki67⁻ T_SCM cells (P = 3 × 10⁻⁷, paired Wilcoxon, N = 25; Fig 1A). We therefore classified T_SCM cells as having either intermediate or high CD95 expression (henceforth CD95int and CD95hi, respectively) with the separation between the populations based on backgating of Ki67 with gates set to provide maximum discrimination of Ki67-positive and Ki67-negative cells (“Materials and methods”). Using this definition, we showed that CD95hi T_SCM cells had significantly higher levels of Ki67 expression than both CD95int T_SCM cells and T_CM (CD8⁺CD3⁺CD45RA⁻CCR7⁺) cells (Fig 1B, P = 0.003, P = 0.002, respectively; paired Wilcoxon test two tailed). To validate this finding, we investigated an independent dataset and again the same pattern was observed with CD95hi T_SCM cells containing a significantly higher proportion of Ki67-expressing cells than both CD95int T_SCM and T_CM (Fig A in S1 Text).

Download:

Fig 1. Ki67⁺ T_SCM cells exhibit increased CD95 expression.

T cells from healthy volunteers were isolated and stained with Ki67 and canonical T cell differentiation markers. (A) Median Fluorescence Intensity (MFI) of CD95 for Ki67⁻ and Ki67⁺ T_SCM cells. Each dot represents one donor (N = 25, Cohort 1); P = 3 × 10⁻⁷ paired Wilcoxon (B) T_SCM cells are classified as CD95hi and CD95int and the percentage of Ki67⁺ cells is compared among T_N, CD95int, CD95hi and T_CM subsets. Each dot represents one donor (N = 15, from Cohort 1). T_N v CD95int P = 0.0005, CD95int v CD95hi P = 0.003, CD95hi v T_CM P = 0.002, T_N v T_CM P = 0.002, paired Wilcoxon, number of independent tests = 3. **P < 0.01, ***P < 0.001, ****P < 0.0001. Bars represents the median, the box the interquartile range (IQR) and whiskers show lower quartile − 1.5 * IQR and upper quartile + 1.5 * IQR; all data points (including outliers) are shown by filled circles. Abbreviations: T_N: T naïve, T_SCM: T stem-cell memory, T_CM: T central memory, MFI: median fluorescence intensity. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g001

CD95hi T_SCM cells exhibit increased self-renewal and functionality compared to CD95int T_SCM cells

Self-renewal, i.e., the ability to maintain phenotype upon division, is a crucial property of stem cells. We investigated the self-renewal properties of CD95int and CD95hi T_SCM cells together with T_CM cells ex vivo. CD8⁺ CD95int, CD95hi and T_CM cells were isolated by FACS from PBMC from healthy donors (see Fig B in S1 Text for representative gating). Cells were cultured for 7 days with homeostatic cytokines (IL-7/IL-15, IL-15 or IL-2) [15–17] and their self-renewal, defined as the percentage of divided cells that had retained their input phenotype, was quantified where divided cells were identified by the dilution of cell trace violet. The self-renewal of CD95hi T_SCM was consistently higher than that of CD95int T_SCM or T_CM in response to all 3 conditions (CD95hi v CD95int: IL-7/IL-15 P = 0.0001, N = 13; IL-15 P = 0.0002, N = 10; IL-2 P = 0.1, N = 5. CD95hi v T_CM: IL-7/IL-15 P = 0.0008, N = 13; IL-15 P = 0.1, N = 10; IL-2 P = 0.5, N = 5; unpaired Wilcoxon), Fig 2A. We conclude that CD95hi T_SCM have higher self-renewal in the context of homeostatic cytokines than CD95int T_SCM and T_CM. The differentiation profile of the sorted cells at day 7 is given in Fig C in S1 Text and the CTV profile in Fig D in S1 Text.

Download:

Fig 2. CD95hi T_SCM cells exhibit increased self-renewal and functionality compared to both CD95int T_SCM cells and T_CM cells.

(A) Self-renewal in response to homeostatic cytokines was investigated by culturing sorted CD95hi, CD95int and T_CM for 7 days with 25 ng/ml IL-7/IL-15, 25 ng/mL IL-15 or 6.1 ng of IL-2 (N = 13, 10, 5, respectively). The self-renewal (defined as percentage of divided cells that have maintained their input phenotype) of each subset on day 7 is shown. The self-renewal of CD95hi T_SCM was consistently higher than that of CD95int T_SCM or T_CM in response to all 3 conditions (CD95hi v CD95int: IL-7/IL-15 P = 0.0001, N = 13; IL-15 P = 0.0002, N = 10; IL-2 P = 0.1, N = 5. CD95hi v T_CM: IL-7/IL-15 P = 0.0008, N = 13; IL-15 P = 0.1, N = 10; IL-2 P = 0.5, N = 5. Unpaired Wilcoxon, number of independent tests per dataset = 2). (B) Self-renewal in response to TCR stimulation was investigated by culturing sorted T cell subpopulations for 7 days with 25 ng/ml IL-7/IL-15, 2 μg/ml aCD3 and 1 μg/ml aCD28. (C) Multipotency in response to TCR stimulation was investigated by culturing sorted CD95hi, CD95int and T_CM for 7 days with 25 ng/ml IL-7/IL-15, 2 μg/ml αCD3 and 1 μg/ml αCD28. The phenotype of the starting (sorted) population is shown in the grey bars above each panel and the subpopulations they differentiated into are plotted. (D) The data from panel C is used to calculate a multipotency index, this captures the diversity in subpopulations that a subpopulation generates and is defined for a subpopulation j as where p_i is the proportion of differentiated cells (i.e., those that have changed phenotype) that are represented by subpopulation i. N = 9 but due to low frequencies of some cell populations N = 4 for CD95hi T_SCM and N = 7 for T_CM. (E) FACS-isolated T cell subsets were stimulated for 12 h with 2 μg/ml αCD3 and 1 μg/ml αCD28. Boxplots on left summarize the percentage of TNFα, IFNγ and IL-2 expressing cells among the different subsets. Boxplot on right shows markers associated with cytotoxicity. N = 5. An extension of panel E to show T_N and T_EM is shown in Fig E in S1 Text. In all cases bars represents the median, the box the interquartile range (IQR) and whiskers show lower quartile − 1.5 * IQR and upper quartile + 1.5 * IQR; all data points (including outliers) are shown by filled circles. All subjects from Cohort 1. A replicate of this figure but with individuals identified by different symbols is provided in Fig S in S1 Text. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g002

In response to αCD3αCD28 stimulation, a strikingly different pattern emerged. In this scenario, both T_SCM subpopulations rapidly differentiated whereas a large proportion of divided T_CM cells retained their phenotype (Fig 2B). Investigating this further, we found that both T_SCM subpopulations were able to differentiate into a number of other phenotypes: T_CM, T_EM and T_EMRA were all generated by both CD95hi and CD95int T_SCM cells in appreciable quantities, Fig 2C. In contrast T_CM were only able to produce T_EM in measurable quantities, Fig 2C. Quantifying this as a “multipotency index” (“Materials and methods”), we found that both CD95hi T_SCM and CD95int T_SCM had a significantly higher multipotency index than T_CM (P = 0.01, N = 4 and P = 0.0004, N = 7, respectively, unpaired Wilcoxon rank sum, Fig 2D).

Together, these results suggests that CD95hi T_SCM are both better able to maintain themselves during homeostasis but also better able to generate a multicellular immune response during challenge when compared with T_CM cells.

Next, we investigated the effector profile of the T_SCM subpopulations following stimulation. CD95int T_SCM and CD95hi T_SCM populations, together with comparator T_N, T_CM, T_EM and T_EMRA cells were sorted at high purity, stimulated with αCD3αCD28 for 12 h and assayed for production of cytokines and molecules associated with cytotoxicity by flow cytometry. We found that a greater proportion of CD95hi T_SCM cells tended to produce IFNγ, IL-2 and TNFα compared to both CD95int T_SCM and T_CM (Fig 2E, Fig E in S1 Text). CD95hi T_SCM also degranulated more readily (measured by CD107a) compared with T_CM; the frequency of granzyme B and perforin-positive cells were very low in all three populations but again followed the same trend (Fig 2E, Fig E in S1 Text).

Finally, given that two recent studies have reported heterogeneity in CD122 (IL-2Rβ) expression and PD1/TIGIT coexpression in the T_SCM population we studied expression of these proteins in the CD95int T_SCM and CD95hi T_SCM subpopulations [18,19]. We found that both CD122⁺ cells and PD1⁺TIGIT⁺ cells were enriched in the CD95hi T_SCM population (Fig F in S1 Text). We further looked at the relationship between Ki67 expression and PD1/TIGIT coexpression. We found no evidence that exclusion of the “exhausted” PD1⁺TIGIT⁺ T_SCM and exhausted PD1⁺TIGIT⁺ T_CM cells made Ki67 expression in the remaining T_SCM and T_CM populations more similar (Fig G in S1 Text). So, at least when considering dynamics, our data are not consistent with the suggestion [19] that PD1⁻TIGIT⁻ T_CM and PD1⁻TIGIT⁻ T_SCM cells are a single homogeneous population.

T_SCM subsets have different gene expression profiles

Next, we investigated gene expression by performing bulk RNAseq analysis of CD8⁺ T_N, CD95int T_SCM, CD95hi T_SCM, T_CM, T_EM and T_EMRA populations sorted from PBMC of healthy donors. Cells were either unstimulated (cultured with RPMI) or stimulated with αCD3αCD28. RNA was extracted and sequenced (“Materials and methods”). PCA representation of the most variable genes confirmed firstly that unstimulated cells of all phenotypes tended to cluster together and away from all stimulated cells regardless of phenotype. Further, for unstimulated cells, CD95int T_SCM clustered near T_N, whereas CD95hi T_SCM clustered with T_CM. Upon stimulation these four populations separated with both T_SCM subpopulations distinct from each other and from T_N and T_CM cells; Fig 3A. Both unstimulated T_EM and T_EMRA cells and stimulated T_EM and T_EMRA cells clustered closely with each other as has previously been noted [17,18]. For the unstimulated CD95int and CD95hi T_SCM subpopulations we calculated which genes (after batch correction) were differentially expressed and then identified which pathways these genes were enriched in; Fig 3B. The results were very clear and self-consistent: pathways that were upregulated in naïve CD8 and CD4 T cells compared to memory T cells were upregulated in CD95int T_SCM compared to CD95hi T_SCM; Fig 3B. The results from repeating the same analysis for the stimulated CD95int and CD95hi T_SCM subpopulations were harder to interpret; Fig 3C. Only one of the pathways with enriched genes was a T cell pathway (GSE21360, PRIMARY versus QUATERNARY MEMORY CD8 TCELL UP from [20]). This pathway suggests that genes upregulated in CD95hi T_SCM are enriched in genes upregulated in memory CD8 T cells following multiple antigen challenge (“QUATERNARY MEMORY CD8 TCELL”) whereas genes upregulated in CD95int T_SCM are associated with memory CD8⁺ T cells following primary antigen challenge (“PRIMARY MEMORY CD8 TCELL”). Heatmaps showing the 70 most differentially expressed genes for both the unstimulated CD95int and CD95hi T_SCM and the stimulated CD95int and CD95hi T_SCM are shown in Panels (A) and (B) of Fig H in S1 Text, respectively.

Download:

Fig 3. Analysis of bulk RNAseq data.

CD8⁺ T_N, CD95int T_SCM, CD95hi T_SCM, T_CM, T_EM and T_EMRA populations were sorted. Cells were either unstimulated or stimulated with αCD3αCD28. RNA was extracted and sequenced. (A) PCA representation of the most variable genes showed that, for unstimulated cells, CD95int T_SCM clustered near T_N, whereas CD95hi T_SCM clustered with T_CM. Upon stimulation these four populations separated with both T_SCM subpopulations distinct from each other and from T_N and T_CM cells. (B) Gene set enrichment analysis of genes differentially expressed between unstimulated CD95int T_SCM and unstimulated CD95hi T_SCM found, transcriptionally CD95int T_SCM were closer to naïve CD4 and CD8⁺ T cells while CD95hi T_SCM were correspondingly closer to various memory populations (i.e., genes that were down in naïve v memory CD4, naïve v effector memory CD4 and naïve v PD1 hi CD8 were also down in CD95int T_SCM). (C) The same analysis for of genes differentially expressed between stimulated CD95int T_SCM and stimulated CD95hi T_SCM was harder to interpret. The only T cell pathway (GSE21360, PRIMARY VS QUATERNARY MEMORY CD8 TCELL UP) implies that genes upregulated in CD95hi T_SCM are enriched in genes upregulated in memory CD8 T cells following multiple antigen challenge whereas genes upregulated in CD95int T_SCM are associated with memory CD8⁺ T cells following primary antigen challenge.

https://doi.org/10.1371/journal.pbio.3003179.g003

We validated these conclusions by qPCR. We focused a priori on 9 genes known to be differentially expressed between T_N and T_SCM cells and between T_SCM cells and T_CM cells [1] (“Materials and methods”). We found that sets of genes associated with T_N cells were more highly expressed by CD95int T_SCM cells than CD95hi T_SCM cells (P = 0.012, Binomial two tailed) while sets of genes associated with T_CM were more highly expressed by CD95hi T_SCM cells compared to CD95int T_SCM cells (P = 0.015, Binomial two tailed), Panels (A) and (B) of Fig I in S1 Text, respectively. Given these trends in the data, we investigated canonical differentiation markers to see if there was a trend from T_N to CD95int, to CD95hi, to T_CM. We quantified CCR7 and CD45RA (protein) expression by flow cytometry and, consistent with the gene profiles, found that for both CCR7 and CD45RA, expression was highest in T_N, significant lower in CD95int than T_N (P = 1 × 10⁻⁶, P = 1 × 10⁻⁶), significantly lower in CD95hi than CD95int (P = 1 × 10⁻⁶, P = 0.0006) and significantly lower in T_CM than in CD95hi (P = 0.0006, P = 1 × 10⁻⁶); i.e., there was a strong and consistent monotonic decrease in CCR7 and CD45RA expression from T_N through CD95int, CD95hi to T_CM (Panels (C) and (D) of Fig I in S1 Text)); N = 21 two-tailed paired Wilcoxon, number of independent tests ≤6.

CD95int and CD95hi T_SCM subsets have different lymph node homing properties

Different T cell subpopulations have different tissue homing properties related to their function [21]. We investigated whether the distribution of CD95int and CD95hi T_SCM cells varied between blood and lymph node after immune challenge by intramuscular injection with a model antigen. Lymph node cells were collected by ultrasound guided fine needle aspiration of axillary lymph nodes from 7 healthy volunteers participating in an experimental synthetic HIV envelope immunogen study (reported elsewhere [22]); matched blood samples from the same individuals at the same time point were also collected. Cells were collected between 7 days and 56 days post-first immunization but before the subsequent boost, a time point at which antigen-specific T cells but not antigen-specific IgG was detectable [22]. We found that the percentage of T_SCM cells that were CD95int was significantly higher in the lymph node than in the blood (median 81.75% in LN versus 71.8% in PBMC, P = 0.029, unpaired Wilcoxon) and conversely that the percentage of T_SCM that were CD95hi tended to be elevated in blood (median 11.6% in LN versus 23.7% in PBMC, P = 0.079, unpaired Wilcoxon), Fig 4.

Download:

Fig 4. CD95hi and CD95int frequencies differ between the blood and the lymph nodes.

PBMCs and lymph node cells isolated from the blood and axillary lymph nodes, respectively, of 7 healthy vaccinated volunteers from Cohort 3 were isolated and stained. Cells were gated on the population of (A) CD95int cells and (B) CD95hi cells. PBMC shown in red and lymph node cells in blue. Each dot represents one sample. CD95int T_SCM (as a proportion of total T_SCM) are significantly enriched in the LN, P = 0.029 unpaired two-tailed Wilcoxon test. Conversely CD95hi T_SCM cells show a non-significant trend to be enriched in the blood P = 0.079. *P < 0.05. Bars represents the median, the box the interquartile range (IQR) and whiskers show lower quartile − 1.5 * IQR and upper quartile + 1.5 * IQR; all data points (including outliers) are shown by circles. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g004

T_SCM subsets have different division histories and different long-term stability

T cell receptor excision circles (TRECs) are small circles of DNA created by DNA excision during V-(D)-J gene rearrangement in the thymus. TRECs are only produced upon T cell receptor rearrangement and are not replicated when a cell divides so they are diluted upon mitosis. TREC content is therefore an indicator of the number of division events that a T cell population has undergone since thymic production. TREC content is typically very low in memory T cells and difficult to quantify by qPCR or even digital droplet PCR (ddPCR) from extracted DNA. We therefore developed a highly sensitive ddPCR assay directly from cell lysate [23]. Using this assay, we quantified the TREC content in high purity FACS sorted populations of CD8⁺ T_N, CD95int T_SCM, CD95hi T_SCM, T_CM, T_EM and T_EMRA (Fig 5A). As expected, TREC content was higher in T_N cells compared to memory cells. Furthermore, the TREC content of CD95int T_SCM cells was significantly higher than that of CD95hi T_SCM cells indicating that these are distinct populations with different division histories (if there was frequent interconversion between CD95hi T_SCM and CD95int T_SCM then these populations would have similar TREC contents) and furthermore that CD95hi T_SCM cells have undergone more rounds of division since thymic production than CD95int T_SCM cells.

Download:

Fig 5. CD95hi cells have a lower TREC content and their frequency increases with age.

(A) TREC (T-cell receptor excision circle) content was quantified by ddPCR for FACS-sorted T cell subpopulations from N = 7 healthy donors, from Cohort 4. Boxplots indicate the number of TREC copies per cell for each subset (bars represents the median, the box the interquartile range (IQR) and whiskers show lower quartile − 1.5 * IQR and upper quartile + 1.5 * IQR; all data points, including outliers, are shown by circles). Graph on right is a zoom in (truncated y axis) to show TREC content of memory populations more clearly. Statistics: Wilcoxon paired test (B) PBMCs isolated from the blood of 21 healthy individuals (23−65 yrs, Cohort 4) were stained to identify CD95hi and CD95int T_SCM cells. The percentage of T_SCM that were CD95hi (orange circles) and CD95int (blue circles) was quantified. The results are plotted against the age of the donor. We found that the proportion of T_SCM that were CD95hi increased significantly with age (Rs = 0.82, P = 6.78 × 10⁻⁶ Spearman). The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g005

Next, we examined the change in CD95int T_SCM and CD95hi T_SCM frequency with donor age. We quantified the frequency of the two subpopulations in a cross-sectional cohort of 21 individuals whose ages ranged from 23 to 65 yrs. We found that the frequency of the T_SCM population that was CD95hi increased significantly with age (Rs = 0.82, P = 7 × 10⁻⁶, Spearman correlation) while the CD95int T_SCM population showed a corresponding decrease (Rs = −0.84, P = 2 × 10⁻⁶, Spearman correlation), Fig 5B.

CD95int and CD95hi T_SCM subsets have different dynamics in humans in vivo

Finally, we investigated the in vivo life span and lineage relationship of CD95int T_SCM cells and CD95hi T_SCM cells using stable isotope labelling. Three individuals received 7 weeks of deuterated “heavy” water (2H₂O, “Materials and methods”); serial saliva and blood samples were taken during and up to 140 days after the start of labelling, Fig 6A. Monocytes were extracted as a rapidly turning over reference population (“Materials and methods”, Figs J–K in S1 Text, Table D in S2 Text). The remaining PBMCs were FACS sorted into CD8⁺ T_N, CD8⁺ CD95int T_SCM and CD8⁺ CD95hi T_SCM subpopulations. Deuterium enrichment in the saliva as well as in the DNA of monocytes and sorted T cell subpopulations was measured by gas chromatography/mass spectrometry (GC/MS) [24], Fig 6B. Strikingly, the stable isotope enrichment for the two T_SCM subpopulations remained distinct for the duration of the labelling experiment (140 days), indicating that these are distinct subpopulations on this timescale.

Download:

Fig 6. The CD95hi subset incorporated more label in vivo compared to CD95int.

(A) Workflow of heavy water labelling technique and timepoints of sample collection. Participants from Cohort 5 drank 70% deuterated water for 7 weeks (100 ml twice daily for one day, then 50 ml twice daily thereafter). Saliva and PBMC were collected at multiple time-points during and after labelling. Cells of interested were sorted and deuterium enrichment was measured by gas chromatography/mass spectrometry (B) M + 1 enrichment of DNA from selected cell populations is shown per individual for all timepoints for the three different subsets. Error bars represent analytic variance and show the standard deviation of 4 replicate measurements. Note, DW19 attained higher levels of label in body water hence the higher enrichment in DNA (the mathematical modelling allows for this variation when calculating cell kinetics). Here lines connect neighboring timepoints for ease of viewing, they do not represent model fits. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g006

We constructed 5 different mathematical models encapsulating five different assumptions about the lineage relationship between the populations (Fig 7, “Materials and methods”) and sought to identify which model and which parameter combinations were most likely to have generated the observed data. In order to further constrain the models and to improve parameter identifiability we fitted two further data sets we had previously obtained [4,12]: one measured the telomere length in T_N and total T_SCM cells from 5 individuals and one quantified the frequency of yellow fever virus (YFV)-specific total T_SCM in a cross sectional cohort of 37 individuals at different time points (ranging from 0.27 to 35.02 yrs) post-YF vaccination. The three datasets were fitted simultaneously. Plots of best fits are given in Fig 8 and Figs L–O in S1 Text with the corresponding parameters in Table 1 in main text and Tables F–I in S2 Text. We found that three of the five models gave good fits: the model with a forked structure in which a fraction of T_N differentiated into CD95hi T_SCM and a fraction into CD95int T_SCM (Fig 7C) and the two linear models (Fig 7D and 7E) (Table K in S2 Text). The independent homogenous and heterogeneous models (Fig 7A and 7B) were not viable as they failed to capture the decline in the YFV-specific population (Panel (C) of Fig L and Panel (C) of Fig M in S1 Text). We conclude that, of the models we considered, the fork model and the two linear models were consistent with the data and could be considered viable options at this stage; the two independent models could be ruled out.

Download:

Table 1. Best fit estimates of parameters obtained by fitting the winning model (Model D, linear model with CD95int T_SCM first) to the labelling, telomere and YFV data. Interquartile range (estimated by bootstrapping the data) shown in square brackets below the corresponding estimates. Note, the clonal half-life corresponds to the half-life of the clone (ln2/d-p) not the half-life of the cell as is typically reported (ln2/d). Best fit estimates for the other 4 models are provided in Tables G–J in S2 Text. We found that CD95int TSCM cells proliferated much more slowly than CD95hi TSCM cells with an average proliferation rate of 0.0008 per day (equivalent to one division every 3.4 yrs) whereas CD95hi TSCM cells had an average proliferation rate of 0.002 per day (equivalent to one division every 1.4 yrs). The data in this table can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.t001

Download:

Fig 7. Models capturing different kinetics and relationships between the cell populations of interest.

(A) Independent homogeneous model. Each population (T_N, CD95int, CD95hi) is independent and homogeneous. (B) Independent heterogeneous model. Each population (T_N, CD95int, CD95hi) is independent and allowed (but not constrained) to be heterogeneous. (C) Forked differentiation model. Upon meeting cognate antigen, T_N cells undergo a clonal burst (of size 2^k); a fraction (f) differentiate into the CD95int pool with the remainder differentiating into the CD95hi pool. (D) Linear differentiation model (CD95int first). Upon meeting cognate antigen, T_N cells undergo a clonal burst (of size 2^k) and differentiate into the CD95int pool which subsequently differentiate into the CD95hi pool. (E) Linear differentiation model (CD95hi first). Upon meeting cognate antigen, T_N cells undergo a clonal burst (of size 2^k) and differentiate into the CD95hi pool which subsequently differentiate into the CD95int pool. Note for compactness, in the cartoons, CD95int T_SCM are designated Tint and CD95hi T_SCM are designated Thi. In every model each population was assumed to be of constant size. The models were fitted to the data to identify which model was most consistent with the experimental observations and to quantify the population parameters.

https://doi.org/10.1371/journal.pbio.3003179.g007

Download:

Fig 8. Fit of the three viable models – Model C (fork), Model D (linear, CD95int first) and Model E (linear, CD95hi first) – to all data (labelling, telomere, YFV) simultaneously.

(A) Fit of models to the labelling data (each individual in a different row, each cell population in a different column); black dots represent data, colored lines represent fitted model predictions (solid: Model C, dashed: Model D, dotted: Model E). (B) Fit of models to the telomere data. Observed differences in mean telomere length between T_N and T_SCM for 5 individuals shown on the left in black, predictions for each individual and each model shown by colored points on the right (DW19 in blue, DW20 in pink, DW25 in orange. squares: Model C, diamonds: Model E, triangles: model D). (C) Fit of the model to the YFV data. Observed data (YFV-specific T_SCM as a % of CD8⁺ T cells) shown by black points, fitted model predictions for each individual shown by colored lines (DW19 in blue, DW20 in pink, DW25 in orange) and for each model shown by line type (solid: Model C, dashed: Model D, dotted: Model E). The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g008

We calculated the clonal life span of CD95int and CD95hi T_SCM cells implied by the three viable models (Table 1 in main text, Tables H and I in S2 Text). All three models indicated that the CD95int T_SCM population was very long-lived at the clonal level (half-life of a CD95int T_SCM clone of the order of 10 yrs) whereas the CD95hi population was shorter-lived (half-life of a CD95hi T_SCM clone of the order of months), Fig 9. We conclude that the CD95int T_SCM subset is most likely to be the population responsible for retaining long-term immune memory.

Download:

Fig 9. Predictions of winning model – Model D (linear, CD95int first) overlaid (NOT fitted) onto a novel data set.

(A) The winning model and best fit parameters were used to predict the long-term dynamics of YFV-specific CD95int T_SCM (left) and CD95hi T_SCM (right) for each individual (shown by colored lines: DW19 in blue, DW20 in pink, DW25 in orange). For CD95int the predictions overlay and only one line can be seen. The predictions are plotted against but not fitted to a proxy for CD95int and CD95hi T_SCM frequencies measured in the cross-sectional cohort post-vaccination (the same cohort for which total T_SCM were fitted). (B) As above but also including 100 predictions obtained by fitting bootstrapped data (sampled with replacement) shown as grey lines. A comparison of all models to this unseen data set is provided in Table L in S2 Text. The data underlying this figure can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003179.g009

Model predictions agree well with an unseen data set

To validate the models and parameters found in the previous section we assessed their ability to predict the long-term dynamics of YFV-specific CD95int T_SCM and CD95hi T_SCM obtained from Fuertes Marraco and colleagues [2]. The flow panels used to analyze the YFV-specific T_SCM dataset in Fuertes Marraco and colleagues did not systematically include CD95 so we were not able to explicitly quantify the frequency of YFV-specific CD95int and CD95hi T_SCM cells over time. However, given the strong association between CD45RA, CCR7 and CD95 we were able to use CD45RA and CCR7 expression to distinguish the two T_SCM subpopulations (Panel (C) and (D) of Fig I in S1 Text, Fig P in S1 Text). We thus use CCR7⁺⁺CD45RA⁺⁺ cells (“naïve-hi” cells in the nomenclature of [2]) as a surrogate for CD95int T_SCM cells and the remaining CD45RA⁺CCR7⁺ cells (“naïve-int” cells in the nomenclature of [2]) as a surrogate for CD95hi T_SCM cells. This new data set was not fitted. We simply predicted the time course of YFV-specific CD95int and CD95hi T_SCM using the three viable models (the fork model, the linear model with T_N differentiating first into CD95int T_SCM and then into CD95hi T_SCM and the linear model with T_N differentiating first into CD95hi T_SCM and then into CD95int T_SCM) and the best fit parameters previously identified and compared them to the new data; it should be noted that the previously fitted data included the time course of the total YFV-specific T_SCM population and so the fitted and the new datasets are not completely independent. The agreement between the prediction and the new data was extremely good for the linear model with CD95int first (Fig 10) but unrealistic for the fork model and the linear model with CD95hi first as they both predicted a very rapid decline in the frequency of CD95hi T_SCM that was not compatible with the observations, (Figs Q–R in S1 Text, Table L in S2 Text). We conclude that the data are best described by the linear (Tint first) model (Model D in Fig 7).

Download:

Fig 10. A population of YFV-specific cells enriched for CD95int T_SCM declines much more slowly over time than a population enriched for CD95hi T_SCM.

Here we reproduce the analysis from Fuertes Marraco and colleagues [2] who investigated the decline of YFV-specific memory subpopulations with time post-vaccination in a cross-sectional cohort. Fuertes Marraco and colleagues note that there is large between-individual variation in the sizes of CD8⁺ memory subpopulations and that this noisy interindividual variation can be reduced by quantifying the frequency of YFV-specific cells within each subpopulation rather than within the total CD8⁺ population (e.g., quantify YFV-specific T_CM CD8⁺ cells as a percent of total T_CM rather than as a percent of total CD8⁺ T cells) (A). Analysis of the raw data shows that a proxy for YFV-specific CD95int T_SCM cells (YFV-specific “naïve hi” cells) – left panel – decline significantly more slowly than a proxy for YFV-specific CD95hi T_SCM cells (YFV-specific “naïve int” cells) – right panel (P = 0.039, linear regression). The blue and orange lines are the best fit straight lines to the CD95int and CD95hi T_SCM data, respectively. (B). Analysis in Fuertes Marraco and colleagues [2] differs slightly in that they chose to exclude three individuals with very low frequencies of YFV-specific CD8⁺ T cells and to include 4 individuals who had received multiple YF vaccinations. Our conclusions are robust to these two steps (if anything they appear to strengthen the conclusion that YFV-specific CD95int T_SCM cells decline more slowly than YFV-specific CD95hi T_SCM cells). Recipients of multiple vaccinations shown by open symbols.

https://doi.org/10.1371/journal.pbio.3003179.g010

Studying the same data in a model-independent way, following the approach of Fuertes Marraco and colleagues [2], we see that the YFV-specific CCR7⁺⁺CD45RA⁺⁺ population (proxy for YFV-specific CD95hi T_SCM cells) decline significantly more rapidly than the proxy for YFV-specific CD95int T_SCM cells, Fig 11; a striking validation of our predictions.

Download:

Fig 11. Schematic of the best performing model and parameters.

An overview of the model and parameters that provides the best description of the data sets. Average residency time is average time a cell spends in that population (“Materials and methods”).

https://doi.org/10.1371/journal.pbio.3003179.g011

This work on a new data set, which was unseen during the fitting process, corroborates the parameterized linear model and provides compelling evidence that YFV-specific CD95hi T_SCM cells have significantly more rapid disappearance kinetics than CD95int T_SCM cells.

Discussion

This study establishes that the human T stem cell population consists of two distinct cell subpopulations which can be distinguished by the level of their CD95 expression. We show that these two populations have different functional profiles ex vivo, different transcriptional patterns, and different tissue distributions consistent with their differential expression of the lymph node homing receptor CCR7 [25]. Most importantly for the maintenance of memory, we show that there are considerable differences in the dynamics of the subpopulations whether measured by TRECs, Ki67 or stable isotope labelling with CD95hi T_SCM proliferating significantly more rapidly than CD95int T_SCM cells. Quantifying the labelling data, we found that CD95int T_SCM tend to harbor very long-lived clones compared to CD95hi T_SCM; this was confirmed by the rapid decline of YFV-specific CD95hi T_SCM post-vaccination compared with CD95int T_SCM. Crucially, using in vivo labelling of human volunteers, we establish that these are long-term distinct populations rather than transient states of the same population.

As well as being distinct from one another, CD95int T_SCM cells and CD95hi T_SCM cells are distinct from all other T cell populations. CD95hi T_SCM cells are most similar to T_CM but there are numerous important differences. Pivotal to the function of a stem cell population is its ability to proliferate without the whole population undergoing differentiation and phenotype change – in which case the parent population is “lost”. This property is known as self-renewal. Under homeostatic conditions ex vivo, CD95hi T_SCM cells have significantly higher levels of self-renewal than T_CM. Conversely, under conditions of TCR activation, these relationships swap with CD95hi T_SCM cells freely differentiating while T_CM retain their phenotype (Fig 2). CD95hi T_SCM cells therefore have significantly higher multipotency index in response to TCR stimulation than T_CM. This suggests a division of labor with CD95hi T_SCM cells responsible for maintaining the memory population during homeostasis but during rechallenge it is CD95hi T_SCM cells rather than T_CM cells that rapidly differentiate and provide the required effector populations. CD95int cells are most similar to T_N but again, there are considerable differences in Ki67 expression, cytokine secretion, TREC content and in vivo division (Figs 1, 2, 5–6). Their slow proliferation rates, very long survival times and lower expression of CD95 may protect them from attrition thus preserving these previously antigen-selected clones for the long-term.

CD95int T_SCM have a transcriptional profile more similar to T_N while CD95hi T_SCM more closely resemble T_CM. However, the simple assumption that CD95int and CD95hi T_SCM cells are on a functional continuum from T_N cells to more highly differentiated T_CM cells is undermined by a number of features that do not vary monotonically across the 4 populations (Fig 12). T_CM and T_N have lower effector profiles than either T_SCM subpopulation (Fig 2E). The ability to maintain phenotype upon proliferation does not vary monotonically either, being higher for CD95hi T_SCM than either CD95int T_SCM or T_CM (Fig 2A). Similarly, Ki67 expression is not monotonic: it is low for T_N and CD95int T_SCM cells, increases dramatically for CD95hi cells and then drops again considerably for T_CM cells (Fig 1B).

Download:

Fig 12. Summary of T_SCM CD95 int and T_SCM CD95hi properties.

As well as being distinct from one another in many respects, CD95int T_SCM cells and CD95hi T_SCM cells are distinct from both T_N and T_CM cells. Furthermore, cellular properties do not vary in a monotonic way from T_N to CD95int to CD95hi to T_CM. A blank cell indicates that combination of property and cell population was not investigated in this study.

https://doi.org/10.1371/journal.pbio.3003179.g012

RNAseq is an extremely powerful tool but it can be problematic for identifying novel subpopulations as it is difficult to tell whether two clusters are different states of the same cell population or two distinct populations. For example, it is well known that cells from disparate cell populations will cluster together and away from their parent population when in cycle. Conversely, the same populations under different conditions can be widely separated (e.g., our Fig 3A). Therefore, existence of a separate cluster does not necessarily mean a separate subpopulation (i.e., there can be false positives) and conversely, failure to cluster does not necessarily mean two cells are from a different subpopulation (i.e., there can be false negatives). Similarly, clusters within other high dimensional datasets, such as those generated by CyTOF, can also be hard to interpret. An unbiased study of the CD8⁺ T cell compartment revealed that, on the basis of a principal components analysis of CyTOF data, there were over 200 CD8⁺ T cell subpopulations [26]. Likewise, a CyTOF-based study of the NK cell compartment found 6,000–30,000 different populations [27]. Given the sheer number of subpopulations identified, it seems likely that at least some of these represent multiple states of the same populations rather than hundreds (or in the case of NK cells, thousands) of long-term distinct populations. Here, by using in vivo labelling, we establish that CD95int T_SCM and CD95hi T_SCM are not transient states of the same cell population but instead are two distinct populations that remain distinct for at least 140 days.

An important study from Galletti and colleagues investigated exhaustion in the CD8⁺ T cell memory compartment [19]. They reported populations of dysfunctional, exhausted cells (defined by coexpression of PD-1 and TIGIT) in the T_SCM and T_CM populations. They found that once PD1⁺TIGIT⁺ cells were excluded from the T_SCM and T_CM pool then, by RNAseq, there were relatively few differentially expressed genes between the PD1⁻TIGIT⁻ T_SCM and the PD1⁻TIGIT⁻ T_CM populations. They suggested that the observed functional differences between T_SCM and T_CM [4,10,11,28,29] may be attributable to different proportions of exhausted cells in these populations and that once these exhausted cells were excluded, the remaining T_SCM and T_CM populations could be considered as a single homogeneous population. This interpretation, that dysfunctional status (PD1⁺TIGIT⁺) and functional status (PD1⁻TIGIT⁻) mark two relatively homogenous populations of T_CM and T_SCM cells does not appear compatible with our data in which two cell populations with similar levels of PD1⁺TIGIT⁺ cells (i.e., CD95int T_SCM and T_CM) differed in many characteristics we measured (Ki67 expression, self-renewal, response to TCR stimulation, production of proinflammatory cytokines, multipotency and TREC content, Fig 12). Moreover, for many features (Ki67 expression, differentiation profile, response to TCR, multipotency, production of proinflammatory cytokines) both CD95hi T_SCM and CD95int T_SCM differed from T_CM in the same direction, again inconsistent with the argument that the only difference between T_CM and T_SCM is in the proportion of exhausted cells they contain. Finally, we quantified Ki67 expression in T_CM, T_SCM, PD1⁻TIGIT⁻ T_CM (i.e., excluding exhausted cells) and PD1⁻TIGIT⁻ T_SCM and found not only that the PD1⁻TIGIT⁻ T_SCM had significantly higher levels of Ki67 than PD1⁻TIGIT⁻ T_CM but that exclusion of the exhausted cells did not make the T_SCM and T_CM populations more similar. Our data support the concept that PD1⁺TIGIT⁺ cells can be found in CD95int T_SCM, CD95hi T_SCM and T_CM populations but not that the differences between T_SCM and T_CM observed by others and ourselves can be explained solely by the proportion of these exhausted cells. Our findings are consistent with those of Fuertes Marraco and colleagues [2,4], who in agreement with Galletti and colleagues, found surprisingly little evidence for differentially expressed genes between T_N, T_SCM and T_CM populations (even before exclusion of exhausted cells) but nevertheless found very significant functional differences, particularly in their dynamics: notably the longevity of the YF-vaccine induced population. This indicates that large differences in functionality are not necessarily accompanied by large numbers of differentially expressed genes.

One potential concern regarding our work is the difficulty of sorting cells that are close in flow space. In particular we were concerned that CD95int T_SCM might simply be T_SCM cells contaminated by T_N and CD95hi T_SCM might simply by T_SCM cells contaminated by T_CM cells. However, this interpretation is not supported by the data. Firstly, for a number of features (self-renewal in response to IL-7/IL-15 and IL-15, differentiation profile in response to IL-2 and IL-15, TNFα, IFNγ and CD107a expression following TCR stimulation), CD95int T_SCM are not intermediate between T_N and CD95hi T_SCM (Fig 12). Similarly, there are multiple features (Ki67 expression, multipotency, self-renewal in response to both cytokines and TCR stimulation, TNFα, IFNγ and CD107a expression following TCR stimulation) for which CD95hi T_SCM are not intermediate between CD95int T_SCM and T_CM (Fig 12). Secondly, the differentiation profiles of CD95int T_SCM and CD95hi T_SCM following culture with either IL-15 or IL-2 are not compatible with contamination (Fig C in S1 Text). Another possible concern is that CD95int and CD95hi T_SCM are not separate populations but transient states of a single population (indeed it could be imagined that CD95int T_SCM are “quiescent T_SCM” and CD95hi T_SCM are “activated T_SCM”). However, two lines of evidence rule this out. First, we find significant differences in TREC content between CD95int and CD95hi T_SCM (Fig 5). If there was frequent interchange between the two subpopulations then these populations would have similar TREC contents. Second, we find that the long-term label enrichment in the CD95int and CD95hi T_SCM populations remains distinct for as long as we follow it – establishing that, at least on a time scale of 140 days, there is not frequent interconversion.

Another potential concern is the between-donor variation in cell labelling (Fig 6B). The main reason for this is that the donors varied in the labelling achieved in their saliva (see Panel (A) of Fig K in S1 Text) possibly due to high consumption of “normal” water which would dilute the label in body water or higher levels of exercise; the reduced saliva labelling is adjusted for in calculation of cell kinetics (“Materials and methods”) so this will not impact our estimates of proliferation or turnover.

Establishing lineage relationship between cell subpopulations in humans is notoriously difficult. Indeed, 25 yrs since the description of the effector memory and central memory subpopulations [21] there is still no consensus on their lineage topology [28,30,31]. One popular approach is to analyze TCR repertoire sharing. However, the very small size of the T_SCM population combined with the highly stochastic nature of TCR repertoire sampling precludes such an approach to studying the relationship between CD95int and CD95hi T_SCM cells. Furthermore, observation of shared clones only establishes a common precursor rather than the lineage relationship, i.e., models C, D and E all predict TCR clone sharing between the CD95int and CD95hi T_SCM populations (provided sufficient blood volumes could be taken). Our data are most consistent with a linear differentiation model from CD95int T_SCM to CD95hi T_SCM. Perhaps the most compelling evidence is prediction of unseen data (considerably more challenging that fitting seen data) where this linear model outperforms all other lineage relationships considered by a considerable margin (Table L in S2 Text). This is further supported by the in vitro differentiation data whereby CD95int T_SCM give rise to CD95hi T_SCM but CD95hi T_SCM tend not to give rise to CD95int T_SCM (Fig C in S1 Text). The TREC data and gene expression data also paint a picture consistent with this model. Nevertheless, other lineage topologies cannot be unequivocally ruled out. However, our central conclusion about the extraordinary longevity of clones in the CD95int T_SCM compartment and the rapid turnover of the CD95hi compartment is robust to the assumptions regarding the underlying lineage topology.

Our finding is reminiscent of a study in mice in which a cell’s division history was estimated in transgenic mice using a synthetic “division-reporter” domain [32]. In this study it was found that murine CD27^HIKLRG1^LO cells (considered a model of human T_CM) consisted of two subpopulations. One subpopulation (with parallels to the CD95hi T_SCM population) had undergone more divisions and had a more effector-like profile while the other was more quiescent and had undergone fewer divisions. Similarly, Zhao and colleagues investigate functional heterogeneity in the human T_SCM population and identify two subpopulations on the basis of CD122 expression [18]. Again, there are parallels with our work (CD95hi T_SCM cells tend to be CD122hi) and in agreement with our work these cells tended to have increased cytokine production whereas CD122lo cells (enriched in our CD95int population) were transcriptionally more similar to T_N cells. Unfortunately, the study of Zhao and colleagues was limited to in vitro flow and RNAseq analysis and did not establish whether CD122hi and CD122lo cells were distinct populations nor their division histories or contribution to the maintenance of memory.

Another interesting parallel is with hematopoiesis. It has been shown that the cells at the apex of the hematopoietic stem cell hierarchy – the hematopoietic stem cells (HSCs) – divide very infrequently with the frequency of division increasing for more mature stem cell populations such as long-term HSCs and multipotent progenitors which also show improved self-renewal [33]. T_SCM cells appear to follow the same schema, with the most immature cells, the CD95int T_SCM subpopulation, dividing very infrequently (on average once every 3 yrs) whereas the more mature CD95hi T_SCM subpopulation divides more frequently (approximately once a year) and have significantly higher levels of self-renewal.

There is considerable interest in T_SCM cells in the clinical and translational space. Response to the adoptive transfer of T cells for treatment of metastatic melanoma patients has been linked to the presence of T_SCM cells in the infused treatment product; and the beneficial effect seems to be tightly linked with these clones ability to survive long term [34]. Similarly stem-like T cells appear to be the population responsible for a successful response to immune-checkpoint blockade (ICB). In a study of responder and non-responder patients receiving ICB, the ratio of stem-like TCF7hi: TCF7lo cells infiltrating the tumor was predictive of responsiveness [35]. T_SCM cells are also being explored as the basis for chimeric antigen receptor (CAR)-T cells [36]. All these approaches leverage the unique longevity and self-renewal properties of T_SCM cells. These same key dynamic properties can have deleterious effects in other contexts. T_SCM cells have been linked to the development of adult T cell leukemia, to the survival of CD4⁺ cells latently infected with HIV-1 as well as to the development of type I diabetes [37–39]. Therapeutic targeting of T_SCM cells – either to enhance the efficacy of cancer immunotherapies, to enhance the longevity of the vaccine-induced response or to block T_SCM-driven diseases – is predicated on an understanding of the cells in question. Our finding that there is marked heterogeneity in the unique dynamic properties of this cell population is crucial information.

CD95 (FAS or APO-1) is a member of the tumor necrosis factor (TNF) receptor family. The FAS-FASL pathway plays a critical role in the control of the immune system; it maintains homeostasis by killing unwanted cells and is also important for activation induced cell death of T cells during a chronic immune response [40]. In common with other members of the TNF family, CD95 has a dual role. It can act as a death inducing receptor leading to apoptosis, while also mediating other functions including proliferation and tissue regeneration [41]. It has been demonstrated that T cell proliferation induced by sub-optimal TCR engagement is enhanced when CD95 is bound [42]; while deletion of CD95 in T cells leads to lymphopenia in mice [43]. In activated T cells, CD95 acts as a dual-function receptor: exposure to low levels of CD95L leads to survival but exposure to high concentrations of CD95L leads to apoptosis [44]. Together, these data demonstrate that CD95 expression is involved in T cell proliferation, survival, and activation.

The data we present here are most consistent with a picture in which T_N cells meet cognate antigen, undergo proliferation and a proportion differentiate into the CD95int T_SCM pool. Cells in this CD95int T_SCM population proliferate slowly (median 0.0008 per day, Table 1 in main text), faster than T_N cells (median 0.0003 per day) but still slower than other memory T cell subpopulation [14]. Clones in this compartment are extraordinarily long-lived (order of decades); indeed, they are longer lived than clones of any memory subpopulation defined to date [14]. A finding that was strikingly confirmed by the YFV tetramer data. This population is predominantly found in the lymph node. In response to TCR stimulation these cells are more reactive than T_N or T_CM (but considerably less than T_EM). These cells have a transcriptional profile similar to T_N cells and their frequency decreases with age. In response to homeostatic cytokines CD95int T_SCM cells have limited self-renewal properties. CD95int T_SCM clones are long-lived not because they self-renew but because they are largely quiescent. In contrast, CD95hi T_SCM cells are short-lived both at the cellular and at the clonal level. When they divide under homeostatic conditions they retain their phenotype (i.e., they self-renew) but in response to TCR stimulation they are highly multipotent and differentiate into all downstream memory subsets (T_CM, T_EM, T_EMRA). CD95hi T_SCM cells are found in higher frequencies in the blood than lymph node, accumulate with age, have a much lower TREC content than CD95int T_SCM and have a transcriptional profile that is more similar to T_CM. In vivo they proliferate rapidly (p = 0.002 per day) but have short lived clones (life span of the order of months) and are sustained by differentiation of cells from the CD95int T_SCM pool. Long-term in vivo labelling, establishes that CD95int T_SCM and CD95hi T_SCM are two distinct populations not transient states of the same cell population.

Stem cells are considered to be cells with three main properties: self-renewal, clonal longevity and multipotency. While both T_SCM subpopulations exhibit considerable multipotency, no single population of T_SCM cells has both the dynamic properties of self-renewal and clonal longevity. Instead, the “stemness” of the T_SCM population is generated by the complementary properties of two distinct subpopulations: CD95int T_SCM which have the property of clonal longevity and CD95hi T_SCM which have the properties of expansion and self-renewal. We suggest that together, these two populations function as a stem cell population.

Materials and methods

Ethics

All study procedures were conducted according to the principles of the Declaration of Helsinki and all participants gave written informed consent following protocols approved by the relevant ethics committee (NRES London 13/LO/0022, ICREC 21IC7146, 15/SC/0089/2, Fulham REC 18/LO/2196). Some of the human samples used in this research project were obtained from the Imperial College Healthcare Tissue and Biobank (ICHTB). ICHTB is supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Imperial College Healthcare NHS Trust and Imperial College London. ICHTB is approved by Wales REC3 to release human material for research (22/WA/0214).

Study subjects

We studied a total of 106 healthy volunteers.

Cohort 1. Ex vivo T_SCM analysis and qPCR (N = 25). Donors were aged 18–60 yrs and healthy at the time of donation with no symptomatic or diagnosed infections, allergies or autoimmune diseases.
Cohort 2. Ex vivo T_SCM analysis: replication (N = 8). Donors were aged 18–60 yrs and healthy at the time of donation with no symptomatic or diagnosed infections, allergies or autoimmune diseases.
Cohort 3. Lymph node fine needle aspirate (N = 7). Lymph node cells were taken from left and/or right lymph nodes from 7 healthy, vaccinated HIV-negative volunteers aged 20–44 yrs enrolled into the EAVI2020_01 study [22].
Cohort 4. RNAseq, TREC and phenotype analysis (N = 21). Donors were aged 27–65 yrs and healthy at the time of donation with no symptomatic or diagnosed infections, allergies or autoimmune diseases.
Cohort 5. In vivo stable isotope labelling (N = 3). Donors were aged 38 yrs (DW19), 42 yrs (DW20) and 56 yrs (DW25) and healthy at the time of donation with no symptomatic or diagnosed infections, allergies or autoimmune diseases.
Cohort 6. Telomere analysis (N = 5). Donors were aged 29−83 yrs. All donors were CMV-seropositive and HIV-1-seronegative. This cohort has previously been described [12].
Cohort 7. Yellow fever vaccination cohort (N = 37). Donors were HLA-A*02⁺ and had received the yellow fever vaccination YF-17D at different timepoints in the past (range 0.27–35.02 yrs). This cohort has previously been described [2,12,29].

Flow cytometry: Analysis

Peripheral blood mononuclear cells were stained with a cocktail of fluorescently labelled antibodies (see Table A in S2 Text for clone, supplier and dilution). For intracellular analysis, cells were fixed/permeabilized using the FoxP3 Transcription Factor Staining Buffer Set (eBioscience, Thermo Fisher Scientific, Massachusetts, US) as per manufacturer’s instructions. Cells were then re-suspended in FACS buffer for acquisition. Proliferation was assessed by measuring the progressive dilution of CellTraceViolet (CTV) dye (Thermo Fisher Scientific). Samples were acquired using a BD LSRFortessa (BD Biosciences, New Jersey, US). Compensation was performed using the UltraComp eBeads (Invitrogen, ThermoFisher Scientific) as single-stained controls. Individual cell populations were gated using fluorescence minus one (FMO) controls. Data were quantitated using FlowJo (v9.0). The separation between CD95int T_SCM and CD95hi T_SCM was defined based on backgating of Ki67 with gates initially set to provide maximum discrimination between Ki67⁺ and Ki67⁻ cells. Henceforth, the same gating was used for all individuals except for some cases where the CD8⁺ subpopulations were visibly shifted and so the CD95hi/int gate was moved based on the location of subpopulations in that individual. On changing to a new machine (e.g., sorter) then this process was repeated to obtain the corresponding cutoff for the new machine.

Flow cytometry: Sorting

T cell subsets were sorted to purity (>80%) using a FACSAria III (BD Biosciences). CD8⁺ T cells were sorted into the following subpopulations based on the expression of canonical differentiation markers: T_N (naïve: CD45RA⁺CCR7⁺CD95⁻), CD95int T_SCM (CD95 intermediate T stem cell like-memory: CD45RA⁺CCR7⁺CD95^INT), CD95hi T_SCM (CD95 high T stem cell like-memory: CD45RA⁺CCR7⁺CD95^HI), T_CM (central memory: CD45RA⁻CCR7⁺), T_EM (effector memory: CD45RA⁻CCR7⁻) and T_EMRA (CD45RA-expressing effector memory: CD45RA⁺CCR7⁻). See Fig B in S1 Text for representative gating.

In vitro culture

FACS-purified CD8⁺ T cell subsets were cultured in RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 10% FBS (heat inactivated, Gibco), 1% penicillin-streptomycin and 2 mM L-glutamine (Sigma–Aldrich, Missouri, US). To evaluate self-renewal under homeostatic conditions, cells were stimulated for 7 d either with recombinant human interleukin 7 (IL-7) and interleukin (IL-15) (both at 25 ng/mL, Miltenyi Biotec, Bergisch Gladbach, Germany) or 25 ng/mL IL-15 or 6.1 ng of IL-2. The cytokines were renewed on day 3. To evaluate differentiation, cells were stimulated for 7 days with purified anti-human plate bound (pb) αCD3 and soluble αCD28 (2 μg/ml and 1 μg/ml, respectively, BD Biosciences) in the presence of IL-7 and IL-15. The plates were pre-coated with pb αCD3 for 2 h at 37 °C. To measure cytokine production and cytotoxicity, cells were stimulated for 12 h with pb αCD3 (2 μg/mL) and soluble αCD28 (1 μg/ml). GolgiStop (1×) and GolgiPlug (1×) (both BD Biosciences) were added during the last 10 h of stimulation. Cells were then removed from the stimulation and stained for surface and intracellular markers as described above. For CD107 staining, the antibody was added at the beginning of the stimulation and cells were not re-stained during the surface staining step. Following Gattinoni and colleagues [1], the multipotency index for a cell subpopulation with index j was calculated using the following formula:

(1)

Where p_i is the proportion of differentiated cells at day 7 that have phenotype i and n is the number of phenotypes considered (n = 6).

Cell trace violet staining

Proliferation was calculated by measuring the progressive dilution of CellTraceViolet (CTV) dye (Thermo Fisher Scientific). CTV was diluted at 5 μM in warm PBS and cells were stained for 20 min at room temperature prior to surface staining. FACS buffer was added at the end of the incubation for 5 min to inactivate the dye.

Gene expression analysis: qPCR

The RNAeasy Micro Kit (Qiagen) was used for RNA isolation as per manufacturer’s instructions. RNA isolation was followed by cDNA conversion on the same day using the TaqMan Reverse transcription reagents (Thermo Fisher Scientific) and a SimpliAmp Thermo Cycler (Applied Biosystems, Massachusetts, US). The cDNA was stored at −20 °C until needed. The transcript levels were assessed via qPCR using TaqMan Fast Advanced Master Mix and TaqMan primer probes (all Thermo Fisher Scientific) as per manufacturer’s protocol. See Table B in S2 Text for a list of genes studied and the corresponding probes. qPCR Assays were run on a QuantStudio 7 Flex Real-Time PCR System (Thermo Fisher Scientific) for 45 cycles and gene expression relative to housekeeping gene 18S rRNA was calculated as 2^−ΔCT; where ΔCT is the difference between the cycle threshold (CT) value for the gene of interest and the CT for the housekeeping gene. All subjects from Cohort 1.

Gene expression analysis: Bulk RNAseq

CD8⁺ T_N, CD95int T_SCM, CD95hi T_SCM, T_CM, T_EM and T_EMRA populations were sorted from PBMC of 12 healthy donors. Cells were cultured in RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 10% FBS (heat-inactivated, Gibco), 1% penicillin-streptomycin and 2 mM L-glutamine (Sigma–Aldrich, Missouri, US) for 12 h (“unstimulated”) or cultured as above with the addition of plate-bound αCD3 (2 μg/mL) and soluble αCD28 (1 μg/ml) at 37 °C (“stimulated”). See Table C in S2 Text for experimental design. RNA extraction was performed utilizing the RNeasy Micro Plus Kit (Qiagen) following the manufacturer’s instructions. The quality and quantity of the isolated RNA were assessed using the RNAseq experiment Agilent RNA 6,000 Pico Kit (Agilent Technologies) following the manufacturer’s protocols. Subsequently, the RNA samples were preserved at −80 °C until they were ready for further processing. RNA sequencing was conducted at the Oxford Genomics Centre, where polyAenriched, strand-specific libraries were generated using the NEBNext Ultra II Directional RNA Library Prep Kits (Illumina). All subjects from Cohort 4.

Generation of gene counts from raw reads.

FASTQ files were processed with nf-core/rnaseq bioinformatics pipeline [45] using the STAR-RSEM alignment/quantification methods. Briefly, reads are aligned to the reference genome (GRCh38) using STAR v2.7.10a [46] and gene expression is quantified using RSEM v1.3.1 to deal with ambiguously mapped reads [47]. Mapping statistics and quality control metrics from FastQC and MultiQC reports indicated high data quality for all samples but for 2, where the number of mapping reads was extremely low. These two samples were discarded. Downstream data analysis was performed in R v4.2.1.

Batch correction of gene counts.

The 80 gene count matrices generated with RSEM were read into R with the tximport() function from tximportData package [48]. Gene features that did not have at least 10 counts in at least 3 samples (the size of the smallest biological subgroup, i.e., stimulated T_cm samples) were discarded. Since samples were obtained from different donors, processed at different times and RNA was extracted in 3 separate batches, batch correction was performed to remove batch effects in an unsupervised manner. Briefly, gene counts were first normalized using the median of ratios method to account for differences in library size and RNA composition between samples [49]. Then, normalized counts were transformed using variance stabilizing transformation with the vst() function from DESeq2 package so differences between samples are not overrepresented by highly expressed genes. To remove batch effects, the sva() function from the SVA package function is applied to the vst gene counts and with model ~ population + condition + population:condition (where population: T_naive, T_scm CD95^hi, T_scm CD95^int, T_cm, T_em and T_emraand condition: stimulated and unstimulated). sva() identified 26 surrogate variables that were regressed out from vst counts using the fsva() function. The batch corrected matrix was used for exploratory analysis and QC to visualize the relationship between the samples. Specifically, three different approaches were used for exploratory analysis: principal component analysis (PCA) with prcomp(), hierarchical clustering with hclust() and Euclidean distances between samples using dist(), all functions from the stats package. Four samples had a highly skewed vst count distribution (D11_Hi_S, D3_Int_S, D13_Int_U and D13_Hi_U) and were excluded from downstream DE analysis.

Differential expression analysis with Wilcoxon Rank Sum test.

The wilcox.exact() function from exactRankTests package was used for statistical testing. Briefly, the DESeqDataSet was first filtered to keep unstimulated samples only. Counts from this subset of unstimulated samples were vst transformed after normalization with respect to library size. Surrogate variables were then re-calculated with sva() on the vst transformed counts. The estimated surrogate variables were regressed out from the vst counts. Unstimulated T_scm CD95hi samples were then compared to unstimulated T_scm CD95int samples with Wilcoxon rank sum test. Only genes that had at least 10 counts in at least 6 samples and had GO annotations were tested. P-values were FDR-adjusted for multiple comparisons using the p.adjust() function. Genes with adjusted p-value < 0.05 were considered differentially expressed. The same steps were repeated to compare stimulated T_scm CD95hi samples to stimulated T_scm CD95int samples.

Enrichment analysis.

Pre-ranked gene set enrichment analysis was performed with the fgsea() function from fgsea package [50] and with the GSEA() function from DOSE package [51]. Gene sets tested for enrichment were downloaded from MSigDB [52,53] and only included immune-related pathways (C7: immunologic signature gene sets, [54]). Gene-concept networks to visualize enrichment results were generated with enrichplot package [55].

Heatmaps.

Heatmaps showing the top 70 DEG were generated using the R packages pheatmap and ComplexHeatmap. Note that SVA corrected counts were scaled to obtain z-scores by row, i.e., by gene. Both samples (columns) and genes (rows) are clustered based on Euclidean distance and complete method and using z-scores.

Lymph node cell collection

Ultrasound guided fine needle aspiration was used to collected axillary lymph node samples from volunteers participating in an experimental vaccine immunogen study using an HIV Env Mosaic immunogen (Cohort 3) [22]. Cells were taken either from the left or right lymphoid tissues, i.e., contralateral and ipsilateral to the site of injection (and in some cases both); matched blood samples from the same individuals at the same time point were also collected. Cells were collected between 7 days and 56 days post-first immunization but before the subsequent boost, a time point at which antigen-specific IgG was undetectable [22]. In cases where data were available from both left and right lymph node we first ascertained that the measurements were independent (no significant correlation) and then pooled the data.

T cell receptor excision circle (TREC) quantification

To quantify TRECs from sorted memory subpopulation we used our in-house droplet digital PCR protocol directly from cell lysate [23]. Cell lysate was made using lysis reagents from the SuperScript IV CellsDirect cDNA Synthesis Kit (ThermoFisher). Cell lysate (30 μl approximately) was prepared according to manufacturer’s instructions, omitting the DNA degradation step. The prepared cell lysate was then heated on a thermocycler at 65 °C for 1 min, 96 °C for 2 min, 65 °C for 4 min, 96 °C for 1 min, 65 °C for 1 min and 96 °C for 30 s. Tubes were spun down for 1 min and cooled at room temperature. From the prepared cell lysate, 7.8 μl were added to the reaction mixture containing 11 μl 2× ddPCR Supermix for probes (no dUTP) (Bio-Rad, Pleasanton, CA, USA), 0.55 μl of TRECs primers (36 μM stock) and probes (10 μM) each, 1.1 μl of RPP30 copy number detection (Bio-Rad), 1 μl of HindIII-HF restriction enzyme (NEB, UK). The sequences of primers and probe used for TRECs are: Forward primer 5′-CAC ATC CCT TTC AAC CAT GCT-3′; Reverse primer 5′-GCC AGC TGC AGG GTT TAG G-3′, Probe Sequence: 5′-ACA CCT CTG GTT TTT GTA AAG GTG CCC ACT-3′ with FAM and BQ-1 quencher. For RPP30, we used the primer probe mix from BioRad. Four replicates were made. During data analysis, these replicates were summed to give a total of 80,000 droplets/sample. To make the droplets, 20 μl reaction mixture was loaded into sample loading wells and 70 μl of droplet oil was loaded in the droplet oil wells in a droplet cartridge (Bio-Rad, USA). The cartridge was covered with a gasket and placed in the droplet generator (Bio-Rad #186-3002). An amount of 40 μl of droplets were transferred to a 96-well plate (Bio-Rad, USA). The plate was put on the thermocycler (Bio-Rad, USA) under the following conditions: 10 min hold at 95 °C, 45 cycles of 95 °C for 15 s then 59 °C for 60 s. Amplified droplets were then read in the QX200 Droplet Reader (Bio-Rad, USA). The data were analyzed in QuantaSoft analysis software version 1.7 (Bio-Rad).

Stable isotope labelling in vivo

We have previously described the labelling protocol in detail [13,56]. Briefly, participants (Cohort 5) were given oral doses of 70% deuterated water (²H₂O) over a 7-week period (100 ml twice daily for one day, then 50 ml twice daily for 48 days). Saliva samples were collected for evaluation of body water labelling. Peripheral blood was collected at successive time-points during and after labelling and PBMC separated by Ficoll gradient centrifugation. Monocytes for normalization, as a cell population expected to approach fully-labelled status during the labelling phase, were sorted from an aliquot of PBMC by CD14 magnetic bead column positive selection (MACS, Miltenyi Biotech, Bisley, UK). PBMC were sorted using a BD FACSAria III flow cytometer into CD8⁺ T_N, CD8⁺ CD95int T_SCM and CD8⁺ CD95hi T_SCM. Deuterium enrichment in the DNA of monocytes and sorted T cell subpopulations was measured by gas chromatography/mass spectrometry of the pentafluorobenzyl derivative using a Shimadzu 2030–QP2020NX with SHIM35MS column (Shimadzu, Milton Keynes, UK) as previously described [24,57].

Single-chromosome telomere length analysis

These data have previously been reported [12]. Briefly, DNA from CD8⁺ T_N and total CD8⁺ T_SCM cells (not separated into CD95hi and CD95int subpopulations) was extracted, and single-telomere length analysis was carried out at the XpYp telomere [58].

Yellow fever vaccine data

Published data were acquired from a cross-sectional study of 37 healthy adults who received a single dose of the yellow fever vaccine YF-17D (Cohort 7) [2]. Time since vaccination ranged from 3 mo to 35 yrs. Four subjects who received multiple YF vaccinations were excluded from the analysis. Total CD8⁺CCR7⁺CD45RA⁺ T_SCM cells (not separated into CD95hi and CD95int subpopulations) specific for the HLA-A*02-restricted YFV NS4b214-222 epitope were quantified by flow cytometry with antibody and tetramer staining.

Modelling the data: (1) Stable isotope labelling data

There are three steps in the modelling to estimate the proliferation and disappearance rate of the T cell subsets. First, we quantify the availability of label in body water by measuring the fraction of heavy water in the saliva. We describe this label availability with an empirical function S(t) with three parts (equation 2) to reflect the three part protocol (full dose 1 day, half dose 48 days, delabel).

(2)

Next, for each individual we model the fraction of label in a rapidly turning over population (monocytes) in order to estimate the amplification factor b_w (also referred to as c [59]); this is a factor that reflects the increase in M + 1 when a cell divides given enrichment S(t) it scales between label enrichment in newly synthesized DNA and precursor availability in body water [14]. We describe the label enrichment in DNA of monocytes using a mechanistic model previously proposed [60,61]:

(3)

Where L_M is the fraction of label in bone marrow monocyte precursors, L_B is the fraction of label in blood monocytes (the observable), p_m is the proliferation rate of precursors, r₁ is the rate of exit from the mitotic pool in bone marrow, Δ is the time spent in the post-mitotic pool in bone marrow, M/B is the ratio of the number of monocytes in the bone marrow to the number of monocytes in the blood, S(t) is the saliva enrichment estimated for that individual in step 1 and b_w is the amplification factor of interest. A schematic of this model is provided in Fig J in S1 Text. We use equilibrium constraints to eliminate p_m and r₁. M/B and Δ were fixed at estimates of 2.6 and 1.6 days, respectively [60,61] (however, we showed that the estimates of b_w are independent of these values; this follows because b_w depends only on the plateau enrichment in blood monocytes). There were therefore two free parameters (b_w and r₂) which were estimated by fitting the model to the data. Finally, we use the parameter estimates (f, δ, b_w) from steps 1 and 2 to describe the label enrichment in T cells.

Five different models of T cell dynamics were developed. Each model encapsulates different assumptions about the relationship between the three T cell populations of interest (T_N, CD95int, CD95hi) and their corresponding dynamics. The aim was to see which model was most consistent with the data and to estimate the corresponding parameters. The models are described below and illustrated in Fig 7.

Model A. Independent homogeneous model.

Each cell population (T_N, CD95int, CD95hi) is independent of the others and kinetically homogeneous (i.e., they do not contain subpopulations with different kinetics). Cell numbers in this model are described by the following equations:

(4)

And label is described by the following:

(5)

Where T_N, T₁, T₂ are the sizes of the T_N, CD95int and CD95hi populations, respectively, L_N, L₁, L₂ is the fraction of label in the T_N, CD95int and CD95hi populations, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; d_N, d₁, d₂ the disappearance rates of the T_N, CD95int and CD95hi populations, respectively; b_w is the amplification factor and S(t) the label in the saliva (equation 2). We use equilibrium constraints (T_N, T₁ and T₂ assumed to be independently in equilibrium in (equation 4)) to eliminate d_N, d₁ and d₂.

Model B. Independent heterogeneous model.

Each cell population (T_N, CD95int and CD95hi) is independent of the others and allowed to be kinetically heterogeneous (i.e., they can contain subpopulations with different kinetics). Cell numbers in this model are described by the following equations:

(6)

Label in this model is described by the following equations:

(7)

Where T_N, T₁, T₂ are the sizes of the T_N, CD95int and CD95hi populations, respectively; L_N, L₁, L₂ is the fraction of label in the T_N, CD95int and CD95hi populations, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; d_N, d₁, d₂ is the disappearance rate of the T_N, CD95int and CD95hi populations, respectively; d*_N, d*₁, d*₂ the disappearance rates of labelled cells in the T_N, CD95int and CD95hi populations, respectively; b_w is the amplification factor and S(t) the label in the saliva. In general, due to heterogeneity [62], the disappearance rate of labelled cells is not equal to the disappearance rate of the whole population (d_N ≠ d*_N etc.) and so equilibrium constraints are not helpful here (since d_N, d₁, d₂ do not appear in equation 7).

Model C. Forked differentiation model.

Upon meeting cognate antigen, T_N cells undergo a clonal burst of size k (i.e., undergo k divisions and produce 2^k cells). A fraction f differentiate into CD95int, the remainder (1 − f) differentiate into CD95hi. Cell numbers in this model are described by

(8)

And label in this model is described by:

(9)

Where T_N, T₁, T₂ are the sizes of the T_N, CD95int and CD95hi populations, respectively; L_N, L₁, L₂ is the fraction of label in the T_N, CD95int and CD95hi populations, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; d_N, d₁, d₂ the disappearance rates of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, k the size of the clonal burst, f the fraction of expanded cells that differentiate into the CD95int population, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells, b_w is the amplification factor and S(t) the label in the saliva. We use equilibrium constraints to eliminate d_N, d₁ and d₂. For the case when we fit only the labelling data we reparameterized this model to improve parameter identifiability. We used

(10)

This allows us to eliminate p₁, p₂ and k leaving us with 5 free parameters for the label only fits (p_N, Π₁, Π₂, f, Δ_N). For the case when we are fitting all the data (labelling, telomere, YFV), then the way in which the parameters appear in different combinations in the models of the different types of data precludes this reparameterization and we use the original parameterization when fitting all the data.

Model D. Linear differentiation model (CD95int first).

Upon meeting cognate antigen, T_N cells undergo a clonal burst of size k (producing 2^k cells), these differentiate first into CD95int and subsequently into CD95hi. Cell numbers are described by the following:

(11)

Label in this model is described by the following equations:

(12)

Where T_N, T₁, T₂ are the sizes of the T_N, CD95int and CD95hi populations, respectively; L_N, L₁, L₂ is the fraction of label in the T_N, CD95int and CD95hi populations, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; d_N, d₁, d₂ the disappearance rates of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, k the size of the clonal burst, Δ₁ the rate of differentiation from CD95int to CD95hi, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells, b_w is the amplification factor and S(t) the label in the saliva. We use equilibrium constraints to eliminate d_N, d₁ and d₂.

Model E. Linear differentiation model (CD95hi first).

Upon meeting cognate antigen, T_N cells undergo a clonal burst of size k (producing 2^k cells), these differentiate first into CD95hi and subsequently into CD95int. Cell numbers in this model are described by the following equations

(13)

Label in this model is described by the following equations:

(14)

Where T_N, T₁, T₂ are the sizes of the T_N, CD95int and CD95hi populations, respectively; L_N, L₁, L₂ is the fraction of label in the T_N, CD95int and CD95hi populations, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; d_N, d₁, d₂ is the disappearance rate of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, k the size of the clonal burst, Δ₁ the rate of differentiation from CD95hi to CD95int, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells b_w is the amplification factor and S(t) the label in the saliva. We use equilibrium constraints to eliminate d_N, d₁ and d₂.

Modelling the data: (2) Telomere data

Equations predicting the difference in mean telomere length between T_N and total T_SCM (θ) were derived following [12,63] for each of the 5 models.

Model A. Independent homogeneous model and Model B. Independent heterogeneous model.

(15)

Where p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells, δ is the number of bases lost per division (assumed to be 50 bp) and t is the age of the individual (in days).

Model C. Forked differentiation model.

(16)

Where μ_N, μ₁, μ₂ is the telomere length of T_N, CD95int and CD95hi T_SCM cells, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, k the size of the clonal burst, f the fraction of expanded cells that differentiate into the CD95int population, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells and δ is the number of bases lost per division (assumed to be 50 bp).

Model D. Linear differentiation model (CD95int first).

(17)

with

Where μ_N, μ₁, μ₂ is the length of telomeres in T_N, CD95int and CD95hi T_SCM cells, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, Δ₁ the rate of differentiation from CD95int to CD95hi, k the size of the clonal burst, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells, δ is the number of bases lost per division (assumed to be 50 bp).

Model E. Linear differentiation model (CD95hi first).

(18)

Where μ_N, μ₁, μ₂ is the length of telomeres in T_N, CD95int T_SCM and CD95hi T_SCM cells, respectively; p_N, p₁, p₂ is the proliferation rate of the T_N, CD95int and CD95hi populations, respectively; Δ_N the fraction of T_N undergoing differentiation per day, Δ₁ the rate of differentiation from CD95hi to CD95int, k the size of the clonal burst, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells and δ is the number of bases lost per division (assumed to be 50 bp).

Modelling the data: (3) Frequency of YFV-specific T_SCM cells

The proportion of tetramer⁺ T_SCM cells expressed as a fraction of CD8⁺ CD16⁻ lymphocytes as a function of time since vaccination was calculated for all 5 models by assuming there was no further flow from the T_N pool into the T_SCM pool (∆_N = 0).

Model A. Independent homogeneous model and Model B. Independent heterogeneous model.

For models A and B, since the populations are independent, there is no flow from the T_N pool and so, since the populations are independently at equilibrium, both models predict that the YFV-specific T_SCM population is of constant size with time post-vaccination.

(19)

Where Y(t) is the frequency of the YFV-specific T_SCM population at time t and 2A is the frequency of the YFV-specific T_SCM population at time 0.

Model C. Forked differentiation model.

(20)

Where Y(t) is the frequency of the YFV-specific T_SCM population at time t, T₁(t) is the frequency of the YFV-specific CD95int T_SCM population at time t, T₂(t) is the frequency of YFV-specific CD95hi T_SCM population at time t and 2A is the frequency of the YFV-specific T_SCM population at time 0. Δ_N is the fraction of T_N undergoing differentiation per day, k the size of the clonal burst, f the fraction of expanded cells that differentiate into the CD95int population, R₁ the ratio of Tint to T_N cells, R₂ the ratio of CD95hi to T_N cells.

Model D. Linear differentiation model (CD95int first).

(21)

Where Y(t) is the frequency of the YFV-specific T_SCM population at time t, T₁(t) is the frequency of the YFV-specific CD95int T_SCM population at time t, T₂(t) is the frequency of YFV-specific CD95hi T_SCM population at time t and 2A is the frequency of the YFV-specific T_SCM population at time 0. Δ_N the fraction of T_N undergoing differentiation per day, Δ₁ the rate of differentiation from CD95int to CD95hi, k the size of the clonal burst, R₁ the ratio of CD95int to T_N cells, R₂ the ratio of CD95hi to T_N cells.

Model E. Linear differentiation model (CD95hi first).

(22)

Where Y(t) is the frequency of the YFV-specific T_SCM population at time t, T₁(t) is the frequency of the YFV-specific CD95int T_SCM population at time t, T₂(t) is the frequency of YFV-specific CD95hi T_SCM population at time t and 2A is the frequency of the YFV-specific T_SCM population at time 0. Δ_N the fraction of T_N undergoing differentiation per day, Δ₁ the rate of differentiation from CD95hi to CD95int, k the size of the clonal burst, R₁ the ratio of Tint to T_N cells, R₂ the ratio of CD95hi to T_N cells.

Fitting the data

A multistep fitting procedure was used. First the saliva data was fitted to estimate f and δ for each individual. Then these values were used to estimate b_w and r₂ for each individual. And then finally this value of b_w was used in the equations to fit the labelling data. For this final step two separate fits were performed. In one all data (labelling, telomere, YF-vaccination) were fitted simultaneously and in the second only the labelling data were fitted. When fitting, fits were performed separately for the three individuals with labelling data (DW19, DW20, DW25). The YF vaccination data was cross-sectional data from 37 individuals (none of whom were labelled). Therefore, for each of the three labelled individuals we aimed to predict the rate of decline of YFV-specific T_SCM based on the individual’s kinetic parameters and this was compared with the observed rate of YFV-specific T_SCM decline in the cross-sectional cohort. Similarly, the individuals who received the stable isotope label were different from the individuals whose telomere lengths were measured. We therefore used the parameters from the individuals who were labelled to predict the average value of θ, this was compared with the five observed values of θ. For Models A and B, θ was dependent on age so for these models we predicted θ for an individual of age 34 yrs (the median age of the individuals providing telomere data). For Models C–E, θ is independent of age as these systems reach equilibrium. We therefore compared equilibrium θ (calculated numerically) with the observed θ. For each type of data (labelling, telomere, YFV) the residuals (difference between the observation and prediction) were normalized by the mean of the data to ensure that the residuals arising from the different types of data were on a similar scale.

Models were fitted by minimizing the sum of squared (normalized) residuals using the global optimizer pseudo from the package FME [64] (v1.3.6.2) in R v 4.2.3. Other packages used in the analysis included deSolve (v1.34), debug (v1.3.189) and pryr (v0.1.6).

Calculations of clonal half-life

We wanted to estimate the half-life of a T cell clone as this is related to the half-life of immune memory, rather than the more typically estimated population half-life which is related to the half-life of cells.

For Model C (fork) the clonal half-lives are given by

(23)

For Model D (linear CD95int first) the clonal half-lives are given by

(24)

For Model E (linear CD95hi first) the clonal half-lives are given by

(25)

In order to calculate clonal half-lives and the uncertainty in the clonal half-lives for each individual the data were randomly sampled with replacement to produce a dataset with the same number of points, the model was fit to the data, the best fit parameters estimated (using 30 different random seeds to find the global minima) and the corresponding clonal half-lives calculated. This process was repeated 100 times. The median and interquartile ranges of the parameters and half-lives are reported. We calculate the clonal half-lives from the bootstrap estimates rather than by using the point estimates of the parameters due to potential correlation between the parameters.

Calculations of residency time

We report the residency time (average time that a cell remains in a given population) for CD95int TSCM and CD95hi TSCM calculated using Model D (linear Tint first).

Model predictions

For each of the 5 models considered, the best fit parameters (Table 1 and Tables F–I in S2 Text) and corresponding model equations were used to predict the dynamics of virus-specific CD95int T_SCM and CD95hi T_SCM over time for 35 yrs. These predictions were compared (simply by overlaying, not fitting) with data from Fuertes Marraco and colleagues [2] in which they quantified “Naïve-hi” (CD8⁺CCR7⁺⁺CD45RA⁺⁺) T_SCM cells specific for the HLA-A*02-restricted YFV NS4b214-222 epitope which are enriched for CD95int T_SCM and the remaining “naïve-int” tetramer-positive T_SCM which will be enriched for CD95hi T_SCM. In both cases the tetramer-positive cells were expressed as a percentage of the total CD8⁺CD3⁺ lymphocyte population. The initial condition for CD95int and CD95hi T_SCM populations was taken to be the median of CD95int and the median of CD95hi T_SCM sampled within 1 yr of vaccination, respectively.

Model independent analysis of cross-sectional YFV-tetramer data

We follow the approach of Fuertes Marraco and colleagues [2] who quantified the frequency of YFV-specific cells within each subpopulation rather than within the total CD8⁺ population (e.g., quantify YFV-specific T_CM CD8⁺ cells as a percent of total T_CM rather than as a percent of total CD8⁺ T cells) in order to reduce between-individual variation in the relative sizes of different CD8⁺ subpopulations. She then expressed the frequency log₁₀ and estimated the rate of decline of the YFV-specific subpopulations by linear regression. She also made two further changes: excluding three individuals with very low frequencies of YFV-specific CD8⁺ T cells and including 4 individuals who had received multiple YF vaccinations. We report results from both approaches.

Statistics

All statistics were calculated in R v4.2.3. All P values reported are two tailed and where multiple tests were conducted the number of independent tests is noted. In the case of box plots, the bar represents the median, the box the interquartile range (IQR) and whiskers show lower quartile − 1.5 * IQR and upper quartile + 1.5 * IQR; all data (including outliers) are shown.

Supporting information

S1 Text. Supplementary Figures.

Fig A. Ki67 varies significantly in an independent dataset Fig B. Representative Gating. Fig C. Phenotype of sorted cells following 7 days ex vivo culture. Fig D. CTV profile of sorted cells following 7 days ex vivo culture. Fig E. CD95hi TSCM cells exhibit increased functionality compared to both CD95int T_SCM cells and T_CM cells. Fig F. Phenotypic characterization of CD8⁺ memory subsets. Fig G. Ki67 expression after exclusion of PD1⁺ TIGIT⁺ cells. Fig H. Heat maps of the 70 most differentially expressed genes. Fig I. CD95hi cells are transcriptionally and translationally closer to T_CM while CD95int more closely resemble TN. Fig J. Schematic of Mechanistic Model to Describe Monocyte Kinetics. Fig K. Best fits of the models to the saliva and monocyte data. (A). Fig L. Fit of Model A (independent homogeneous populations) to all data (labelling, telomere, YFV) simultaneously. Fig M. Fit of Model B (independent heterogeneous populations) to all data (labelling, telomere, YFV) simultaneously. Fig N. Fit of Model C (fork) to all data (labelling, telomere, YFV) simultaneously. Fig O. Fit of Model E (linear, CD95hi first) to all data (labelling, telomere, YFV) simultaneously. Fig P. CD45A and CCR7 expression of the CD95hi T_SCM and CD95int T_SCM population. Fig Q. Predictions of Model C (fork) overlaid (NOT fitted) onto a novel data set. Fig R. Predictions of Model E (linear Thi first) overlaid (NOT fitted) onto a novel data set. Fig S. Duplicate of Fig 2 showing unique identifier for each donor in response to request from reviewer.

https://doi.org/10.1371/journal.pbio.3003179.s001

(PDF)

S2 Text. Supplementary Tables.

Table A. Antibodies used for flow cytometry and sorting. Table B. Human TaqMan probes used for RT-PCR. Table C. RNAseq Experimental Design. Table D. Saliva and monocyte parameters. Table E. Parameters in the models for fitting labelling, telomere and YF-vaccination data. Table F. Parameters obtained by fitting model A (indep homog) to labelling, telomere and YFV data. Table G. Parameters obtained by fitting model B (indep heterog) to labelling, telomere and YFV data. Table H. Parameters obtained by fitting model C (fork) to labelling, telomere and YFV data. Table I. Parameters obtained by fitting model E (linear, CD95hi first) to labelling, telomere and YFV data. Table J. Summary of the fits of the 5 models to the labelling, telomere and YFV vaccination data. Table K. Summary of the fits of the 5 models to the labelling data only. Table L. Summary of the comparison of the 5 model predictions with the unseen, unfitted YFV vaccination data.

https://doi.org/10.1371/journal.pbio.3003179.s002

(PDF)

S1 Data. Underlying data.

https://doi.org/10.1371/journal.pbio.3003179.s003

(XLSX)

S1 File. Flow data underlying supplementary figures (PART I).

https://doi.org/10.1371/journal.pbio.3003179.s004

(ZIP)

S2 File. Flow data underlying supplementary figures (PART II).

https://doi.org/10.1371/journal.pbio.3003179.s005

(ZIP)

S3 File. Flow data underlying Fig 2 (main text).

https://doi.org/10.1371/journal.pbio.3003179.s006

(ZIP)

Acknowledgments

We are very grateful to Silvia Fuertes Marraco for sharing her published data.

References

1. Gattinoni L, Lugli E, Ji Y, Pos Z, Paulos CM, Quigley MF, et al. A human memory T cell subset with stem cell-like properties. Nat Med. 2011;17(10):1290–7. pmid:21926977.
- View Article
- PubMed/NCBI
- Google Scholar
2. Fuertes Marraco SA, Soneson C, Cagnon L, Gannon PO, Allard M, Abed Maillard S, et al. Long-lasting stem cell-like memory CD8⁺ T cells with a naïve-like profile upon yellow fever vaccination. Sci Transl Med. 2015;7(282):282ra48. pmid:25855494.
- View Article
- PubMed/NCBI
- Google Scholar
3. Jung S, Jung JH, Noh JY, Kim WJ, Yoon SY, Jung J, et al. The generation of stem cell-like memory cells early after BNT162b2 vaccination is associated with durability of memory CD8⁺ T cell responses. Cell Rep. 2022;40(4):111138. pmid:35839774.
- View Article
- PubMed/NCBI
- Google Scholar
4. Gattinoni L, Speiser DE, Lichterfeld M, Bonini C. T memory stem cells in health and disease. Nat Med. 2017;23(1):18–27. pmid:28060797.
- View Article
- PubMed/NCBI
- Google Scholar
5. Klebanoff CA, Gattinoni L, Torabi-Parizi P, Kerstann K, Cardones AR, Finkelstein SE, et al. Central memory self/tumor-reactive CD8⁺ T cells confer superior antitumor immunity compared with effector memory T cells. Proc Natl Acad Sci U S A. 2005;102(27):9571–6. pmid:15980149.
- View Article
- PubMed/NCBI
- Google Scholar
6. Zhang Y, Joe G, Hexner E, Zhu J, Emerson SG. Host-reactive CD8⁺ memory stem cells in graft-versus-host disease. Nat Med. 2005;11(12):1299–305. pmid:16288282.
- View Article
- PubMed/NCBI
- Google Scholar
7. Gattinoni L, Zhong X-S, Palmer DC, Ji Y, Hinrichs CS, Yu Z, et al. Wnt signaling arrests effector T cell differentiation and generates CD8⁺ memory stem cells. Nat Med. 2009;15(7):808–13. pmid:19525962.
- View Article
- PubMed/NCBI
- Google Scholar
8. Graef P, Buchholz VR, Stemberger C, Flossdorf M, Henkel L, Schiemann M, et al. Serial transfer of single-cell-derived immunocompetence reveals stemness of CD8⁺ central memory T cells. Immunity. 2014;41(1):116–26. pmid:25035956.
- View Article
- PubMed/NCBI
- Google Scholar
9. Gerlach C, Rohr JC, Perié L, van Rooij N, van Heijst JWJ, Velds A, et al. Heterogeneous differentiation patterns of individual CD8⁺ T cells. Science. 2013;340(6132):635–9. pmid:23493421.
- View Article
- PubMed/NCBI
- Google Scholar
10. Lugli E, Dominguez MH, Gattinoni L, Chattopadhyay PK, Bolton DL, Song K, et al. Superior T memory stem cell persistence supports long-lived T cell memory. J Clin Invest. 2013;123(2):594–9. pmid:23281401.
- View Article
- PubMed/NCBI
- Google Scholar
11. Lugli E, Gattinoni L, Roberto A, Mavilio D, Price DA, Restifo NP, et al. Identification, isolation and in vitro expansion of human and nonhuman primate T stem cell memory cells. Nat Protoc. 2013;8(1):33–42. pmid:23222456.
- View Article
- PubMed/NCBI
- Google Scholar
12. Costa Del Amo P, Lahoz-Beneytez J, Boelen L, Ahmed R, Miners KL, Zhang Y, et al. Human TSCM cell dynamics in vivo are compatible with long-lived immunological memory and stemness. PLoS Biol. 2018;16(6):e2005523. pmid:29933397.
- View Article
- PubMed/NCBI
- Google Scholar
13. Ahmed R, Roger L, Costa Del Amo P, Miners KL, Jones RE, Boelen L, et al. Human stem cell-like memory T cells are maintained in a state of dynamic flux. Cell Rep. 2016;17(11):2811–8. pmid:27974195.
- View Article
- PubMed/NCBI
- Google Scholar
14. Macallan DC, Busch R, Asquith B. Current estimates of T cell kinetics in humans. Curr Opin Syst Biol. 2019;18:77–86. pmid:31922055.
- View Article
- PubMed/NCBI
- Google Scholar
15. Surh CD, Boyman O, Purton JF, Sprent J. Homeostasis of memory T cells. Immunol Rev. 2006;211:154–63. pmid:16824125.
- View Article
- PubMed/NCBI
- Google Scholar
16. Prlic M, Lefrancois L, Jameson SC. Multiple choices: regulation of memory CD8 T cell generation and homeostasis by interleukin (IL)-7 and IL-15. J Exp Med. 2002;195(12):F49-52. pmid:12070294.
- View Article
- PubMed/NCBI
- Google Scholar
17. Lugli E, Goldman CK, Perera LP, Smedley J, Pung R, Yovandich JL, et al. Transient and persistent effects of IL-15 on lymphocyte homeostasis in nonhuman primates. Blood. 2010;116(17):3238–48. pmid:20631381.
- View Article
- PubMed/NCBI
- Google Scholar
18. Zhao Y, Cai C, Samir J, Palgen J-L, Keoshkerian E, Li H, et al. Human CD8 T-stem cell memory subsets phenotypic and functional characterization are defined by expression of CD122 or CXCR3. Eur J Immunol. 2021;51(7):1732–47. pmid:33844287.
- View Article
- PubMed/NCBI
- Google Scholar
19. Galletti G, De Simone G, Mazza EMC, Puccio S, Mezzanotte C, Bi TM, et al. Two subsets of stem-like CD8⁺ memory T cell progenitors with distinct fate commitments in humans. Nat Immunol. 2020;21(12):1552–62. pmid:33046887.
- View Article
- PubMed/NCBI
- Google Scholar
20. Wirth TC, Xue H-H, Rai D, Sabel JT, Bair T, Harty JT, et al. Repetitive antigen stimulation induces stepwise transcriptome diversification but preserves a core signature of memory CD8⁺ T cell differentiation. Immunity. 2010;33(1):128–40. pmid:20619696.
- View Article
- PubMed/NCBI
- Google Scholar
21. Sallusto F, Lenig D, Förster R, Lipp M, Lanzavecchia A. Two subsets of memory T lymphocytes with distinct homing potentials and effector functions. Nature. 1999;401(6754):708–12. pmid:10537110.
- View Article
- PubMed/NCBI
- Google Scholar
22. Day S, Kaur C, Cheeseman HM, de Groot E, McFarlane LR, Tanaka M, et al. Comparison of blood and lymph node cells after intramuscular injection with HIV envelope immunogens. Front Immunol. 2022;13:991509. pmid:36275655.
- View Article
- PubMed/NCBI
- Google Scholar
23. Kaur C, Adams S, Kibirige CN, Asquith B. Absolute quantification of rare gene targets in limited samples using crude lysate and ddPCR. Cold Spring Harbor Laboratory; 2024. https://doi.org/10.1101/2024.03.27.586936
24. Busch R, Neese RA, Awada M, Hayes GM, Hellerstein MK. Measurement of cell proliferation by heavy water labeling. Nat Protoc. 2007;2(12):3045–57. pmid:18079703.
- View Article
- PubMed/NCBI
- Google Scholar
25. Choi H, Song H, Jung YW. The roles of CCR7 for the homing of memory CD8⁺ T cells into their survival niches. Immune Netw. 2020;20(3):e20. pmid:32655968.
- View Article
- PubMed/NCBI
- Google Scholar
26. Newell EW, Sigal N, Bendall SC, Nolan GP, Davis MM. Cytometry by time-of-flight shows combinatorial cytokine expression and virus-specific cell niches within a continuum of CD8⁺ T cell phenotypes. Immunity. 2012;36(1):142–52. pmid:22265676.
- View Article
- PubMed/NCBI
- Google Scholar
27. Horowitz A, Strauss-Albee DM, Leipold M, Kubo J, Nemat-Gorgani N, Dogan OC, et al. Genetic and environmental determinants of human NK cell diversity revealed by mass cytometry. Sci Transl Med. 2013;5(208):208ra145. pmid:24154599.
- View Article
- PubMed/NCBI
- Google Scholar
28. Crompton JG, Narayanan M, Cuddapah S, Roychoudhuri R, Ji Y, Yang W, et al. Lineage relationship of CD8⁺ T cell subsets is revealed by progressive changes in the epigenetic landscape. Cell Mol Immunol. 2016;13(4):502–13. pmid:25914936.
- View Article
- PubMed/NCBI
- Google Scholar
29. Fuertes Marraco SA, Soneson C, Delorenzi M, Speiser DE. Genome-wide RNA profiling of long-lasting stem cell-like memory CD8 T cells induced by Yellow Fever vaccination in humans. Genom Data. 2015;5:297–301. pmid:26484272.
- View Article
- PubMed/NCBI
- Google Scholar
30. Restifo NP, Gattinoni L. Lineage relationship of effector and memory T cells. Curr Opin Immunol. 2013;25(5):556–63. pmid:24148236.
- View Article
- PubMed/NCBI
- Google Scholar
31. Koh C-H, Lee S, Kwak M, Kim B-S, Chung Y. CD8 T-cell subsets: heterogeneity, functions, and therapeutic potential. Exp Mol Med. 2023;55(11):2287–99. pmid:37907738.
- View Article
- PubMed/NCBI
- Google Scholar
32. Bresser K, Kok L, Swain AC, King LA, Jacobs L, Weber TS, et al. Replicative history marks transcriptional and functional disparity in the CD8⁺ T cell memory pool. Nat Immunol. 2022;23(5):791–801. pmid:35393592.
- View Article
- PubMed/NCBI
- Google Scholar
33. Johnson C, Belluschi S, Laurenti E. Beyond “to divide or not to divide”: kinetics matters in hematopoietic stem cells. Exp Hematol. 2020;92:1-10.e2. pmid:33188890.
- View Article
- PubMed/NCBI
- Google Scholar
34. Krishna S, Lowery FJ, Copeland AR, Bahadiroglu E, Mukherjee R, Jia L, et al. Stem-like CD8 T cells mediate response of adoptive cell immunotherapy against human cancer. Science. 2020;370(6522):1328–34. pmid:33303615.
- View Article
- PubMed/NCBI
- Google Scholar
35. Sade-Feldman M, Yizhak K, Bjorgaard SL, Ray JP, de Boer CG, Jenkins RW, et al. Defining T cell states associated with response to checkpoint immunotherapy in melanoma. Cell. 2018;175(4):998-1013.e20. pmid:30388456.
- View Article
- PubMed/NCBI
- Google Scholar
36. Arcangeli S, Falcone L, Camisa B, De Girardi F, Biondi M, Giglio F, et al. Next-generation manufacturing protocols enriching TSCM CAR T cells can overcome disease-specific T cell defects in cancer patients. Front Immunol. 2020;11:1217. pmid:32636841.
- View Article
- PubMed/NCBI
- Google Scholar
37. Gearty SV, Dündar F, Zumbo P, Espinosa-Carrasco G, Shakiba M, Sanchez-Rivera FJ, et al. An autoimmune stem-like CD8 T cell population drives type 1 diabetes. Nature. 2022;602(7895):156–61. pmid:34847567.
- View Article
- PubMed/NCBI
- Google Scholar
38. Jaafoura S, de Goër de Herve MG, Hernandez-Vargas EA, Hendel-Chavez H, Abdoh M, Mateo MC, et al. Progressive contraction of the latent HIV reservoir around a core of less-differentiated CD4⁺ memory T cells. Nat Commun. 2014;5:5407. pmid:25382623.
- View Article
- PubMed/NCBI
- Google Scholar
39. Nagai Y, Kawahara M, Hishizawa M, Shimazu Y, Sugino N, Fujii S, et al. T memory stem cells are the hierarchical apex of adult T-cell leukemia. Blood. 2015;125(23):3527–35. pmid:25847015.
- View Article
- PubMed/NCBI
- Google Scholar
40. Strasser A, Jost PJ, Nagata S. The many roles of FAS receptor signaling in the immune system. Immunity. 2009;30(2):180–92. pmid:19239902.
- View Article
- PubMed/NCBI
- Google Scholar
41. Peter ME, Budd RC, Desbarats J, Hedrick SM, Hueber A-O, Newell MK, et al. The CD95 receptor: apoptosis revisited. Cell. 2007;129(3):447–50. pmid:17482535.
- View Article
- PubMed/NCBI
- Google Scholar
42. Alderson MR, Armitage RJ, Maraskovsky E, Tough TW, Roux E, Schooley K, et al. Fas transduces activation signals in normal human T lymphocytes. J Exp Med. 1993;178(6):2231–5. pmid:7504062.
- View Article
- PubMed/NCBI
- Google Scholar
43. Hao Z, Hampel B, Yagita H, Rajewsky K. T cell-specific ablation of Fas leads to Fas ligand-mediated lymphocyte depletion and inflammatory pulmonary fibrosis. J Exp Med. 2004;199(10):1355–65. pmid:15148335.
- View Article
- PubMed/NCBI
- Google Scholar
44. Paulsen M, Janssen O. Pro- and anti-apoptotic CD95 signaling in T cells. Cell Commun Signal. 2011;9:7. pmid:21477291.
- View Article
- PubMed/NCBI
- Google Scholar
45. Harshil P, Phil E, Alexander P, Jonathan M, Olga B, Gregor S. nf-core/rnaseq: nf-core/rnaseq v3.14.0 - Hassium Honey Badger. Zenodo; 2024.
46. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21. pmid:23104886.
- View Article
- PubMed/NCBI
- Google Scholar
47. Li B, Dewey CN. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics. 2011;12:323. pmid:21816040.
- View Article
- PubMed/NCBI
- Google Scholar
48. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550. pmid:25516281.
- View Article
- PubMed/NCBI
- Google Scholar
49. Anders S, Huber W. Differential expression analysis for sequence count data. Genome Biol. 2010;11(10):R106. pmid:20979621.
- View Article
- PubMed/NCBI
- Google Scholar
50. Sergushichev CA. Fgsea. Bioconductor; 2017.
51. Yu G, Wang L-G, Yan G-R, He Q-Y. DOSE: an R/Bioconductor package for disease ontology semantic and enrichment analysis. Bioinformatics. 2015;31(4):608–9. pmid:25677125.
- View Article
- PubMed/NCBI
- Google Scholar
52. Liberzon A, Subramanian A, Pinchback R, Thorvaldsdóttir H, Tamayo P, Mesirov JP. Molecular signatures database (MSigDB) 3.0. Bioinformatics. 2011;27(12):1739–40. pmid:21546393.
- View Article
- PubMed/NCBI
- Google Scholar
53. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–50. pmid:16199517.
- View Article
- PubMed/NCBI
- Google Scholar
54. Godec J, Tan Y, Liberzon A, Tamayo P, Bhattacharya S, Butte AJ, et al. Compendium of immune signatures identifies conserved and species-specific biology in response to inflammation. Immunity. 2016;44(1):194–206. pmid:26795250.
- View Article
- PubMed/NCBI
- Google Scholar
55. Guangchuang Y. Enrichplot. Bioconductor; 2018.
56. Zhang Y, Yan AW, Boelen L, Hadcocks L, Salam A, Gispert DP, et al. KIR-HLA interactions extend human CD8⁺ T cell lifespan in vivo. J Clin Invest. 2023;133(12):e169496. pmid:37071474.
- View Article
- PubMed/NCBI
- Google Scholar
57. Westera L, Zhang Y, Tesselaar K, Borghans JAM, Macallan DC. Quantitating lymphocyte homeostasis in vivo in humans using stable isotope tracers. Methods Mol Biol. 2013;979:107–31. pmid:23397392.
- View Article
- PubMed/NCBI
- Google Scholar
58. Capper R, Britt-Compton B, Tankimanova M, Rowson J, Letsolo B, Man S, et al. The nature of telomere fusion and a definition of the critical telomere length in human cells. Genes Dev. 2007;21(19):2495–508. pmid:17908935.
- View Article
- PubMed/NCBI
- Google Scholar
59. Vrisekoop N, den Braber I, de Boer AB, Ruiter AFC, Ackermans MT, van der Crabben SN, et al. Sparse production but preferential incorporation of recently produced naive T cells in the human peripheral pool. Proc Natl Acad Sci U S A. 2008;105(16):6115–20. pmid:18420820.
- View Article
- PubMed/NCBI
- Google Scholar
60. Patel AA, Zhang Y, Fullerton JN, Boelen L, Rongvaux A, Maini AA, et al. The fate and lifespan of human monocyte subsets in steady state and systemic inflammation. J Exp Med. 2017;214(7):1913–23. pmid:28606987.
- View Article
- PubMed/NCBI
- Google Scholar
61. Lahoz-Beneytez J, Elemans M, Zhang Y, Ahmed R, Salam A, Block M, et al. Human neutrophil kinetics: modeling of stable isotope labeling data supports short blood neutrophil half-lives. Blood. 2016;127(26):3431–8. pmid:27136946.
- View Article
- PubMed/NCBI
- Google Scholar
62. Asquith B, Debacq C, Macallan DC, Willems L, Bangham CRM. Lymphocyte kinetics: the interpretation of labelling data. Trends Immunol. 2002;23(12):596–601. pmid:12464572.
- View Article
- PubMed/NCBI
- Google Scholar
63. De Boer RJ, Noest AJ. T cell renewal rates, telomerase, and telomere length shortening. J Immunol. 1998;160(12):5832–7 pmid:9637494.
- View Article
- PubMed/NCBI
- Google Scholar
64. Soetaert K. R package FME: Inverse modelling, sensitivity, Monte Carlo – applied to a dynamic simulation model. n.d.

[ref1] 1. Gattinoni L, Lugli E, Ji Y, Pos Z, Paulos CM, Quigley MF, et al. A human memory T cell subset with stem cell-like properties. Nat Med. 2011;17(10):1290–7. pmid:21926977.
View Article
PubMed/NCBI
Google Scholar

[2] View Article

[3] PubMed/NCBI

[4] Google Scholar

[ref2] 2. Fuertes Marraco SA, Soneson C, Cagnon L, Gannon PO, Allard M, Abed Maillard S, et al. Long-lasting stem cell-like memory CD8⁺ T cells with a naïve-like profile upon yellow fever vaccination. Sci Transl Med. 2015;7(282):282ra48. pmid:25855494.
View Article
PubMed/NCBI
Google Scholar

[6] View Article

[7] PubMed/NCBI

[8] Google Scholar

[ref3] 3. Jung S, Jung JH, Noh JY, Kim WJ, Yoon SY, Jung J, et al. The generation of stem cell-like memory cells early after BNT162b2 vaccination is associated with durability of memory CD8⁺ T cell responses. Cell Rep. 2022;40(4):111138. pmid:35839774.
View Article
PubMed/NCBI
Google Scholar

[10] View Article

[11] PubMed/NCBI

[12] Google Scholar

[ref4] 4. Gattinoni L, Speiser DE, Lichterfeld M, Bonini C. T memory stem cells in health and disease. Nat Med. 2017;23(1):18–27. pmid:28060797.
View Article
PubMed/NCBI
Google Scholar

[14] View Article

[15] PubMed/NCBI

[16] Google Scholar

[ref5] 5. Klebanoff CA, Gattinoni L, Torabi-Parizi P, Kerstann K, Cardones AR, Finkelstein SE, et al. Central memory self/tumor-reactive CD8⁺ T cells confer superior antitumor immunity compared with effector memory T cells. Proc Natl Acad Sci U S A. 2005;102(27):9571–6. pmid:15980149.
View Article
PubMed/NCBI
Google Scholar

[18] View Article

[19] PubMed/NCBI

[20] Google Scholar

[ref6] 6. Zhang Y, Joe G, Hexner E, Zhu J, Emerson SG. Host-reactive CD8⁺ memory stem cells in graft-versus-host disease. Nat Med. 2005;11(12):1299–305. pmid:16288282.
View Article
PubMed/NCBI
Google Scholar

[22] View Article

[23] PubMed/NCBI

[24] Google Scholar

[ref7] 7. Gattinoni L, Zhong X-S, Palmer DC, Ji Y, Hinrichs CS, Yu Z, et al. Wnt signaling arrests effector T cell differentiation and generates CD8⁺ memory stem cells. Nat Med. 2009;15(7):808–13. pmid:19525962.
View Article
PubMed/NCBI
Google Scholar

[26] View Article

[27] PubMed/NCBI

[28] Google Scholar

[ref8] 8. Graef P, Buchholz VR, Stemberger C, Flossdorf M, Henkel L, Schiemann M, et al. Serial transfer of single-cell-derived immunocompetence reveals stemness of CD8⁺ central memory T cells. Immunity. 2014;41(1):116–26. pmid:25035956.
View Article
PubMed/NCBI
Google Scholar

[30] View Article

[31] PubMed/NCBI

[32] Google Scholar

[ref9] 9. Gerlach C, Rohr JC, Perié L, van Rooij N, van Heijst JWJ, Velds A, et al. Heterogeneous differentiation patterns of individual CD8⁺ T cells. Science. 2013;340(6132):635–9. pmid:23493421.
View Article
PubMed/NCBI
Google Scholar

[34] View Article

[35] PubMed/NCBI

[36] Google Scholar

[ref10] 10. Lugli E, Dominguez MH, Gattinoni L, Chattopadhyay PK, Bolton DL, Song K, et al. Superior T memory stem cell persistence supports long-lived T cell memory. J Clin Invest. 2013;123(2):594–9. pmid:23281401.
View Article
PubMed/NCBI
Google Scholar

[38] View Article

[39] PubMed/NCBI

[40] Google Scholar

[ref11] 11. Lugli E, Gattinoni L, Roberto A, Mavilio D, Price DA, Restifo NP, et al. Identification, isolation and in vitro expansion of human and nonhuman primate T stem cell memory cells. Nat Protoc. 2013;8(1):33–42. pmid:23222456.
View Article
PubMed/NCBI
Google Scholar

[42] View Article

[43] PubMed/NCBI

[44] Google Scholar

[ref12] 12. Costa Del Amo P, Lahoz-Beneytez J, Boelen L, Ahmed R, Miners KL, Zhang Y, et al. Human TSCM cell dynamics in vivo are compatible with long-lived immunological memory and stemness. PLoS Biol. 2018;16(6):e2005523. pmid:29933397.
View Article
PubMed/NCBI
Google Scholar

[46] View Article

[47] PubMed/NCBI

[48] Google Scholar

[ref13] 13. Ahmed R, Roger L, Costa Del Amo P, Miners KL, Jones RE, Boelen L, et al. Human stem cell-like memory T cells are maintained in a state of dynamic flux. Cell Rep. 2016;17(11):2811–8. pmid:27974195.
View Article
PubMed/NCBI
Google Scholar

[50] View Article

[51] PubMed/NCBI

[52] Google Scholar

[ref14] 14. Macallan DC, Busch R, Asquith B. Current estimates of T cell kinetics in humans. Curr Opin Syst Biol. 2019;18:77–86. pmid:31922055.
View Article
PubMed/NCBI
Google Scholar

[54] View Article

[55] PubMed/NCBI

[56] Google Scholar

[ref15] 15. Surh CD, Boyman O, Purton JF, Sprent J. Homeostasis of memory T cells. Immunol Rev. 2006;211:154–63. pmid:16824125.
View Article
PubMed/NCBI
Google Scholar

[58] View Article

[59] PubMed/NCBI

[60] Google Scholar

[ref16] 16. Prlic M, Lefrancois L, Jameson SC. Multiple choices: regulation of memory CD8 T cell generation and homeostasis by interleukin (IL)-7 and IL-15. J Exp Med. 2002;195(12):F49-52. pmid:12070294.
View Article
PubMed/NCBI
Google Scholar

[62] View Article

[63] PubMed/NCBI

[64] Google Scholar

[ref17] 17. Lugli E, Goldman CK, Perera LP, Smedley J, Pung R, Yovandich JL, et al. Transient and persistent effects of IL-15 on lymphocyte homeostasis in nonhuman primates. Blood. 2010;116(17):3238–48. pmid:20631381.
View Article
PubMed/NCBI
Google Scholar

[66] View Article

[67] PubMed/NCBI

[68] Google Scholar

[ref18] 18. Zhao Y, Cai C, Samir J, Palgen J-L, Keoshkerian E, Li H, et al. Human CD8 T-stem cell memory subsets phenotypic and functional characterization are defined by expression of CD122 or CXCR3. Eur J Immunol. 2021;51(7):1732–47. pmid:33844287.
View Article
PubMed/NCBI
Google Scholar

[70] View Article

[71] PubMed/NCBI

[72] Google Scholar

[ref19] 19. Galletti G, De Simone G, Mazza EMC, Puccio S, Mezzanotte C, Bi TM, et al. Two subsets of stem-like CD8⁺ memory T cell progenitors with distinct fate commitments in humans. Nat Immunol. 2020;21(12):1552–62. pmid:33046887.
View Article
PubMed/NCBI
Google Scholar

[74] View Article

[75] PubMed/NCBI

[76] Google Scholar

[ref20] 20. Wirth TC, Xue H-H, Rai D, Sabel JT, Bair T, Harty JT, et al. Repetitive antigen stimulation induces stepwise transcriptome diversification but preserves a core signature of memory CD8⁺ T cell differentiation. Immunity. 2010;33(1):128–40. pmid:20619696.
View Article
PubMed/NCBI
Google Scholar

[78] View Article

[79] PubMed/NCBI

[80] Google Scholar

[ref21] 21. Sallusto F, Lenig D, Förster R, Lipp M, Lanzavecchia A. Two subsets of memory T lymphocytes with distinct homing potentials and effector functions. Nature. 1999;401(6754):708–12. pmid:10537110.
View Article
PubMed/NCBI
Google Scholar

[82] View Article

[83] PubMed/NCBI

[84] Google Scholar

[ref22] 22. Day S, Kaur C, Cheeseman HM, de Groot E, McFarlane LR, Tanaka M, et al. Comparison of blood and lymph node cells after intramuscular injection with HIV envelope immunogens. Front Immunol. 2022;13:991509. pmid:36275655.
View Article
PubMed/NCBI
Google Scholar

[86] View Article

[87] PubMed/NCBI

[88] Google Scholar

[ref23] 23. Kaur C, Adams S, Kibirige CN, Asquith B. Absolute quantification of rare gene targets in limited samples using crude lysate and ddPCR. Cold Spring Harbor Laboratory; 2024. https://doi.org/10.1101/2024.03.27.586936

[ref24] 24. Busch R, Neese RA, Awada M, Hayes GM, Hellerstein MK. Measurement of cell proliferation by heavy water labeling. Nat Protoc. 2007;2(12):3045–57. pmid:18079703.
View Article
PubMed/NCBI
Google Scholar

[91] View Article

[92] PubMed/NCBI

[93] Google Scholar

[ref25] 25. Choi H, Song H, Jung YW. The roles of CCR7 for the homing of memory CD8⁺ T cells into their survival niches. Immune Netw. 2020;20(3):e20. pmid:32655968.
View Article
PubMed/NCBI
Google Scholar

[95] View Article

[96] PubMed/NCBI

[97] Google Scholar

[ref26] 26. Newell EW, Sigal N, Bendall SC, Nolan GP, Davis MM. Cytometry by time-of-flight shows combinatorial cytokine expression and virus-specific cell niches within a continuum of CD8⁺ T cell phenotypes. Immunity. 2012;36(1):142–52. pmid:22265676.
View Article
PubMed/NCBI
Google Scholar

[99] View Article

[100] PubMed/NCBI

[101] Google Scholar

[ref27] 27. Horowitz A, Strauss-Albee DM, Leipold M, Kubo J, Nemat-Gorgani N, Dogan OC, et al. Genetic and environmental determinants of human NK cell diversity revealed by mass cytometry. Sci Transl Med. 2013;5(208):208ra145. pmid:24154599.
View Article
PubMed/NCBI
Google Scholar

[103] View Article

[104] PubMed/NCBI

[105] Google Scholar

[ref28] 28. Crompton JG, Narayanan M, Cuddapah S, Roychoudhuri R, Ji Y, Yang W, et al. Lineage relationship of CD8⁺ T cell subsets is revealed by progressive changes in the epigenetic landscape. Cell Mol Immunol. 2016;13(4):502–13. pmid:25914936.
View Article
PubMed/NCBI
Google Scholar

[107] View Article

[108] PubMed/NCBI

[109] Google Scholar

[ref29] 29. Fuertes Marraco SA, Soneson C, Delorenzi M, Speiser DE. Genome-wide RNA profiling of long-lasting stem cell-like memory CD8 T cells induced by Yellow Fever vaccination in humans. Genom Data. 2015;5:297–301. pmid:26484272.
View Article
PubMed/NCBI
Google Scholar

[111] View Article

[112] PubMed/NCBI

[113] Google Scholar

[ref30] 30. Restifo NP, Gattinoni L. Lineage relationship of effector and memory T cells. Curr Opin Immunol. 2013;25(5):556–63. pmid:24148236.
View Article
PubMed/NCBI
Google Scholar

[115] View Article

[116] PubMed/NCBI

[117] Google Scholar

[ref31] 31. Koh C-H, Lee S, Kwak M, Kim B-S, Chung Y. CD8 T-cell subsets: heterogeneity, functions, and therapeutic potential. Exp Mol Med. 2023;55(11):2287–99. pmid:37907738.
View Article
PubMed/NCBI
Google Scholar

[119] View Article

[120] PubMed/NCBI

[121] Google Scholar

[ref32] 32. Bresser K, Kok L, Swain AC, King LA, Jacobs L, Weber TS, et al. Replicative history marks transcriptional and functional disparity in the CD8⁺ T cell memory pool. Nat Immunol. 2022;23(5):791–801. pmid:35393592.
View Article
PubMed/NCBI
Google Scholar

[123] View Article

[124] PubMed/NCBI

[125] Google Scholar

[ref33] 33. Johnson C, Belluschi S, Laurenti E. Beyond “to divide or not to divide”: kinetics matters in hematopoietic stem cells. Exp Hematol. 2020;92:1-10.e2. pmid:33188890.
View Article
PubMed/NCBI
Google Scholar

[127] View Article

[128] PubMed/NCBI

[129] Google Scholar

[ref34] 34. Krishna S, Lowery FJ, Copeland AR, Bahadiroglu E, Mukherjee R, Jia L, et al. Stem-like CD8 T cells mediate response of adoptive cell immunotherapy against human cancer. Science. 2020;370(6522):1328–34. pmid:33303615.
View Article
PubMed/NCBI
Google Scholar

[131] View Article

[132] PubMed/NCBI

[133] Google Scholar

[ref35] 35. Sade-Feldman M, Yizhak K, Bjorgaard SL, Ray JP, de Boer CG, Jenkins RW, et al. Defining T cell states associated with response to checkpoint immunotherapy in melanoma. Cell. 2018;175(4):998-1013.e20. pmid:30388456.
View Article
PubMed/NCBI
Google Scholar

[135] View Article

[136] PubMed/NCBI

[137] Google Scholar

[ref36] 36. Arcangeli S, Falcone L, Camisa B, De Girardi F, Biondi M, Giglio F, et al. Next-generation manufacturing protocols enriching TSCM CAR T cells can overcome disease-specific T cell defects in cancer patients. Front Immunol. 2020;11:1217. pmid:32636841.
View Article
PubMed/NCBI
Google Scholar

[139] View Article

[140] PubMed/NCBI

[141] Google Scholar

[ref37] 37. Gearty SV, Dündar F, Zumbo P, Espinosa-Carrasco G, Shakiba M, Sanchez-Rivera FJ, et al. An autoimmune stem-like CD8 T cell population drives type 1 diabetes. Nature. 2022;602(7895):156–61. pmid:34847567.
View Article
PubMed/NCBI
Google Scholar

[143] View Article

[144] PubMed/NCBI

[145] Google Scholar

[ref38] 38. Jaafoura S, de Goër de Herve MG, Hernandez-Vargas EA, Hendel-Chavez H, Abdoh M, Mateo MC, et al. Progressive contraction of the latent HIV reservoir around a core of less-differentiated CD4⁺ memory T cells. Nat Commun. 2014;5:5407. pmid:25382623.
View Article
PubMed/NCBI
Google Scholar

[147] View Article

[148] PubMed/NCBI

[149] Google Scholar

[ref39] 39. Nagai Y, Kawahara M, Hishizawa M, Shimazu Y, Sugino N, Fujii S, et al. T memory stem cells are the hierarchical apex of adult T-cell leukemia. Blood. 2015;125(23):3527–35. pmid:25847015.
View Article
PubMed/NCBI
Google Scholar

[151] View Article

[152] PubMed/NCBI

[153] Google Scholar

[ref40] 40. Strasser A, Jost PJ, Nagata S. The many roles of FAS receptor signaling in the immune system. Immunity. 2009;30(2):180–92. pmid:19239902.
View Article
PubMed/NCBI
Google Scholar

[155] View Article

[156] PubMed/NCBI

[157] Google Scholar

[ref41] 41. Peter ME, Budd RC, Desbarats J, Hedrick SM, Hueber A-O, Newell MK, et al. The CD95 receptor: apoptosis revisited. Cell. 2007;129(3):447–50. pmid:17482535.
View Article
PubMed/NCBI
Google Scholar

[159] View Article

[160] PubMed/NCBI

[161] Google Scholar

[ref42] 42. Alderson MR, Armitage RJ, Maraskovsky E, Tough TW, Roux E, Schooley K, et al. Fas transduces activation signals in normal human T lymphocytes. J Exp Med. 1993;178(6):2231–5. pmid:7504062.
View Article
PubMed/NCBI
Google Scholar

[163] View Article

[164] PubMed/NCBI

[165] Google Scholar

[ref43] 43. Hao Z, Hampel B, Yagita H, Rajewsky K. T cell-specific ablation of Fas leads to Fas ligand-mediated lymphocyte depletion and inflammatory pulmonary fibrosis. J Exp Med. 2004;199(10):1355–65. pmid:15148335.
View Article
PubMed/NCBI
Google Scholar

[167] View Article

[168] PubMed/NCBI

[169] Google Scholar

[ref44] 44. Paulsen M, Janssen O. Pro- and anti-apoptotic CD95 signaling in T cells. Cell Commun Signal. 2011;9:7. pmid:21477291.
View Article
PubMed/NCBI
Google Scholar

[171] View Article

[172] PubMed/NCBI

[173] Google Scholar

[ref45] 45. Harshil P, Phil E, Alexander P, Jonathan M, Olga B, Gregor S. nf-core/rnaseq: nf-core/rnaseq v3.14.0 - Hassium Honey Badger. Zenodo; 2024.

[ref46] 46. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics. 2013;29(1):15–21. pmid:23104886.
View Article
PubMed/NCBI
Google Scholar

[176] View Article

[177] PubMed/NCBI

[178] Google Scholar

[ref47] 47. Li B, Dewey CN. RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome. BMC Bioinformatics. 2011;12:323. pmid:21816040.
View Article
PubMed/NCBI
Google Scholar

[180] View Article

[181] PubMed/NCBI

[182] Google Scholar

[ref48] 48. Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550. pmid:25516281.
View Article
PubMed/NCBI
Google Scholar

[184] View Article

[185] PubMed/NCBI

[186] Google Scholar

[ref49] 49. Anders S, Huber W. Differential expression analysis for sequence count data. Genome Biol. 2010;11(10):R106. pmid:20979621.
View Article
PubMed/NCBI
Google Scholar

[188] View Article

[189] PubMed/NCBI

[190] Google Scholar

[ref50] 50. Sergushichev CA. Fgsea. Bioconductor; 2017.

[ref51] 51. Yu G, Wang L-G, Yan G-R, He Q-Y. DOSE: an R/Bioconductor package for disease ontology semantic and enrichment analysis. Bioinformatics. 2015;31(4):608–9. pmid:25677125.
View Article
PubMed/NCBI
Google Scholar

[193] View Article

[194] PubMed/NCBI

[195] Google Scholar

[ref52] 52. Liberzon A, Subramanian A, Pinchback R, Thorvaldsdóttir H, Tamayo P, Mesirov JP. Molecular signatures database (MSigDB) 3.0. Bioinformatics. 2011;27(12):1739–40. pmid:21546393.
View Article
PubMed/NCBI
Google Scholar

[197] View Article

[198] PubMed/NCBI

[199] Google Scholar

[ref53] 53. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A. 2005;102(43):15545–50. pmid:16199517.
View Article
PubMed/NCBI
Google Scholar

[201] View Article

[202] PubMed/NCBI

[203] Google Scholar

[ref54] 54. Godec J, Tan Y, Liberzon A, Tamayo P, Bhattacharya S, Butte AJ, et al. Compendium of immune signatures identifies conserved and species-specific biology in response to inflammation. Immunity. 2016;44(1):194–206. pmid:26795250.
View Article
PubMed/NCBI
Google Scholar

[205] View Article

[206] PubMed/NCBI

[207] Google Scholar

[ref55] 55. Guangchuang Y. Enrichplot. Bioconductor; 2018.

[ref56] 56. Zhang Y, Yan AW, Boelen L, Hadcocks L, Salam A, Gispert DP, et al. KIR-HLA interactions extend human CD8⁺ T cell lifespan in vivo. J Clin Invest. 2023;133(12):e169496. pmid:37071474.
View Article
PubMed/NCBI
Google Scholar

[210] View Article

[211] PubMed/NCBI

[212] Google Scholar

[ref57] 57. Westera L, Zhang Y, Tesselaar K, Borghans JAM, Macallan DC. Quantitating lymphocyte homeostasis in vivo in humans using stable isotope tracers. Methods Mol Biol. 2013;979:107–31. pmid:23397392.
View Article
PubMed/NCBI
Google Scholar

[214] View Article

[215] PubMed/NCBI

[216] Google Scholar

[ref58] 58. Capper R, Britt-Compton B, Tankimanova M, Rowson J, Letsolo B, Man S, et al. The nature of telomere fusion and a definition of the critical telomere length in human cells. Genes Dev. 2007;21(19):2495–508. pmid:17908935.
View Article
PubMed/NCBI
Google Scholar

[218] View Article

[219] PubMed/NCBI

[220] Google Scholar

[ref59] 59. Vrisekoop N, den Braber I, de Boer AB, Ruiter AFC, Ackermans MT, van der Crabben SN, et al. Sparse production but preferential incorporation of recently produced naive T cells in the human peripheral pool. Proc Natl Acad Sci U S A. 2008;105(16):6115–20. pmid:18420820.
View Article
PubMed/NCBI
Google Scholar

[222] View Article

[223] PubMed/NCBI

[224] Google Scholar

[ref60] 60. Patel AA, Zhang Y, Fullerton JN, Boelen L, Rongvaux A, Maini AA, et al. The fate and lifespan of human monocyte subsets in steady state and systemic inflammation. J Exp Med. 2017;214(7):1913–23. pmid:28606987.
View Article
PubMed/NCBI
Google Scholar

[226] View Article

[227] PubMed/NCBI

[228] Google Scholar

[ref61] 61. Lahoz-Beneytez J, Elemans M, Zhang Y, Ahmed R, Salam A, Block M, et al. Human neutrophil kinetics: modeling of stable isotope labeling data supports short blood neutrophil half-lives. Blood. 2016;127(26):3431–8. pmid:27136946.
View Article
PubMed/NCBI
Google Scholar

[230] View Article

[231] PubMed/NCBI

[232] Google Scholar

[ref62] 62. Asquith B, Debacq C, Macallan DC, Willems L, Bangham CRM. Lymphocyte kinetics: the interpretation of labelling data. Trends Immunol. 2002;23(12):596–601. pmid:12464572.
View Article
PubMed/NCBI
Google Scholar

[234] View Article

[235] PubMed/NCBI

[236] Google Scholar

[ref63] 63. De Boer RJ, Noest AJ. T cell renewal rates, telomerase, and telomere length shortening. J Immunol. 1998;160(12):5832–7 pmid:9637494.
View Article
PubMed/NCBI
Google Scholar

[238] View Article

[239] PubMed/NCBI

[240] Google Scholar

[ref64] 64. Soetaert K. R package FME: Inverse modelling, sensitivity, Monte Carlo – applied to a dynamic simulation model. n.d.

Figures

Abstract

Introduction

Results

CD95 expression identifies TSCM subpopulations with differential ex vivo proliferation

CD95hi TSCM cells exhibit increased self-renewal and functionality compared to CD95int TSCM cells

TSCM subsets have different gene expression profiles

CD95int and CD95hi TSCM subsets have different lymph node homing properties

TSCM subsets have different division histories and different long-term stability

CD95int and CD95hi TSCM subsets have different dynamics in humans in vivo

Model predictions agree well with an unseen data set

Discussion

Materials and methods

Ethics

Study subjects

Flow cytometry: Analysis

Flow cytometry: Sorting

In vitro culture

Cell trace violet staining

Gene expression analysis: qPCR

Gene expression analysis: Bulk RNAseq

Generation of gene counts from raw reads.

Batch correction of gene counts.

Differential expression analysis with Wilcoxon Rank Sum test.

Enrichment analysis.

Heatmaps.

Lymph node cell collection

T cell receptor excision circle (TREC) quantification

Stable isotope labelling in vivo

Single-chromosome telomere length analysis

Yellow fever vaccine data

Modelling the data: (1) Stable isotope labelling data

Model A. Independent homogeneous model.

Model B. Independent heterogeneous model.

Model C. Forked differentiation model.

Model D. Linear differentiation model (CD95int first).

Model E. Linear differentiation model (CD95hi first).

Modelling the data: (2) Telomere data

Model A. Independent homogeneous model and Model B. Independent heterogeneous model.

Model C. Forked differentiation model.

Model D. Linear differentiation model (CD95int first).

Model E. Linear differentiation model (CD95hi first).

Modelling the data: (3) Frequency of YFV-specific TSCM cells

Model A. Independent homogeneous model and Model B. Independent heterogeneous model.

Model C. Forked differentiation model.

Model D. Linear differentiation model (CD95int first).

Model E. Linear differentiation model (CD95hi first).

Fitting the data

Calculations of clonal half-life

Calculations of residency time

Model predictions

Model independent analysis of cross-sectional YFV-tetramer data

Statistics

Supporting information

S1 Text. Supplementary Figures.

S2 Text. Supplementary Tables.

S1 Data. Underlying data.

S1 File. Flow data underlying supplementary figures (PART I).

S2 File. Flow data underlying supplementary figures (PART II).

S3 File. Flow data underlying Fig 2 (main text).

Acknowledgments

References

CD95 expression identifies T_SCM subpopulations with differential ex vivo proliferation

CD95hi T_SCM cells exhibit increased self-renewal and functionality compared to CD95int T_SCM cells

T_SCM subsets have different gene expression profiles

CD95int and CD95hi T_SCM subsets have different lymph node homing properties

T_SCM subsets have different division histories and different long-term stability

CD95int and CD95hi T_SCM subsets have different dynamics in humans in vivo

Modelling the data: (3) Frequency of YFV-specific T_SCM cells