Causes of variation in malaria infection dynamics: insights from theory and data

Abstract

Parasite strategies for exploiting host resources are key determinants of disease severity (i.e., virulence) and infectiousness (i.e., transmission between hosts). By iterating the development of theory and empirical tests, we investigated whether variation in parasite traits across two genetically distinct clones of the rodent malaria parasite, Plasmodium chabaudi, explains differences in within-host infection dynamics and virulence. First, we experimentally tested key predictions of our earlier modeling work. As predicted, the more virulent genotype produced more progeny parasites per infected cell (burst size), but in contrast to predictions, invasion rates of red blood cells (RBCs) did not differ between the genotypes studied. Second, we further developed theory by confronting our earlier model with these new data, testing a new set of models that incorporate more biological realism, and developing novel theoretical tools for identifying differences between parasite genotypes. Overall, we found robust evidence that differences in burst sizes contribute to variation in dynamics and that differential interactions between parasites and host immune responses also play a role. In contrast to previous work, our model predicts that RBC age structure is not important for explaining dynamics. Integrating theory and empirical tests is a potentially powerful way of progressing understanding of disease biology.

Figure 1. Schematic of immune clearance rates of parasitized red blood cells (RBCs).

Fitted model parameters are labeled for one of the two “windows” of immune activity. These include day postinfection when immune clearance begins (s_p), maximum clearance rate (c_p), rise time to maximum rate (r_p), and total duration of immune activity window (l_p). We allow for two windows of immune activity since experimental infection data show two clear peaks of parasites that could each potentially have a unique, specific immune response. For further details, see appendix A.

Figure 2. Predictions from the model of Mideo et al. (; left) and experimental data (right) for invasion rates (A) and burst sizes (B).

A, left, The invasion rate of reticulocytes (Retic) is predicted to be similar for both parasite genotypes and much lower than for mature red blood cells (RBCs; Normo). The more virulent genotype (AS, triangles) is predicted to have a higher invasion rate of mature RBCs than the less virulent genotype (DK, circles). A, right, We find little support for our predictions. While invasion rates of both genotypes are similar in reticulocytes, they are an order of magnitude higher than expected, and higher invasion rates in mature RBCs across genotypes were not observed. Means of log₁₀-transformed invasion rate estimates × merozoite life span (assumed to be 30 min) are presented from four or five individual infections. Values along the Y-axis have units of cells per microliter. B, left, Burst sizes are predicted to be higher for AS (more virulent) than for DK (less virulent). The model predicts that AS has a higher burst size in reticulocytes than in mature RBCs, so the average burst size in AS infections should increase when the proportion of infected RBCs that are reticulocytes is experimentally increased in a subset of hosts, while the opposite is predicted for DK. B, right, As predicted, AS had a higher burst size in control infections, but increasing the proportion of reticulocytes did not result in the predicted changes. Means are calculated from average burst sizes in four or five individual infections from each treatment. Error bars = ±1 SEM.

Figure 3. Experimental and predicted infection dynamics of a relatively more virulent genotype (AS, top row) and less virulent genotype (DK, bottom row).

A, Average densities ± 1 SEM for red blood cells (RBCs; solid lines, filled circles) and parasites (dashed lines, open circles). B, Infection dynamics as predicted by the model of Mideo et al. (2008b), using experimentally estimated invasion rates and average burst sizes (in control mice, col. 3, table 1). All other parameters are set to the genotype-specific values published in Mideo et al. (2008b). Secondary parasite peaks are predicted to occur with greater frequency and magnitude than is seen in the data. Even more extreme and less congruent results occur when burst sizes are set to the age-specific values listed in the last two columns of table 1, since these high reticulocyte burst sizes further enhance parasite growth when RBC densities start to rebound and the relative density of reticulocytes increases (not shown). C, Infection dynamics as predicted when invasion rates are set to the published best estimates in Mideo et al. (2008b) rather than experimentally estimated values. It this case, the reticulocyte invasion rate is an order of magnitude lower than for mature RBCs, which slows and dampens secondary parasite peaks.

Figure 4. The next iteration of model fitting.

Fits of reduced hybrid model (the “most likely” model) to newly collected parasite and red blood cell (RBC) densities. Crosses are data, gray regions correspond to 95% posterior predictive intervals of the predicted dynamics, and solid lines give the best-fit solutions for each individual mouse.

Figure 5. Parasite genotype-specific model inferences.

Plots show marginal distributions of parasite genotype-level (hyper)parameters. Histograms represent 4,000 best-fit estimates from refitting the model to all data. By comparing the fit of the most likely model that allows for these genotype-specific parameters, to the model that includes only a single parameter for both genotypes, the support for genotype-specific differences can be determined. We find that there is important variation in burst sizes across genotypes, but not in the upregulation of host red blood cell production or in invasion rates.