Turnover of Murine Cytomegalovirus–Expanded CD8⁺ T Cells Is Similar to That of Memory Phenotype T Cells and Independent of the Magnitude of the Response

Memory CD8⁺ T cells are a crucial component of the adaptive immune response to viruses. Ag-specific memory CD8⁺ T cells convey immune protection against viral infections that may last for long periods of time, sometimes even life-long. There is ample evidence that memory T cells isolated from the blood and lymph nodes are relatively short-lived. Their lifespan is much shorter than that of naive T (T_N) cells, and far shorter than the long-term immune protection they convey (1–10). Memory T cell populations are heterogeneous, both phenotypically and functionally. They consist of phenotypically defined subpopulations, such as central memory T (T_CM) and effector memory T (T_EM) cells, and of subsets that differ in terms of exposure to their cognate Ag. In vivo deuterium labeling studies have shown that different subsets of memory T cells can have different kinetics. CD4⁺ T_EM cells were shown to have shorter lifespans than did T_CM cells (10), and yellow fever virus (YFV)–specific memory T cells generated by vaccination, which can persist for years, were found to have longer lifespans than did bulk memory-phenotype cells (11).

CMV infection is a persistent, chronic infection, which, in contrast to YFV vaccination, results in continual Ag presentation. CMV is under constant immune surveillance, and it triggers ongoing CD8⁺ T cell responses that provide effective viral control for long periods of time. A hallmark of the CD8⁺ T cell response to CMV infection is the steady maintenance or accumulation of large populations of virus-specific effector CD8⁺ T cells over time, a phenomenon termed memory inflation (12). Expanded CD8⁺ T cell populations specific for unique CMV epitopes can become extraordinarily large, composing up to 20% of the total memory T (T_M) CD8⁺ cell pool (13–18). These large CMV-specific T cell responses turned out to be maintained dynamically, through continuous production of relatively short-lived cells (19). Nevertheless, a face-to-face comparison of the in vivo dynamics of CMV-specific and bulk memory-phenotype CD8⁺ T cells suggested that inflating CMV-specific memory CD8⁺ T cell responses are composed of cells that are longer-lived than other memory cells. This has led to the hypothesis that T cell inflation arises from the accumulation of relatively long-lived CMV-specific memory T cells (20).

In this study, we addressed this hypothesis using in vivo deuterium labeling and mathematical modeling, the state-of-the-art techniques to quantify lymphocyte turnover, in the setting of murine CMV (MCMV) infection, a relevant experimental model to study memory T cell inflation (14). In contrast to the postulated hypothesis that CMV-specific T cells may have extended lifespans, we found no significant difference in the expected lifespans of MCMV-specific CD8⁺ T cells and bulk memory-phenotype CD8⁺ T cells. Using recombinant viruses inducing inflationary CD8⁺ T cell responses of different magnitudes, we found that MCMV-specific T cells composing small and large inflationary T cell responses had very similar turnover rates.

129S2/SvPas Crl (129/Sv) mice were purchased from Charles River Laboratories (Sulzfeld, Germany). Mice were housed and handled in accordance with good animal practice as defined by the Federation of Laboratory Animal Science Associations and the national animal welfare body Die Gesellschaft für Versuchstierkunde (Society of Laboratory Animal Science). All animal experiments were approved by the responsible state office (Lower Saxony State Office of Consumer Protection and Food Safety, Germany; permit number 33.19-42502-04-15/1836 and by the Animal Experiments Committee of Utrecht University, IVD Utrecht, the Netherlands; DEC AVD115002016714).

Bacterial artificial chromosome–derived recombinant viruses MCMV^ie2SL and MCMV^M45SL were generated and propagated as described previously (18), and the recombinant virus MCMV^ie2KNL was generated and propagated as described in Borkner et al. (21).

Female 8-wk-old mice were infected with 2 × 10⁵ PFU of MCMV^ie2SL (n = 40), MCMV^M45SL (n = 37), or MCMV^ie2KNL (n = 41) and housed in specific pathogen-free conditions throughout the experiment. Noninfected sex- and age-matched mice were used as controls (n = 10).

One hundred twenty days after MCMV infection, mice received 8% deuterated water (99.8% ²H₂O, Cambridge Isotope Laboratories) in their drinking water for 28 days. At day 4, mice were given an i.p. boost injection of 15 ml/kg ²H₂O in PBS. To determine deuterium enrichment in the body water, EDTA plasma was collected during the up- and down-labeling phase and was frozen and stored at −80°C until analysis.

Spleen, thymus, and blood were isolated at different time points during and after label administration. Blood was collected in EDTA tubes. Single-cell suspensions from blood, spleen, and thymus were obtained as described previously (22).

To determine the fraction of SSIEFARL (SL) and KCSRNRQYL (KNL) epitope-specific T cells, single-cell suspensions from blood and spleen were stained with allophycocyanin-conjugated SL-K^b or KNL-D^b tetramers for 15 min at 4°C. Samples were further stained for 30 min at 4°C with anti–CD3-allophycocyanin-eFluor 780 (clone 17A2; eBioscience), anti–CD3e-FITC (clone 145-2C11; BD Phamingen), anti-CD3-V500 (clone 500A2; BD), anti–CD4-Pacific Blue (clone GK1.5; BioLegend), anti–CD4-Brilliant Violet 650 (clone GK1.5; BD Horizon), anti–CD4-allophycocyanin-H7 (GK1.5; BD), anti–CD8a-PerCP/Cy5.5 (clone 53-6.7; BioLegend), anti–CD8a-Brillian Violet 786 (clone 53-6.7; BD), anti–CD44-Alexa Fluor 700 (clone IM7; BioLegend), anti–CD44-Alexa Fluor 450 (IM7; eBioscience), anti–CD62L-eVolve 605 (clone MEL-14; eBioscience), anti–CD62L-FITC (MEL-14; eBioscience), and anti–CD127-PE/Cy7 (clone A7R34; BioLegend) mAbs. For intracellular staining, cells were subsequently fixed for 20 min at room temperature with 100 µl of fixation/permeabilization buffer of the FoxP3 Transcription Factor Staining Set (eBioscience), permeabilized for 15 min at room temperature in 100 µl of permeabilization buffer (eBioscience), and stained with Ki-67-PE (clone 16A8; BioLegend) in 100 µl of permeabilization buffer for 30 min at room temperature. Cells were analyzed on an LSRFortessa flow cytometer using FACSDiva software (BD Biosciences) and FlowJo software (version 9.8.3). For infected mice, tetramer-positive (Tet⁺) (CD3⁺CD8⁺Tet⁺) T cells and tetramer-negative T_N (CD3⁺CD8⁺Tet⁻CD62L⁺CD44⁻), T_CM (CD3⁺CD8⁺Tet⁻CD62L⁺CD44⁺), and T_EM (CD3⁺CD8⁺Tet⁻CD62L⁻CD44⁺) cells were sorted from spleen on a FACSAria II SORP (BD Biosciences), FACSAria III (BD Biosciences), or MoFlo XDP cell sorter (Supplemental Fig. 1A). For uninfected mice, T_N, T_CM, and T_EM cells were sorted from spleen on a FACSAria III (BD Biosciences).

Genomic DNA was isolated from total thymocytes and from sorted T cell subsets from MCMV-infected and uninfected mice according to the manufacturer’s instructions using the NucleoSpin Blood QuickPure kit (Macherey-Nagel), and stored at −20°C until further analysis.

Deuterium enrichment in plasma and DNA was analyzed by gas chromatography/mass spectrometry (GC/MS) using an Agilent 5973/6890 GC/MS system (Agilent Technologies). Plasma was derivatized to acetylene (C₂H₂) as previously described (5). The derivative was injected into the GC/MS system equipped with a PoraPLOT Q 25 × 0.32 column (Varian) and measured in selected ion monitoring mode monitoring ions m/z 26 (M+0) and m/z 27 (M+1). From the ratio of ions, plasma deuterium enrichment was calculated by calibration against standard samples of known enrichment. DNA obtained from sorted lymphocytes and granulocytes was hydrolyzed to deoxyribonucleotides and derivatized to pentafluoro tri-acetate (5). The derivative was injected into the GC/MS system equipped with a DB-17 column (Agilent Technologies) and measured in selected ion monitoring mode monitoring ions m/z 435 (M+0) and m/z 436 (M+1). From the ratio of ions, we calculated DNA deuterium enrichment by calibration against deoxyadenosine standards of known enrichment, as previously described (23).

We deduced the dynamics of tetramer-negative T_N, tetramer-negative T_CM, tetramer-negative T_EM, and tetramer-negative T_M (calculated as the weighted average of T_CM and T_EM cells) CD8⁺ cells, and of Tet⁺ CD8⁺ T cells from the deuterium labeling data using previously published mathematical models (3, 24). In brief, to monitor the changing levels of deuterium in body water during the course of the experiment, a simple label enrichment/decay curve was fitted to the deuterium enrichment in plasma (3):

where S(t) is the fraction of deuterium in plasma at time t (in days), f is the predicted plateau value of deuterium enrichment in the plasma, δ is the turnover rate of body water per day, S₀ is the plasma enrichment level attained due to the i.p. ²H₂O boost, and ²H₂O administration was stopped at τ = 28 days. The level of label incorporation in the different cell subsets was described by

Here, l(t) is the fraction of labeled DNA in the cell subset, p is the average (per capita) production rate of the cells, d* is the average (per capita) rate at which labeled cells are lost [which need not be equal to the average loss rate of cells in the population (24)], and c is an amplification factor, which accounts for the multiple hydrogen atoms that can be replaced by deuterium (see Ref. 3). To estimate the value of c, we first fitted Eq. 2 for a kinetically homogeneous population (p = d*) (5) to the level of deuterium enrichment in the DNA of total thymocytes, as they are known to have a high turnover rate (5). The resulting estimated value of c was subsequently fixed to estimate the turnover rates of T_N, T_CM, T_EM, and T_M cells and Tet⁺ CD8⁺ T cells. The best fits to the plasma and thymocyte data are shown in Supplemental Fig. 2 (see Supplemental Table I for the estimated parameter values). When modeling the deuterium enrichment levels of T_N cells, a time delay (Δ) was introduced between T cell production in the thymus and the appearance of labeled DNA in T_N cells in the spleen, based on previous observations (5). This was done by incorporating a delayed labeling curve of the deuterium enrichment in plasma [i.e., S(t − Δ)] in Eq. 2 when fitting the dynamics of T_N cells.

To estimate the rate of change in cell numbers, r, in the T_N, T_CM, T_EM, T_M, and Tet⁺ T cell populations, we used a simple exponential growth/decay model, $dN(t)dt = rN(t) = (p−d)N(t)$⁠, which we fitted to the cell number data from the start of the experiment [i.e., 120 days postinfection (dpi)] until 550 days later (see Table II and Supplemental Fig. 3). Results were very similar when r was estimated based on cell numbers during the first 140 days of the experiment. Based on the resulting value of r and the estimated value of p from Eq. 2, the cellular loss rates, p − r, were calculated (see Table II). The expected lifespans of cells can be calculated as the inverse of their average loss rates, that is, as 1/d.

Best fits to the labeling and cell number data were determined by minimizing the sum of squared residuals using the R function modCost() of the FME package (25). The fractions of labeled DNA, x, were transformed using the function arcsin(sqrt(x)) before the fitting procedure. Fitting the cell number data yielded estimates for the initial cell number at the start of the experiment at 120 dpi, N(0), and the exponential growth rate, r. The 95% confidence intervals (CIs) on the estimated parameters for both labeling and cell number data were determined using a bootstrap method where the data points were resampled 500 times. Fitting the exponential growth/decay model to these 500 data samples yielded 500 bootstrap trajectories. The 95% CI trajectories for the cell numbers were calculated by taking the 95% CI of these 500 bootstraps at each time point.

Statistical analyses were performed using GraphPad Prism. Comparisons between two and more groups were performed using Kruskal–Wallis and Dunn’s multiple comparison tests. A p-value <0.05 was considered significant.

To study the kinetics of MCMV-specific and inflationary CD8⁺ T cell responses during the memory phase of MCMV infection, we made use of 129/Sv mice to benefit from the well-defined H^2b MHC class I haplotype and the well-characterized arrays of epitopes associated with it, while circumventing a protective dominant role of NK cells in controlling the infection (18, 26). The avidity of the viral epitope together with its context of gene expression define the kinetics and magnitude of the cognate inflationary CD8⁺ T cell response (18). We used three well-characterized MCMV mutants expressing low-avidity or high-avidity epitopes in different genetic contexts: the recombinant MCMV^ie2SL, which expresses the high-avidity HSV-1 epitope SL inserted at the C terminus of the immediate-early 2 (ie2) gene (18); the MCMV^ie2KNL mutant expressing the low-avidity epitope KNL also inserted at the C terminus of the ie2 gene (21); and the MCMV^m45SL recombinant, which expresses the same epitope as MCMV^ie2SL inserted in a different genetic context, the early m45 gene (18).

SL-specific and KNL-specific CD8⁺ T cells were analyzed 120 dpi using tetramer staining to determine the magnitude and the phenotype of the inflating T cell response. As previously described (21), the SL and the KNL epitopes expressed within the ie2 gene induced larger inflationary T cell responses than does the SL epitope expressed within the m45 gene (Fig. 1A). MCMV^ie2SL induced the inflationary response of the highest magnitude; a median of 13% of total CD8⁺ T cells were SL Tet⁺. The size of the specific response to MCMV^ie2KNL was ∼9% of the CD8⁺ T cell pool, and significantly larger than the response to MCMV^m45SL, which remained below 5%. Yet, even the latter recombinant virus induced a clearly detectable Tet⁺ CD8⁺ T cell population (Fig. 1A).

Ag-specific CD8⁺ T cells composing an inflationary response typically present an effector phenotype and maintain their effector function (14). Accordingly, the vast majority (>80%) of Tet⁺ CD8⁺ T cells composing large inflationary responses (MCMV^ie2SL and MCMV^ie2KNL) had a T_EM cell phenotype (CD44⁺CD62L^–). In contrast, only 60% of the Tet⁺ CD8⁺ T cells induced by MCMV^m45SL presented a T_EM cell phenotype, whereas the remaining 40% expressed T_CM cell markers (CD44⁺CD62L⁺) (Fig. 1B, Supplemental Fig. 1B, 1C). Less than 2% of the Tet⁺ T cells had a T_N cell phenotype (CD44⁻CD62L⁺).

To study the dynamics of CD8⁺ T cells in the stable phase of chronic MCMV infection, we first determined cell proliferation by measuring Ki-67 expression. The fraction of Ki-67^hi cells within T_N, T_CM, and T_EM CD8⁺ cells was not significantly different between uninfected mice and chronically infected mice for all three viruses (Fig. 2). In line with previous reports (27), we found that the percentage of Ki-67^hi cells was the lowest within T_N cells (median over all groups of 0.4%), intermediate within T_CM cells (median of 4.5%), and the highest within T_EM cells (median of 12%) (Fig. 2B). Approximately 5% of the Tet⁺ CD8⁺ T cells were Ki-67^hi. For each viral infection, the total fraction of Ki-67^hi cells within the Tet⁺ cells was not significantly different from that of memory phenotype T cells, and it was in fact between the Ki-67 expression levels of T_EM and T_CM cells (Fig. 2B). Based on Ki-67 expression, we thus found no indication that inflated MCMV-specific CD8⁺ T cells have different proliferation rates than do bulk memory phenotype CD8⁺ T cells. Because Ki-67 only provides a snapshot marker of T cell proliferation, we next studied the average production and loss rates of the cells using in vivo deuterium labeling.

The in vivo kinetics of memory CD8⁺ T cells have primarily been studied in bulk memory phenotype T cells (5). Here, we quantified the dynamics of T_N, T_CM, T_EM and T_M CD8⁺ cell subsets in chronically infected and uninfected 129/Sv mice. Mice received ²H₂O for 4 wk and were sacrificed at different time points during the labeling and the de-labeling period. We subsequently used previously published mathematical models (5) (see Materials and Methods) to quantify the average production and loss rates of T_N, T_CM, T_EM, and T_M CD8⁺ cells based on their deuterium labeling data.

Deuterium enrichment curves of T_N, T_CM, T_EM, and T_M cells were very similar in MCMV^ie2SL-, MCMV^ie2KNL-, and MCMV^m45SL-infected animals (Fig. 3). In line with this, the best fits of the model to the data yielded similar estimates for the average production rates p and the average loss rates of labeled cells d* within the T_N, T_CM, T_EM, and T_M CD8⁺ cell populations for the three different viruses (Fig. 4A, Table I). The estimated average production rate p of T_N cells during chronic MCMV infection was 0.0090 per day for MCMV^ie2SL, 0.0095 per day for MCMV^ie2KNL, and 0.0077 per day for MCMV^m45SL, suggesting that T_N cells turn over relatively little. In contrast, T_CM cells turned over significantly, with average production rates of 0.0132 per day for MCMV^ie2SL, 0.0151 per day for MCMV^ie2KNL, and 0.0139 per day for MCMV^m45SL. The average production rates of T_EM cells were consistently the highest, with 0.0204 per day for MCMV^ie2SL, 0.0229 per day for MCMV^ie2KNL, and 0.0205 per day for MCMV^m45SL (Table I).

To investigate how the size of an MCMV-specific memory T cell response is related to its turnover (20), we quantified the turnover rates of Tet⁺ CD8⁺ T cells composing large (MCMV^ie2SL) and intermediate (MCMV^ie2KNL) inflationary responses, as well as a low inflationary response (MCMV^m45SL). Despite the clear differences in the height of the Ag-specific responses induced by these three viruses, the corresponding deuterium enrichment curves of Tet⁺ CD8⁺ T cells were very similar (Fig. 3). The best fits of the model to the deuterium-enrichment data of SL Tet⁺ CD8⁺ T cells in MCMV^ie2SL- and MCMV^m45SL-infected mice (Fig. 3) confirmed that, despite the 3-fold difference in the size of these inflationary responses (Fig. 1A) and their different T_CM and T_EM compositions (Fig. 1B), their average production rates p and loss rates of labeled cells d* were not significantly different (Fig. 4A, Table I). We estimated that SL Tet⁺ CD8⁺ T cells had an average production rate of 0.0169 per day in MCMV^ie2SL-infected mice and 0.0161 per day in MCMV^m45SL-infected mice. For KNL-specific T cells, which comprised the intermediate inflationary response, we found an average production rate of 0.0195 per day (Table I). Thus, there were no substantial differences in the average turnover rates of Tet⁺ CD8⁺ T cells composing large, intermediate, and low inflationary responses.

Finally, to investigate the hypothesis that accumulation of inflationary responses in MCMV is due to accumulation of long-lived cells, we compared the average turnover rates of MCMV-specific (Tet⁺) CD8⁺ T cells to those of tetramer-negative T_M CD8⁺ cells (see Materials and Methods). When comparing the best fits of the individual datasets, we found no statistical indication that T_M cells and Tet⁺ T cells had different production rates p or loss rates of labeled cells d* (Fig. 4A, Table I).

Since the average loss rate d* of labeled cells may not be representative of the cell population as a whole (24), we used additional information on absolute cell numbers to compare the average loss rates d of Tet⁺ and T_M cells. Although these absolute cell numbers are notoriously noisy, we estimated a slight increase in cell numbers in the T_CM, T_EM, T_M, and Tet⁺ CD8⁺ T cell populations (Table II, Supplemental Fig. 3). The average production rates p may thus not be equal to the average loss rates d of cells. Even when accounting for this rate of increase, r, in cell numbers, we found very similar average loss rates d (where d = p − r, see Materials and Methods) of T_M and Tet⁺ cells in MCMV^ie2SL-, MCMV^ie2KNL-, and MCMV^m45SL-infected mice (Fig. 4B, Table II). We thus found no evidence for the previously proposed idea that CMV-specific T cells are longer-lived than other memory T cells (20).

Although an advantage of the current study is the face-to-face comparison of Tet⁺ and tetramer-negative cells in the same mouse, it is more than likely that the tetramer-negative T cell populations still contained MCMV-specific CD8⁺ T cells specific for other MCMV epitopes (17, 18, 21). We wondered whether this could have masked possible differences in the turnover of MCMV-specific and non–MCMV-specific memory CD8⁺ T cells. To investigate this, we compared the deuterium enrichment levels of T_N, T_CM, T_EM, and T_M cells in MCMV-infected mice to those in uninfected mice. Because these levels were very similar (Supplemental Fig. 4), we conclude that the expected lifespan of MCMV-specific memory CD8⁺ T cells is not significantly different from that of other memory-phenotype CD8⁺ T cells.

During the chronic phase of MCMV infection, we found no evidence that MCMV-specific CD8⁺ T cells are longer-lived or produced at higher rates than do bulk memory-phenotype CD8⁺ T cells. These findings are in line with our recent findings in humans, which showed that CMV-specific CD8⁺ T cells had similar turnover rates as bulk memory CD8⁺ T cells (28). Both outcomes are remarkable in the light of a previously published deuterated-glucose labeling study in humans, which reported that CMV-specific CD8⁺ T cells incorporated less deuterium than did CD45RO⁺ (memory) T cells (20), which led to the hypothesis that inflating responses are composed of relatively long-lived memory T cells. We found that the turnover rates of Ag-specific T cells composing inflationary responses that varied up to 3-fold in size were not significantly different. This adds further support to our conclusion that the magnitude of inflationary responses is not explained by extended lifespans of MCMV-specific T cells.

To interpret the deuterium labeling data, we used a previously proposed kinetic heterogeneity model (24), which yields the average production rate p of cells, as well the average loss rate d* of labeled cells. It was previously explained that cell populations in steady state typically yield d* > p, because p is representative of all cells in the population, whereas d* is biased toward cells that have just divided (24). The estimated value of p can thus safely be interpreted as the average production rate of cells, which apparently does not differ between inflationary CMV-specific T cell responses and bulk memory T cells. The cellular production that we measured by in vivo deuterium labeling captures both T cell division as well as a possible influx of cells from the naive compartment. Although it has previously been shown that naive T cells can continuously be recruited into the MCMV-specific T cell response (19), the contribution of this influx is probably relatively small, as memory inflation in mice was shown to be hardly affected by thymectomy (29, 30). Thus, assuming that the vast majority of MCMV-specific cells are formed by peripheral T cell division, and not by continuous recruitment of new naive MCMV-specific T cells into the memory pool, the similar production rates of Tet⁺ and T_M cells imply that MCMV-specific T cells do not divide more frequently than do other memory T cells, which is supported by their similar Ki-67 expression levels. It has recently been shown that the MCMV-specific inflationary T cell response is “fueled” by a subset of Tcf1⁺ MCMV-specific T cells (31), and that continual, stochastic encounters with MCMV maintain the inflationary response (32).

To compare the average loss rates of Tet⁺ and T_M cells, we used additional information on absolute cell numbers, which confirmed that also the expected lifespans of Tet⁺ and T_M cells are very similar. We found that the expected lifespan (calculated as 1/d) of bulk memory-phenotype T cells was ∼65 days, whereas that of MCMV-specific T cells was 73 days for the large inflationary response in MCMV^ie2SL-infected mice, 54 days for the intermediate inflationary response in MCMV^ie2KNL-infected mice, and 68 days for the low inflationary response in MCMV^m45SL-infected mice. These estimates are well in line with previous studies showing the dynamic behavior of inflating responses (19). We thus found no evidence for the previously proposed idea that CMV-specific T cells are longer-lived than other memory T cells (20). Instead, our data suggest that the explanation for the size differences between MCMV-specific CD8⁺ T cell responses and for memory inflation in general should be sought earlier during infection. In line with this, several studies have shown that the inflationary potential of CMV-specific T cells is set early, during the acute phase of the response (19, 30), and is linked to the number of primed KLRG1- CMV-specific CD8⁺ T cells (33) and to the early transcriptomic profile and T_CM cell precursor content of the CMV-specific CD8⁺ T cells (34).

Nonsteady cell numbers in some of the populations may explain why we sometimes found values of d* lower than p (see Table I). This is typically not observed in deuterium labeling experiments (24), and it suggests that cells that have recently divided live longer than other cells in the population. Alternatively, for populations that are not in steady state, if cellular turnover is dependent on cell densities, average production rates may decrease during the labeling experiment, although cell numbers are increasing. This could explain why, contrary to what is typically observed for populations in steady state, some turnover rates during the de-labeling phase were lower than during the up-labeling phase (A.C. Swain, J. Drylewicz, J.A.M. Borghans, and R.J. de Boer, manuscript in preparation). In this light, it is interesting to note that the T cell repertoire against CMV has recently been shown to continuously evolve during chronic infection, in that the relative immunodominance of high-affinity clones declines during chronic infection, most likely due to cellular senescence (35).

A previous study in humans reported that YFV-specific CD8⁺ T cells triggered upon YFV vaccination divide sporadically, approximately every 666 days (11), and less than bulk memory T cells. It was argued that the shorter intermitotic times of bulk memory T cells probably reflect their continuous Ag stimulation (11). In line with this hypothesis, we found that MCMV-specific CD8⁺ T cells, which are repeatedly exposed to their cognate Ag, have similar lifespans as bulk memory-phenotype CD8⁺ T cells, which may also be continuously exposed to commensal and environmental Ags. However, the observation that lymphocytic choriomeningitis virus–specific memory CD8⁺ T cells transferred into naive mice had similar turnover rates to those of bulk memory-phenotype CD8⁺ T cells (36) suggests that even Ag-specific T cells maintained in the absence of cognate Ag can turnover as fast as bulk memory-phenotype T cells. Although we cannot exclude the possibility that the differences in the maintenance of YFV-, MCMV-, and lymphocytic choriomeningitis virus–specific memory CD8⁺ T cells are due to mouse and human differences, the characteristics of different Ag-specific memory T cells may also depend on the nature of the infection, the duration of the stimulus, and the concomitant response to other Ags. It is therefore perhaps not surprising that Ag-specific T cell responses against different infections have different dynamics (37). Future studies into the dynamics of memory T cells specific for Ags that are presented persistently (chronic), intermittently (latent reactivating), or only once (acute) are needed to gain more insight into how Ag-specific memory T cell responses are maintained in mice and humans.

Chronic CMV infection in both mice and humans is under constant immune surveillance and triggers ongoing CD8⁺ T cell responses. It is thought that CMV infection modulates the peripheral lymphoid pool (38, 39) and affects T cell differentiation and function (40), not only of CMV-specific T cells but also of T cells with other specificities (41). Under this hypothesis, we directly compared the dynamics of T_N, T_CM, T_EM, and T_M cells in uninfected and chronically MCMV-infected mice (Supplemental Fig. 4) and found no significant differences in their kinetics. Despite differences in the composition of the T cell pool, our results therefore suggest that the dynamics of non–MCMV-specific CD8⁺ T cells are not substantially affected during chronic MCMV infection. We previously observed that also cellular immune function was maintained during latency, as responses to heterologous virus infection and immune protection were not diminished in mice latently infected with MCMV or other herpesviruses (42).

The large prevalence of chronic CMV infection in the human population (>50%) (43) and its effect on healthy aging (44–46), together with the emerging interest in CMV-based vector vaccines (15), highlight the need to understand how CMV-specific CD8⁺ T cell responses are maintained. In vivo MCMV infection provided us with the means to address fundamental questions about the maintenance and turnover of inflated CD8⁺ T cell responses. The finding that the maintenance of inflationary MCMV-specific CD8⁺ T cells does not differ from that of low inflationary memory CD8⁺ T cells suggests that inflationary CD8⁺ T cell responses, such as those induced by CMV-based vector vaccines, may also result in a memory CD8⁺ T cell response of high magnitude without substantial alterations in the dynamics of the cells.

We thank Ramon Arens for generously providing all MHC-peptide tetramers, Nienke Vrisekoop for help and support with the animal experiments, Derek Macallan and Linda Hadcocks for measuring ²H₂O enrichment in the body water of the uninfected mice, and Janine Schreiber and Lothar Gröbe for technical assistance.

The authors have no financial conflicts of interest.