The huge interest in Cytomegalovirus (CMV) as a driver of T-cell proliferation and differentiation originates from its apparent role in human immune system aging. During adult life, large numbers of CMV-specific T-cells are thought to undergo homeostatic proliferation and so maintain the overall size of the memory T-cell pool in the face of declining thymic output (1). In addition, frequent reactivation of CMV from latency provides repetitive antigen stimulation, thus further inflating the CMV-specific memory T-cell pool (2).

However, not only do CMV-specific memory T-cells increase with age in CMV-infected people, some authors have proposed that CMV infection also drives non-CMV-specific T-cells to differentiate towards a more terminal phenotype and so potentially modifies responses to other antigens in CMV-infected individuals (3). This was proposed because the difference in the numbers of advanced memory T-cells between CMV-infected and uninfected people appears to exceed what might reasonably be explained by the presence of CMV-specific T-cells alone.

To explore this phenomenon, Fletcher et al. examined CMV-specific T-cells as well as T-cells specific for M. tuberculosis, Epstein–Barr virus, varicella-zoster virus, and herpes simplex virus in both CMV- and CMV+ individuals (4). They showed that responses to these other infections were not only bigger among CMV+ individuals (in terms of CD4 T-cell percentage) but also more differentiated (larger proportions of CD27-/CD28- CD4 T-cells), suggesting that this might be direct evidence of an effect of CMV infection on non-CMV-specific T-cells. In vitro experiments suggested that such a bystander effect could be mediated by the production of IFN-α produced by plasmacytoid dendritic cells, and the authors speculated that this might occur in vivo in the T-cell areas of lymph nodes in the presence of CMV.

Given that >60% of the older population in Western Europe are CMV+ (with higher prevalence in most other parts of the world), understanding if CMV infection is responsible for the differentiation of non-CMV-specific T-cells is important because this might open up new therapeutic strategies focusing on CMV for counteracting inflammation. CMV-specific T-cells can be identified and enumerated using a number of different assays, with T-cell activation-based assays such as Elispot and flow cytometry being the most frequently used approaches (5, 6). These assays detect responses to different types of CMV antigen preparations such as CMV lysates and recombinant proteins, partial or complete proteins, and peptide pools. Irrespective of the type of assay, it is not possible, however, to measure the size of the immune response to the totality of CMV proteins in a single assay unless all expressed proteins are used at the same time. This is technically not feasible. As a result, in every CMV antigen-stimulated assay, there is always a population of CMV-specific T-cells of unknown size that was not activated by the antigens selected for stimulation. It is not possible to determine the relative size of that population. However, it is possible to run many assays covering multiple different CMV antigens in parallel in order to get close to the real proportion of CMV-specific T-cells in the blood. Using such an approach in 2005, we published the probably most comprehensive study ever done in terms of the CMV protein-specific T-cell response repertoire (7). Stimulations with >200 peptide pools covering all putative open reading frames (ORFs) of the CMV genome translating into amino acid (aa) chains >30 aa were run in parallel. The responses from all stimulations were added up to derive a summated, overall response to CMV in terms of CD4 and CD8 T-cell proportions. This produced the best possible (minimum) estimate of the entire CMV-specific T-cell response. One of the most important findings of this study was that this entire response can be approximated very closely using the sum of a reduced set of protein-specific responses. These included just six proteins for the CD4 and 15 proteins for the CD8 T-cell response, the sum of which showed correlations of R = 0.93 and R = 0.91, respectively, with the summated response to all proteins among CD4 and CD8 T-cells (7).

The present analysis was driven by the idea to capture the overall memory subset distributions of non-CMV-antigen-specific CD4 and CD8 T-cells in CMV+ people and compare these to the overall CD4 and CD8 T-cell distributions in CMV-negative people. This would show beyond doubt if there was a general effect of CMV infection on non-CMV-specific T-cells beyond responses to selected chronic infections. We attempt to do this by, first, estimating the size of the entire CD4 and CD8 T-cell responses to CMV (in % of all CD4 and all CD8 T-cells, respectively) based on the measured responses to a reduced set of CMV antigens as described above. In the second step, we determine the overall distributions of CD4 and CD8 T-cells in terms of naïve (T_Na) central memory (T_CM), effector memory (T_EM), and revertant T-cells (T_Rv) across all measured CMV antigen-specific responses. Third, by multiplying the estimated sizes of the entire CMV-specific CD4 and CD8 T-cell responses with each respective memory subset proportion, the contribution of CMV-specific T-cells to each of the CD4 and CD8 T-cell memory subsets (in terms of all CD4 or CD8 T-cells) is determined. Finally, by subtracting these estimated contributions from each of the memory subsets, the proportion of non-CMV-specific T-cells within each of them is revealed.

This analysis was carried out in 96 CMV-infected individuals, and the extrapolated memory subset distributions of the non-CMV-specific T-cells were compared to those in 81 CMV-uninfected individuals. We believe that our analysis provides an estimate of the memory subset distribution of non-CMV-specific T-cells in CMV-infected people that is as accurate as possible and corroborates the idea that CMV drives the differentiation of T-cells outside the CMV-specific compartment. At the same time, our analysis puts this effect into perspective and shows that strong effects are limited to a minority of CMV+ individuals.

We used two datasets for this analysis: first, our most recent dataset (“dataset 1”), which included 96 CMV+ and 81 CMV- people with CMV-specific responses measured relative to 16 CMV antigens in terms of IL-2, TNF, and IFN-γ as well as basic memory phenotype data (CD45RA and CCR7 expression), and second, we used the original dataset from 2005 (“dataset 2”), including T-cell responses to 213 putative CMV proteins (CMV ORFs translated into protein sequences, represented by 13,687 peptides arranged in pools) in terms of IFN-γ. This dataset included 16 female and 17 male CMV+ participants of mixed biogeographical ancestry (for details, see “Materials and methods”).

We initially analyzed the distribution of canonical memory T-cell subsets (9) in dataset 1 to assess the differences between CMV+ and CMV- people in detail. Figure 3A shows a comparison between the percentages of naïve T-cells (CD45RA+/CCR7+, T_Na), central memory T-cells (CD45RA-/CCR7+, T_CM), effector memory T-cells (CD45RA-/CCR7-, T_EM), and CD45RA-revertant memory T-cells (CD45RA+/CCR7-, T_Rv) among the CD4 (left) and CD8 (right) T-cell compartments in CMV- and CMV+ people. Differences between CMV- and CMV+ individuals were statistically significant with respect to all of these subsets and in both T-cell compartments, except for CD4 T_CM cells. The biggest differences among CD4 T-cells were found with respect to the T_EM compartment, whereas among CD8 T-cells the biggest differences were seen in the T_Rv compartment. We also visualized the overall distributions of T-cell frequencies across these compartments using histograms (Figure 3B). In keeping with our previous work (10), we used CMV- individuals to determine the upper boundaries for outliers and applied them to both CMV- and CMV+ individuals as this approach provides a sense of how often CMV+ individuals show distributions that would not be expected among CMV- individuals. Upper outlier boundaries were calculated as upper quartile + 1.5 * interquartile range (UQ + 1.5 * IQR). For the most part, the T-cell frequencies measured in each compartment in CMV+ donors were inside the outlier boundaries defined in CMV- individuals (indicated as vertical red lines for T_EM and T_Rv cells) (Figure 3B). Among CD4 T-cells, the T_EM and T_Rv compartments showed the most striking differences between CMV+ and CMV- people. The distribution of T_EM and T_Rv CD4 T-cells among CMV+ individuals compared with CMV- people appeared more skewed (i.e., increase of the width of the distribution and loss of basic curve shape) than shifted (i.e., change of mean while retaining the basic shape of the curve) and showed many outliers (Figure 3B). CD8 T_Rv cells also showed strong population skewing with a majority of CMV+ individuals, however, not showing outlier values (Figure 3B). This initial analysis demonstrated that an effect of CMV on the T-cell memory phenotype distribution does not occur in all CMV+ people to the same extent. A majority of individuals do not show such a marked effect at all (Table 2). Of note is that, in approximately 2/3 of CMV+ people, the proportions of T-cells remained within the respective outlier limits found in CMV- people in all analyzed compartments (64% in CD4 and 65% among CD8).

For the present report, we explored the extent to which CMV infection drives memory T-cell differentiation within and outside of the CMV-specific T-cell compartment. A strong effect of CMV on peripheral blood lymphocyte subset distributions, including strong skewing of T-cell subsets towards a late(r) differentiation stage, has been reported by many authors (12, 13). Looking in more detail, however, in almost 100 CMV+ individuals, the present analysis showed that the memory subset distribution of the overall CD4 and CD8 T-cell compartment is strongly skewed only in a minority of CMV-infected people. Rather than a shift of the entire curve (indicating that all peripheral blood T-cells are affected in an unspecific way), we observed a larger spread of the size of CD4 T_EM and CD8 T_Rv subsets with much increased values in a minority of individuals.

To provide a more objective comparison between CMV- and CMV+ people and in analogy to our previous report on response sizes and outliers among CMV-specific T-cell responses (10), we determined non-parametric boundaries for outliers as upper quartile (UQ) + 1.5 * interquartile range (IQR) in CMV- people and then checked if the corresponding percentages of these subsets in CMV+ people were outliers or not. Based on these outlier limits, the percentage of the T_Rv subset was most likely to be an outlier with respect to both CD4 and CD8 T-cells. However, in almost two-thirds of CMV+ people, none of the subset percentages were outliers irrespective of T-cell compartment. While such skewed distributions may be less frequent in CMV- people, our data shows that the skewing can be just as strong in the absence of CMV infection. We acknowledge that there are some observable changes in CMV+ people, in particular among the CD8 T_Rv population (Figure 3B), that do not exceed the outlier threshold. These more subtle changes are not considered by our approach, which may be a limitation. However, it is still true that subset sizes up to the threshold are not uncommon among CMV- people.

To understand if the perceived differences between CMV- and CMV+ people with respect to the overall distribution of peripheral blood T-cells into the canonical T_Na, T_CM, T_EM, and T_Rv compartments were the effect of (predominantly late differentiated) CMV-specific T-cells simply mixing with non-CMV-specific T-cells, we had to estimate both the size and the memory subset distribution of all CMV-specific T-cells in dataset 1. This included revisiting dataset 2 for testing the ability of the protein-specific responses recorded in dataset 1 to predict the entire CMV-specific T-cell response. As a result of optimizing the selection of proteins to predict the entire CMV-response as closely as possible, we were able to adopt the mechanism for estimating the overall response proportions as originally published in 2005 (7). We are aware that the individuals analyzed in the original study were a different mix of genetic backgrounds (HLA background); however, there was no alternative dataset to be used. We believe that this will only make a small difference since there are, to our knowledge, no reports showing that CMV-specific T-cell proportions significantly differ between populations of different biogeographical ancestry.

Following the estimation of the size and memory subset distribution of the entirety of CMV-specific CD4 and CD8 T-cells found in the peripheral blood of CMV+ individuals, we subtracted the CMV-specific portion of each memory compartment from its overall size (measured by a simple surface phenotype stain) in terms of CD4 or CD8 T-cell percentages, which left us with an estimate of the percentages of non-CMV-specific T-cells in each compartment. The distribution of these supposedly non-CMV-specific T-cells both among CD4 and CD8 T-cells still looked quite different from the memory subset distributions found in CMV- people. We also compared CD4 and CD8 T-cell memory subset distributions between individuals that did or did not have outlier responses with respect to the CD4 T_EM and T_Rv and the CD8 T_Rv compartments, respectively. This comparison showed that the distribution of non-CMV-specific memory subsets in those without outliers was more similar to the ones in CMV- people. To explore if the remaining differences were at least partly due to differences in age and sex between the CMV- and CMV+ groups, we used logistic regression. This revealed the effects of age and sex in addition to CMV serology on the extrapolated sizes of the non-CMV-specific T_CM, T_EM, and T_RV compartments. Importantly, it confirmed a significant positive effect of CMV serostatus on the CD4 T_EM and CD8 T_CM, T_EM, and T_RV compartments. The detected significant negative effect of female sex on the CD4 T_EM and CD8 T_Rv populations is in agreement with the fact that, in women, peripheral blood T-cells tend to show less terminal differentiation profile than in men (11).

The contribution of CMV-specific T-cells to the most differentiated compartments was visualized by plotting the estimated contributions of CMV-specific and non-CMV-specific CD4 T_EM, CD4 T_RV, and CD8 T_EM cells to the respective overall memory compartments (Figure 8). This showed that, in some cases, CMV-specific T-cells only made a minimal contribution to a very large T_EM or T_Rv compartment. In some cases, large CMV-specific responses went with a small proportion of non-CMV-specific T-cells and vice versa, indicating that the effect on the non-CMV-specific T-cells is not driven by the size of the CMV-specific response.

Because more CMV protein coding content has been discovered since the original analysis (14), the method for estimating overall CMV-specific response size devised in the original analysis in 2005 may result in an underestimation of overall response size. The use of a single cytokine readout, i.e., IFN-γ, rather than a combination of activation markers is also likely to underestimate a response size. This is because not all CMV-specific T-cells will produce IFN-γ in response to stimulation with CMV antigens or other antigens (15). Studies using MHC tetramers for single CD8 T-cell epitopes suggested that only around approximately 60% of T-cells, on average, will produce IFN-γ (16, 17). This indicates that we may still have underestimated the proportion of CMV-specific T-cells in our sample. With respect to the memory subset distribution of the entire CMV-specific T-cell response, we chose to use the weighted average for each subset across all positive responses to estimate the overall CMV-specific memory subset distribution in each individual. This approach accounts for the fact that larger responses make a larger contribution to the overall distribution. If the CMV-specific T-cell response was still much larger than our estimate, the “corrected” memory subset distribution of non-CMV-specific T-cells would be different; however, the apparent effect of CMV on non-CMV-specific T-cells would not disappear.

The wide variation of overall memory compartment distributions across CMV+ individuals as well as the lack of correlation between the relative sizes (in % of CD4 or CD8 T-cells) of corresponding memory subsets among CMV-specific and non-CMV-specific T-cells suggested that different mechanisms drive the effect of CMV infection on these populations. In order to be able to compare overall T-cell differentiation between populations more comprehensively, we performed a bivariate correlation analysis using an overall CMV-specific response size and the previously introduced T-cell differentiation index (DSc) (11) for the various populations of interest. These analyses confirmed that the differentiation of CMV-specific T-cells had no noticeable effect on the differentiation of non-CMV-specific T-cells. The finding that the overall proportion of CMV-specific CD4 T-cells correlated moderately with the differentiation score of all CD4 T-cells (Table 5) is in agreement with our previous report of a moderate positive and a moderate negative correlation between the size of the CMV-specific CD4 T-cell response and the size of the T_CM and T_Rv compartments (referred to as T_EMRA in the cited publication), respectively, of the tuberculin-specific CD4 T-cell response (18). However, these weak to moderate correlations offer no explanation as to what the mechanism behind these correlations might be and if these are direct or indirect effects. We cannot exclude that the situation might look different if we had access to other compartments—for example, secondary lymphoid organs or the lung, where many CMV-specific T-cells are found as shown in rhesus macaques (19). This is a limitation that most studies into human pathogen-specific T-cells face.

Our analysis was designed to explore if the skewed overall memory T-cell subset distribution observed in CMV+ individuals is simply the result of mixing a population of CMV-specific T-cells with large T_EM and T_EMRA components with a population of non-CMV-specific T-cells, many of which are T_Na or T_CM. While we may still have somewhat underestimated the size and skewness of the memory subset distribution of CMV-specific T-cells, we conclude that the pronounced skewing of the overall memory subset distribution observed in approximately 30% of CMV+ people cannot simply be explained by the presence of CMV-specific T-cells in the mix. The fact that strong overall memory subset skewing is only present in a minority of CMV+ people suggests that there must be predisposing factors rendering individuals susceptible to these changes if they are infected with CMV. While the mechanisms proposed by Fletcher et al. (4) could, in theory, be effective in this minority of individuals, it remains unclear why the size of the effect would differ so much among CMV+ people. Moreover, Fletcher’s explanation would appear to imply that (reactivated) CMV is ordinarily (or at least frequently) present in lymph nodes at the time when T-cells for other infectious antigens are primed. While it is generally accepted that CMV infection causes immune activation driving immune system differentiation, a convincing correlate of a general CMV-driven immune activation beyond the lymphocyte compartment has remained elusive. In the future, it will be important to focus on those individuals in whom CMV has the largest impact on shaping the T-cell compartment as these changes are associated with immune system aging.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving humans were approved by UK National Research Ethics Service (NRES) ‘London Centre’ (Reference 13/LO/1270). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

SF: Conceptualization, Methodology, Visualization, Writing – original draft, Writing – review & editing. BR: Data curation, Formal Analysis, Methodology, Software, Writing – original draft, Writing – review & editing. OF: Supervision, Writing – review & editing. AP: Formal Analysis, Methodology, Writing – review & editing. LP: Supervision, Writing – review & editing. FK: Conceptualization, Formal Analysis, Investigation, Methodology, Project administration, Supervision, Visualization, Writing – original draft, Writing – review & editing.