Plasticity and genetic variation in traits underpinning asexual replication of the rodent malaria parasite, Plasmodium chabaudi

Fifty-six C57BL/6 female mice (aged 6–8 weeks) were obtained in-house (University of Edinburgh) and infected with P. chabaudi clones AJ, AS, CR and ER from the Edinburgh Malaria reagent repository (University of Edinburgh). Plasmodium chabaudi was isolated between 1948 and 1974 from African thicket rats, Thamnomys spp., in Central Africa [29]. After cloning, the parasite genotypes have been cryopreserved and undergone regular transmission through mosquitoes to maintain their wild type phenotypes [30]. The four P. chabaudi genotypes used span the diversity of virulence (in terms of host anaemia and weight loss) reported from previous experiments [31‐33]. We first describe the main experiment exploring genotype by environment interactions in asexual traits, and then describe a parallel experiment testing the effect of our environmental perturbation on the immune environment.

First, hosts were allocated to anaemia or control treatment groups. Anaemia was induced by administering 30 mg/kg phenylhydrazine (PHZ) in PBS carrier via intra-peritoneal injection and controls received only the carrier. PHZ causes premature lysis of RBCs through denaturation of haemoglobin, and, in consequence, induces an inflow of immature RBCs (reticulocytes) with an EPO-mediated feedback loop [34]. Inducing host anaemia with PHZ is a useful environmental perturbation because, unlike simply following infections through their natural course, the availability and age-structure of RBC resources can be manipulated independently of confounders resulting from the progression of infections (e.g., immune responses, parasite density). Four days post PHZ or carrier administration, all mice were infected with 5 × 10⁶ parasitized RBCs. Briefly, 7 hosts for each of the two treatment groups were infected with clones AJ, AS, CR or ER. All PHZ treatments and genotypes were randomized across 10 cages, with 8 cages each containing 6 mice and 2 cages each containing 4 mice. From day 1 to 5 post infection (PI), mice were monitored daily at 08.00 (GMT) by taking 2 μl of blood to quantify RBC density (Coulter Technologies) and making a thin blood smear to quantify asexual stages, the age structure of RBCs (measured as the proportion of RBCs that were reticulocytes, which are larger and stain bluer with Giemsa than mature RBCs), and the proportion of infected reticulocytes. At midnight on day 4 PI, a blood smear was made from each mouse to count the number of merozoites inside schizonts (burst size). From 08.00 on day 3 PI to 08.00 on day 5 PI, blood smears were made for each mouse every 8 h to assess the relative abundance of the developmental stages (staging them as either ring stage, trophozoite or schizont) to estimate the duration of the asexual cycle.

Next, a separate set of forty mice, host strain-, age-, and sex-matched, were injected with 30 mg/kg PHZ (n = 23) or PBS carrier (n = 17, control treatment) to test whether PHZ has an effect on the innate immune environment encountered by parasites on the day of infection. Specifically, the cytokines TNF-alpha and IFN-gamma, which kill asexual stages either directly or indirectly through the recruitment of macrophages [35], were assayed by ELISA (see Additional file 1 ‘Immune assays’ for further details).

Asexual densities were calculated by multiplying the proportion of infected RBCs (estimated from blood smears) with the density of RBCs per ml of blood. Infections in PHZ-treated hosts reached peak on day 4 PI, thus to ensure that all infections were examined in their growth phase, densities from day 5 are not included in the replication rate analyses. Burst size was quantified for between 10 and 15 schizonts on each smear. Asexual cycle duration was estimated using a Bayesian-based computational model. Briefly, the model is based on a mathematical description of the phase distribution of parasites within their developmental cycle, classifying each as one of three stages (ring stage, trophozoite or schizont) at 8-h sampling points over 2 days. The joint likelihood of the model parameters was calculated using broad priors, which contributed little to the posterior because of the periodicity and the quantity of data. Finally, a Markov Chain Monte Carlo method was used to sample the posterior, which is highly suitable for sampling complex posteriors of high-dimensional, non-linear, multivariate dynamical systems (see Additional file 1 ‘Estimating cycle duration’ and ‘Inference of model parameters’ for further details). Invasion rates were calculated for reticulocytes and normocytes separately using equations 1 and 2 in [3] and reflect the probability that a parasite will invade an RBC and survive long enough to complete its cycle. These calculations account for the age structure and density of the available RBCs before and after invasion, as well as of the density of merozoites before invasion. Invasion rates were calculated only on day 4 PI because this was when the proportion of infected reticulocytes was high enough to return reliable estimates. To estimate the density of merozoites, the average burst size for each infection was multiplied by the density of ring stage parasites on day 3 PI. This model assumes that merozoite lifespan is equal across genotypes and PHZ combinations and that no death of uninfected RBCs takes place before or during parasite invasion, so that the total density of RBCs on day 4 PI is equal to the density of RBCs that the parasites were exposed to at the start of the invasion cycle.

All statistical analyses were carried out using R (version 3.2.3). For burst size and asexual cycle duration, linear models (2-way-ANOVAs) were constructed relating the dependent variable of interest to genotype and PHZ treatment. The density of schizonts with a high burst size and invasion rates required square root transformation whereas reticulocyte proportion, asexual density and replication rate required log10 transformation. Maximum-likelihood based deletion tests were used to assess the effects of excluding each fixed effect. The dynamics (patterns observed during infections) of the proportion of RBCs that are reticulocytes, density of RBCs/ml blood, invasion rates, and asexual density/ml blood were analysed with linear mixed effect (LME) models, including PHZ treatment, genotype, and day PI, as fixed effects and with mouse identity as a random effect. The significance of fixed effects in LME models was determined by comparing the model including and the model excluding the fixed effect of interest using a likelihood ratio test. In minimized models where genotype or any interaction with genotype remained. Finally, a linear model was constructed to estimate the relative importance of traits (burst size, cycle duration, invasion rates, genotype) contributing to replication rate. For each statistical model, diagnostic plots of residuals were examined to assess whether linear modelling assumptions were satisfied [36, 37].

First, the experiment assumed parasites in PHZ-treated hosts enter and experience more anaemic within-host environments than parasites in control hosts. This requirement was met: all four genotypes of P. chabaudi experienced non-overlapping RBC environments between control and PHZ treatments (Fig. 1, Table 1). On the day before infection, regardless of which parasite genotype hosts were destined to be infected with, PHZ hosts exhibited significantly reduced RBC density and significantly increased proportion of RBCs that were reticulocytes (Fig. 1a, b). Specifically, mean total RBC density in control hosts was 8.42 billion cells (± 0.20) per ml, compared to 6.87 billion (± 0.22) in PHZ-treated hosts. Reticulocyte proportion was increased on average from 0.023 (± 0.006) to 0.077 (± 0.007) by PHZ. The RBC environments also remained significantly different between control and PHZ-treated hosts during the experiment (Fig. 1c, d). There was however, a significant genotype by day interaction for reticulocyte proportion dynamics, driven by a faster increase in AJ/AS infections compared to CR/ER infections, in PHZ-treated hosts.

Second, the assumption that PHZ caused a much stronger perturbation to RBC metrics than to immune responses was met. The standards in each ELISA assay yielded the expected concentrations of IFN-gamma and TNF-alpha. For IFN-gamma, one mouse was excluded because of poor repeatability between duplicates and two mice were positive but very low concentrations were detected (2.79 and 0.299 pg/ml) and belonged to a PHZ-treated group and a control group, respectively. For TNF-alpha, all mice showed good repeatability and were positive but very low concentrations were detected (15.6 and 20 pg/ml) and belonged to a PHZ-treated group and a control group, respectively. Note, the mice in which IFN-gamma was detected differed from those in which TNF-alpha was detected. Thus, parasites within PHZ or control groups all encountered very similar inflammatory cytokine environments upon and during infections.

Asexual replication was significantly affected by whether parasites infected PHZ-treated or control hosts in a genotype-specific manner (genotype by PHZ interaction, Fig. 2, Table 2), with AJ exhibiting the greatest increase in densities in anaemic hosts.

Burst size could not be estimated for 3 mice because fewer than 10 mature schizonts were found on their blood smears. The effect of host anaemia on burst size did not differ across the genotypes (genotype by PHZ treatment F(3,51) = 0.45, p = 0.718, Table 3). However, burst size was significantly greater in PHZ-treated hosts (PHZ: F(1,55) = 46.25, p < 0.001, Fig. 3a), and lower in CR than the other genotypes (t value = − 2.57, p = 0.01). The minimal model explained 45.1% of the observed variance in burst sizes.

To investigate whether higher burst sizes can be explained by a higher proportion of reticulocytes, the proportion of infected reticulocytes observed on day 4 PI was compared to the density of schizonts with burst size > 6. There is a non-significant effect of reticulocyte proportion on the density of schizonts with high burst size (F(1,21) = 3.36, p = 0.08), with the proportion of infected reticulocytes in the blood only explaining 10.1% of the variance in the density of high burst-size schizonts. Overall, the proportion of infected reticulocytes in the blood of PHZ-treated hosts was on average only 1.1% (± 0.2), yet the total proportion of infected RBCs was 24.9% (± 0.03) by day 4 PI, suggesting it is unlikely that burst size is only determined by the age of the RBC a parasite resides in.

The duration of the asexual cycle is approximately 24 h in P. chabaudi [21, 38]. Asexual cycle duration varies depending on PHZ treatment (F(1,52) = 5.91, p = 0.019) and parasite genotype (F(3,54) = 6.22, p = 0.001), but with no significant interaction between the two (F(3,51) = 1.54, p = 0.22) (Fig. 3b). In general, the asexual cycle duration is at least 30 min longer in PHZ-treated hosts (especially for AJ, AS and CR) except ER which has a faster cycle than the others (Table 4). Overall, the minimal model explained 27.2% of the total variance in cycle duration.

The method used to calculate invasion rates can cause some estimates to be negative due to associated random measurement error between variables. In this data set, 5 out of 56 mice returned negative rates for both reticulocytes and normocytes and one mouse negative for only normocytes and were therefore excluded from this particular analysis. However, excluding or including these data points affected the results only quantitatively, not qualitatively. Invasion rates for both normocytes (Fig. 4a) and reticulocytes (Fig. 4b) varied but could not be explained by RBC age, PHZ treatment or parasite genotype, fitted either as main effects or within interactions (Table 5).

A model was constructed to ask whether the plasticity and genetic variation detected in asexual traits explain the different asexual dynamics exhibited by the genotypes in control and PHZ-treated hosts. Asexual dynamics were summarized as the change in density between day 1 to 4 PI (referred to as replication rate) and the model included main effects for PHZ, burst size and cycle duration, reticulocyte and normocyte invasion rates, and their interactions with parasite genotype (higher order interaction terms were omitted so as to guarantee sufficient degrees of freedom). The model simplified to include only main effects for genotype and burst size (Table 6), explaining 46.4% of the variance observed in replication rate. Asexual replication was positively associated with burst size for all genotypes, with AJ and AS generally achieving higher replication rates than CR and ER (see Fig. 2).