Memories that last: Dynamics of memory T cells throughout the body

Memory T cells form an essential part of immunological memory, which can last for years or even a lifetime. Much experimental work has shown that the individual cells that make up the memory T‐cell pool are in fact relatively short‐lived. Memory T cells isolated from the blood of humans, or the lymph nodes and spleen of mice, live about 5–10 fold shorter than naive T cells, and much shorter than the immunological memory they convey. The commonly accepted view is, therefore, that long‐term T‐cell memory is maintained dynamically rather than by long‐lived cells. This view is largely based on memory T cells in the circulation, identified using rather broad phenotypic markers, and on research in mice living in overly clean conditions. We wondered to what extent there may be heterogeneity in the dynamics and lifespans of memory T cells. We here review what is currently known about the dynamics of memory T cells in different memory subsets, locations in the body and conditions of microbial exposure, and discuss how this may be related to immunometabolism and how this knowledge can be used in various clinical settings.

radiotherapy, cells with unstable chromosomes were rapidly lost from the CD45R0 + (i.e., memory) but not from the CD45RA + (naive) T-cell population. 3 Likewise, in patients who had undergone chemotherapy, CD45R0 + T cells reconstituted faster than CD45RA + T cells. 4 In an effort to quantify the rates of cell division of memory T cells, later studies used in vivo stable isotope labeling, in the form of heavy glucose ( 2 H-glucose) or heavy water ( 2 H 2 O), and estimated that human memory T cells live between 34 and 340 days, [5][6][7][8] whereas naive T cells can live up to 2.8-9.1 years. 6,[9][10][11] Possible explanations for the differences in these estimates between laboratories and techniques have been reviewed previously. 12,13 Are these estimated lifespans representative for all memory T cells? Over the years, it has become increasingly clear that memory of the total amount of T cells in the body-the other 98% reside in lymphoid and nonlymphoid tissues. 14 There is much evidence that in homeostasis, tissue-resident memory T cells tend to stay in the tissue they have homed to. If indeed these tissue-resident memory T cells hardly recirculate, then most literature falls short on predicting the lifespan of these cells. Thirdly, it is important to realize that our current knowledge is biased towards studies in "clean" laboratory mice. Although these studies have revealed essential fundamental principles about the generation and maintenance of T-cell memory, extrapolation of these findings to humans is not always trivial. The immune system of "dirtier" mice might better reflect the immune system of humans that naturally live in pathogen-rich environments, and hence might better reflect the true dynamics of memory T cells.
Lastly, different subsets of memory T cells employ different metabolic strategies, which might influence the capacity of these cells to proliferate or survive.
There are various clinical settings that could benefit from a better understanding of the dynamics of memory T-cell subsets-that is, whether cells are maintained by proliferation, cellular longevity, migration or differentiation of T cells from other subsets. For instance, in cancer T-cell therapy, a memory T-cell subset that maintains itself by proliferation might be preferred over one that needs constant input from activated naive T cells. Likewise, interference with T-cell mediated diseases-including psoriasis, inflammatory bowel disease, and graft-versus-host disease-requires quantitative insights into the population dynamics of memory T cells at relevant locations in the body. The effect of therapies that interfere with T-cell proliferation or migration is crucially dependent on the proliferative capacity or circulatory behavior of the memory T cells that are causing disease. Finally, some novel vaccination strategies aim to induce longterm T-cell memory at the site of pathogen entry. 15,16 The success of such vaccinations depends on the long-term maintenance of memory T cells at specific sites in the body. A better understanding of the in vivo dynamics of different T-cell memory populations throughout the human body is, therefore, needed. Here, we review the steps that have been made so far to unravel the dynamics and lifespans of the full breadth of memory T cells.

| HOW TO ME A SURE THE DYNAMI C S OF T CELL S?
Let us define the lifespan of a cell as the time between the moment the cell was born and the moment it dies or divides. A stationary cell, like an enterocyte lining the small intestine, can be followed relatively easily throughout its life. A T cell, however, wanders through the body. Following an individual T cell would require sampling a whole organism, which is impossible in practice. Yet, several methods have been used to quantify the dynamics of T cells and their lifespans, all coming with their own strengths and limitations. These methods can roughly be categorized in three groups: monitoring cell population sizes, analyzing cellular markers, and labeling cells.

| Monitoring cell population sizes
To get a rough estimate of memory T-cell dynamics, changes in numbers of cells can be followed over time. In mice, the numerical response of T cells to infections has been quantified extensively. After an initial proliferation of pathogen-specific T cells, more than 90% of effector T cells die, and a small population of pathogen-specific memory T cells persist long-term. 17 Another method to follow cell numbers in mice is parabiosis, the surgical joining of the vasculature of two mice. This method has been used to study the exchange of tissue-resident memory T-cell populations between specific sites. In humans, certain clinical interventions that disturb the normal homeostasis of memory T cells, including radiotherapy, 3 chemotherapy and conditioning before hematopoietic stem cell transplantation, 18 have enabled the investigation of memory T-cell dynamics. Moreover, the follow-up of donor T cells that come along with transplanted tissues allows for rough estimations on the stability of T-cell populations. [19][20][21] Under the assumption that cells are not lost through, for example, cell death, increases in cell numbers provide a lower-bound on the rate of cell production. Likewise, decreases in cell numbers provide a lower-bound on the rate of cellular loss, assuming that there is no formation of new cells. At the cellular level, however, monitoring of cell population sizes comes with some important limitations.
Firstly, the interpretation of cell numbers is complicated by the multiple mechanisms that underlie population dynamics: for example, proliferation, cell death, import and export of cells from other compartments. Secondly, radiotherapy, chemotherapy, and parabiosis are all nonphysiological situations in which (low) levels of inflammation or cell damage and reduced competition between cells can alter cell dynamics. 22 Thirdly, cell numbers can vary significantly, which may in part be explained by the circadian rhythm through which Tcell numbers fluctuate during the day. 23,24

| Analyzing cellular markers
A commonly used and relatively easy way to evaluate how many T cells are proliferating is to measure the fraction of cells expressing the intranuclear protein Ki67. Ki67 is absent during the resting phase (G0), and accumulates during all active phases of the cell cycle (G1, S, G2, and M). Although useful, the interpretation of Ki67 data is complicated by the fact that Ki67 can be expressed for up to 5-7 days after cell division. 25,26 Dual staining of Ki67 and DNA, therefore, provide more accurate readouts. 27 Other flow cytometric markers, like markers for T-cell senescence, apoptosis or quiescence, can also be used to define the status of T cells. Although measurements of these markers and comparison between different conditions can give valuable information, it remains difficult how they should be translated to cellular lifespans.

| Labeling cells
Methods that directly assess the division and lifespan of cells rely on labeling of cells. Cells can be labeled with intracellular dyes, nucleoside analogues, or stable isotopes. Intracellular dyes, like cell trace violet (CTV) or carboxyfluorescein diacetate succinimidyl ester (CFSE), are serially diluted upon every division. Up to seven divisions can be tracked reliably before the fluorescence of the dye returns to background levels. 28 More recently, a genetic cell division tracer system has been developed in mice, which can be used to follow large numbers of cell divisions through expression of red-fluorescent protein. 29 T-cell turnover can also be studied using nucleoside analogues, like bromodeoxyuridine (BrdU) or ethynyldeoxyuridine (EdU). These are incorporated instead of thymidine into the DNA during cell division. Correct interpretation of nucleoside analogue labeling data is again not trivial and requires mathematical modeling. 30 A disadvantage of these labeling techniques is that they can be toxic, 31 may in fact influence leukocyte turnover, [32][33][34] and cannot be used to study the in vivo dynamics of cells in humans.
Stable isotopes, like deuterium in deuterated water ( 2 H 2 O) or deuterated glucose ( 2 H 2 -glucose), are also incorporated into newly formed DNA during cell division. In contrast to nucleoside analogues, deuterated water and glucose can safely be given to humans and animals without influencing leukocyte turnover. 32 Deuterium is measured in isolated DNA from sorted cell populations using a combination of gas chromatography and mass spectrometry (GC/MS), and mathematical models are used to deduce the turnover rates of cells from the resulting up-and downlabeling curves.
Given the strengths and weaknesses of the different techniques to study T-cell dynamics, the most reliable information comes from studies that combine different methods. In the sections below, we describe how these different approaches have provided insights into the dynamics and lifespans of memory T cells throughout the body.

| DYNAMIC S OF CIRCUL ATING MEMORY T-CELL SUBS E TS
It is now quite commonly accepted that memory T cells need not be long-lived to convey long-term immunological memory. Several labelling studies have pointed out, however, that there is considerable heterogeneity in the circulating memory T-cell pool, in terms of cellular lifespans, proliferative capacity, and dependence on input from other T-cell subsets. 5,9,10,35,36 Another issue that remains debated is the role of antigen. It has been suggested that memory T cells spe-

| Memory T-cell subsets based on phenotypic markers
It is common practice to define T-cell subsets based on the expres- blood-derived T-cell subsets. They found that the more effectordifferentiated the cells were, the more they expanded after cytokinestimulation and the less they expressed Bcl-2, suggesting a higher in vivo turnover of more effector-differentiated memory T-cell subpopulations. In humans, the dynamics of cells in phenotypically separated memory T-cell subsets have been studied using shortterm in vivo deuterium labelling experiments for CD4 + T cells, 38 and long-term labelling experiments for CD4 + and CD8 + T cells. 8,39,40 Short-term labelling experiments suggested that CD4 + T EM cells in humans live approximately 3-fold shorter than T CM cells. 38  depending on the co-expression of CD28. 8 In a long-term labelling study among cytomegalovirus (CMV) + and CMV − healthy individuals, we found only a subtle trend, but no significant difference in the average lifespan of cells in the combined T EM + T EMRA population compared to cells in the T CM population. These findings were quite consistent between CMV + and CMV − individuals. 39 The fact that we did not find significant differences may have been due to the inclusion of T EMRA cells. Although based on data from only one healthy individual, Ladell and colleagues reported that CD8 + T EMRA cells were very long-lived, with a half-life of nearly 10,000 days. 8 Bacchus-Souffan et al., 40  Recent in vivo labelling studies have also revealed the dynamic properties of memory T cells expressing the senescence marker CD57. Based on the finding that CD57 + T cells hardly proliferate in vitro after antigen stimulation, it was long assumed that memory T cells expressing this senescence marker would hardly proliferate in vivo. 44 Labelling data showed, however, that memory T cells expressing CD57 undergo as much turnover as memory T cells that do not express CD57, 45 again suggesting that even the most differentiated memory T cells can sustain themselves by self-renewal ( Figure 1). In line with this, a very recent study 46 demonstrated that T cells that had undergone up to 50 rounds of clonal expansion, expressed a whole array of common exhaustion markers, but still maintained the capacity to expand upon vaccination.
While memory T cells thus do not seem to have an intrinsic chronological constraint or proliferative limit, it is commonly thought that aging does affect the dynamics of memory T cells. Indeed, the replacement rate of memory T cells in old mice is lower than that in young mice. 35 It remains unclear whether this is due to less supply from the naive compartment or due to increased resistance to replacement by the existing memory T cells. In humans, there appears to be no age-related influence on the turnover of memory cells. 11 In an in vivo heavy water labeling study, we found that the expected lifespans of memory T cells in older individuals (aged 66-75) were very similar to those in young individuals (aged 20-25). It should be noted, however, that the older individuals in the latter study were extremely healthy and likely had little or no long-term exposure to low-grade chronic inflammation, or "inflammaging", which is believed to contribute to many age-related changes in the immune system.

| Antigen-experienced memory T cells
The development of T-cell receptor (TCR) transgenic mouse models and major histocompatibility complex (MHC) multimers has facilitated the study of true cognate antigen-specific memory T cells.
Researchers have thereby been able to investigate the dynamics of memory T cells specific for different antigens, and to compare these to the dynamics of different memory T-cell subsets based on phenotypic markers.
One of the first examples in which the dynamics of antigenspecific memory T cells were studied in vivo in humans is the response to CMV. CMV is well-known for the large number of memory T cells it induces, which can even gradually increase over

F I G U R E 1 Influence of phenotype and antigen specificity on the lifespan of memory T cells. The differentiation state is generally correlated to T-cell lifespan, which varies from very long-lived naive T cells (T N ) to shorter-lived central memory (T CM ) cells and even shorter-lived effector memory (T EM ) T cells. The lifespan of stem cell memory T cells (T SCM )
is still under debate: both relatively long and short lifespans have been suggested. Memory T cells expressing the senescence marker CD57 divide as often as memory T cells that do not express CD57. The influence of antigen specificity seems to differ based on the antigen: yellow fever virus (YFV)-specific memory T cells have been suggested to live longer or about as long as bulk memory T cells, while cytomegalovirus (CMV)-specific memory T cells, in both mice and men, have been shown to live as long a bulk memory T cells. Created with BioRe nder.com. time, a process termed "memory inflation". [47][48][49] Based on shortterm in vivo deuterated glucose labelling, it has been proposed that CMV-specific CD8 + T cells may be longer-lived than other memory T cells, which may explain their accumulation over time. 50 Our more recent data do not support this hypothesis. We performed long-term in vivo labeling experiments in CMV-infected humans 39 and MCMV-infected mice 41 and quantified the turnover of CMV-specific memory T cells in the memory phase of infection.
In both humans and mice, we found no indication that the turnover of CMV-specific cells differed from that of bulk memory T cells.
The explanation for the large size of CMV-specific T-cell responses is thus not found in longer lifespans or higher proliferation rates of these cells, and should possibly be sought early after CMV infection, although the mechanisms driving these expansions remain to be elucidated. 39,41 The observation that antigen-specific T cells specific for a latent/ chronic infection such as CMV undergo the same levels of turnover as bulk memory phenotype cells raises the question whether the relatively high rates of turnover of cells in the circulating memory T-cell pool may in fact reflect the continuous stimulation of cells by antigen. In other words, what is the evidence that true memory T cells, which are not continuously seeing their cognate antigen, are also maintained in a dynamic way? CFSE labelling of adoptively transferred lymphocytic choriomeningitis virus (LCMV)-specific memory T cells into naive mice has suggested that even antigen-specific memory T cells that stably persist in the absence of antigenic stimulation have relatively short lifespans (~50 days), independent of their epitope-specificity and similar to those of bulk memory phenotype T cells. 51 In humans, the dynamics of antigen-specific memory T cells in the absence of continuous antigenic stimulation has been beautifully studied using the live-attenuated yellow fever virus (YFV) vaccine. 52 The vaccine induces immunological memory in the form of antibodies and CD8 + memory T cells, which can be detected up to decades after vaccination. Long-term in vivo deuterium labeling of YFV-vaccinated individuals in the memory phase of the response (i.e., 4-9 months post-vaccination) suggested that YFVspecific memory T cells live for about 140 days, similar to bulk memory T cells in previous studies. Nevertheless, in another labeling experiment in which deuterium was administered shortly after vaccination, the majority of the YFV-specific memory T cells that were still present 1-2 years after vaccination originated from cells that were labeled shortly after vaccination. These cells had a quiescent phenotype with features of stem cell-like and effectorlike memory CD8 + T cells. The authors, therefore, suggested that at least part of the antigen-specific memory T-cell pool may consist of long-lived memory cells.
Future studies into the dynamics of antigen-specific memory T cells, varying from those that are continuously exposed to their antigen to those whose antigen is no longer present, are needed to gain better insight into how desired long-term memory T-cell responses against pathogens and undesired memory T-cell responses against self-antigens are maintained.

| DYNAMI C S OF MEMORY T CELL S IN TISSUE S
To date, most research on human memory T cells has been performed on cells isolated from the blood. Most immune responses, however, occur in tissues, whether directed against cancer cells, pathogens or self-antigens. Recent studies have shown that memory T cells in tissues not only vastly outnumber those in blood, 17 but may also provide superior protection. 15,18,53 Because many CD8 + T cells in tissues hardly recirculate through the blood, these "tissueresident" memory T (T RM ) cells 15  Across species and tissues, T RM cells are broadly unified by a transcriptional and phenotypical core signature, 18,57,58 although T RM cells also show signs of tissue-specific adaptation. 59 Perhaps the most widely used T RM marker is CD69, 56 which prevents the egress of T cells from tissues. 60 The adhesion molecule CD103 is often used as an additional marker, mainly for CD8 + T cells in epithelial tissues such as the intestine or skin. 18,58 Additionally, T RM cells may express molecules that provide tissue-homing cues such as CXCR6, avoid exit cues by downregulating CD62L, and generally upregulate inhibitory molecules such as PD-1 and CD101. 58 Memory T cells in tissues are generally assumed to be longlived. However, studying this is quite challenging as it is difficult to track individual T RM cells in time and space, particularly in humans.
Therefore, various approaches have been used to infer the lifespan of memory T cells in tissues, based on changes in cell numbers over time, expression of cellular proliferation and death markers, and quantification of cellular lifespans using stable isotope labeling.

| Persistence of T RM cell populations
One approach that has been instrumental to follow the dynamics of memory T cells in tissues is parabiosis. A key paper 53  The fate of memory T cells in tissues has also been investigated by following donor memory T cells that have been transplanted along with solid organs. In one study, researchers followed HLAmismatched donor T cells in 5 patients after full facial skin transplantation. 66 Donor T cells could be found back up to 23 months after transplantation-the latest time point of investigation. 66 Similarly, HLA-mismatched donor T cells were followed in biopsies from patients with diabetes mellitus type 1 after a pancreatic-duodenal transplantation. 19,20 After 1 year, about 70% of CD103 − and CD103 + CD4 + T cells in tissue biopsies were still of donor origin. 19 For CD8 + T cells, patients varied substantially in the amount of replacement.
CD103 + CD8 + T cells had a stronger tendency to persist than CD103 − cells: 1 year after transplantation, donor CD103 + CD8 + T cells still made up 60% of CD8 + T cells in the transplant, while donor CD103 − CD8 + T cells had almost entirely been replaced by host CD103 − CD8 + T cells. 20 Interestingly, there was no evidence for lateral migration of donor T cells into adjacent recipient duodenum. 19,20 In another study of HLA-mismatched liver transplants, small populations (~2%) of donor CD8 + T cells persisted for as long as 11 years in the transplanted liver and displayed a stronger T RM phenotype (higher expression of CXCR3) than recipient-derived CD8 + T cells. 67 Lastly, donor T cells were still present after 1 year in the bronchoalveolar lavage of HLA-mismatched lung transplant patients, albeit with substantial interindividual differences (between 0.5-55% for CD4 + T cells and 0.7-85% for CD8 + T cells) and there seemed to be a preferential maintenance of donor CD8 + over CD4 + T cells. 21 Donor T cells were not found in the peripheral blood, arguing against T RM cells exiting the tissue. 21 Taken together, these studies suggest that also in humans, memory T cells can form distinct populations in tissues, which can persist for prolonged periods of time, even in the absence of influx from other compartments.

| The skin: a classic example of T RM cells
The local and long-term persistence of tissue-resident memory T cells demonstrates the stability of the T RM cell population in situ, but it does not prove that individual T RM cells are long-lived. How T RM cells are maintained requires additional information about cell proliferation or cell death. As a readout for the proliferation or death of memory T cells, multiple studies have used the staining of Ki67, Bcl-2, and DNA content, or labeling with CFSE, nucleoside analogues (like bromodeoxyuridine, BrdU) or stable isotopes. Since the mechanisms to maintain a stable T RM population might differ per organ, we will discuss a few exemplary organs separately.
Of all tissues, the dynamics of skin T RM cells are perhaps the most extensively studied. This is rather surprising, as the skin of C57Bl/6 mice contains hardly any αβ T cells, and nearly only γδ T cells. 68 Conventional αβ T RM populations have experimentally been induced than LCMV-specific liver T RM cells (around 15 ± 1%). 59 In healthy adult humans, <5% of skin T cells were Ki67 + , 65,71 in line with the relatively low level of Ki67 expression of murine antigen-induced skin T RM cells. Interestingly, the expression of Ki67 by fetal skin T cells was much higher: about 20% at 17-19 weeks and 10-20% at 23 weeks gestational age. 71 This suggests that T cells proliferate when they seed the skin early in life, but maintain their numbers later in life by low levels of proliferation.
Taken together, the work on skin T RM cells in mice and humans paints a "classic" picture of the dynamics of tissue-residency.
Memory T cells are recruited to the skin to fight off an infection, maintain a stable population without significant input, and can be called upon during a secondary immune response (Figure 2).

| Dynamic maintenance of T RM cells in the lung
Unlike the skin, observations regarding the maintenance of T RM cells in the lung seem more contradictory. After viral clearance, murine antigen-specific memory T cells were present in the lung for over a year. 72 These lung T RM cells seemed to consist of two distinct subpopulations: a large population (90%), which was seeded early during the acute response and showed minimal cell division, and a small population (10%), which was maintained over a year by homeostatic proliferation. 73 A later study 74 found that the half-life of the majority of lung T RM cells was only 10-14 days, and suggested that the lung T RM population is continuously replaced by cells recruited from the circulation. 74 In this model, the continuous activation and recruitment of circulatory T cells could be mediated by long-term persistence of viral antigen in lung-draining mediastinal lymph nodes. 75 Yet another study 76 challenged this view. Using a combination of intravascular monoclonal antibody staining and parabiosis, a rapid equilibrium of antigen-specific memory T cells established between the infected and naive parabiont in the lung vasculature and spleen, but not in the airways, lung interstitium and parenchyma or draining lymph nodes. 76 After 2 weeks of parabiosis, about 20% of lung T RM cells were replaced by donor T cells, and this percentage did not change significantly even after 7 weeks of parabiosis. Furthermore, when host T cells were depleted with an anti-Thy1.1 antibody prior to parabiosis, the ratio of the lung T RM cells was not affected: 80% of T RM cells were still of host origin after 2 weeks, showing that the host lung T RM population was not continuously fed by the circulation. 76 The authors thus concluded that continuous recruitment of T cells is not necessary to maintain a stable T RM population in the lung.
The conclusions of Takamura and colleagues 76 could, however, be a consequence of the relatively short follow-up after parabiosis.
To study the long-term maintenance of memory T cells, two mice- In humans, lung CD69 + memory T cells had a ~1.5-fold lower level of Ki67 expression than recirculating CD69 − memory T cells: for CD4 + T cells about 6% versus 9%, and for CD8 + T cells about 3% versus 5%, respectively. 58 Interestingly, the levels of Ki67 expression in the lung were systematically higher than those in the spleen, especially for CD4 + T cells. 20 Taken together, the mechanisms through which lung T RM cells are maintained seem conflicting-some studies argue for and other against a continuous influx of cells (Figure 2). A possible explanation for the discrepancies between studies could lie in the definition of the T RM cell populations. First, the use of intravascular monoclonal antibody staining helps to discriminate between T cells in the lung vasculature and those that are truly in the lung tissue. 79 Second, the definition of T RM cells has been refined over the years adding markers like CD69, CD103, and CD49a. 58 Third, even within one tissue, memory T cells could show distinct survival strategies depending on the specific niche in which they are kept. 76

| Is the bone marrow a special niche for longlived memory T cells?
The bone marrow is the main site of leukocyte production in the body, but is also thought to function as a long-term storage for immunological memory. Stromal cells provide mature immune cells with necessary survival signals, a concept that is generally accepted for plasma cells. 80,81 A similar concept might be true for memory T cells. The bone marrow supports a sizeable population of CD69 + T RM cells-around 30% for CD4 + and 60% for CD8 + human memory T cells. 82 It was suggested that certain antigen-specific memory T cells are enriched in the bone marrow in comparison to blood, 82 although another study did not find this correlation. 83 How memory T cells are maintained in the bone marrow, is a matter of active discussion. 34,84,85 Some studies in mice have favored the bone marrow as the preferential site for proliferation over other tissues. In these studies, a higher fraction of bone marrow memory CD8 + T cells incorporated BrdU [86][87][88] and had an increased DNA content 86,87 as compared to memory CD8 + T cells in spleen.
Additionally, when adoptively transferred CFSE-labeled cells were recovered from various organs, the ones from bone marrow showed the highest dilution of CFSE. 86,88 Together, these experiments suggested that the bone marrow is a place where CD8 + memory T cells are actively dividing.
Other studies in mice have favored the bone marrow as a privileged site for T cells to rest. Early after infection, CD4 + LCMVspecific memory T-cell numbers peaked in the spleen and lymph nodes, to then decrease in numbers, while the bone marrow still harbored a considerable pool of LCMV-specific T cells a month after infection. 89 The amount of BrdU incorporation and the DNA content of CD4 + memory T cells were lower in bone marrow than in a variety of other tissues and there was a downregulation of the general transcriptome of CD4 + memory T cells in bone marrow compared to the spleen. 89 Different antigen-specific CD8 + memory T cells had a lower level of Ki67 expression in bone marrow than their counterparts in spleen. 33 Moreover, cyclophosphamide, a DNA cross-linker and common chemotherapeutic, reduced the number of CD8 + memory T cells in spleen but not in bone marrow. 90 Together, these results suggested the opposite picture: bone marrow memory T cells may in fact be resting and protected from cell death to allow for their long-term maintenance in bone marrow niches.
Also in humans, bone marrow CD4 + and CD8 + T cells were interpreted to be resting, as their expression of Ki67 was lower than that of their counterparts in peripheral blood: 1.2% versus 4.2% for CD4 + T cells and 1.7% versus 5.0% for CD8 + T cells, respectively. 82 Ki67-epxression was largely confined to the CD69 − sub-population of bone marrow memory T cells. 82 Because the transcriptome of T RM cells mimicked that of resting cells, it was concluded that these cells maintain themselves through cellular longevity. 82 We have previously measured the lifespan of memory T cells in adult mice 91 and goats. 92 In both species, we corroborated the finding that bone marrow memory T cells show lower expression of Ki67 than memory T cells from blood, 91,92 and in mice also lower expression than in lymph nodes and spleen. 91 At first sight, these findings seem to suggest that bone marrow memory T cells are indeed longer-lived than circulating memory T cells. However, when we quantified the lifespans of blood and tissue-derived memory T cells using in vivo heavy water labeling, we found similar lifespan estimates for memory T cells in the circulation and those in the bone marrow in both goats and mice. 91,92 Whether the bone marrow provides a niche for long-lived memory T cells thus seems not completely resolved. While the expression of Ki67 by memory T cells seems reproducibly lower in the bone marrow compared to the blood and spleen, lifespan estimates of memory T cells in the bone marrow are not convincingly longer than those of memory T cells in the circulation. How can these two different observations be reconciled? The ongoing debate could be related to the inherent difficulty of translating Ki67 expression into proliferation rates. If 5% of a cell population is Ki67 + , is that enough to imply homeostatic turnover? Or does 95% of Ki67 − cells imply

| EFFEC TS OF MICROB IAL E XP OSURE ON T-CELL MEMORY
Laboratory mice have been invaluable for studying the maintenance of T-cell memory, and the role of antigen and location therein, yet the pitfalls of using clean laboratory mice have become increasingly clear. Studying T-cell memory in laboratory mice is highly artificial because of (1) the restricted microbial exposure in laboratory mice, and (2) the fact that T-cell responses are studied in isolation rather than in the context of a history of infections. Indeed, large differences have been observed between the immune system of wild and laboratory mice. 93 While the immune system of wild mice may better reflect that of humans, using wild mice comes at the price of significant variation between mice, both in terms of genetics and environmental triggers. Additionally, many immunological tools have been specifically developed for laboratory mice. To tackle these problems and still study the mouse immune system in more natural, antigenrich circumstances, several new mouse models have recently been developed, each bringing the mouse immune system closer to its natural circumstances. 94,95 In one of the first models, laboratory C57Bl/6 mice were cohoused with feral or pet store mice. Cohoused mice showed major differences in their T-cell pool compared to clean laboratory mice. 96 Among others, the percentage of naive T cells was significantly reduced, while percentages of T CM and T EM cells were significantly increased. Likewise, investigators have sequentially exposed laboratory mice to infections to which humans are frequently exposed, including herpes viruses, influenza and helminth infections. 97 While this approach is more controlled than cohousing, it largely recapitulated the findings in cohoused animals.
Alternatively, laboratory mice have been "rewilded" by releasing them into outdoor enclosures, 98,99 or have been exposed to fecal transplants from wild mice, 100 and wildling mice have been created by transferring C57Bl/6 embryos into pseudo-pregnant wild mice, which then gave birth to C57Bl/6 pups with a natural microbiome inherited from their wild mothers. 101 The overall picture generated by these more natural mouse models is quite a consistent one: in the presence of increased/natural levels of microbial exposure, the T-cell compartment is more mature and differentiated, with higher levels of T-cell activation and increased fractions of memory T cells. Clear differences have also been observed in the T RM population. While clean laboratory mice had only modest T RM populations in nonlymphoid tissues, large T RM populations were observed in the female reproductive tract, salivary glands, liver and kidney of cohoused mice, 96 and in the intestines of rewilded mice. 99 Overall, the T-cell pool of mice exposed to microbes better resembles that of human adults, while the T-cell pool of clean laboratory mice in fact resembles that of newborn humans. 96,97 Importantly, these changes also lead to altered functional responses to vaccination, infection and cancer. 97 for the small intestine, in which cohousing led to a 6-fold drop in the number of LCMV-specific T RM cells. The overall picture generated by these experiments is that tissues have a remarkable flexibility to accommodate more memory T cells when exposed to a larger variety of antigens. Importantly, through parabiosis of congenically distinct mice, which were cohoused with pet store mice, these researchers also demonstrated that the phenomenon of tissue-residency is not an artifact of clean laboratory mice: even upon cohousing, T RM cells in different tissues remained in the host mouse after parabiosis. 77 Future studies in more natural mouse models should point out to what extent T RM cell populations in different tissues maintain themselves through cell division or cellular longevity.

| THE EFFEC T OF T-CELL ME TABOLIS M ON SURVIVAL
Coming back to the example of sharks: Why does a Greenland shark live so much longer than a gray reef shark? The puzzle of longevity has always captivated humans. An intuitive solution is that a lifespan shortens when more energy is expended: a candle that burns brighter also burns faster, a mechanical object that is used more intensely wears more quickly. These common-day observations fueled the "rate-of-living" theory, dating back almost a century. 102 It was observed that, in rest, large animals expend less energy compared to small animals, yet also that large animals live much longer than small animals. The product of resting energy expenditure and lifespan was thought to be constant, and it would thus seem that the total "amount" of living an organism does is fixed: life can be used up slowly and sparingly, or fast and excessively. Later research questioned the absoluteness of the "rate-of-living" theory. 103 Plenty of small animals have surprisingly long lifespans, like different bird and bat species.
Even though the relationship between metabolism and lifespan might not be easily captured, it is tempting to speculate that also for immune cells there is a relationship between metabolism and lifespan. Multiple studies have shown the importance of certain metabolic components for the development, fitness, and survival of memory T cells. One of these studies linked a well-known factor for memory T-cell survival, interleukin (IL)-7, to fat metabolism. 104   induced the expression of the glycerol channel aquaporin 9 (AQP9), specifically on memory but not on naive CD8 + T cells. 104 This channel allows cells to import glycerol and use it to synthesize and store triglycerides. Knocking-out the Aqp9 gene in LCMV-specific T cells did not prevent T-cell expansion, but did impair their long-term survival. 104 Similarly, in memory but not effector CD8 + T cells, IL-15 was shown to enhance the mitochondrial spare respiratory capacity-the reserve ability of cells to produce energy in conditions of stress or extra work. 105 IL-15 promoted mitochondrial biogenesis and expression of an enzyme that controls fatty acid oxidation. 105 Both examples illustrate how cytokines that have long been associated with memory T-cell survival [106][107][108][109][110] might mediate their effects through subset-specific modulation of T-cell metabolism.
Likewise, the postulated longevity of T RM cells might depend on the available metabolic substrates in the niche they inhabit. In the skin of mice, the long-term establishment and maintenance of vaccinia-specific CD8 + T RM cells was shown to be critically dependent upon (the oxidation of) exogenous free fatty acids. 111 Loss of the fatty acid binding proteins FABP4 and FABP5 did not hamper T CM cell survival in the spleen, but resulted in a significant longterm reduction in the number of T RM cells in the skin. 111 In contrast, splenic memory T cells relied on extracellular glucose to support fatty acid oxidation and oxidative phosphorylation. 112 Studies in humans corroborated the finding that skin T RM cells rely on fatty acid metabolism and oxidative phosphorylation. 63 This suggests that the metabolic strategies that memory T cells employ to sustain their numbers long-term might be conserved across species. It would be interesting to use in vivo stable isotope labeling to explore the lifespan of memory T separated by their metabolic profiles.

| CON CLUS I ON S AND FUTURE DIREC TIONS
The current view that T-cell memory is maintained dynamically is largely based on memory T cells isolated from the circulation, identi- Despite the observed differences in lifespans and proliferative behavior of cells in different memory T-cell subsets and at different sites of the body, the current overall picture is still one of dynamic maintenance. We have found no evidence so far for truly long-lived memory T cells, with lifespans as long as those of naive T cells for example. This raises the question what could be the selective advantage of storing immunological memory in cells undergoing relatively rapid turnover. Based on in silico simulations, in which memory T cells compete for survival or proliferation factors, we have recently shown that the diversity of the memory T-cell repertoire tends to reach higher levels if memory T cells are relatively short-lived. A relatively rapid turnover of memory T cells creates more flexibility to store new memories, and hence makes the immune system more agile. 113 The observed differences between clean laboratory mice and mice exposed to wild microbiota raise the question how memory T-cell dynamics may differ between humans living in environments with different levels of exposure to pathogens. Weisman et al. 114

CO N FLI C T O F I NTE R E S T S TATE M E NT
The authors have declared no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analyzed in this study.