Antigen affinity, costimulation, and cytokine inputs sum linearly to amplify T cell expansion

See allHide authors and affiliations

Science  28 Nov 2014:
Vol. 346, Issue 6213, pp. 1123-1127
DOI: 10.1126/science.1260044


T cell responses are initiated by antigen and promoted by a range of costimulatory signals. Understanding how T cells integrate alternative signal combinations and make decisions affecting immune response strength or tolerance poses a considerable theoretical challenge. Here, we report that T cell receptor (TCR) and costimulatory signals imprint an early, cell-intrinsic, division fate, whereby cells effectively count through generations before returning automatically to a quiescent state. This autonomous program can be extended by cytokines. Signals from the TCR, costimulatory receptors, and cytokines add together using a linear division calculus, allowing the strength of a T cell response to be predicted from the sum of the underlying signal components. These data resolve a long-standing costimulation paradox and provide a quantitative paradigm for therapeutically manipulating immune response strength.

Stimulatory signals add up for T cells

T cell activation is a dynamic process. T cells encounter multiple input signals such as antigens, costimulatory molecules, and cytokines at different times and anatomical locations during an infection. But how do T cells integrate this information to determine the extent to which they divide? To find out, Marchingo et al. stimulated mouse T cells in culture with different combinations of inputs and also tracked antigen-specific T cell responses in mice infected with influenza virus. They found that T cells linearly sum the various stimulatory inputs they receive to determine their “division destiny.”

Science, this issue p. 1123

Upon infection, pathogen-specific CD8+ T cells undergo a characteristic kinetic sequence: rapid proliferation and expansion followed by population contraction due to cell death (1). While short-term stimulation is sufficient to trigger CD8+ T cell proliferation (2, 3), further exposure to stimulatory signals is required for an effective response (2, 4, 5). Although multiple attempts have been made to create a theory relating the integration of stimulatory signals to T cell response strength, all have been qualitative (6) and thus have lacked the power to predict the quantitative effect of altering stimulatory combinations and strength. The current qualitative paradigm describes T cell activation and response magnitude as the outcome of three requisite signals: signal 1, T cell receptor (TCR) (1); signal 2, membrane-bound antigen-presenting cell (APC)–delivered costimuli (4); and signal 3, cytokines from inflammatory, homeostatic, or autocrine sources (5). The importance of these signals for T cell expansion is highly context-dependent, as classic in vitro studies have identified many “critical” signal 2 and 3 molecules (5, 79); however, gene deletion typically yields only moderate defects in the in vivo CD8+ T cell response (1013), implying considerable redundancy.

Recent studies in B cells report an automated return to quiescence after a series of division rounds (1416). The number of mitotic cycles B cells undergo varies and is influenced by the strength of stimulation. We hypothesized that T cells might be programmed in a similar manner with the final number of divisions (N) (Fig. 1A, left panel) a function of the sum of inputs from signals 1, 2, and 3 (illustrated in Fig. 1A). If correct, it may be possible to determine the calculus of addition to serve as the basis for a quantitative framework for T cell costimulation.

Fig. 1 CD8+ T cells undergo a program of proliferation and quiescence in vivo and in vitro.

(A) Quantitative T cell expansion hypothesis. By this model, the number of mitotic cycles a T cell undergoes after activation (N) varies and is determined by a sum of the individual inputs it receives. In the example shown, signal 1, 2, and 3 stimuli each individually elicit a small increase in mean population division number. The cumulative effect of these contributions, when summed linearly, would lead to geometric increases in total cell number at the peak response. Analysis of OT-I/FucciRG CD8+ T cells [(B) to (E)] transferred to HKx31-N4–infected recipients on day 2 after infection or [(E) to (G)] in vitro stimulated with N4 in the presence of mIL-2 blocking antibody (S4B6). (B) Number and (C) percentage of OT-I/FucciRG CD8+ T cells expressing FucciR and FucciG reporter proteins pooled from mediastinal lymph node, spleen, and lungs at the indicated time points after transfer. (D) Fitted total (black) and quiescent (red, FucciR+G + small FucciRG cells) cell numbers using the Cyton model (19). (E) The estimated DD distribution from Cyton fitting to in vivo (D) and in vitro (G) stimulated cells. (F) Percentage of FucciR and FucciG expression. (G) Fitted total (black) and quiescent (red, FucciR+G + small FucciRG cells) cell numbers (19). [(B) to (E)] n = 5 to 10 mice per time point, pooled from two independent experiments; mean ± SEM. [(E) to (G)] Representative of three independent experiments; mean ± SEM from triplicate culture wells.

To test this hypothesis, we first measured the onset of quiescence in CD8+ T cells using TCR-transgenic OT-I mice [which recognize chicken ovalbumin SIINFEKL (N4) peptide bound to H2Kb] crossed with FucciRG mice in which cells fluoresce red (FucciR) during G0/G1 and green (FucciG) for the duration of the S/G2/M cell cycle phases (17). Quiescent (FucciR+G), recently divided (FucciRG), and actively dividing (FucciG+) cells can be distinguished, because cells that have reverted to a quiescent state (G0) accumulate higher levels of FucciR (17, 18). We define the number of generations of division before returning to quiescence as the cell’s division destiny (DD) (1416, 19). OT-I/FucciRG CD8+ T cells were transferred into mice infected with recombinant HKx31 influenza virus expressing N4 (HKx31-N4) (20). During early expansion, most OT-I/FucciRG CD8+ T cells were proliferative, with <10% reverting to a quiescent state by day 3 of the response. This proportion of quiescent cells increased steadily, reaching ~75% of all OT-I/FucciRG CD8+ T cells by the onset of contraction at day 7 (Fig. 1, B and C). To estimate the number of divisions T cells underwent before dropping out of cycle, the Cyton model (15, 21) was fitted to total and quiescent cell numbers (Fig. 1D and table S1). Results were consistent with a T cell DD range spanning ~10 generations (Fig. 1E). If DD is carried through each cell lineage (16), this result predicts up to 1000-fold differences in T cell family size, consistent with recent single-cell tracking studies (22, 23).

To further explore the regulation of division progression, we developed a minimal in vitro stimulation system using CellTrace Violet (CTV)–labeled OT-I/FucciRG CD8+ T cells. The contribution of signals 2 and 3 was reduced by using peptide self-presentation by purified CD8+ T cells (24). The strong effect of autocrine interleukin-2 (IL-2) was controlled by adding blocking antibody (clone S4B6) and using human IL-2 (hIL-2), resistant to S4B6, when required (25). Superficially the in vitro pattern of early proliferation with a gradual onset of quiescence recapitulated the in vivo response (Fig. 1, C and F), with the major differences being the DD and subsequent time to die (Fig. 1, D, E, and G, and table S1).

We speculated that the three known sources of regulation—TCR affinity, costimuli, and cytokines—might combine to convert the low DD observed in vitro into the extensive outcome possible in vivo (Fig. 1E).

To improve estimation of mean DD (mDD) in our in vitro CTV division tracking assay, we used OT-I CD8+ T cells deficient in the proapoptotic molecule Bim (OT-I/Bcl2l11−/−) for all experiments. These cells reported the same mDD as OT-I/Bcl2l11+/+ CD8+ T cells (Fig. 2A; fig. S1, A to C; and table S2), but the enhanced survival upon reverting to quiescence facilitated DD measurement at later times, consistent with previous studies in B cells (15, 16).

Fig. 2 Costimuli and cytokines program changes to DD.

(A) CTV-labeled OT-I/Bcl2l11+/+ or OT-I/Bcl2l11−/− CD8+ T cells were stimulated with N4; total cell numbers (left), mean division number (middle), and an estimation of the percentage of the starting cells whose progeny are contributing to the response at that time point, calculated by removing the effect of cell expansion at each time point [percent cohort number, right; as described in (19)] were determined. Mean DD (mDD) on each graph is indicated by dotted lines. (B) Mean division number of CTV-labeled OT-I/Bcl2l11−/− CD8+ T cells cultured in the presence (+, solid lines) or absence (–, dotted lines) antibodies to of CD28 (2 μg/mL) and CD27 (immobilized, 5 μg/mL) (blue), IL-12 (1 ng/mL), human IL-2 (hIL-2, 31.6 U/mL), or IL-4 (1000 U/mL) (red) for 26 hours, washed and further cultured with (dark) or without (light) costimulation. (C) CTV-labeled OT-I/Bcl2l11−/− CD8+ T cells stimulated with N4 in the presence of hIL-2 at the indicated concentrations were subcultured into fresh hIL-2 every ~48 hours and the cell number was fitted using the Cyton model (dotted lines) (19). (D and E) Dose-response curve of mDD (D) and DD distributions (E) from Cyton fitting in (C). All cultures contained 25 μg/mL S4B6. Representative of at least two independent experiments; mean ± SEM from triplicate culture wells.

Fig. S2, A to F, shows the effect of a range of T cell stimuli on mDD. TCR affinity, several agonists representative of cell-contact–mediated costimulation, and some, but not all, cytokines tested were able to regulate DD in a dose-dependent manner. To determine when DD was most susceptible to regulation, cells from cultures where stimulation was removed immediately before the first division (24) were compared to cells with constant costimulation. Agonist antibodies to CD28 and CD27 principally acted before the first division. In contrast, ongoing exposure to IL-2, IL-4, and IL-12 was required for maximal proliferation (Fig. 2B). Higher levels of IL-2 or IL-4 caused T cells to divide beyond CTV resolution (fig. S3) and the culture capacity. Therefore, to investigate the potential of IL-2 and IL-4 to extend DD when cytokine levels were maintained, OT-I/Bcl2l11−/− CD8+ T cells were subcultured every 48 hours in hIL-2 (Fig. 2C) or IL-4 (fig. S4A), and total cell numbers were calculated using splitting ratios. Cyton fitting revealed that hIL-2 and IL-4 can increase the mDD by up to ~11 and 7 divisions, respectively (Fig. 2, C to E; fig. S4, A and B; and tables S3 and S4). Titration of hIL-2 showed the effect on mDD to be dose dependent (Fig. 2D) and that this increase in mDD was associated with an increase in variance (Fig. 2E and fig. S4C). Together these results demonstrated that DD can be intrinsically programmed by early signals, but also has the flexibility to be “reprogrammed” or extended by extrinsic stimuli as the T cells divide.

We then determined how T cells integrated multiple contributors to DD. In Fig. 3, A and B, we show the increase in mDD imprinted before the first division for low concentrations of CD28 and CD27 agonist antibodies, and IL-12 protein. The combination of antibodies to CD28 and CD27 programmed an mDD that was equivalent to the sum of each individual effect, with IL-12 giving a slightly greater than additive increase in mDD (Fig. 3, C and D). Importantly, no single “second signal” appeared obligatory, but rather multiple small arithmetic effects on DD culminated in large geometric differences in the cell numbers produced (Fig. 1A, center and right panels). Thus, an increase of ~2.2 divisions in mDD (Fig. 3D) with the accompanying ~0.5 division increase in standard deviation (fig. S5, A to D, and table S5) summed from three weak costimuli resulted in a net ~8-fold increase in the peak cell number (Fig. 3, E and F), with the additional difference in response magnitude attributable to small variations in the starting cell number (fig. S5E and table S5). Early programming was cell intrinsic because cells imprinted with an mDD of ~1 or ~3.4 generations gave the same outcome irrespective of whether they were subsequently cultured separately or together (fig. S6, A and B). The approximately additive effect of stimuli on DD also applied for a range of combinations when stimuli were retained in culture during subsequent division rounds (fig. S7, A to D).

Fig. 3 Summation of DD from multiple costimuli geometrically amplifies the T cell response.

Percentage cohort number (Fig. 2A) (19) versus mean division number for N4-stimulated, CTV-labeled OT-I/Bcl2l11−/− CD8+ T cells cultured with (A) antibodies to CD27 (immobilized, 5 μg/mL) and CD28 (2 μg/mL) or (B) IL-12 protein (1 ng/mL) alone, (C) antibodies to CD27 and CD28 together, or (D) a combination of all three costimuli for 26 hours, washed and recultured without further stimulation. Relationship between cell number and either (E) mean division number or (F) time for data in [(A) to (D)]. Arrows represent the effect of individual stimuli on mDD. All cultures contained S4B6 at 25 μg/mL. Graphs are representative of three independent experiments; mean ± SEM of triplicate culture wells.

Taken together, this series of experiments reveals two stages of regulation of T cell DD. In the first stage, signal 1 and a series of signal 2 and 3 stimuli of different strengths and combinations can additively “program” a heritable number of division rounds before the first cell division. In the second stage, exposure to external signals, mainly cytokines, can be processed and added to the DD. These features are consistent with a molecular mechanism whereby each stimulatory signal contributes a quantum of mitosis-promoting protein or complex that is diluted by division until a submitotic concentration is reached and division ceases (26).

The two-stage DD programming model makes two key predictions for the role of extrinsic factors, such as IL-2, that could be tested during in vivo CD8+ T cell responses: (i) the major physiological role of autocrine IL-2 is in maintaining division, and therefore it will be more important away from the initial site of CD8+ T cell priming; and (ii) the effect of IL-2 on DD will sum with other stimuli, allowing the prediction of CD8+ T cell expansion kinetics when IL-2 and other stimuli are combined.

We tested the first prediction by comparing the expansion of IL-2 receptor α–deficient OT-I CD8+ T cells (OT-I/Il2ra−/−) with OT-I/Il2ra+/+ CD8+ T cells in two different in vivo systems, namely an anti-influenza response and islet graft rejection model. Similar numbers of OT-I/Il2ra+/+ and OT-I/Il2ra−/− CD8+ T cells were detected at the site of priming [mediastinal lymph node (mLN)] during the expansion phase when cotransferred into HKx31-N4 infected mice (Fig. 4A). In contrast, a bias toward expansion of OT-I/Il2ra+/+ CD8+ T cells was observed in the spleen and lungs, consistent with a role for IL-2 in the maintenance of cell expansion (Fig. 4A). OT-I/Il2ra+/+ CD8+ T cells also outcompeted OT-I/Il2ra−/− CD8+ T cells at the effector site during an anti-islet graft response (Fig. 4B). The proportion of bromodeoxyuridine-positive (BrdU+) OT-I/Il2ra+/+ CD8+ T cells after a 1-hour in vivo BrdU pulse was ~2.5 times as high as for OT-I/Il2ra−/− CD8+ T cells in the graft (Fig. 4C), confirming that this bias was attributable to proliferation in the effector site and not due to migration alone (27).

Fig. 4 In vivo summation of DD can predict T cell expansion.

(A) Ratio of OT-I/Il2ra+/+:OT-I/Il2ra−/− CD8+ T cells recovered from mLN, spleen, and lungs after cotransfer of equal numbers of each into HKx31-N4–infected recipients. Mean ± SEM, n = 5 to 10 mice per time point, pooled from two independent experiments; two-way analysis of variance (ANOVA). (B and C) CTV-labeled OT-I/Il2ra+/+ and OT-I/Il2ra–/– CD8+ T cells were co-transferred to recipient mice at a 30:70 ratio 1 day before engraftment under the renal capsule with islets expressing membrane-bound ovalbumin under the rat insulin promoter. Mice were pulsed with BrdU 1 hour before organ harvesting on day 6 after engraftment, and the ratio of OT-I/Il2ra+/+:OT-I/Il2ra–/– of (B) divided CD8+ T cells and (C) the percentage of divided CD8+ T cells positive for BrdU+ within their respective genotype populations was measured in the renal (draining) lymph node (rLN) and graft. Mean ± SEM, n = 4 mice, representative of two independent experiments; two-way ANOVA and one-tailed t test, respectively. For all experiments, data points were excluded from ratio and percentage calculations when less than 100 cells were detected. Dotted lines in (A) and (B) represent the OT-I/Il2ra+/+:OT-I/Il2ra−/− transfer ratio. (D) The Cyton model was fitted to OT-I/Il2ra+/+ and OT-I/Il2ra−/−CD8+ T cell numbers from HKx31-Q4–infected mice and OT-I/Il2ra−/− CD8+ T cells from HKx31-N4–infected mice pooled from the mLN, spleen, and lungs after cotransfer of equal numbers of cells into recipient mice, and (E) the division destiny distribution was determined (top) (19). By summation of mean and variance of these DD distributions, the cumulative effect of increasing TCR-antigen affinity and IL-2 signaling on the DD distribution (i.e., the DD distribution for OT-I/Il2ra+/+ CD8+ T cells in HKx31-N4–infected mice) was predicted (bottom). (F) Using this mean and variance, the cell number over time was predicted for OT-I/Il2ra+/+ CD8+ T cells in HKx31-N4–infected mice (19). Mean ± SEM; n = 5 mice per time point, representative of two independent experiments. ****P < 0.0001; ***P < 0.001; **P < 0.01; *P < 0.05.

To investigate the additive nature of T cell stimuli in vivo, OT-I/Il2ra+/+ and OT-I/Il2ra−/− CD8+ T cells were cotransferred into mice infected with either high-affinity (HKx31-N4) or low-affinity (HKx31-Q4) influenza virus (20). The Cyton model was fitted to T cell numbers to estimate the increase in DD due to TCR affinity or IL-2 signaling alone (Fig. 4, D and E, top, and table S6). By summation of these individual contributions to the mean and variance of the DD distribution, we predicted the effect of a combined increase in TCR affinity and IL-2 signaling on DD (Fig. 4E, bottom) and successfully recreated the expansion kinetics of OT-I/Il2ra+/+ CD8+ T cells during an HKx31-N4 infection (Fig. 4F).

Manipulating T cell responses by costimulation and cytokine signaling is an important emerging therapeutic regimen (28, 29), and a quantitative framework will facilitate the rational development of optimal interventions. To support this goal, we propose a quantitative paradigm where no one signal is obligatory but rather combinations of inputs add together to geometrically enhance outcomes (Fig. 1A). Thus, combinations of different costimulatory and cytokine signals provide many alternative paths to generate T cell responses of similar magnitude. This framework reconciles long-standing discrepancies between in vivo and in vitro results for IL-2 and costimulatory signals and reveals a quantitative basis for current switch-inspired two- and three-signal models of activation. Further studies measuring simultaneous differentiation changes to effector and memory states associated with cell division would complete the T cell calculating framework and further enhance our ability to predict therapeutic strategies for immunomodulation.

Supplementary Materials

Materials and Methods

Figs. S1 to S10

Tables S1 to S9

References (3032)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Because OT-I division times are relatively homogeneous, a division-based dilution model is difficult to distinguish from a mechanism where the loss of division motivation ceases in all descendants at the same time after activation. Further molecular studies are needed to distinguish between these and other possible mechanisms.
  3. Acknowledgments: We thank S. Turner for providing the recombinant influenza virus and discussion and comments on the manuscript; L. Corcoran and V. Bryant for discussion and comments on the manuscript; A. Giang for discussion of experiments; M. Camilleri, L. Mackiewicz, M. Dayton, and D. Pavlyshyn for technical assistance; and P. Bouillet for Bcl2l11−/− mice. Fucci Red and Fucci Green mice were provided by the Riken BioResource Centre through the National BioResource Project of the Ministry of Education, Culture, Sports, Science, and Technology of Japan. We also thank K. Lafferty and his two-signal theory for inspiring this work. The data are tabulated in the main paper and in the supplementary materials. Experimental data and the code for fitting the Cyton model will be made available upon request. This work was supported by the National Health and Medical Research Council (NHMRC) through project grants 1010654 and 1043414, program grants 1054925 and 1037321, and fellowships to M.R.D., A.M.L., and P.D.H., K.R.D. and P.D.H. were supported by Human Frontier Science Program grant RGP0060/2012, G.T.B. was supported by an Australian Research Council Future Fellowship, and K.R.D. was supported by Science Foundation Ireland grant 12IP1263. This work was made possible through Victorian State Government Operational Infrastructure Support and Australian Government NHMRC Independent Research Institutes Infrastructure Support Scheme. J.M.M. was the recipient of an Australian Postgraduate Award and a Walter and Eliza Hall Institute Edith Moffat Scholarship.
View Abstract

Stay Connected to Science

Navigate This Article