Decomposing virulence to understand bacterial clearance in persistent infections

Abstract

Following an infection, hosts cannot always clear the pathogen, instead either dying or surviving with a persistent infection. Such variation is ecologically and evolutionarily important because it can affect infection prevalence and transmission, and virulence evolution. However, the factors causing variation in infection outcomes, and the relationship between clearance and virulence are not well understood. Here we show that sustained persistent infection and clearance are both possible outcomes across bacterial species showing a range of virulence in Drosophila melanogaster. Variation in virulence arises because of differences in the two components of virulence: bacterial infection intensity inside the host (exploitation), and the amount of damage caused per bacterium (per parasite pathogenicity). As early-phase exploitation increased, clearance rates later in the infection decreased, whereas there was no apparent effect of per parasite pathogenicity on clearance rates. Variation in infection outcomes is thereby determined by how virulence - and its components - relate to the rate of pathogen clearance. Taken together we demonstrate that the virulence decomposition framework is broadly applicable and can provide valuable insights into host-pathogen interactions.

Fig. 1. Decomposing virulence.

Both exploitation and per parasite pathogenicity (PPP) can harm the host and thereby contribute towards virulence. Exploitation describes the infection intensity, or parasite/pathogen load, inside the hos PPP describes the damage per parasite that an infection does to the host. Here there is variation among parasite species in exploitation and PPP, as illustrated by hypothetical relationships between host fitness and infection intensity for two species of parasite infecting the same host genetic background. a The parasite species have the same PPP but differ in exploitation. Parasite 1 has lower exploitation compared with parasite 2, because it causes a lower infection intensity. b In contrast with a the parasite species have the same average exploitation but species 1 has lower PPP because its reaction norm has a shallower slope. This means that compared with species 2, species 1 causes less damage to the host with increasing parasite load. Figure modified from Råberg.

Fig. 2. Framework and hypotheses for the relationship between per parasite pathogenicity (PPP), exploitation, virulence and clearance rate.

a–e Host efforts to clear the pathogen generate costs and benefits via changes in a clearance rate and c immunity costs. b, d Clearance rate and immunity costs affect host survival and, e, in combination lead to an optimal host clearance effort (vertical dotted line) that maximises host survival. Host clearance effort is conceptualised as the strength of the resistance mechanisms that act towards pathogen clearance. f Schematic for how PPP, exploitation, and host clearance effort affect virulence and clearance rate: combined, these determine the three infection outcomes. This scheme extends the framework proposed by Råberg & Stjernman, which proposes that PPP and exploitation are different determinants of virulence. The green and yellow labelled arrows represent our hypotheses and predictions. g–j Illustration of hypotheses H1 and H2, which are based on the idea that PPP and exploitation can each increase virulence. Two scenarios are depicted: low PPP or exploitation, and high PPP or exploitation (see legend). g The relationship between host clearance effort and clearance rate is not affected by high/low PPP or exploitation. h However, increased PPP or exploitation lower host survival. i These differential survival benefits of host clearance effort lead to the hypotheses that increases in PPP (H1) and exploitation (H2) lead to an increase in the optimal host clearance effort. j As a result, increased clearance rates are predicted with increasing PPP and exploitation (predictions P1 and P2 respectively). k–n Illustration of hypothesis H3, which is based on the idea that in addition to affecting virulence (g–j), higher exploitation makes it harder for the host to clear the infection. k Thus, for a given host clearance effort the clearance rate should be lower for higher exploitation, which results in lower host survival (l). m When considering total host survival, these differential survival benefits do not necessarily affect the optimal host clearance effort. n However, due to its direct influence on clearance rate, higher exploitation is predicted to lead to a lower clearance rate (P3). See Supplementary Note.

Fig. 3. Fly survival and virulence after injection with one of four bacterial species.

a–d Survival curves after injection with four bacterial species, each at one of five doses. Controls were injected with Ringer’s solution or received no injection (naïve). The legend in panel a shows the treatments for all survival curves, where CFU denotes colony forming units. Flies were housed in vials in groups of six. In the following, n = the number of flies examined over three independent experiments and n is given in the order of the treatments as shown in the legend: Enterobacter cloacae n = 99, 86, 79, 98, 96, 89, 92; Providencia burhodogranariea n = 98, 91, 101, 89, 103, 103, 105; Lactococcus lactis n = 96, 97, 102, 104, 108, 106, 107; Pseudomonas entomophila n = 92, 97, 101, 102, 95, 96, 102. e Virulence measured as the natural log of maximum hazard for all bacterial species, where each data point is the maximum hazard between zero- and 20-days post injection, calculated from one experimental replicate per bacterial dose. In the following, n = the number of experimental replicates/doses per bacteria: E. cloacae n = 14, all other bacteria n  =  15. Ringer’s injected and naïve flies are not included. Black lines show means and standard errors. We fitted a linear model (two-tailed) where the main effect of bacterial species was p < 0.0001 (see main text). See Supplementary Table 1 for multiple comparisons and effect sizes. Different letters denote means that are significantly different from one another.

Fig. 4. Bacterial load per living fly after injection with one of four bacterial species.

a–d Flies were injected with either Enterobacter cloacae, Providencia burhodogranariea, Lactococcus lactis or Pseudomonas entomophila and then homogenised at between 1- and 35-days post injection. The injection dose legend for all panels is shown in d where CFU denotes colony forming units. The arrows on the y-axis indicate the approximate injection doses. Each data point is the bacterial load of one fly. Reducing data points on the x-axis are due to increasing fly death over time. Sample sizes can be found in Supplementary Table 3a–d. Black lines show geometric means.

Fig. 5. Virulence decomposition.

a Pathogen exploitation given as infection intensity/bacterial load across species. Each data point is one of five injection doses per bacterial species, per experimental replicate, and gives the geometric mean of bacterial load for days one and two post injection (denoted as _1,2), where Enterobacter cloacae n = 13, Providencia burhodogranariea = 15, Lactococcus lactis = 10. The circles are jittered along the x-axis to aid visualisation of overlapping data points. Black lines show means and standard errors. We fitted a linear model (two-tailed) where the main effect of bacterial species was p < 0.0001 (see main text). Different letters denote means that are significantly different from one another. See Supplementary Table 4 for multiple comparisons and effect sizes. b PPP given as the relationship between bacterial load and the inverse of maximum hazard, so that the virulence increases with proximity to the x-axis. The bacterial load data is the same as that given in a but with the addition of the Ringer’s control group, giving the following total sample sizes Enterobacter cloacae n = 15, Providencia burhodogranariea = 18, Lactococcus lactis = 12. To allow inclusion of the uninfected Ringer’s control group to the figure, we added one CFU to all mean bacterial load values. The natural log of maximum hazard data is estimated from survival data for the corresponding injection doses and experimental replicates (as in Fig. 3e). Lines show linear regressions with 95 % confidence intervals.

Fig. 6. Bacterial clearance by living flies.

The proportion of live flies that were found to be uninfected (for bacteria detection limit see Methods) at different times post injection with a Enterobacter cloacae, b Providencia burhodogranariea, c Lactococcus lactis or d Pseudomonas entomophila. Each column shows a different injection dose. Numbers above the bars indicate the total numbers of flies from which the proportions were calculated, i.e., the total numbers of flies homogenised. Zeros on the x-axis mean that there were no flies alive from which to assess clearance.

Fig. 7. Effect of bacterial species and exploitation on clearance rate.

For all figures, each data point is from one injection dose per bacteria, per experimental replicate, and gives the mean proportion of cleared infections (out of the initial infected population) on days three and four (clearance index_3,4) or days seven, 14 and 21 (clearance index_7,14,21). a Mean species differences in clearance index_3,4. For Pseudomonas entomophila n = 6, for Enterobacter cloacae, Providencia burhodogranariea and Lactococcus lactis n = 15. The circles are jittered along the x-axis to aid visualisation of overlapping data points. Black lines show means and standard errors. We fitted a linear model (two-tailed) where the main effect of bacterial species was p = 0.0016 (see main text). Different letters denote means that are significantly different from one another (Mann-Whitney-U post hoc tests, two-tailed tests, see main text for statistical results). The effect of pathogen exploitation, given as bacterial load, upon b mean clearance index_3,4 and c mean clearance index_7,14,21. The geometric mean of bacterial load was calculated from days 1 and 2 post injection (denoted as _1,2), i.e., the same values as in Fig. 4. The negative relationships between the two variables are shown with 95 % confidence intervals. Statistics are given in the main text and the species legend for both panels is shown in c. The negative relationships in b and c support H3 and P3 (Fig. 2k, n).