Research ArticleInfectious Disease

Long-lasting stem cell–like memory CD8+ T cells with a naïve-like profile upon yellow fever vaccination

See allHide authors and affiliations

Science Translational Medicine  08 Apr 2015:
Vol. 7, Issue 282, pp. 282ra48
DOI: 10.1126/scitranslmed.aaa3700

Yellow fever vaccine induces long-term naïve-like memory

In the ongoing quest to find better models of human disease, humans themselves are frequently overlooked. New vaccines for viral infections have hit barriers in translating attempts to induce protective immunity by producing long-lasting memory T cell responses. Now, Fuertes Marraco et al. report that individuals who receive the current yellow fever vaccine develop just that. They found that yellow fever–specific CD8+ T cells with a naïve-like phenotype persisted in vaccinated individuals for more than 25 years. These cells were capable of self-renewal and resembled the stem cell–like memory subset. Thus, by studying vaccinated individuals and building on their own success, researcher may learn—in people—what exactly makes long-term memory T cells tick.


Efficient and persisting immune memory is essential for long-term protection from infectious and malignant diseases. The yellow fever (YF) vaccine is a live attenuated virus that mediates lifelong protection, with recent studies showing that the CD8+ T cell response is particularly robust. Yet, limited data exist regarding the long-term CD8+ T cell response, with no studies beyond 5 years after vaccination. We investigated 41 vaccinees, spanning 0.27 to 35 years after vaccination. YF-specific CD8+ T cells were readily detected in almost all donors (38 of 41), with frequencies decreasing with time. As previously described, effector cells dominated the response early after vaccination. We detected a population of naïve-like YF-specific CD8+ T cells that was stably maintained for more than 25 years and was capable of self-renewal ex vivo. In-depth analyses of markers and genome-wide mRNA profiling showed that naïve-like YF-specific CD8+ T cells in vaccinees (i) were distinct from genuine naïve cells in unvaccinated donors, (ii) resembled the recently described stem cell–like memory subset (Tscm), and (iii) among all differentiated subsets, had profiles closest to naïve cells. Our findings reveal that CD8+ Tscm are efficiently induced by a vaccine in humans, persist for decades, and preserve a naïveness-like profile. These data support YF vaccination as an optimal mechanistic model for the study of long-lasting memory CD8+ T cells in humans.


Cytotoxic CD8+ T cells are critical to the destruction and clearance of pathological cells that harbor intracellular infections or aberrant alterations such as tumors. Therefore, CD8+ T cells are at the center stage in the design of immunotherapeutic strategies to fight cancer and chronic infectious diseases. Compared to established therapies (antimicrobrial drugs, chemo- or radiotherapies), a major advantage of immunotherapy lies in the “memory” quality of adaptive immune responses, with potential for long-term and recall effects. Great efforts are ongoing to better understand the mechanisms underlying the development and long-term maintenance of memory for protection from disease (15).

The study of human CD8+ T cell responses has largely relied on antiviral responses to highly prevalent or medically relevant viruses. These conventional references include the lifelong persisting herpes viruses cytomegalovirus (CMV) and Epstein-Barr virus (EBV), the acute and seasonal influenza (flu) virus, and chronic viruses that cause severe pathology such as hepatitis C virus (HCV) and HIV (6). For instance, the “CEF” (CMV, EBV, flu) peptide pool is among the most frequently used positive controls to monitor human CD8+ T cells (7). However, these viral specificities correspond to infections that are difficult to track in terms of the timing of the exposure to the pathogen, including recurrence in the case of chronic infections. Conversely, active vaccination offers a scenario of synchronized and supervised induction of the immune response. Hence, vaccines represent, in practice, the best available models for the controlled study of an immune response in humans (8).

Although prevention of infection per se is primarily guaranteed by the induction of neutralizing antibodies, the cytotoxic CD8+ T cell responses are of particular importance in the clearance of pathogens (9). Several vaccines and, in particular, live attenuated formulae such as the yellow fever (YF) and smallpox vaccines induce robust T cell responses considered as major contributors to the protection conferred by vaccination (8, 10). The live attenuated YF vaccine (YF-17D and YF-17DD strains) stands out as one of the most successful in terms of public health impact, as well as the robustness and quality of the immune response elicited (8, 10, 11). In use since 1937, the vaccine is known to induce neutralizing antibodies that persist at least up to 38 years (12). In recent years, the YF-17D vaccine has also attracted major research interest for its capacity to induce a particularly strong CD8+ T cell response. This response features high magnitude, broad specificity with overlapping epitopes spanning multiple human leukocyte antigen (HLA) class I and class II motifs (conferring prevalent and high immunogenicity in the human population), polyfunctionality, high proliferative potential, and long-term persistence (1316). In addition, the particular innate response to the live attenuated virus may contribute to the success of the YF vaccine, featuring the activation of multiple dendritic cell subsets via several pattern recognition receptors to orchestrate such a competent adaptive immune response (1719).

Upon vaccination, there is a peak of viremia that resolves by day 15. The CD8+ T cell response peaks between 14 and 30 days and can reach up to 10% of the circulating CD8+ T cells even for a single specificity such as the NS4b214–222 epitope (13, 15, 16). In terms of the differentiation phenotype, YF-specific CD8+ T cells show an effector phenotype (CCR7 CD45RA) very early in the response (day 14). However, CD45RA is reexpressed from day 30 onward (13, 16, 20). This EMRA population shows properties of polyfunctional memory cells (CD28+/−, CD27+, proliferation), which defies the conventional classification of EMRA (CCR7 CD45RA+) as terminally differentiated effector CD8+ T cells (10, 13, 16).

In the light of these recent reports, it has become apparent that YF vaccination induces a distinct activation and memory differentiation of human CD8+ T cells that may explain its particular efficacy; however, reports so far have addressed the phenotype of YF-specific CD8+ T cells with particular emphasis in the first 90 days after vaccine administration, and very limited data exist at later time points. The EMRA phenotype for instance has been observed in limited numbers of donors and only up until 46 months after vaccination (13, 16, 20). Here, we aimed to thoroughly characterize the memory CD8+ T cells that persist in the long term, in the range of decades after YF vaccination.


The YF-17D vaccine induces a naïve-like population of antigen-experienced CD8+ T cells that is stably maintained for more than 25 years

We studied a cohort of 41 healthy volunteers vaccinated with YF-17D, between 3 months and 35 years ago, including four individuals having received multiple vaccines (table S1). To detect YF-specific CD8+ T cells, we used HLA-A*02 tetramers to stain cells bearing a T cell receptor (TCR) specific for the highly prevalent NS4b214−222 epitope, hereafter referred as A2/NS4b (Fig. 1A) (13, 14, 16). A2/NS4b+ CD8+ T cells were detected in the large majority of vaccinees, with only 3 of 41 donors having frequencies below 0.01% and considered negative (Fig. 1B). Remarkably, A2/NS4b+ CD8+ T cells were detected over at least 25 years after vaccination, albeit at decreasing frequencies with time (Fig. 1B).

Fig. 1. A2/NS4b+ CD8+ T cells persist long term after YF vaccination, featuring a naïve-like population.

Forty-one vaccinees were studied, covering 0.27 to 35 years after vaccination (table S1). Unvaccinated donors (UN; n = 10) were studied as controls. (A) Representative flow cytometry plots showing A2/NS4b tetramer staining of total CD8+ T cells. (B) Frequencies of A2/NS4b+ cells within total CD8+ T in unvaccinated controls and in vaccinees versus years since vaccination. Donors below 0.01% were considered negative. •, vaccinees (n = 34); ˚, vaccinees with multiple vaccines (n = 4; time since last vaccination is taken into account); ⋄, unvaccinated controls (UN; n = 3 of 10); ×, excluded donors [vaccinees (n = 3 of 41) or unvaccinated controls (n = 7 of 10)]. Linear regressions correspond to the single-shot vaccinees group (•), indicating goodness of fit (R2) and 95% confidence intervals. (C) Gating strategy to discriminate differentiation subsets based on the conventional markers CCR7 and CD45RA, as indicated. (D) Representative flow cytometry plots showing subsets (CCR7 versus CD45RA) in A2/NS4b+ CD8+ T cells from various vaccinees. (E) Subset distribution (%) within A2/NS4b-specific CD8+ T cells across vaccinees, ordered vertically with increasing “years since vaccination.”

To study the differentiation status of YF-specific CD8+ T cells, we defined subsets based on the expression of conventional markers CCR7 and CD45RA, namely, naïve (CCR7+ CD45RA+), CM (CCR7+ CD45RA), EM (CCR7 CD45RA), and EMRA (CCR7 CD45RA+) subsets (6), as indicated in Fig. 1C. A2/NS4b+ CD8+ T cells showed considerable heterogeneity in subset distribution among donors. Overall, only a few donors showed substantial CM populations, and the largest proportion of cells were found either in the EMRA subset or, surprisingly, falling within the conventional naïve gate (Fig. 1, D and E). We hereafter termed these CCR7+ CD45RA+ A2/NS4b+ CD8+ T cells “naïve-like” because they appeared in the conventional naïve gate, but their “genuine naïve versus memory” nature was to be determined, as addressed throughout the experiments that follow. Moreover, we observed that these A2/NS4b+ naïve-like CD8+ T cells displayed variable levels for CCR7 and CD45RA. There was either CCR7high CD45RAhigh expression (hereafter termed “naïvehigh”) or intermediate CCR7 and/or CD45RA expression (hereafter termed “naïveint”). The detailed gating is shown in Fig. 1 (C and D) and quantified in Fig. 1E.

Next, we addressed the relationship between the differentiation status of YF-specific CD8+ T cells in relation to the years since vaccination. Notably, there was an inherent trend toward longer vaccine history corresponding to vaccinees of older age (fig. S1A). Increasing age is known to correlate with fewer naïve cells and slightly more differentiated cells (EM/EMRA) (fig. S1, B and C) (21, 22). To normalize interdonor and age-related variations, we quantified the frequency of A2/NS4b+ CD8+ T cells within each differentiation subset, rendering the data independent of the personal composition of subsets in total CD8+ T cells as well as independent of the age of donors (fig. S1D).

YF vaccination induced a large population of differentiated CD8+ T cells, comprising up to 6.45% of EMRA early after vaccination, which steadily decreased with time (Fig. 2A, panel “EMRA,” and Fig. 2C, slope = −0.093). Although generally less frequent, the CM and EM populations detected also decreased with time (Fig. 2A, panels “EM” and “CM”) but less steeply (Fig. 2C). The population of naïve-like A2/NS4b+ CD8+ T cells was also clearly increased in vaccinees compared to unvaccinated individuals, hinting that this naïve-like population is mobilized by the vaccine (Fig. 2A, panel “Naïve”). Yet most strikingly, the frequency of naïve-like A2/NS4b+ CD8+ T cells was stably maintained over the years after vaccination, even after 25 years (Fig. 2C, slope = −0.014).

Fig. 2. The naïve-like population of A2/NS4b+ CD8+ T cells induced by the YF-17D vaccine is stable long term.

(A and B) Quantification of the frequency of A2/NS4b+ cells among the various conventional subsets showing naïve, CM, EM, and EMRA (A), as well as naïvehigh and naïveint (B) versus increasing years since vaccination. ˚, vaccinees (n = 34); ˚, vaccinees with multiple vaccines (n = 4; time since last vaccination is taken into account); ⋄, unvaccinated controls (n = 3). Linear regressions shown correspond to the single-shot vaccinees group (•) with details on R2, line, and 95% confidence intervals. (C) Compilation of the slopes with SE from the linear regressions shown in (A) and (B); the slope reflects the trend for change in the log10 frequency of A2/NS4b+ cells with increasing time since vaccination.

Further analysis of the A2/NS4b+ CD8+ T cells within the conventional naïve gate showed different behaviors for the naïvehigh versus the naïveint cells. The latter were sharply induced by the vaccine, reaching frequencies up to 2.46%, with a slow but significant decrease with time (Fig. 2B, panel “Naïveint” and Fig. 2C, slope = −0.034). In contrast, the naïvehigh A2/NS4b+ CD8+ T cells barely increased, only by an average 1.5-fold in vaccinated versus detectable unvaccinated donors [note also that most unvaccinated donors (7 of 10) had A2/NS4b+ frequencies below the detection limit of <0.01%]. The naïvehigh A2/NS4b+ CD8+ T cells also showed a strikingly flat time course (Fig. 2B, panel “naïvehigh” and Fig. 2C, slope = −0.00009). As an additional observation, the four vaccinees with a history of multiple vaccines (table S1) showed frequencies and phenotype of A2/NS4b+ CD8+ T comparable to single-shot vaccinees (Figs. 1B and 2).

Naïve-like YF-specific CD8+ T cells induced by the YF-17D vaccine are distinct from genuine naïve cells and resemble the Tscm subset

We next focused on the characterization of these newly identified naïve-like CD8+ T cells induced by vaccination with YF-17D. It was critical to evaluate the memory nature of such cells that fall in the conventional naïve gate. Recently, a new memory subset that also coexpresses CCR7 and CD45RA has been reported and termed the stem cell–like memory (Tscm) subset (2325). Tscm have properties of differentiated cells yet retain high stemness and phenotypical proximity to naïve cells—they were also originally described as naïve-like before coining Tscm. We therefore performed a detailed characterization of YF-specific CD8+ T cells with a twofold focus: (i) to address whether they are truly distinct from genuine naïve cells and (ii) to assess similarities to the recently reported Tscm subset.

We performed a thorough screen using a wide panel of markers to study YF-specific CD8+ T cells among the characteristic profiles of the differentiation subsets. This panel included conventional and previously reported differentiation markers [CD27, CD28, CD45RO, interleukin-7 receptor α (IL-7Rα)/CD127, CD62L, CCR5, granzymes A and B, perforin, BTLA, PD1, PDL1, KLRG1, and 2B4] (6, 26), markers that have been used to distinguish the Tscm subsets [CD58, CD95, CXCR3, IL-2Rβ/CD122, CD11a (α chain of LFA-1), CD161, IL-18Rα, and ABC-B1] (23, 25, 27, 28), markers that showed potential relevance in differentiation and migration (CCR4 and CLA) (29), resident memory markers (CD69 and CD103) (30), activation markers (HLA-DR and CD38) (13, 16), and other related markers (IL-2Rα/CD25, IL-15Rα/CD215, and ICOS). This wide marker screen was performed on a representative selection of 16 vaccinees, spanning up to 17 years after vaccination (fig. S2A). As shown above, most A2/NS4b+ cells were either EMRA or naïve-like.

First, to assess whether the naïve-like A2/NS4b+ CD8+ T cells in vaccinees were truly distinct from genuine naïve cells, A2/NS4b+ CD8+ T cells in unvaccinated individuals were studied in more detail. This was technically challenging because of the low frequencies in unvaccinated individuals, which were generally close to the detection threshold of 0.01% of total CD8+ T cells (Fig. 1B). Analysis was nevertheless possible in five unvaccinated donors with sufficient A2/NS4b+ CD8+ T cells (>35 cells detected) and showed that these fell in the naïve gate (Fig. 3, A and B). Pertinently, these also expressed CD28 and CD27 but did not express markers characteristic of the Tscm subset, notably CD58, CD95, and CXCR3 (6, 23, 25, 27) (Fig. 3, C and D). In contrast, naïve-like A2/NS4b+ CD8+ T cells from vaccinees did show positive expression for the Tscm markers CD58, CD95, and CXCR3 (Fig. 3, C and D). Therefore, the naïve-like A2/NS4b+ CD8+ T cells in vaccinees were truly distinct from genuinely naïve A2/NS4b+ CD8+ T cells found in unvaccinated individuals and rather corresponded to the Tscm subset. In agreement, the total Tscm subset was analyzed on the basis of gating the minor population of cells expressing CD58 and CD95 within the total naïve population (Fig. 3E). This showed a prominent population of A2/NS4b+ Tscm cells in vaccinees (average of 0.98% A2/NS4b+ cells within Tscm) as opposed to unvaccinated individuals, where A2/NS4b+ cells were not detectable within Tscm (Fig. 3F).

Fig. 3. Naïve-like A2/NS4b+ CD8+ T cells induced by the YF vaccine are clearly distinct from genuine naïve cells and resemble the Tscm subset.

(A) Representative subset analysis (CCR7 versus CD45RA) of an unvaccinated donor within A2/NS4b+ or total CD8+ T cells. (B) Subset distribution in A2/NS4b+ CD8+ T cells from unvaccinated controls (UN; n = 5) versus vaccinees (VAC; n = 16) showing mean and SE. (C and D) The indicated markers were compared in detectable A2/NS4b+ subsets in unvaccinated controls (n = 3) or vaccinees (n = 16) versus the reference subsets in total CD8+ T cells. Shown are representative off-set overlay histograms (C) and corresponding heatmap-based quantifications of percent positive cells (D) with Bonferroni-adjusted P values following two-way analysis of variance (ANOVA) indicated for unvaccinated controls versus vaccinees. (E) Representative gating of the Tscm subset as CD58+ CD95+ within total naïve (CCR7+ CD45RA+) cells. (F) Frequency of A2/NS4b+ cells within total Tscm in vaccinees or unvaccinated controls (mean and SD); nd, not detectable. (G and H) As in (C) and (D), for 18 markers further assessed in vaccinees (n = 16), with Bonferroni-adjusted P values following two-way ANOVA indicated per marker. Wilcoxon matched-pairs test to compare profiles “overall” are indicated at the bottom (complemented in fig. S2). ns, not significant.

A wide variety of markers were further analyzed in vaccinees, comparing A2/NS4b+ CD8+ T populations to the respective reference populations in total CD8+ T cells. In addition to CD58, CD95, and CXCR3 mentioned above, A2/NS4b+ naïve-like CD8+ T cells distinctly expressed IL-2Rβ, CD11ahigh, KLRG1, granzyme A, IL-18Rα, and CD45RO, in clear contrast to the total naïve population (Fig. 3, G and H). The profile of A2/NS4b+ EMRA cells was overall close to the total EMRA population but displayed particularly elevated levels of CD28, CD27, IL-7Ra, CXCR3, CD62L, BTLA, and IL-18Rα. The latter is in agreement with previous studies where YF vaccination was reported to raise a particular type of EMRA cells that defy the terminally differentiated phenotype generally attributed to the EMRA subset (10, 13, 31, 32). Although most A2/NS4b+ CD8+ T cells were naïve-like or EMRA, there were sufficient events in EM and CM gates of A2/NS4b+ CD8+ T cells in several donors for flow cytometry analyses; the complete data series, including the 31 markers and all detectable subsets, are detailed in fig. S2B. Considering the markers overall, it was only the A2/NS4b+ naïve-like CD8+ T cells that showed a significantly different profile as compared to the corresponding reference population (that is, versus total naïve; Wilcoxon comparisons in Fig. 3H and fig. S2B).

Additionally, to investigate changes over time since vaccination, we selected the eight markers that showed greatest differences between vaccine-induced A2/NS4b+ naïve-like CD8+ T cells and the total naïve: CD58, CD95, CXCR3, KLRG1, CD11ahigh, IL-18Rα, granzyme A, and IL-2Rβ. The time span of vaccination in the selection of 16 donors was ≈17.5 years. Considering the years since vaccination (fig. S2C), the levels of all markers within naïve-like A2/NS4b+ CD8+ T cells had a slight tendency to decrease, statistically significant for IL-2Rβ and CXCR3. The reference total naïve cells also displayed a tendency for lower CXCR3 and higher CD58 with time, pointing toward interdonor variation for these two markers. Nevertheless, for the eight markers, the A2/NS4b+ naïve-like cells expressed distinctly superior levels as compared to the total naïve population, even in the second decade after vaccination.

Naïve-like Tscm are also found in other viral antigen specificities

T cells with various viral specificities and particular T cell differentiation stages have been thoroughly studied in humans and are frequently used as reference populations. These specificities pertain to relatively prevalent chronic or acute viral infections such as HCV, flu, EBV, HIV, and CMV, which, in this order, display increasing differentiation status (6).

To compare YF-specific CD8+ T cells to other antigen specificities, we further analyzed the 16 donors for CD8+ T cells specific for flu, EBV, and CMV (Fig. 4). In addition, we studied Melan-A–specific CD8+ T cells as a population of cells that displays a truly naïve quality in healthy donors not bearing melanoma nor diagnosed with vitiligo (31, 32). All 16 vaccinees had detectable populations specific for Melan-A and flu, 13 were positive for EBV, and 5 were positive for CMV (Fig. 4B). As expected, Melan-A–specific CD8+ T cells were predominantly naïve, whereas flu-, EBV-, and CMV-specific CD8+ T cells showed subset distributions with increasing differentiation (CM→EM→EMRA) (Fig. 4C). All specificities showed cells falling in the conventional naïve gate (Fig. 4, A and C). Further analysis of CD58, CD95, and CXCR3 revealed that, within the naïve gate, Melan-A–specific cells were mostly CD58 CD95 CXCR3low, supporting the notion that these cells are genuinely naïve. In contrast, all viral antigen–specific cells showed naïve-like cells triple positive for CD58, CD95, and CXCR3, albeit at varying degrees. Flu-specific naïve-like CD8+ T cells were more variable depending on the donor, whereas EBV- and CMV-specific cells were particularly high in CD58 and CD95 expression. For CXCR3, YF- and CMV-specific CD8+ T cells showed higher expression (Fig. 4D). Flu-specific naïve-like CD8+ T cells often showed a split population for CD58 and CD95. This may be due to the presence of both genuine naïve and naïve-like cells, presumably depending on the history of flu vaccination and/or infection. The latter was, however, difficult to track in our donors, given the endemic nature of flu worldwide. Our data overall indicated that the naïve-like subset was also detected in other viral specificities and generally differed from the reference naïve population in total CD8+ T cells and from self-antigen–specific naïve CD8+ T cells.

Fig. 4. Comparison of A2/NS4b+ CD8+ T cells to other antigen specificities.

(A) A selection of 16 vaccinees (as in Fig. 3) was analyzed for various antigen-specific populations on the basis of two combinatorial tetramer stainings, including YF NS4b–specific [*, phycoerythrin (PE) in staining I and PE and allophycocyanin (APC) in staining II], EBV-specific (A2/EBV-PE in staining II), flu-specific (A2/Flu-MA APC in staining II), CMV-specific (A2/NLV-PE and A2/NLV-APC in staining I), and the self-antigen Melan-A–specific (A2/ELA APC in staining I). Total CD8+ T cells and antigen-specific populations were analyzed for subset distribution (CCR7 versus CD45RA) and subsets in turn for CD58, CD95, and CXCR3. One representative vaccinee (positive for CMV) is shown. (B) Frequencies of antigen-specific populations in the 16 vaccinees: only 5 donors were positive for CMV-specific CD8+ T cells, whereas all donors showed detectable levels of Melan-A–specific CD8+ T cells (fig. S5). (C) Subset distribution in the various antigen-specific CD8+ T populations across donors (mean and SE). (D) Frequencies of cells positive for CD58, CD95, and CXCR3, across differentiation subsets within the various antigen specificities or total CD8+ T cells, as indicated.

In addition, we addressed whether naïve-like CD8+ T cells were also generated with specificities other than the HLA-A*02–restricted NS4b214–222 (LLWNGPMAV). Previous studies using stimulation of YF vaccinee samples with peptide pools have shown that the CD8+ T cell response can be diverse and highly variable across donors, yet HLA-A*02–positive donors consistently display responses to NS4b214–222 (13, 14, 16). Only a limited number of epitopes other than A2/NS4b, or peptide pools activating vaccinee samples, have been described (13, 14, 16, 20). On the basis of these studies and HLA-binding predictions (BIMAS and SYFPEITHI), we generated a selection of nonamer or decamer epitopes alternative to the A2/NS4b that could potentially yield tetramer stainings in YF vaccinees. After the HLA-typing information available in our cohort, we could test non–HLA-A*02–restricted epitopes (“B35/HPF” in three vaccinees, “B7/RPI” in four vaccinees, and “A24/VYM” in one vaccinee) plus HLA-A*02–restricted epitopes in 25 vaccinees (“A2/AMD,” “A2/VML,” “A2/VCY,” and “A2/GIL”) (fig. S3). Notably, many vaccinees in our cohort did not have the appropriate HLA type to analyze the aforementioned selection of YF epitopes. Of 108 samples tested, this screen resulted in 4 positive samples (tetramer-positive cells >0.01% in total CD8+ T cells), including 2 for A2/VML and 2 for B7/RPI (fig. S3A). Analyses of these tetramer-positive cells subsequently showed that B7/RPI+ CD8+ T cells could also display a naïve-like Tscm phenotype (falling in the conventional naïve gate yet expressing CD58 and CD95) (fig. S3B). Donor LAU 5005 had particularly high frequencies of naïve-like B7/RPI+ CD8+ T cells (90%), whereas the two A2/VML-positive samples showed very few naïve-like cells (0.5 to 2.8%) (fig. S3B). Nonetheless, the screen showed that naïve-like Tscm subsets were also generated against epitopes other than A2/NS4b, in particular the B7/RPI. Together with the evidence on naïve-like CD8+ T cells found in other viral specificities (Fig. 4), this shows that generation of naïve-like Tscm is not exclusive to A2/NS4b+ CD8+ T cells.

Naïve-like YF-specific CD8+ T cells display mRNA profiles that are close to Tscm cells and with high preservation of “naïveness”

To comprehend the global particularities of YF vaccine–induced naïve-like CD8+ T cells, we analyzed their genome-wide mRNA profile and compared it to the conventional subsets in total CD8+ T cells. The latter included naïve, CM, and a pool of EM and EMRA referred to as “effectors.” We also analyzed the Tscm subset (23, 25), gated as the minor population expressing CD58 and CD95 within the naïve gate of total CD8+ T cells. These five populations were isolated from each of eight selected vaccinees, spanning more than 17 years of vaccination history. The sorting strategy and the subset distribution of cells from vaccinees are shown in fig. S4.

Most remarkably, principal components analysis (PCA) clearly showed a gradient of differentiation along the PC1, from naïve to a mixture of Tscm/CM to effectors. Within this gradient, the A2/NS4b+ naïve-like CD8+ T cells clustered near the Tscm/CM samples (Fig. 5, A to C). Moreover, among all differentiated populations, A2/NS4b+ naïve-like CD8+ T cells lied distinctly closest to the naïve subset (Fig. 5, A to C). These observations were further confirmed quantitatively on the basis of pairwise comparisons to calculate intersample distances along PC1. The naïve-like A2/NS4b+ CD8+ T cells were most closely related to the Tscm subset (Fig. 5D, left panel). In addition, the naïve-like A2/NS4b+ CD8+ T cells were closest in comparison to the reference naïve subset (Fig. 5D, right panel). The total Tscm samples were generally found interspersed with the total CM samples (Fig. 5, B and C). Pertinently, by considering the gene sets that define “differentiation” in CD8+ T subsets as well as the differences in “Tscm versus naïve” as described in the report of the Tscm subset (23), we could similarly observe that the A2/NS4b+ naïve-like CD8+ T cells were closely clustered together and displayed a profile between the CM/Tscm and naïve cells (fig. S4, E and F).

Fig. 5. The mRNA profiles of naïve-like A2/NS4b+ CD8+ T cells are similar to Tscm and display highest naïveness.

(A) Whole-genome expression profiles were assessed on the five populations shown color-coded, each purified from n = 8 vaccinees (sorting strategy in fig. S6). (B) Sample distribution in the first two principal components (PC1 versus PC2) of the PCA considering the top 10% most variable genes. (C) Heatmap-based display of the main contributors to PC1 in the PCA considering the top 10% variable genes, showing the 40 samples ordered by their coordinate on PC1. (Note: the corresponding genes are listed in table S2). (D) Distribution of intersample Euclidean distances along PC1 in the PCA considering the top 10% variable genes. Each boxplot summarizes the pairwise distances between samples in naïve-like A2/NS4b+ (left panel) or total naïve (right panel) versus the population indicated on the vertical axis.

Very few genes were significantly differentially expressed between the A2/NS4b+ naïve-like CD8+ T cells and the total naïve, total Tscm, or total CM because of intersample variation in the context of relatively small mRNA differences and in contrast to the numerous significant genes in comparison to the total effectors (table S3). The analysis of mRNA profiles altogether showed that contributions from large numbers of genes globally differentiated the various subsets. These global differences reflected a gradient of differentiation that suggests that the A2/NS4b+ naïve-like CD8+ T cells, among all differentiated subsets, have globally preserved most naïveness.

Long-term persisting YF-specific CD8+ T cells respond to cognate peptide and show homeostatic proliferation with IL-15 with a proliferative advantage for the naïve-like phenotype

As compared to naïve cells that have never been primed, memory CD8+ T cells readily respond to cognate antigen and are capable of homeostatic proliferation in the presence of IL-15 (23, 33). Therefore, we assessed the proliferation capacity of YF-specific CD8+ T cells and its link to the naïve-like phenotype. Because of the limited bioavailability of samples (technical limitation detailed in Discussion), we could not analyze individually isolated subsets of A2/NS4b+ CD8+ T cells from large numbers of donors and stimulate with multiple conditions. We therefore first stimulated peripheral blood mononuclear cells (PBMCs) from vaccinees with the NS4b peptide in the presence of IL-2 or IL-15 or treated with either cytokine alone. PBMCs from three unvaccinated individuals served as controls. In the presence of IL-2 or IL-15, A2/NS4b+ CD8+ T cells from most vaccinees proliferated efficiently in response to cognate NS4b peptide (Fig. 6A). Whereas IL-2 alone generally resulted in no expansion, IL-15 alone induced considerable expansion, demonstrating IL-15–driven homeostatic proliferation characteristic of memory cells (Fig. 6A). Pertinently, the A2/NS4b+ CD8+ T cells from the three unvaccinated individuals did not show proliferation, in agreement with their genuine naïve status (Fig. 6A).

Fig. 6. Functional characteristics of A2/NS4b+ CD8+ T cells: Responses to cognate peptide and homeostatic IL-15 and self-renewal.

(A and B) PBMCs from vaccinees (n = 35) and unvaccinated controls (n = 3) were stimulated with NS4b peptide and cytokines (IL-2 or IL-15) as indicated for 7 days. (A) Expansion based on fold counts to start (mean and SD; Wilcoxon P values paired per vaccinee). (B) Expansion versus the starting frequency of A2/NS4b+ CD8+ T cells with a naïve-like phenotype. Undetectable samples (<15 counts) were excluded. R, Spearman correlations. (C to E) Purified A2/NS4b+ CD8+ T cells, either naïve-like (n = 5) or non-naïve (n = 4), were isolated from vaccinees and expanded with PHA, IL-2, and irradiated feeders for 14 days. (C) Gating strategy. (D) Frequencies of expanded cells with a naïve-like phenotype showing mean, SD, and P value from Mann-Whitney test. (E) Population doubling of naïve-like cells. (F) CD8+ T cells from vaccinees (n = 7) were stimulated with anti-CD3 and anti-CD28 beads for 24 hours. Naïve-like and non-naïve A2/NS4b+ CD8+ T cells were analyzed for Ki67 and HLA-DR expression in comparison to the reference subsets in total CD8+ T (gating shown in fig. S5F). Shown is the mean and SD, with P values from paired Wilcoxon tests.

To assess the link between the naïve-like phenotype and proliferation capacity, we compared the expansion to the starting frequency of naïve-like cells (Fig. 6B). Despite the variability among samples, a higher starting frequency of naïve-like cells significantly correlated with greater expansion in response to peptide and IL-15, and a tendency in this direction was seen with peptide and IL-2 (P = 0.068). A higher starting naïve-like frequency was also significantly associated with greater homeostatic proliferation with IL-15 alone and with better survival with IL-2 alone (Fig. 6B). In addition, the capacity to respond to the peptide was relatively stable long term after vaccination, more than 15 years after vaccination (fig. S5A).

Naïve-like YF-specific CD8+ T cells expanded in vitro generate effectors and show self-renewal

The defining property of stem cells is their capacity to self-renew (34, 35). A fundamental question was therefore to assess the quality of the progeny of naïve-like A2/NS4b+ CD8+ T cells. To this end, naïve-like A2/NS4b+ CD8+ T cells were purified and expanded in vitro, in comparison to non-naïve counterparts. Because we started with very low numbers of cells, we stimulated with phytohemagglutinin (PHA), IL-2, and irradiated feeders, a protocol that is used for T cell cloning because it provides robust polyclonal stimulation. By day 14, sufficient cells were yielded to analyze the progeny generated from either naïve-like (n = 5) or non-naïve (n = 4) A2/NS4b+ CD8+ T cells. Both progenies displayed CCR7 effectors (Fig. 6C). The non-naïve progeny showed a slightly increased population doubling and higher proportion of EM as opposed to EMRA in the progeny of naïve-like (fig. S5, B to D). The effector progeny from naïve-like showed loss of naïve markers CD28, CD27, and IL-7Rα, demonstrating typical differentiation and not merely loss of CCR7 (fig. S5, B and E). Critically, the purified naïve-like A2/NS4b+ CD8+ T cells, and not the purified non-naïve, generated a small fraction of expanded cells that retained the naïve-like phenotype (Fig. 6, C and D). This was not due to the lack of stimulation of a fraction of starting naïve-like cells because the numbers of naïve-like cells were slightly, yet effectively, increased as compared to the input (Fig. 6E). Therefore, the naïve-like A2/NS4b+ CD8+ T cells demonstrated self-renewal, in support of their stemness potential.

In addition, we assessed naïve-like or non-naïve A2/NS4b+ CD8+ T cells for short-term functional readouts, in comparison to the reference subsets in total CD8+ T cells, using polyclonal anti-CD3 and anti-CD28 stimulation. First, we analyzed the proliferative marker Ki67 and the activation marker HLA-DR as early as possible, that is, at 24 hours, to avoid substantial changes in subset composition, because it naturally occurs within the first days of T cell activation (26). As previously shown, using isolated subsets from nonhuman primates (24), the total Tscm subset showed highest Ki67 expression among all subsets, and only naïve cells did not up-regulate HLA-DR (Fig. 6F and fig. S5F). Intriguingly, within A2/NS4b+ CD8+ T cells, the naïve-like population behaved similarly to the reference total naïve, whereas non-naïve A2/NS4b+ cells were comparable to total EM and EMRA subsets (Fig. 6F). Secretion of interferon-γ (IFN-γ), tumor necrosis factor–α (TNF-α), and IL-2 was also assessed 4 hours after anti-CD3 and anti-CD28 stimulation. However, the medium control showed substantial background secretion in A2/NS4b+ populations, which could be due to activation by the tetramer staining (unavoidable for the experiment). This technical limitation precluded conclusions when comparing A2/NS4b+ to the reference subsets in total CD8+ T cells (not stained with tetramer) (fig. S6). Overall, these short-term functional assays showed that the behavior of naïve-like A2/NS4b+ CD8+ T cells was close to the reference total naïve subset, in agreement with their high degree of naïveness observed in the whole-transcriptome analyses described in Fig. 5.


Here, we report two major findings that provide new insights into the biology of long-term persisting human memory CD8+ T cells. First, the YF-17D vaccine induced a population of naïve-like memory CD8+ T cells that resembles the recently described Tscm subset and preserves a high degree of naïveness. This was detected in a controlled setting, that is, in an antigen-specific setting with a known time of antigen priming in humans. Second, the A2/NS4b+ naïve-like population appeared stable over at least 25 years of vaccination history. This setting provided proof for the longevity and stability of CD8+ Tscm in humans, being in the range of decades.

Several studies have previously analyzed the frequencies and differentiation status of YF vaccine–induced CD8+ T cells early after vaccination. These include consideration of several epitopes up to 90 days after vaccination (16), analyses on various differentiation markers mostly within the first 90 days and with limited numbers of donors up to 27 months (13), as well as describing the TCR repertoire in two donors up to 54 months after vaccination (20). Overall, these studies found a predominant EMRA population that is highly polyfunctional, which is unexpected for EMRA cells, because they are normally considered to be terminally differentiated (10). Notably, similar observations were made in CD8+ T cells induced by the smallpox vaccine, a second live attenuated vaccine that triggers very effective CD8+ T cells in humans (10, 15).

Relevant to our findings, the flow cytometry data presented in these previous reports also show evidence for a CD45RA+ CCR7+ double positive population in YF-specific CD8+ T cells either within the first 90 days (n = up to 15) (13, 16) or in limited numbers of donors up to 46 months (n = 1 to 3) after vaccination (13, 20). This would correspond to the naïve-like A2/NS4b+ CD8+ T cells that we hereby report. However, this detection of naïve-like memory CD8+ T cells was neglected amidst the focus on the effector populations raised by the vaccine. Our study analyzed A2/NS4b+ CD8+ T cells both beyond 5 years after vaccination and in a considerably large cohort (n > 20). We thoroughly characterized this naïve-like subset raised by the YF vaccine, and discovered that it resembles Tscm and becomes particularly predominant in the long term (range of decades).

The phenotypic observation that A2/NS4b+ CD8+ T cells from vaccinees falling in the conventional naïve gate can express varying levels of CD45RA and CCR7 remains intriguing. The A2/NS4b+ naïvehigh population only increased by 1.5-fold in frequency between detectable unvaccinated controls and vaccinees. Yet, there was a clear change in phenotype, from genuine naïve (CD58 CD95) in unvaccinated controls to Tscm (CD58+ CD95+) in vaccinees. This particular phenotypic shift and the threshold of tetramer staining used (0.01% and >20 events; that is, considerably high) support that the cells analyzed were not merely background. Background events would have shown a distribution of subsets close to the total population of cells. Regarding the various populations characterized, our study is primarily descriptive of the phenotypic observations that can be made in the study of YF-specific CD8+ T cells in vaccinees and in relation to the memory subsets that have been described. In Table 1, we summarize the populations that are referred to in this study, including conventional subsets in total CD8+ T cells, the recently reported Tscm subset, and the A2/NS4b+ populations that we characterized in unvaccinated individuals and vaccinees. The fact that Tscm populations (whether the reference total Tscm or the naïve-like A2/NS4b+ CD8+ T cells) are found within the conventional naïve gate raises concerns on our current definition of memory subsets and naïve cells, popularly defined by CCR7 and CD45RA (6).

Table 1. Summary of the CD8+ T cell populations referred to in this study.
View this table:

Out of all memory subsets, Tscm are thought to have retained the most “stemness,” to be closest to the naïve cells, and thus represent a subset of utmost interest for long-term immunity (4, 24, 36). In the framework of adoptive T cell therapy, the least differentiated memory cells are thought to be the best source to generate a progeny of potent antitumor CD8+ T cells and self-renew for long-term therapeutic efficacy (4, 24, 3739). Recently, serial adoptive transfers of single CD62L+ CM cells in mice demonstrated better stemness as opposed to more differentiated EM cells by reconstituting immunocompetence against infection with Listeria monocytogenes (40). In humans and nonhuman primates, the quantity and quality of the progeny of the various subsets have been assessed by proliferation assays in vitro. It has been found that, in particular, naïve cells and Tscm can self-renew and reconstitute the other differentiated populations (including Tscm when starting with naïve) (23, 24). Naïve cells represent the mature T cell type that is ready to encounter antigen with the highest stemness and pluripotency, potentially producing all types of effector and memory progenies (36, 37, 41). Yet, naïve cells remain to be optimally primed to respond to their cognate antigen, and adoptive T cell transfers rely on the isolation and amplification of T cells, which inevitably induces differentiation. There is thus a quest to optimally prime and minimally differentiate T cells to generate memory cells that preserve highest proximity to naïve and thus immunotherapeutic potential.

Intriguingly, from our genome-wide mRNA profiling data, there is no apparent set of genes or a signature that minimally and significantly defines the A2/NS4b+ naïve-like CD8+ T cells. Very few genes were significantly differentially expressed between naïve, CM, Tscm, or the A2/NA4b+ naïve-like CD8+ T cells. Rather, the analyses across samples highlighted a differentiation gradient defined globally by contributions from several genes, transitioning from naïve to A2/SN4b+ naïve-like to a mixture of Tscm/CM to effectors. This supports the notion that CD8+ T cells undergo differentiation programs upon priming, within which memory cells may retain varying proximity to the naïve cells. In the course of an immune response, memory precursors that are rescued at the earliest differentiation stage would generate memory populations that preserve higher stemness and multipotency (36, 42). Thus, the proximity to the naïve cells, that is, the naïveness of a memory cell, may be regarded as the parameter that reflects early rescue and may therefore serve as a measure of memory quality. In this regard, naïveness is beyond the property of stemness and reflects globally the preservation of naïve properties. Overall, we found that naïve-like A2/NS4b+ CD8+ T cells (i) phenotypically resembled the Tscm subset by expressing markers such as CD58, CD95, CXCR3, IL-2Rβ, and high LFA-1 (23); (ii) efficiently responded to cognate peptide and IL-15–driven homeostatic proliferation, in contrast to genuine naïve cells; and (iii) showed self-renewal capacity. Yet, A2/NS4b+ naïve-like CD8+ T cells were not identical to the reference total Tscm and also shared certain behaviors with the reference naïve subset, as observed in the genome-wide transcriptome profile and short-term functional characteristics (early Ki67 and HLA-DR up-regulation upon activation). In the latter analyses, the total Tscm were very close to total CM. Together, it is possible that there is substantial variation in the quality of various antigen-specific Tscm cells (and memory subsets in general), with different histories of priming and antigen exposure that may affect phenotype and functionality. The high degree of naïveness of YF-specific CD8+ T cells may reflect the absence of chronic or recent antigen exposure, as opposed to the total Tscm subset that may contain variably and possibly relatively recently activated cells. This high naïveness also suggests that naïve-like A2/NS4b+ CD8+ T cells are a particularly high-quality population within the Tscm subset.

As addressed in Results, the Tscm subset described by Gattinoni et al. has also been detected in the context of cancer or viral specificities, namely, Melan-A, CMV, and flu (23). We also detected naïve-like CD8+ T cells in CMV-, EBV-, and flu-specific CD8+ T cells, albeit with variation in the expression of Tscm markers such as CD58, CXCR3, and CD95. Yet, the key point in our study is that the exposure to flu, CMV, and EBV viruses is difficult to track and supervise experimentally, given the chronic and/or prevalent quality of these viral infections in the human population. This contrasts to the vaccination setting, such as with YF-17D, where the parameter “time since antigen priming” is known. The latter is the central feature to demonstrate the longevity of the vaccine-induced naïve-like memory subset that we hereby report.

Only a few studies so far have provided proof for persistence of CD8+ T cells of known antigen specificity in the long term, in the range of years to decades. For instance, these concern detection of CD8+ and CD4+ T cell responses to polyomavirus in vaccinated individuals over two decades (43) and the detection of measles-specific CD8+ T cells up to 34 years after vaccination (44), but no phenotypic characterization was performed. A recently published study has gathered evidence that Tscm cells may survive at least 12 years in humans, based on the follow-up of T cell clones in cohorts of patients treated against inherited adenosine deaminase immunodeficiency with genetically corrected hematopoietic stem cells or peripheral blood leukocytes (45). Within our cohort of vaccinees, the A2/NS4b+ naïve-like CD8+ T cells were detectable for at least 25 years. The in-depth analyses showed that A2/NS4b+ naïve-like CD8+ T cells maintained higher levels of the eight markers (CD58, CD95, CXCR3, KLRG1, CD11ahigh, IL-18Rα, granzyme A, and IL-2Rβ) that prominently distinguished them from total naïve, over the 17 years after vaccination studied. Moreover, we did not observe a particular correlation between expansion in vitro and vaccination history. These points altogether suggest that the qualities of the naïve-like A2/NS4b+ CD8+ T cells become apparent early after vaccination (the minimum vaccination history studied was 0.27 years) and remain relatively stable over the years to decades after vaccination.

Our study was confronted with two major limitations. First, the study of antigen-specific CD8+ T cells is technically difficult because of their low frequencies in blood, which impose the requirement of large quantities of cells per analysis. For instance, isolation of 1000 naïve-like A2/NS4b+ CD8+ T cells (such as samples for microarrays) required 100 × 106 to 500 × 106 PBMCs, depending on the donor. A standard blood donation yields 400 × 106 to 1000 × 106 PBMCs, and a leukapheresis-based donation (one cycle) yields 2000 × 106 to 8000 × 106 PBMCs. For an experiment using, for example, 4 samples of 1000 purified A2/NS4b+ CD8+ T cells, this would easily require the use and processing of a full blood donation.

A second limitation was the fact that the study is not longitudinal, and we need to consider the confounding variable of interdonor variability. Only longitudinal studies may comprehend the development and stability of this A2/NS4b+ CD8+ Tscm population during YF vaccination, including the clonotypic composition, expansion and interrelations among subsets, and the reaction upon administration of a recall vaccine. The diversity of YF epitopes across vaccinees and the high prevalence A2/NS4b+ CD8+ T cells in HLA-A*02 positive donors (fig. S3) (13, 14) raise questions on the immunodominance of YF epitopes, whether host-related characteristics (such as HLA allele) influence the outcome of the immune response and long-term memory elicited by the YF vaccine, and whether there are links to side effects or the efficacy of protection to infection. In view of the results of our screen on CD8+ T cell YF specificities other than A2/NS4b, very large number of donors (beyond the size of our cohort) would be necessary to accumulate positive samples for the various HLA restrictions and epitopes. In agreement with previous studies (13, 14), our tetramer screen illustrated the high prevalence of the A2/NS4b specificity (LLWNGPMAV) and thus its practicality as a model antigen in humans. Nonetheless, our study has the practical advantage of avoiding the awaiting of decades to study cohorts of vaccinees with very long vaccination times. Together, our results support that YF vaccination is particularly suitable to investigate innate and specific immune mechanisms responsible for the generation and maintenance of self-renewing and very long-lasting memory cells in humans.


Study design, population, and ethics statement

The study was open to healthy volunteers aged 18 to 65 years having received the YF-17D vaccine (Stamaril, Sanofi Pasteur) with no limit on minimum or maximum vaccination history. Vaccinees were grouped according to years since vaccination: <1 year (I), 1 to 5 years (II), 5 to 10 years (III), and >10 years (IV). The targeted sample size was n = 10 per group, and this was met for all groups except group I. No outliers were excluded from analyses. The study protocol was approved by the Human Research Ethics Committee of the Canton de Vaud (protocol 329/12) with healthy volunteers participating under written informed consent. Eligible volunteers donated blood according to the standards of the Blood Transfusion Center (Service Vaudois de Transfusion Sanguine), and the leukocyte-rich fraction (buffy coat) was recovered for the study. For further in-depth analyses requiring larger number of cells, some of the volunteers (among those aged up to 50 years) were also invited for a leukapheresis-based donation. Samples from unvaccinated blood donors were also obtained from the Blood Transfusion Center.

Peripheral blood collection and preparation

PBMCs were obtained from leukocyte-rich blood samples after density gradient fractionation using Lymphoprep. All samples were immediately cryopreserved in RPMI 1640 supplemented with 40% fetal calf serum (FCS) and 10% dimethyl sulfoxide awaiting experimental use.

Flow cytometry

The tetramers and antibodies used are detailed in table S4. For the analysis of antigen-specific populations, CD8+ T cells were first enriched from cryopreserved samples using the human CD8+ T cell enrichment kit from StemCell Technologies (negative selection, per manufacturer’s instructions). Stainings were performed using phosphate-buffered saline with 5 mM EDTA, 0.2% bovine serum albumin, and 0.2% sodium azide [fluorescence-activated cell sorting (FACS) buffer]. Tetramer stainings were performed for 40 min at room temperature. Surface antibody staining was then performed, followed by staining with the fixable dead cell marker Vivid Aqua (Molecular Probes, Invitrogen), each step at 4°C for 30 min. Cells were fixed overnight in 1% formaldehyde (supplemented with 2% glucose and 5 mM sodium azide). Intracellular staining was performed last, using antibodies in FACS buffer with 0.1% saponin for 30 min at 4°C. For Ki67 intracellular/nuclear staining, the fixation and permeabilization was performed using the reagents from the Foxp3 staining kit from eBioscience. Samples were acquired using a Gallios flow cytometer (Beckman Coulter, three-laser configuration) with antibody panels limited to 10 colors. The data were processed with FlowJo (Tree Star Inc., v9.5.2). Samples with antigen-specific populations below 0.01% tetramer-positive cells in total CD8+ T cells were considered negative. Populations consisting of less than 20 events were not considered eligible for further analysis (for example, no analysis of markers on non-naïve populations of A2/NS4b+ CD8+ T cells in unvaccinated individuals). Isolation of cells for microarray analyses (described further below) was performed using BD FACSAria I flow cytometer.

Whole-transcriptome microarrays: mRNA sample preparation and analysis

CD8+ T cells were enriched from cryopreserved PBMCs using the negative enrichment kit from StemCell. For the flow cytometry–based purification, samples were stained with A2/NS4b-PE tetramer for 40 min, followed by surface staining for 30 min (CD8+, CD45RA, CCR7, CD16, CD58, CD95; referenced above) and live/dead staining for 30 min. For mRNA analysis on whole transcriptome, 1000 cells were isolated from each of the five different populations using the strategy shown in fig. S6, per donor (n = 8 donors, D1 to D8 ordered with increasing vaccination history), by flow cytometry–based purification directly into 13.5 μl of SuperAmp lysis buffer (Miltenyi Biotec). From CD8+ T enrichment, all manipulations including cell sorting were carried at 4°C. After sorting, RNA lysates were incubated for 10 min at 45°C and immediately frozen at −20°C. Purifications were performed on two separate days (sort I: D1, D3, D4, and D6; sort II: D2, D5, D7, and D7).

Thereafter, frozen RNA samples were shipped, processed, and analyzed together by the Genomic Services of Miltenyi Biotec according to their SuperAmp technology and by hybridization onto Agilent Whole Human Genome Microarray 8x60K (v2; one color, Cy3). The integrity of complementary DNA was checked via the Agilent 2100 Bioanalyzer. Fluorescence signals of the hybridized Agilent microarrays were detected using Agilent Microarray Scanner System (Agilent Technologies). Raw output data were generated via the Agilent Feature Extraction software. The raw data were background-corrected and quantile-normalized using the backgroundCorrect and normalizeBetweenArrays functions in the limma R package (version 3.20.8, R version 3.1.0). Control probes and probes whose normalized expression value did not exceed the background level in any of the samples were filtered out, and the expression of replicated probes was averaged, leaving 41,923 probes that were used for further analysis. Exploratory principal components analysis was performed with the prcomp function in R. Because the differences between the cell populations were supposed to affect only a relatively small subset of the genes, an independent filtering procedure was applied to exclude the probes with the lowest variance across all samples, keeping only the 10% most variable probes. All probes were standardized by subtracting the mean values and dividing by the SD across all samples before applying the PCA. We extracted the top 50 probes with positive and negative loadings, respectively, on the PC1 and constructed a heatmap of their expression levels across the samples (Fig. 5C). To further illustrate the similarities between samples from the different cell populations, we calculated the Euclidean distance between all pairs of samples along the PC1. The distributions of pairwise distances between samples from each pair of distinct subgroups, as well as pairwise distances between samples within each subgroup, are summarized with boxplots in Fig. 5D (“intersample distances”). Differential expression analysis was performed with the limma R package (version 3.20.8). For each pair of cell populations, we performed a gene-wise moderated paired t test, accounting for interdonor differences. The nominal P values were adjusted for multiple comparisons using the Benjamini-Hochberg procedure (46). Hierarchical clustering was performed using Euclidean distance and complete linkage.

Proliferation assays and in vitro stimulations

The complete medium used was RPMI 1640, complemented with 10% heat-inactivated FCS, 1% nonessential amino acids (Gibco), 1% l-glutamine (Gibco), Hepes (10 mM), and penicillin/streptomycin (10,000 U/ml; Gibco). For the proliferation assays using NS4b peptide and cytokines, PBMCs were thawed and cultured at 1.0 × 106 to 1.5 × 106 cells per ml per 2 cm2 in flat-bottom plates, using NS4b214–222 peptide at 1 μg/ml (LLWNGPMAV), human IL-2 at 100 U/ml (Proleukin, Roche Pharma), and recombinant human IL-15 at 20 U/ml (10 ng/ml, PeproTech). For the quantification of proliferation, the “input (counts at start)” of A2/NS4b+ CD8+ T cells was calculated on the basis of the number of PBMCs seeded and the percentage of A2/NS4b+ CD8+ T cells determined by flow cytometry. The “counts at day 7” were quantified on the basis of absolute numbers of A2/NS4b+ CD8+ T cells upon full acquisition of samples by flow cytometry. The “expansion” was calculated by dividing the counts at day 7 by the input.

For the expansion of purified naïve-like or non-naïve YF-specific CD8+ T cells, A2/NS4b+ CD8+ T cells were isolated by flow cytometry cell sorting (same strategy as the samples used for whole-transcriptome analyses). Cells (200 to 4000) (depending on the donor’s respective frequencies of A2/NS4b+ and subsequent naïve-like/non-naïve phenotype) were isolated and stimulated in U-bottom 96-well plates with PHA (1 μg/ml), IL-2 (150 U/ml), and 106/ml “feeder” cells. These feeders were prepared by mixing freshly prepared PBMCs from two independent blood donors, irradiated with 30 Gy. The medium composition was the same as above, except human serum was used instead of FCS. Medium was renewed, and cells were split periodically until sufficient cells had expanded to allow analysis (cell counting to assess population doubling and flow cytometry), that is, at day 14. For the experiments using anti-CD3 and anti-CD28 stimulation to assess short-term function, PBMCs were first thawed and rested overnight in complete medium (in the absence of cytokines) at a density of 0.75 × 106 to 1 × 106 cells per cm2 per 0.5 ml. Next day, CD8+ T cells were enriched using the negative selection kit from StemCell Technologies (per manufacturer’s instructions). CD8+ T cells were stained with A2/NS4b APC tetramer before and after the stimulation with beads to minimize the loss of detection of tetramer-positive cells because of the TCR internalization that inherently occurs upon T cell activation; notably, this phenomenon affects the quality of the tetramer staining after T cell stimulation and renders the population of tetramer-positive cells less distinct as opposed to ex vivo analyses (fig. S5, A and F: activated, versus Fig. 1A: ex vivo). Tetramer-stained CD8+ T cells were plated at 2.5 × 106 cells per 2.5 ml per 4 cm2 (12-well plate) per condition and stimulated with anti-CD3 and anti-CD28 beads (Miltenyi Biotec) at 1:1 ratio. For the intracellular cytokine readout at 4 hours, brefeldin A was added at 10 μg/ml (no cytokines). For the readout at 24 hours, the medium was supplemented with IL-2 (100 U/ml).

Quantifications and statistical analysis

Quantifications were made on the basis of the FlowJo, Microsoft Excel, GraphPad Prism, and SPICE softwares. Statistical values were obtained using the analyses and tests (including normality tests) as detailed in the figure legends; where indicated, *P < 0.05; **P < 0.01; ***P < 0.001. For statistical comparison of pie charts generated using SPICE, the built-in test in SPICE software (v5.3) was used (using 10,000 permutations) (47).


Fig. S1. Influence of interdonor and age-related variability and its normalization for the study of A2/NS4b+ CD8+ T cell subsets.

Fig. S2. Data complementary to Fig. 3.

Fig. S3. Analyses of CD8+ T cells specific for YF epitopes alternative to the HLA-A*02–restricted NS4b.

Fig. S4. Data complementary to Fig. 5.

Fig. S5. Data complementary to Fig. 6.

Fig. S6. Cytokine production by NS4b-specific CD8+ T cell subsets in comparison to the reference subsets in total CD8+ T cells.

Table S1. Summary of the cohort of YF-17D vaccines.

Table S2. Main genes contributing to PC1.

Table S3. Differentially expressed genes comparing naïve-like A2/NS4b+ CD8+ T cells to the other four subsets listed (Excel).

Table S4. Tetramers and antibodies used for flow cytometry analyses.


Acknowledgments: We thank all blood donors for their participation, the personnel of the Blood Transfusion Center in Epalinges (C. Thibaud for coordinating the leukaphereses and J. Conne for the laboratory processing), D. Labes (Flow Cytometry Facility of the Ludwig Cancer Center), and all the members of our laboratories. Funding: Ludwig Cancer Research Center, the Cancer Vaccine Collaborative, the Cancer Research Institute (all New York), the Swiss Cancer League (02836-08-2011), the Swiss National Science Foundation (310030_135553, 320030_152856, and CRSII3_141879), and a grant from the Swiss State Secretariat for Education, Research and Innovation to the SIB for service and infrastructure resources. Author contributions: S.A.F.M., M.A., and D.E.S. conceived and designed the experiments. S.A.F.M., L.C., S.A.M., S.W., and D.E.S. elaborated the clinical protocol. S.A.F.M., P.O.G., M.A., and N.M. performed the experiments. S.A.F.M., C.S., P.O.G., M.A., M.D., and D.E.S. analyzed the data (statistical analyses: S.A.F.M., C.S., P.O.G., and M.A.). S.A.F.M., C.S., P.O.G., M.A., N.R., and D.E.S. wrote the paper. All authors revised and accepted the final version of the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The microarray data of this study have been deposited in the Gene Expression Omnibus (GEO) public repository with the GEO accession number GSE65804.
View Abstract

Navigate This Article