Expression of IL‐7Rα and KLRG1 defines functionally distinct CD8+ T‐cell populations in humans

Abstract During acute viral infections in mice, IL‐7Rα and KLRG1 together are used to distinguish the short‐lived effector cells (SLEC; IL‐7RαloKLRGhi) from the precursors of persisting memory cells (MPEC; IL‐7RαhiKLRG1lo). We here show that these markers can be used to define distinct subsets in the circulation and lymph nodes during the acute phase and in “steady state” in humans. In contrast to the T cells in the circulation, T cells derived from lymph nodes hardly contain any KLRG1‐expressing cells. The four populations defined by IL‐7Rα and KLRG1 differ markedly in transcription factor, granzyme and chemokine receptor expression. When studying renal transplant recipients experiencing a primary hCMV and EBV infection, we also found that after viral control, during latency, Ki‐67‐negative SLEC can be found in the peripheral blood in considerable numbers. Thus, combined analyses of IL‐7Rα and KLRG1 expression on human herpes virus‐specific CD8+ T cells can be used to separate functionally distinct subsets in humans. As a noncycling IL‐7RαloKLRG1hi population is abundant in healthy humans, we conclude that this combination of markers not only defines short‐lived effector cells during the acute response but also stable effector cells that are formed and remain present during latent herpes infections.


Introduction
Following viral challenge, CD8 + T cells with different phenotypes and functions can be found in experimentally infected mice. Acute In contrast IL-7Rα hi KLRG1 lo cells have an increased propensity to persist in the memory phase, and were thus named memoryprecursor effector cells (MPECs) [1][2][3]. Further, early during the acute phase, recently primed T cells displaying an IL-7Rα lo KLRG1 lo early effector cell (EEC) phenotype are found from which both MPECs and SLECs can form [4]. This latter observation accords well with complementary data from different approaches demonstrating that effector and memory cells can be generated from a single antigen-activated precursor [5][6][7]. The fate of virus-specific CD8 + T cells is determined by transcriptional regulation. Transcription factors, like T-bet and eomesodermin (Eomes), are crucial regulators of effector and memory formation [8]. Despite the fact that T-bet and Eomes display a high degree of structural homology, as well as functional redundancy, murine T-bet was found to be more relevant to the generation of effector cells during the acute phase of infection, whereas Eomes is important to the persistence of memory cells and the generation of secondary responses [9,10].
Importantly, mice and men differ substantially with regard to both lifespan and the range of pathogens with which they have coevolved. These factors possibly also have affected T-cell memory formation and memory homeostasis, especially when it concerns responses to persistent viruses such as EBV and human CMV (hCMV) that establish lifelong latent infection after the acute phase. Moreover, given the large time span in which such viruses have coevolved with humans and the distinct modes of infection exhibited by each virus, the antiviral T-cell responses may have adapted in order to provide a defence specifically tailored to best control these distinct infections. Supporting evidence for this notion is, among others, provided by our observation that HOBIT, a BLIMP-1 homolog regulating effector functions in murine NKT cells [11], is specifically expressed in human cytolytic CD8 + CD45RA + CD27¯T cells, a population selectively induced in hCMV-infected people [12].
Combinations of different cell surface markers have been used to define total and virus-specific human CD8 + T-cell subsets. Combinations of CD45RA, CCR7, CD28, and CD27 are commonly being employed to define naïve (CD45RA + CCR7 + CD28 + CD27 + ), central memory (CD45RA¯CCR7 + CD28 + CD27 + ), effector memory (CD45RA¯CCR7¯CD28 + CD27 + ), and effector-type cells (CD45RA + CCR7¯CD28¯CD27¯, also referred to as T EM RA, for a review see [13]). However, since this marker combination cannot be used for the definition of murine T cells subsets, a comparison of subset development between experimentally infected mice and naturally infected humans is cumbersome. For this reason, we here determined IL-7Rα/KLRG1 phenotypes of human CD8 + T cells targeting different persisting viruses in both latently infected healthy individuals and primary hCMV-and EBV-infected kidney allograft recipients. In addition, we analyzed if different IL-7Rα/KLRG1 expression patterns associate with different functional profiles of cells.
CD8 + T cells with all possible combinations of IL-7Rα/KLRG1 expression were detectable in blood and lymph nodes of healthy adults, of which the vast majority was Ki-67-negative (noncycling). Furthermore, IL-7Rα/KLRG1 expression patterns define different functional CD8 + T cells subtypes with unique transcriptomes. IL-7Rα-negative cells were generally expressing markers that befit a cytotoxic functional profile. Among these, the relatively small IL-7Rα lo KLRG1 lo population uniquely harbors Ki-67 + (cycling) cells in healthy asymptomatic individuals. Further, in all study cohorts, substantial numbers of CD8 + T cells were detected with an IL-7Rα lo KLRG1 hi "SLEC" phenotype, expressing high amounts of T-bet and granzyme B, that persist in a Ki-67-negative "steady state" long after viral control. Importantly, these cells are enriched in the hCMV-specific population and are largely excluded from the lymph nodes. Thus, in humans, a stable pool of quiescent CD8 + "vigilant" effector-type cells, phenotypically corresponding to murine short-lived effector cells, is abundant in the circulating compartment. Such cells were seen significantly less often among human EBV-specific memory populations, and by monitoring T-cell responses during primary infection we found that these distinctions per herpes virus-targeted response are installed early during the acute phase.

IL-7Rα and KLRG1 expression patterns identify cytotoxic and noncytotoxic memory cells
We first enumerated the four IL-7Rα/KLRG1-defined subsets within the total circulating CD8 + T-cell pool of EBV and hCMVseropositive healthy adults (after exclusion of the naïve population). We specifically selected these individuals since both hCMV and EBV significantly impact the makeup of the circulating CD8 + T-cell pool (i.e. smaller naïve and central-memory populations and large T EM RA populations) and because the majority of the adults is latently infected with these viruses (hCMV ß50-60% and EBV 80-90%) [14][15][16]. In these individuals that had no clinical and/or laboratory signs of acute viral infection, we detected substantial numbers of KLRG1-expressing cells. These were either displaying an IL-7Rα hi double-positive effector cell (DPEC) phenotype or an IL-7Rα lo (in mice defined as SLEC) phenotype, in the circulating nonnaïve pool (Fig. 1A).
A considerable proportion of the cells contained within the KLRG1 + subsets, especially those with an IL-7Rα lo phenotype, express effector molecules like T-bet and granzymes ( Fig. 1B and Supporting Information Figs. 2 and 3). Therefore, these cells closely approximate CD28-negative effector type (T EM RA) cells [14]. IL-7Rα hi KLRG1 lo MPEC cells instead displayed markers that fit a classic noncytotoxic memory phenotype, with a considerable amount expressing the lymphoid tissue-homing receptor CCR7, the costimulatory receptors CD28 and CD27, while showing a low abundance of T-bet, CD57 and granzymes (Fig. 1B, C, Supporting Information Figs. S2, S3, and S4) [17][18][19][20]. Indeed, these cells therefore closely approximate the central-memory and early-differentiated effector-memory subsets [14].
A fourth IL-7Rα lo KLRG1 lo subset (previously coined EEC in mice [21]), was also detected in the circulation of these healthy individuals, although at much lower frequencies than the other  subsets (Fig. 1A). In accordance with their name, these cells indeed displayed all the traits associated with cytotoxic effector function and were in this respect more similar to the KLRG1 + subsets than to the IL-7Rα hi KLRG1 lo MPECs ( Fig. 1B and Supporting Information Figs. 2, S3, and S4). Nevertheless, these cells also shared some features with MPECs with regard to a high expression of CCR7 and CCR4 (Fig. 1B, C, Supporting Information Figs. 2, 3, and 4). Interestingly, when compared to the other subsets, the IL-7Rα lo KLRG1 lo cells uniquely contained a subpopulation of Ki-67 + (actively cycling) cells (Fig. 1B, C, Supporting Information Figs. 2, 3, and 4). Finally, these cells also highly expressed HLA-DR and CD38 (Supporting Information Figs. 3 and 4), further indicative of an activated state.

Transcriptomic analysis reveals distinct functional specializations of IL-7Rα/KLRG1-defined subsets
We determined the transcriptional profile in peripheral blood directly ex vivo of the four populations that could be distinguished by the surface expression of IL-7Rα and KLRG1. Multidimensional scaling ( Fig. 2A) revealed that the expression profiles of all populations segregated according to phenotype rather than by donor. While transcriptional variation among populations sorted based on the expression of at least one marker was minimal, variation among those defined as the double negative population was substantial. Hierarchical clustering (Fig. 2B) demonstrated clustering of samples according to phenotype. Interestingly KLRG1 expression appeared to be the largest denominator as IL-7Rα hi KLRG1 hi clustered together with IL-7Rα lo KLRG1 hi rather than IL-7Rα hi KLRG1 lo populations. Likewise, IL-7Rα lo KLRG1 lo clustered with IL-7Rα hi KLRG1 lo populations despite their difference in cytotoxic effector traits. Hierarchical clustering revealed seven clusters of differentially expressed genes among the primary T-cell populations (Fig. 2C). Subsequently, we performed a gene set enrichment aimed at linking gene clusters with GO terms (Fig. 2D). With the exception of cluster 3, all clusters revealed to harbor unique functional characteristics. As such, cluster 1, highly expressed by IL-7Rα lo KLRG1 hi cells was enriched for genes mediating leukocyte activation and cluster 4, highly expressed by IL-7Rα lo KLRG1 lo , was enriched for genes participating in cell cycle. Indeed, befitting the previously observed granzyme B and Ki-67 + (cycling) protein profiles observed in these subsets as described above (Fig. 1).

The Ki-67 + IL-7Rα lo KLRG1 lo effector population concerns activated cells with an effector phenotype
Given the unique presence of actively cycling cells found in immunocompetent asymptomatic individuals in the IL-7Rα lo KLRG1 lo population, we were curious to what extent the Ki-67 + EECs differed from the Ki-67-negative EECs.
Only few proteins were expressed differentially by these populations in an apparently biologically relevant manner, being: HLA-DR, a class II MHC protein and CD151, a tetraspanin, both markers of T-cell activation; CD28 and ICOS, both costimulatory receptors and CCR5, a C-C chemokine receptor, all molecules that are expressed to a much greater extent by the Ki-67 + cells (Fig. 3A, B and Supporting Information Fig. 4). As such, with the exception of these molecules, Ki-67-expressing and Ki-67-negative EECs are highly similar. Importantly, both subpopulations contain a high degree of granzyme B, but also express T-bet, CX 3 CR1, perforin, and granulysin ( Fig. 3A and Supporting Information Fig. 4, see also Fig. 1B and C), all molecules associated with cytotoxic effector and effector-memory T cells [17,18,22].

IL-7Rα hi KLRG1 lo cells with noncytotoxic memory traits are enriched in lymph nodes
Previously, Sallusto et al. described how CD8 + T cells can be distinguished according to CCR7-expression, establishing the central-memory (i.e. LN-homing) and effector-memory (peripherally active) subset dichotomy [19]. Since we found a clear predominance of CCR7 expression by the IL-7Rα hi KLRG1 lo MPEC population ( Fig. 1B), we were interested in the enumerations of the IL-7Rα/KLRG1-defined subsets in human LNs compared to the circulation. For this purpose, we isolated CD8 + T-cell populations from peri-aortal LN samples acquired from renal transplant recipients (RTRs) just before implantation of the transplant.
Importantly, the composition of the circulating pool of these patients was comparable to that of the healthy individuals ( Fig. 1A and Fig. 4A). It was also apparent that IL-7Rα hi KLRG1 lo MPECs predominated in LNs as compared to DPECs and SLECs that predominate in the circulation (Fig. 4A). EECs were also detected in low frequencies in the LNs, again uniquely holding a Ki-67 expressing subpopulation (Fig. 4A, B, and C). KLRG1-expressing cells were much less often detected in LNs.
In lymph nodes, IL-7Rα/KLRG1 expression patterns also separated functionally distinct sets of cells ( Fig. 4B-H and Supporting Information Fig. 6). However, we found differences between lymph nodes and peripheral blood. Notably, the expression of T-bet was lower in all subsets retrieved from the lymph nodes, including those of the IL-7Rα hi KLRG1 hi DPECs and IL-7Rα lo KLRG1 hi SLECS, which was paralleled by a lower expression of granzyme B (Fig. 4D and E). In contrast, CD28 and granzyme K expression was higher by all the non-MPEC LN subsets when compared to the circulatory subsets ( Fig. 4F and G).

The IL-7Rα/KLRG1 division defines functionally distinct pools of herpes-virus specific CD8 + T cells
The finding that healthy individuals have high numbers of circulating CD8 + T cells with an IL-7Rα lo KLRG1 hi SLEC phenotype could mean that they have encountered viruses to which an immune response was recently mounted, or rather, that these populations have arisen as a consequence of persistent infections, such as herpes viruses. To address these distinct options, we used   Flow cytometric analysis of IL-7Rα lo KLRG1 lo nonnaïve CD8 + T cells from the PBMCs from ten buffy coats. Box and whiskers min to max plot of the fold change of the geomean fluorescence intensity (GMFI). The gray area represents changes less than twofold. Only when Ki-67-expressing IL-7Rα lo KLRG1 lo CD8 + T cells had a more than twofold higher or lower mean expression (outside the gray area) expression was considered different. (B) MFI of Ki67, CD151, CCR5, CD28, ICOS, and HLA-DR on ki-67-negative (-) and ki-67-expressing (+) IL-7Rα lo KLRG1 lo nonnaïve CD8 + T cells. Representative histograms and graphs of the individual fold changes of the GMFI of each of the four KLRG1/IL-7Rα-defined subsets in nonnaïve CD8 + T cells of ten buffy coats compared to the naïve (CD27 + CD45RA + ) CD8 + T cells of the same sample can be found in Supporting Information Figure 4. Data shown (mean ± SD) are from one independent experiments with n = 10 donors per experiment. Statistical analysis used: Wilcoxon matched-pairs signed rank test.
In RTRs and healthy individuals we found that both circulatory hCMV-and EBV BLZF1/BMLF-specific cells predominantly displayed an IL-7Rα hi KLRG1 hi DPECor IL-7Rα lo KLRG1 hi SLEC phenotype (Fig. 5A, Supporting Information Figs. 7 (HI) and 8 (RTR)). For all hCMV and lytic EBV (BMLF/BZLF) epitopes, a low percentage of IL-7Rα hi KLRG1 lo MPEC phenotype cells was found. In stark contrast, analysis of influenza-specific CD8 + T cells, as an example of CD8 + T cells directed against viruses that are effectively cleared by the immune system, revealed that virtually all cells expressed IL-7Rα, including a significant IL-7Rα hi KLRG1 lo MPEC population (Fig. 5A, Supporting Information Figs. 7 (HI) and 8 (RTR)). When further phenotyping these virus-specific memory populations, we noticed that the expression of molecules associated with a distinct functional profile (Fig. 5B, Supporting Information Figs. 7 and 8). It must be emphasized that virtually all virus-specific CD8 + T cells were Ki-67-negative (Fig. 5B, Supporting Information Figs. 7 and 8). Differences on the virus-specific level between circulating T cells and T cells in the lymph nodes,  with regard to T-bet and granzyme B expression, were as expected, given the findings presented above (Figs. 4 and 5B and Supporting Information Fig. 8). However, each herpes virus-specific population displayed a similar IL-7Rα/KLRG1 subset distribution in the circulation as in the lymph nodes. Furthermore, similar to what was seen on nonantigen-specific level, also on the virus-specific level, decreased levels of T-bet and granzyme B are present in the LN, regardless of IL-7Rα/KLRG1 phenotype (Fig. 5C).

Signatures of hCMV-and EBV-specific CD8 + T cells arise during primary infection
The different traits of hCMV and EBV-specific CD8 + T cells, as well as the functional properties of the cells with an SLEC phenotype found during viral latency, urged us to analyse whether virus-specific T cells found early in the primary antiviral response also differ with respect to their IL-7Rα/KLRG1 expression pattern  and functional traits. To this end, we followed three RTRs experiencing primary hCMV and/or EBV infection in the first year of transplantation. These patients were included during the preprophylaxis treatment era. Consequently, hCMV viral loads were detectable early after transplantation. Since these patients developed no significant hCMV disease, no antiviral treatment was administered during follow-up. Immunological control, as defined by the absence of detectable viral loads and the appearance of virus-specific IgG antibodies, was achieved for hCMV within the three months after transplantation and ßthree months later for EBV (Fig. 6A, Supporting Information Figs. 9A, and 10A).
Human CMV and EBV BZLF1/BMLF population dynamics were very much alike with regard to both bigger T-cell population sizes, as well as a predominance of KLRG1 expressing cells, when compared to EBV EBNA-specific populations. In contrast EBV EBNAspecific cells were much more often expressing IL-7Rα ( Fig. 6A  of specificity, were present initially in the response but generally became a minor fraction after longer follow-up (Fig. 6B, Supporting Information Figs. 7B and 8B). The hCMV-specific response was characterized from early on by an abundance of IL-7Rα lo KLRG1 hi SLEC cells (Fig. 6B, Supporting Information Figs. 9B and 10B). This was markedly different from the EBV-specific populations that contained more IL-7Rα-expressing cells that were not always coexpressing KLRG1 (Fig. 6B and C, Supporting Information Fig. 9B and C).
When looking at the functional traits of the IL-7Rα/KLRG1 defined subsets within the herpes virus-specific pools, it became evident that the functional separation found during latency appears to be installed very early in the response. Specifically, early hCMV responses were dominated by CD8 + T cells with an EEC or SLEC phenotype abundantly expressing T-bet and granzyme B (Fig. 6B, 7A, B, Supporting Information Figs. 9-15), whereas EBV-specific pools hardly contained high T-bet and granzyme B expressing cells (Fig. 7A, B, Supporting Information Figs. [11][12][13]. Curiously, hCMV-specific MPECs also expressed considerable amounts of T-bet and granzyme B, which is unexpected considering the generally infrequent expression of these molecules by MPECs detected during viral latency (Fig. 7A, B, Supporting Information Figs. [11][12][13]. Indeed, expression of T-bet and granzyme B by MPECs drops markedly as time progresses and patients enter the latency phase. A similar finding concerned Eomes and granzyme K, which was expressed at really highly level already at the earliest time point by EBV-specific populations, regardless of the IL-7Rα/KLRG1 phenotype (Fig. 7C, D, Supporting Information Figs. 11-13).

Discussion
In conclusion, we found that a subset definition based on IL-7Rα/KLRG1 expression can be used to separate functional sets of (virus-specific) CD8 + T cells in humans during viral latency. From a functional point of view, the definition derived from the mouse data appears to be largely in accord with the findings in humans. IL-7Rα hi KLRG1 lo (in mice defined as MPEC) concern noncytotoxic cells with memory functions, having high levels of CCR7 and low expression of T-bet and granzyme B. Remarkably, the expression of Eomes, a factor strongly implicated in memory cell generation in mice, is relatively low in these cells and in these anatomical compartments. In contrast, IL-7Rα lo KLRG1 hi cells (in mice defined as SLEC) found during latency strongly resemble cytotoxic effector-type cells, abundantly expressing both T-bet and granzyme B. It is however unlikely that human IL-7Rα lo KLRG1 hi cells found during latency are short-lived as initially was described for their murine counterparts. None of the latently infected subjects analysed had clinical or laboratory signs of viral replication. Further, in accordance with the absence of antigen-driven proliferative T-cell responses, no Ki-67 expression was found in the IL-7Rα lo KLRG1 hi pool. Finally, studies analysing the turn-over of T-cell subsets in humans have shown that the population of cells with constitutive effector functions actually has a long lifespan [23,24]. Therefore, we argue that to coin the term SLEC to human CD8 + T cells with an IL-7Rα lo KLRG1 hi phenotype would be a misnomer. Rather, the expression of KLRG1 on human CD8 + T cells appears to be associated with the acquisition of immediate effector functions during acute and latent viral infection. In fact, also in the mouse antigen-specific IL-7Rα lo KLRG1 hi CD8 + T cells can be detected long-time after infection [25,26].
IL-7Rα lo KLRG1 lo cells (in mice defined as EEC) are similar to the IL-7Rα lo KLRG1 hi cells with the big difference that this population holds a subset of activated and actively cycling cells. Nevertheless, by their expression of CCR7 and CCR4, they also resemble the noncytotoxic IL-7Rα hi KLRG1 lo MPECs. We therefore hypothesize that these cells concern recently activated naïve or MPEC cells that have lost expression of IL-7Rα while having acquired expression of typical effector molecules. Future investigations should focus on their functional role in maintaining immunological control over latent viral infections. The finding of substantial populations of IL-7Rα hi KLRG1 hi DPECs in healthy individuals is also interesting. These cells express large amounts of granzyme K and granzyme B but are still susceptible to IL-7R-mediated signalling, a feature that is normally associated with maintenance of noncytotoxic memory populations in mice [27]. Also here, further investigations are indicated into their functional role, especially in controlling EBV infection as the EBV-specific CD8 + T cells harbour particularly large amounts of these cells.
Additionally, analyses of IL-7Rα/KLRG1 expression allows for a clear separation of functionally different human CD8+ T-cell subsets also in LNs. In line with the previous findings of Sallusto et al. [19], CCR7-expressing cells, mainly displaying an IL-7Rα hi KLRG1 lo phenotype, predominate in this anatomical compartment. However, the distinctions between noncytotoxic (memory) and cytotoxic (effector) traits are less apparent in the LNs, since typical effector traits are expressed significantly less often in this environment than in the circulation, even by the KLRG1expressing subsets. These findings also hold true on the antigen (virus)-specific level and the analyses we performed here reveal a number of additional specific features of virus-specific cells. Depending on the anatomical location and on the anti-viral specificity, the distribution of the IL-7Rα/KLRG1-defined subsets differs markedly. We regard it likely that the differences between virusspecific populations with regard to IL-7Rα/KLRG1 expression and phenotype and function in a general sense, are the result of a complex combination of signals (e.g. as transduced via the T-cell receptor, costimulatory and coinhibitory receptors, and cytokine receptor interactions) that is delivered to the T cells in a certain spatiotemporal context that is distinct for each viral infection. This may not be so difficult to understand given the fact that each virus exhibits its own specific mode of infection, primarily targeting different organ systems/cell types, were some viruses are cleared after infection while others establish latency. Even immunomodulatory viral aspects, such as the hCMV-mediated downregulation of MHC class I molecules may be involved in this process [28]. Also on the virus-specific level, KLRG1-expressing cells are largely excluded from lymph nodes and therefore may function primarily at nonlymphoid compartments. As our studies in the primary infection showed that, already early in the response, distinct viruses elicit differential T-cell responses and analyses of the T cells during acute infection might provide early qualitative information on the T-cell response that provides protection during the latency phase. This insight is of particular importance to the creation of effective vaccination strategies in the future where one would aim to establish a functionally distinct memory T-cell population that provides protection against a pathogen on the long term.

Subjects
In order to study the dynamics and phenotype of acute phase hCMV-and EBV-specific CD8 + T cells compared to those present during the latency phase, hCMV-specific CD8 + T cells from 2 hCMV-and EBV-seronegative recipients of a hCMV-and EBV-seropositive kidney transplant were analyzed longitudinally (Table 1, patient 1-2). To eliminate the influence of a primary EBV infection on the hCMV-specific CD8 + T cells we also studied the hCMV-specific CD8 + T cells from one kidney transplant recipient who was hCMV-seronegative prior to transplantation and received a hCMV-seropositive kidney (Table 1, patient 3). The treatment of each patient as well as other relevant information can be found in Table 1.
PBMCs were isolated prior to transplantation and lymph node mononuclear cells (LNMC) were isolated during kidney transplantation. All patients were treated with quadruple immunosuppression, consisting of CD25mAb induction therapy, Prednisolone, calcineurin-inhibitor, and mycophenolic acid. Except for CD25mAb, immunosuppressive treatment was started after transplantation. At the time the lymph node was gathered the first dose of CD25mAb was already administered. However, we have demonstrated that CD25mAb (basiliximab, Novartis Pharma, Basel, Switzerland) was not detectable in LNMC and that ex vivo CD25 expression on LNMC could be blocked with CD25mAb [29].

Ethics statement
The medical ethics committee of the Academic Medical Center, Amsterdam, approved the study and all subjects gave written informed consent.

Isolation of mononuclear cells from peripheral blood and lymph nodes
PBMCs were isolated using standard density gradient centrifugation. Lymph nodes were collected from kidney transplant recipients during living donor kidney transplantation. Briefly, LNMCs were isolated from surgical residual material of the recipient, gathered during the implantation of the transplanted kidney. Before anastomizing the arteria and vena renalis, the iliac artery and vein are dissected free. The residual tissue removed in this procedure, to obtain adequate vascular exposure, often contains lymph nodes. Directly after extraction, the gathered lymph nodes were chopped in small pieces. A cell suspension was obtained by grinding the material through a flow-through chamber. PBMCs and LNMCs were subsequently cryopreserved until the day of analysis.

Virological analysis
Quantitative PCR for hCMV was performed in EDTA (ethylenediaminetetraacetic acid) whole blood samples, as described (2).
To determine CMV serostatus, anti-CMV immunoglobulinG (IgG) was measured in serum using the AxSYM microparticleenzyme immunoassay (Abbott Laboratories, Abbott Park, IL, USA) according to the manufacturer's instructions. Measurements were calibrated relative to a standard serum. EBV serostatus was determined by qualitative measurement of specific IgG against the viral capsid antigen (VCA) and against nuclear antigen of EBV using respectively the anti-EBV VCA IgG ELISA and the anti-EBV nuclear antigen of EBV IgG ELISA (Biotest, Dreieich, Germany). All tests were performed following the instructions of the manufacturers.

RNA-seq data analysis
CD8 + T cells were isolated from buffycoat derived PBMC (from six different donors) with CD8 + T-cell beads (Miltenyi Biotec B.V.). Subsequently they were stained with a-CD3 eF450 (cloneSK7), a-CD127 (IL7Rα) PE-Cy7 (eBioRDR5) (eBioscience), a-CD8 APC (RPA-T8), a-CD45RA PE (HI100) (BD Biosciences), a-CD27 BV510 (O323) (Biolegend), a-KLRG1 AF488 (kind gift of HP Pircher) and live/dead fixable near IR (Invitrogen) and sorted into all four KLRG1/IL-7Rα-defined CD8 + CD3 + subsets on a FACS ARIA III sorter (BD) (Supporting Information Fig. 3). RNA was isolated with the Qiagen RNeasy Plus Micro kit (Qiagen Benelux B.V., the Netherlands) according to manufacturer's instructions and subsequently sequenced with Illumina NextSeq 500 platform, single-end protocol with a read length of 76 nt. The reads were aligned to hg38 reference genome, with STAR (v2.4.0j) aligner software using defaults settings. The reads that aligned to (only exonic parts of) each gene were quantified using feature Counts (v1.4.3-p1) software and Ensembl annotation. After visual inspection of MDS plots and (log2 counts) boxplots of all the samples, in combination of Bioanalyzer RIN scores, we decided to remove six samples. Log2 transformed CPM (counts per million + 0.5) were made in edgeR (v3.14.0), genes with less than 1 CPM in 3 or more samples were removed from further analysis. Subsequently, limma (v3.28.21) trend was used to test for significantly expressed genes between the groups, making use of a block (aka paired) design that takes donor information into account. An FDRadjusted p-value < 0.05 was considered significant. Hierarchical clustering was done using only genes that were significant in at least one of six limma comparisons (in total 2890 genes), using z-scores of log2 transformed RPKM (reads per kilobase per million reads + 0.5) values. Using Ward.D clustering method with Euclidean distance in gplots (v3.0.1), the genes were divided in seven groups. Enrichment analysis of GO terms was done for each group, using clusterProfiler (v3.0.5). Analyses with bioconductor packages were all done in R (v3.3.2). Data are available from the GEO under accession number GSE113098.

Statistical analysis
The Wilcoxon matched-pairs signed rank test was used for analysis of differences between paired samples. A p-value of < 0.05 was considered statistically significant.