Heterogeneity in killing efficacy of individual effector CD8+ T cells against Plasmodium liver stages

Abstract

Vaccination strategies in mice inducing high numbers of memory CD8⁺ T cells specific to a single epitope are able to provide sterilizing protection against infection with Plasmodium sporozoites. We have recently found that Plasmodium-specific CD8⁺ T cells cluster around sporozoite-infected hepatocytes but whether such clusters are important in elimination of the parasite remains incompletely understood. Here, we used our previously generated data in which we employed intravital microscopy to longitudinally image 32 green fluorescent protein (GFP)-expressing Plasmodium yoelii parasites in livers of mice that had received activated Plasmodium-specific CD8⁺ T cells after sporozoite infection. We found significant heterogeneity in the dynamics of the normalized GFP signal from the parasites (termed 'vitality index' or VI) that was weakly correlated with the number of T cells near the parasite. We also found that a simple model assuming mass-action, additive killing by T cells well describes the VI dynamics for most parasites and predicts a highly variable killing efficacy by individual T cells. Given our estimated median per capita kill rate of k = 0.031/h we predict that a single T cell is typically incapable of killing a parasite within the 48 h lifespan of the liver stage in mice. Stochastic simulations of T cell clustering and killing of the liver stage also suggested that: (i) three or more T cells per infected hepatocyte are required to ensure sterilizing protection; (ii) both variability in killing efficacy of individual T cells and resistance to killing by individual parasites may contribute to the observed variability in VI decline, and (iii) the stable VI of some clustered parasites cannot be explained by measurement noise. Taken together, our analysis for the first time provides estimates of efficiency at which individual CD8⁺ T cells eliminate intracellular parasitic infection in vivo.

Keywords: CD8+ T cell; Plasmodium sporozoites; intravital imaging; liver stages; mathematical modelling.

Figure 1.

Experimental design to measure T cell impact on Plasmodium liver stages. (a) Experimental design. Mice were infected with 3 × 10⁵ GFP-expressing Py sporozoites and divided into two groups as control or treatment. In the control group, mice were left until imaging started 24–28 h after the infection. In the treatment group, 9 × 10⁶ Py-specific activated CD8⁺ T cells (PyTCR) were transferred into infected mice 20 h after infection. Four to eight hours later, murine livers were surgically exposed and z-stacks of images were taken using a spinning disc confocal microscope [17]. The VI of a given parasite was calculated as the ${\log }_{10}$ of the ratio between brightness of the GFP signal from the parasite to the brightness of the background based on maximal z projections. The number of T cells within 40 μm of each liver stage was also calculated. (b(i),b(ii)): Example of the VI dynamics for control (b(i)) or treatment (b(ii)) groups. ‘Superficial’ and ‘deep’ denote location of the parasite based on the z coordinate in the imaging data. (b(iii)) Change in the number of Py-specific CD8⁺ T cells near the Py-infected hepatocytes over time since T cell transfer. See electronic supplementary material, figure S1 for all data graphs.

Figure 2.

Higher CD8⁺ T cell numbers per parasite are negatively correlated with parasite VI. (a) Correlation between parasite VI and T cell numbers per parasite is shown for individual mice. The Spearman rank correlation coefficients ρ and corresponding p-value were calculated using the scipy.stat package in python and are shown on individual panels. (b) Spearman rank correlation between parasite VI and CD8⁺ T cell numbers per parasite is shown for all parasites for which such correlation existed along with the estimated p-value from the correlation test (19/32). Dashed horizontal line denotes p = 0.05. (c(i)) Prediction of the best-fit linear mixed-effects model assuming T cell numbers per parasite as a random effect (equation (2.6)) are shown for individual T cell number : mouse combinations. Negative values for effect (weight) correspond to T cell numbers that reduce VI over time. (c(ii)) Prediction of the best fit linear mixed-effects model assuming mouse as a random effect (equation (2.5)) are shown for individual mice (negative and positive effect corresponds to the parasite’s VI declining or increasing, respectively). In panels (c), grey and red colours denote positive and negative weight values, respectively.

Figure 3.

The best-fit model suggests a relatively low killing efficacy of CD8⁺ T cells in vivo. (a) Schematic of the model depicting the essential interaction between CD8⁺ T cells (T) and VI (V_t) where k, r and n correspond to killing efficiency per T cell, growth rate of the parasites, and parameter indicating competition (n < 1) or cooperation (n > 1) between T cells located near the same parasite, respectively. Alternative models (defined in equation (2.12) with n = 1 and n ≠ 1) were fitted to the data on VI dynamics while taking into account change in the number of CD8⁺ T cells per parasite as predicted by the DDR model (electronic supplementary material, figure S7). (b–f) Experimental data on VI change in a subset of parasites are shown by markers and lines are the best fit predictions of the simplest, mass-action model (equation (2.12)). Estimated killing rate k is shown on individual panels, and all fits are shown in electronic supplementary material, figure S8. The dotted horizontal line denotes the death threshold VI = 0.2. (g) Estimates of per capita T cell killing efficacy along with predicted 95% confidence intervals is plotted against the time-average of the number of T cells near each liver stage for n = 24 parasites which had at least one T cell at some time during imaging. Estimates with k = 0 were assigned a value of 10⁻³/h (limit of detection, LOD) to display on the log-scale. Median estimates for the killing rate for individual mice or all data together (dashed line) are indicated on the panel. (h) Distribution of the estimated killing rates from g shown as a histogram. (i) Stochastic simulations on resampling of estimated killing efficacy per T cell from h predicting the probability of liver stage elimination within different time intervals (48 h, 24 h or 12 h) as the function of the number of T cells in the cluster. Horizontal dashed line denotes 80% probability.

Figure 4.

Constancy of parasite VI is not consistent with stochastic simulations of killing by T cells. We simulated parasite VI dynamics using the Gillespie algorithm for 20 h with a small growth rate r = 0.0001/h and then added killing by T cells at a per capita rate k = 0.031/h with the total kill rate being determined by the number of T cells per parasite increasing with time (see Materials and methods for more detail). (a,c) We plot trajectory from 250 simulations (grey lines) with the average change (red line) calculated from every time step of 0.02 hour. T cell dynamics was added in the simulations using equation (2.11) with parameters estimated for individual parasites ((a,b) ID = 4, (c,d) ID = 12). Black lines in panels (a,c) show the actual change in VI over time as was observed in the data for parasite IDs 4 (a) and 12 (c). (b,d) For every simulation we calculate the decay rate of the VI and show the distribution of decay rate with probability density function (blue line). The actual decay rate for the VI in the data for these two parasites is shown by the horizontal black dotted line. Inserts show the T cell dynamics as assumed by the clustering model. Grey shaded areas denote the imaging period which we used to calculate the VI decay rate. Simulations were done using package gillespy2 with Tau-Hybrid solver in python. None of the VI slopes in simulations were above the VI slope in the data suggesting p < 0.01 for comparing simulations with the data.