High Production Rates Sustain In Vivo Levels of PD-1high Simian Immunodeficiency Virus-Specific CD8 T Cells in the Face of Rapid Clearance

Programmed Death 1 (PD-1) expression by human/simian immunodeficiency virus (HIV/SIV)-specific CD8 T cells has been associated with defective cytokine production and reduced in vitro proliferation capacity. However, the cellular mechanisms that sustain PD-1high virus-specific CD8 T cell responses during chronic infection are unknown. Here, we show that the PD-1high phenotype is associated with accelerated in vivo CD8 T cell turnover in SIV-infected rhesus macaques, especially within the SIV-specific CD8 T cell pool. Mathematical modeling of 5-bromo-2′ deoxyuridine (BrdU) labeling dynamics demonstrated a significantly increased generation rate of PD-1high compared to PD-1low CD8 T cells in all memory compartments. Simultaneous analysis of Ki67 and BrdU kinetics revealed a complex in vivo turnover profile whereby only a small fraction of PD-1high cells, but virtually all PD-1low cells, returned to rest after activation. Similar kinetics operated in both chronic and acute SIV infection. Our data suggest that the persistence of PD-1high SIV-specific CD8 T cells in chronic infection is maintained in vivo by a mechanism involving high production coupled with a high disappearance rate.

Several studies have described in vivo T cell dynamics during HIV and SIV infection (11)(12)(13)(14)(15)(16)(17). It is clear from this body of work that both HIV and SIV infection lead to increased turnover of CD4 and CD8 T cells in vivo. Although the interpretation of labeling studies that focused on CD4 T cell kinetics was confounded by the fact that these cells are the major targets for cytopathic viral infection, it became clear that increased turnover of CD4 and CD8 T cells alike is primarily a physiological consequence of incessant immune activation, both specific and nonspecific (18). The study of CD8 T cell dynamics is therefore important, not only in its own right, but also as a paradigm for understanding the consequences of ongoing immune activation.
It is unclear how a large population of PD-1 ϩ antigen-specific CD8 T cells is maintained in the presence of continuous HIV/SIV replication, especially when all data suggest a critical role for chronic antigen stimulation, which should also induce apoptosis. One possibility is that PD-1 ϩ virus-specific CD8 T cells are maintained via slow production and slow clearance, suggesting that they do not actually undergo apoptosis in response to in vivo stimulation as they do in vitro, perhaps due to an altered cytokine environment. Alternatively, these cells could persist in a more dynamic state of high production and high clearance, constantly replenished from a population of PD-1 Ϫ virus-specific CD8 T cells. To distinguish between these two possibilities, we investigated the turnover of bulk and SIV-specific CD8 T cell populations by analyzing the cellular incorporation of 5-bromo-2=deoxyuridine (BrdU) in SIV-infected rhesus macaques.

MATERIALS AND METHODS
Animals. Four colony-bred rhesus macaques (Covance Research Products), housed and handled in accordance with the standards of the American Association for the Accreditation of Laboratory Animal Care, were infected intravenously (i.v.) with SIVmac251. All animal studies were approved by the Animal Care and Use Committees of the Vaccine Research Center, National Institute of Allergy and Infectious Diseases, NIH. Viral loads were measured using a QIAamp Viral RNA Mini Kit (Qiagen) at Duke Human Vaccine Institute, Durham, NC. Peripheral blood mononuclear cells (PBMCs) were isolated from whole blood by centrifugation over an isotonic discontinuous Percoll (Sigma) density gradient (35% to 60% [vol/vol]). After washing, the cells were cryopreserved until further use.
Flow cytometry. Briefly, 2 ϫ 10 6 to 3 ϫ 10 6 cells were washed and stained sequentially with ViViD, ␣PD-1biotin, and fluorochrome-labeled streptavidin. After two further washes, the cells were incubated with tetramer at 37°C for 20 min, washed twice, and surface stained with ␣CD8, ␣CD28, and ␣CD95 for 20 min, and then fixed with 1ϫ fluorescence-activated cell sorter (FACS) lysing solution (BD Biosciences) and permeabilized with a Cytofix/Cytoperm Kit (BD Biosciences). The cells were then treated with DNase (APC BrdU Flow Kit; BD Biosciences) for 30 min at 37°C, washed with Perm/Wash buffer (Cytofix/Cytoperm Kit; BD Biosciences), and stained intracellularly with ␣CD3, ␣BrdU, and ␣Ki67. The gating scheme is shown in Fig. S1A in the supplemental material. The CD28 dim CD95 low (here referred to as naive) CD8 T cell population was used to define PD-1 high and PD-1 low expression. Cells were analyzed using a modified LSRII flow cytometer (BD Immunocytometry Systems). Between 500,000 and 10 6 events were acquired for each condition. Antibody capture beads (BD Biosciences) stained separately with the individual MAbs used in the test samples were used for electronic compensation. Data analysis was performed using FlowJo version 9.0.1 (TreeStar). The forward scatter area (FSC-A) versus forward scatter height (FSC-H) profile was used to gate out cell aggregates; ViViD was used to exclude apoptotic cells. After selection based on CD3 positivity, BrdU and Ki67 expression was measured in gated CD4, CD8, and tetramer ϩ CD8 T cells with respect to differentiation status and PD-1 expression.
In vivo administration of BrdU. BrdU (Sigma) was dissolved in Hanks balanced salt solution (Life Technologies) at 10 mg/ml, pH 7.4, and sterile filtered into autoclaved bottles. Macaques received 30 mg/kg body weight of BrdU daily via i.v. injection on four consecutive days. Blood was collected on days 1 (pre-BrdU injection, basal level), 2, 3, 4, 7, 9, and 14 for acute-phase studies; blood samples were collected on days 22 and 31, in addition, for chronic-phase studies. The same macaques were studied in both acute-and chronic-phase SIV infection using separate courses of BrdU administration.
Mathematical model. A system of coupled ordinary differential equations (ODE) was defined to explore the kinetic behavior of three different cell states defined as resting (R), activated (A), and effector (E). In this model, resting cells can become activated at rate a, and activated cells can divide at rate p or differentiate (at rate e) into effectors that can disappear at rate d or return to rest (state R) at rate r. The differential equations for the time evolution of the population sizes are as follows: The BrdU labeling and delabeling process was simulated by assuming that during the labeling phase, cells that divided became BrdU high , whereas during the delabeling phase, cells that divided lost BrdU on average in four steps, meaning that four divisions were necessary to transition from BrdU high to an unlabeled state. This "minimal model" simplifies the dilution process because the BrdU population generated during the labeling period is heterogeneous in terms of labeling intensity; thus, some labeled cells that continue to divide become negative earlier than others. Mathematically, this was captured by using five subgroups of R, A, and E, corresponding to four BrdU ϩ groups and one BrdU Ϫ group. The param-eters r, a, e, p, and d of the model were fitted against the experimental data (fractions of BrdU ϩ cells, total and Ki67 Ϫ , in blood over time) using the above ODEs in MATLAB (details can be found in the supplemental material). Steady state was assumed and confirmed experimentally (see Fig.  S1B in the supplemental material).
Statistical analysis. All statistical analyses were performed using GraphPad Prism (GraphPad Software). P values were calculated using the Mann-Whitney U test, and values of Ͻ0.05 were considered significant. The mean values Ϯ standard errors are also presented. BrdU decay rates were calculated assuming first-order kinetics. The natural log of the percentage of BrdU high CD8 T cells was plotted against time. These data were fitted to a straight line using the method of least squares and the first-order rate constant determined from the slope of the line.

High turnover of CM9 ؉ PD-1 high CD8 T cells during the chronic phase of SIV infection.
To investigate the in vivo dynamics of rigorously defined T cell populations in rhesus macaques 3 to 4 months after SIVmac251 infection, we conducted a serial analysis of BrdU incorporation using flow cytometry and mathematical modeling (Fig. 1a). No BrdU integration was detected in naive CD8 T cells, indicating very slow in vivo turnover within the compartment (Fig. 1b). In agreement with previously published data (6), almost all CM9-specific CD8 T cells were found to express a PD-1 high phenotype (see Fig. S1A and B in the supplemental material). BrdU incorporation was higher in the CD28 high CD95 high population of CM9 ϩ PD-1 high CD8 T cells than in the corresponding CD28 low CD95 high population (57.7 Ϯ 2.9 versus 22.5 Ϯ 2.1, maximum percentage of BrdU high cells; P ϭ 0.0294) (Fig. 1b). The accumulation of BrdU was also accelerated in the CM9 ϩ PD-1 high CD28 high CD95 high population (peak on day 3 or 4) compared to the CM9 ϩ PD-1 high CD28 low CD95 high population (peak on day 5 or 6) (  1b). The frequency of TL8 ϩ SIV-specific CD8 T cells during the chronic phase was very low (see Fig. S1B in the supplemental material), making the analysis of BrdU incorporation problematic. Analysis of bulk memory CD8 T cell subsets revealed higher in vivo incorporation of BrdU in the PD-1 high CD28 high CD95 high population than in the PD-1 high CD28 low CD95 high population (26.2 Ϯ 2.4 versus 15.4 Ϯ 1.5, maximum percentage of BrdU high cells; P ϭ 0.0286) (Fig. 1b). Similarly to CM9 ϩ CD8 T cells, the kinetics of BrdU incorporation differed between these two populations; BrdU accumulation peaked at day 3 or 4 for the PD-1 high CD28 high CD95 high population, while the peak for PD-1 high CD28 low CD95 high cells occurred on day 5 or 6. PD-1 low cells were characterized by lower in vivo incorporation of BrdU than PD-1 high cells in both memory compartments ( BrdU decay rates were calculated assuming first-order kinetics. As expected, no decay was apparent in the naive population ( Fig.  1c, left). The decay rate was significantly higher in CM9 ϩ cells (0.418 Ϯ 0.028 per day) than in the bulk CD28 high CD95 high (0.234 Ϯ 0.044 per day; P ϭ 0.013) and CD28 low CD95 high (0.177 Ϯ 0.016 per day; P ϭ 0.0003) CD8 T cell populations (Fig.  1c, left). We further analyzed BrdU decay rates in individual mem-ory populations with respect to PD-1 levels (Fig. 1c, right). Again, significantly higher decay rates were found for CM9 ϩ cells than for bulk PD-1 high CD8 T cells in both memory populations (0.651 Ϯ 0.09 versus 0.338 Ϯ 0.032 per day, CD28 high CD95 high , P ϭ 0.021; 0.453 Ϯ 0.02 versus 0.231 Ϯ 0.015 per day, CD28 low CD95 high , P ϭ 0.0004). Furthermore, a trend toward higher decay rates in CM9 ϩ CD28 high CD95 high cells than in CM9 ϩ CD28 low CD95 high cells was observed (Fig. 1c, right). Interestingly, bulk PD-1 high CD8 T cells were characterized by significantly higher BrdU decay rates than for PD-1 low CD8 T cells in both memory compartments (0.338 Ϯ 0.032 versus 0.147 Ϯ 0.014 per day, CD28 high CD95 high , P ϭ 0.0026; 0.231 Ϯ 0.015 versus 0.121 Ϯ 0.018 per day, CD28 low CD95 high , P ϭ 0.004) (Fig. 1c, right). Overall, our data point to accelerated in vivo turnover of CM9 ϩ PD-1 high CD8 T cells and indicate that PD-1 low CD8 T cells have the lowest turnover among the populations tested.
CM9 ؉ PD-1 high CD8 T cells are characterized by high in vivo generation rates. Despite the high in vivo turnover of CM9 ϩ CD8 T cells, the percentage (ϳ0.7%) of these cells in the periphery remained relatively constant over the 31 days of observation (see Fig. S1B in the supplemental material). Analysis of PD-1 and Ki67 expression revealed a similar steady state for all populations tested (see Fig. S1B in the supplemental material). These are important prerequisites for the modeling approach used here, which is therefore not well suited to acute-phase analyses. To explore the proliferation and disappearance kinetics of the various populations more thoroughly, we analyzed in vivo turnover using a simple five-parameter ODE model (Fig. 2a). Due to the limited number of data points, we opted for a simple model that nevertheless accounts for the fact that activated cells will first proliferate and then contract (differentiate, die, and disappear). A simpler model with only four parameters, according to which cells were assumed to proliferate and disappear simultaneously, was able to fit the observed total BrdU uptake and BrdU loss kinetics for most cell populations but failed to reproduce the kinetics of the appearance of Ki67 Ϫ BrdU ϩ cells (Fig. 2a, bottom); this observation emphasizes the importance of taking into account the time-structured "cohort" behavior of activated cells (18,20). A representative "fitting curve," generated by fitting the parameters of this model to one of our data sets, is shown in Fig. 2a (bottom). The complete set of fits is provided in Fig. S2 in the supplemental material. CM9 ϩ CD8 T cells were characterized by higher activation/proliferation rates than bulk CD8 T cells in both the CD28 high CD95 high and CD28 low CD95 high compartments (Fig. 2b). The generation rate was calculated based on the product of the proliferation rate and the fraction of activated cells in each particular population. CM9 ϩ CD8 T cells had a significantly increased generation rate compared to PD-1 high bulk memory (CD28 high CD95 high and CD28 low CD95 high ) CD8 T cells (Fig. 2b). Cells with low PD-1 expression had the lowest generation rates in both memory populations. A striking hierarchy was also apparent with respect to the fraction of resting cells. In particular, CM9 ϩ CD28 high CD95 high CD8 T cells displayed the smallest fraction of cells in the resting state (40%), while substantially larger fractions were observed in the bulk memory populations, especially those with low levels of PD-1 expression. Taken together, our data indicate that a high-generation/high-disappearance mechanism underlies the maintenance of antigen-specific and bulk PD-1 high CD8 T cell populations in chronic SIV infection.
Only a small fraction of activated PD-1 high CD8 T cells return to a resting state. Next, we analyzed the in vivo kinetics of individual CD8 T cell populations according to the expression of Ki67, an activation/proliferation marker that is downregulated when cells return to a resting state. The majority of CM9 ϩ cells displayed a Ki67 high phenotype in both the CD28 high CD95 high (79.69 Ϯ 1.78) and CD28 low CD95 high (52.78 Ϯ 1.28) populations (see Fig.  S1B in the supplemental material). Among BrdU ϩ cells, coexpression of Ki67 can be divided into three populations (Ki67 high BrdU high , Ki67 dim BrdU high , and Ki67 low BrdU high ) and tracked over time (Fig. 3a). Analyzing the in vivo BrdU label contents in these populations for both CM9 ϩ and total CD8 T cells allowed us to follow the kinetics of activation and return to rest. In the CM9 ϩ CD28 high CD95 high population, BrdU peaked at day 3 or 4 in the Ki67 high BrdU high state, followed by the Ki67 dim BrdU high (peak at day 6 or 7) and Ki67 low BrdU high (peak at days 13 to 20) populations (Fig. 3b, top). It is noteworthy, however, that these state transitions do not take place in the peripheral blood, where cells appear only very transiently; rather, the blood samples provide a window into the kinetics that take place in the lymph nodes. Importantly, more cells expressed the Ki67 high BrdU high phenotype than progressed to the subsequent populations with lower levels of Ki67 expression (Fig. 3b, top). This phenomenon was more prominent in the CM9 ϩ CD28 high CD95 high population than in the CM9 ϩ CD28 low CD95 high population (Fig. 3b, bottom), thereby confirming our computational analysis, which had indicated a clear hierarchy in the ability of cells to return to rest after activation (Fig. 2b). Analysis of bulk CD8 T cells revealed a similar loss (through death or migration into the tissues) of cells during transition from an activated to a resting state in the PD-1 high compartment, especially in the CD28 high CD95 high population (Fig. 3b). Interestingly, this was not the case for bulk CD8 T cells expressing a PD-1 low phenotype, where BrdU high cells transitioned through the phases of Ki67 expression without substantial loss in total cell numbers (Fig. 3b). Next, the BrdU decay rates were calculated for CD8 T cell populations in the Ki67 high BrdU high and Ki67 dim BrdU high compartment. Again, the CM9 ϩ CD8 T cell population showed the highest decay rate among all populations tested, reaching statistical significance in the Ki67 high BrdU high CD28 low CD95 high (0.811 Ϯ 0.011 versus 1.044 Ϯ 0.09 per day, bulk PD-1 high versus CM9 ϩ CD8 T cells; P ϭ 0.043) and Ki67 dim BrdU high CD28 high CD95 high (1.001 Ϯ 0.04 versus 1.417 Ϯ 0.09 per day, bulk PD-1 high versus CM9 ϩ CD8 T cells; P ϭ 0.007) cellular compartments (Fig. 3c). Bulk PD-1 high CD8 T cells were consistently characterized by higher BrdU decay rates than bulk PD-1 low CD8 T cells in all populations tested (Fig. 3c), reflecting rapid proliferation-induced label dilution and replacement of labeled cells by unlabeled cells in the PD-1 high compartment.
High turnover of CM9 ؉ PD-1 high CD8 T cells during the acute phase of SIV infection. In further experiments, we investigated the in vivo dynamics of BrdU incorporation 4 weeks after SIVmac251 infection (Fig. 4a). A low, transient integration of BrdU was observed in the CD28 dim CD95 low CD8 T cell population (Fig. 4b). Similar to the chronic phase of SIV infection, BrdU incorporation was highest in the CM9 ϩ compartments of both memory CD8 T cell subsets ( Fig. 4b; see Fig. S3 in the supplemental material); furthermore, BrdU loss was accelerated in the CM9 ϩ population compared to matched bulk memory CD8 T cells (Fig.  4b). In contrast to the situation in chronic SIV infection, however, bulk PD-1 high CD28 high CD95 high and PD-1 high CD28 low CD95 high CD8 T cells showed similar maximum levels of BrdU incorporation (33 Ϯ 6.6 versus 32.7 Ϯ 7.6, maximum percentage of BrdU high cells) (Fig. 4b). BrdU decay rates were significantly higher in CM9 ϩ CD8 T cells (0.496 Ϯ 0.05) than in bulk CD8 T cells in both the CD28 high CD95 high (0.183 Ϯ 0.03 per day; P ϭ 0.004) and CD28 low CD95 high (0.195 Ϯ 0.01 per day; P ϭ 0.002) compartments. Additionally, the PD-1 low population displayed lower decay rates than the PD-1 high population in both the  CD28 high CD95 high (0.143 Ϯ 0.03 versus 0.368 Ϯ 0.04 per day; P ϭ 0.004) and CD28 low CD95 high (0.094 Ϯ 0.017 versus 0.185 Ϯ 0.039 per day; P ϭ 0.07) subsets. As in the chronic phase, simultaneous analysis of BrdU and Ki67 expression revealed complex BrdU dynamics in both the CM9 ϩ and bulk CD8 T cell populations (Fig.  4c). A large difference in maximum BrdU incorporation was apparent between Ki67 high and Ki67 dim CM9 ϩ cells (Fig. 4c) Fig. 4c and data not shown). Again, CM9 ϩ CD8 T cells showed the highest decay rates of all populations tested, even during acute infection (Fig. 4d). Therefore, the in vivo dynamic profile of CD8 T cell populations during the acute phase is similar to that observed in the chronic phase of SIV infection.

DISCUSSION
Recently published work has shown that PD-1 could serve as a regulator of antigen-specific CD8 T cell survival (9,21,22). In particular, the level of PD-1 expression correlates with in vitro sensitivity to cell death in both HIV (9) and SIV (6) infection. This in vitro phenotype raises the question of what cellular mechanism(s) supports the sustained PD-1 high HIV/SIV-specific CD8 T cell populations observed in vivo. We considered two possibilities. First, the rapid disappearance of PD-1 high virus-specific CD8 T cells may be counterbalanced by rapid production (high generation/high disappearance). Second, PD-1 high CD8 T cells may not be as susceptible to apoptosis in vivo as they are in vitro, in which case a low generation/low disappearance rate could sustain PD-1 high virus-specific CD8 ϩ T cells. The current study was designed to investigate the relative impacts of these mechanisms in mediating the potential proapoptotic function of PD-1. Accordingly, the in vivo turnover rates of several rigorously defined CD8 T cell populations with respect to PD-1 expression were determined in SIV-infected rhesus macaques.
Several previous studies have shown a significantly increased in vivo turnover of CD8 T cells in SIV-infected monkeys compared to noninfected animals (12,14,17). Furthermore, memory CD8 T cells were found to express a phenotype characterized by accelerated in vivo accumulation of BrdU, followed by rapid loss compared to naive CD8 T cells (12,17), indicative of increased proliferation/disappearance of the former cells. In our study, no BrdU integration was found in the "naive" CD28 dim CD95 low CD8 T cell compartment during the chronic phase, and only transient BrdU integration in a small fraction of these cells was observed during the acute phase. Furthermore, only very low levels (Ͻ1%) of BrdU incorporation were detected before SIV infection, even in the PD-1 high bulk memory CD8 T cell compartment. This finding is in line with the very low levels of Ki67 expression previously reported in this population of cells in non-SIV-infected animals (23) and indicates that PD-1 does not demarcate proliferation by itself. Furthermore, the short time window (4 days) of BrdU administration in this study likely only permits the detection of cell populations with high turnover rates, in contrast to previous studies in which BrdU was given for 2 to 4 weeks (12,14,17). As expected, SIVspecific CD8 T cells were characterized by higher turnover rates than bulk CD8 T cells in both memory compartments tested. Additionally, high expression of PD-1 was consistently associated with higher in vivo turnover in all populations tested. This was more prominent in the less differentiated CD28 high CD95 high population than in the CD28 low CD95 high population for both total and SIV-specific CD8 T cells. The kinetics of BrdU integration were also found to be different in these two memory subsets, but only in the PD-1 high compartment. Thus, PD-1 high CD28 high CD95 high cells accumulated BrdU faster and reached maximum integration earlier than PD-1 high CD28 low CD95 high cells. Although the peak of BrdU integration was different, similar kinetics applied to SIV-specific and bulk CD8 T cells in the same memory compartment. These findings indicate that the expression of coinhibitory receptors, such as PD-1, as well as differentiation status, plays an important role in the regulation of CD8 T cell proliferation/disappearance in vivo.
The relative representation of phenotypically distinct CD8 T cell populations was found to be quite constant during the period of investigation in both the chronic and acute phases of SIV infection. An important aspect of this profile relates to whether a cell that starts out in one phenotype can be assumed to end in another phenotype. To explore this issue further, we would require far more data, because the models would include parameters for phenotypic transitions (differentiation steps). As our model does not allow for such transitions, we assumed the presence of a population that can receive a certain degree of stimulation and appear as PD-1 ϩ in the blood. The data and the model show that most of these cells received a potent stimulus and rapidly underwent several divisions, after which they quickly disappeared. In agreement with our previous data (6), the vast majority of SIV-specific CD8 T cells expressed a PD-1 high phenotype during the chronic phase. This profile was associated with significantly higher Ki67 expression than in bulk PD-1 high CD8 T cells in both memory compartments tested, while PD-1 low cells expressed the lowest levels of Ki67. The application of a mathematical model revealed significantly higher generation rates in the SIV-specific CD8 T cell compartment than in bulk PD-1 high populations; PD-1 low cells displayed the lowest generation rates. Expression of Ki67, a marker of CD8 T cell activation, therefore, correlates with the in vivo turnover of total and virus-specific CD8 T cells during chronic SIV infection. This finding agrees with previously reported data from studies of HIV infection (15).
In the case of SIV-specific CD8 T cells, the loss of BrdU was found to follow a biphasic mode, with a rapid decline (before day 9) followed by a slower phase (after day 9). Such a biphasic mode was previously reported in other studies of HIV (13) and SIV (24) infection. The loss of BrdU could be due to cell death, migration to different anatomical compartments, and/or dilution in cells that continue to divide. Loss of Ki67 expression by BrdU-labeled cells after discontinuation of BrdU administration identifies cells that have stopped dividing, while the persistence of a population of cells with high levels of Ki67 expression but diminishing BrdU intensity helps to identify cells that continue to divide (16,24). We therefore investigated the in vivo turnover of CD8 T cell populations further by analyzing the kinetics of cells coexpressing BrdU and Ki67. BrdU dynamics were found to be dramatically different between SIV-specific CD8 T cell populations with respect to Ki67 expression. A hierarchy was observed both for peak height and the rate of decline of BrdU in the different populations tested. The Ki67 high BrdU high population exhibited the highest integration/ decline rate of BrdU, followed by the Ki67 dim BrdU high and Ki67 low BrdU high populations. This profile was more prominent in the less differentiated CD28 high CD95 high memory cell subset. A similar profile was observed in the bulk PD-1 high compartment. In the PD-1 low compartment, all three populations (Ki67 high BrdU high , Ki67 dim BrdU high , and Ki67 low BrdU high ) exhibited similar peak heights. Label loss was faster in the PD-1 high than in the PD-1 low compartment regardless of Ki67 expression, supporting a potential proapoptotic role for PD-1 in cells expressing a PD-1 high phenotype. Furthermore, the slower decline of PD-1 high BrdUlabeled cells in the Ki67 dim and Ki67 low populations than in the Ki67 high population likely stems from a greater disappearance of BrdU high cells in the PD-1 high than in the PD-1 low compartment. This conclusion is further supported by the results of our computational analysis, which indicates a considerably lower rate of return to rest for PD-1 high cells.
PD-1 is a marker of potent stimulation and demarcates cells that have received proproliferative signals. In the experimental data, we cannot distinguish cells that received these stimuli over a prolonged period of time, performed several divisions, and then assumed a state that corresponds to desensitization from cells that simply upregulated PD-1 due to recent stimulation, thereby retaining the ability to undergo rapid division pending receipt of strong inhibitory signals via PD-1. For the modeling process, however, this lack of distinction was not critical, because we only assumed the existence of a population of cells that can receive a certain degree of stimulation and appear as PD-1 high in the blood. Again, our data show that the vast majority of PD-1 high cells received strong stimulation, rapidly underwent several divisions, and then disappeared.
Our simple computational analysis assumed a self-contained PD-1 high population whose homeostasis is maintained by a (small) fraction of cells returning to rest after activation and proliferation. Reconciling our findings of vigorous proliferation with a previously assumed role of PD-1 as a marker of defective responsiveness to antigenic stimulation, one could hypothesize that PD-1 high T cells, after leaving a proliferative state, do not return to a pool of cells that can be restimulated to undergo further rounds of division. Thus, the virusspecific Ki67 Ϫ BrdU ϩ PD-1 high cells observed in our study might not be contributing to the long-term persistence of the PD-1 high phenotype. One possibility is that new PD-1 high virus-specific CD8 T cells are continuously generated, especially in the setting of chronic infection. Alternatively, PD-1 low cells could continuously convert to a PD-1 high phenotype by upregulating PD-1 upon stimulation/activation. A more finely resolved kinetic analysis of BrdU incorporation and phenotypic transition will be necessary to distinguish between these possibilities.