Stochastic variation in the initial phase of bacterial infection predicts the probability of survival in D. melanogaster

  1. David Duneau  Is a corresponding author
  2. Jean-Baptiste Ferdy
  3. Jonathan Revah
  4. Hannah Kondolf
  5. Gerardo A Ortiz
  6. Brian P Lazzaro
  7. Nicolas Buchon  Is a corresponding author
  1. Cornell University, United States
  2. UMR5174 EDB, CNRS, ENSFEA, Université Toulouse 3 Paul Sabatier, France
7 figures, 2 tables and 2 additional files

Figures

Figure 1 with 2 supplements
The outcome of infection ranges from 100% to 0% survival.

(A) Canton S flies were injected with the same inoculum (OD600 = 1, ca. 30,000 bacteria) of different bacteria species. Providencia alcalifaciens (n = 29) and Serratia marcescens (n = 30) are lethal …

https://doi.org/10.7554/eLife.28298.003
Figure 1—figure supplement 1
The variability of outcome in infection is not due to variation in the time of exposure to CO2 before and after injection.

We monitored the survival after infection of groups either exposed to CO2 for 2 min during the injection, exposed to 15 min of CO2 before an injection time frame of 2 min, or injected during a time …

https://doi.org/10.7554/eLife.28298.004
Figure 1—figure supplement 2
The chronic phase of infection is characterised by the Set Point Bacterial Load (SPBL).

(A) The Set-Point Bacterial Load (SPBL) is stable up to 10 days post-injection (linear regression: Time: df = 1, F = 0.11, p=0.73). The grey dot represents the dose initiating the infection and the …

https://doi.org/10.7554/eLife.28298.005
Figure 2 with 1 supplement
Hosts die at a set bacterial burden, the BLUD, which varies with host and pathogen.

(A) The bacterial load in flies within 30 min of death from P. rettgeri infection is a constant and is higher than that of flies that did not die from the infection. This is true when comparing dead …

https://doi.org/10.7554/eLife.28298.009
Figure 2—figure supplement 1
The BLUD is constant over time and does not depend on the initial dose.

(A) BLUD several days post-injection. The BLUD is consistent even over 10 days post-injection. (B) Effect of initial dose on BLUD. For both Canton S and Oregon R hosts, the BLUD for P. rettgeri was …

https://doi.org/10.7554/eLife.28298.010
Figure 3 with 1 supplement
The infection outcome depends on inter-individual variation in within-host bacterial proliferation.

(A) For bacteria that do not kill any infected hosts, bacterial loads decreased soon after the beginning of the infection. Grey dots on the x-axis represent individuals without detectable bacteria. …

https://doi.org/10.7554/eLife.28298.013
Figure 3—figure supplement 1
In vivo proliferation in the early phase of infection is equivalent to the rate of in vitro growth in LB medium.

During the eight first hours, bacterial proliferation in LB medium (Culture 1 to 4) did not differ from the bacterial proliferation within the host Canton S. This was confirmed by the …

https://doi.org/10.7554/eLife.28298.014
Early intra-host bacterial evolution does not explain the dual outcome of infection.

Bacterial evolution inside the fly does not explain the binary outcomes of infection. (A) Quantification of bacterial load to determine bacterial populations killing the host (bacterial load close …

https://doi.org/10.7554/eLife.28298.017
Role of the immune system in the infection dynamic.

(A) Within-host P. rettgeri loads at different times post-injection for wild-type (Canton S) flies and flies deficient in phagocytosis (Hml-Gal4 >UAS GFP; UAS-Bax, Gal80ts). Distinct groups of …

https://doi.org/10.7554/eLife.28298.019
Figure 6 with 1 supplement
A generalized model of infection.

(A) A schematic representation of a conceptual model for within-host bacterial growth dynamics that, upon three phases, lead either to host survival or death. In a first early phase, the bacterial …

https://doi.org/10.7554/eLife.28298.021
Figure 6—figure supplement 1
A simulation of the WHD model.

(A) Red curves correspond to the predicted average log2-transformed bacterial load in the Baranyi model (the sigmoid curve) and in the bacterial decrease model. Shades of grey indicate the …

https://doi.org/10.7554/eLife.28298.022
Figure 7 with 1 supplement
The time to effective control by the immune response determines the outcome of infection.

(A) Bacterial load over time in hosts with pre-activated immune systems (Gal80TS; daughterless-Gal4 > UAS Imd) and in the corresponding control flies (Gal80TS; daughterless-Gal4 > Canton s). Early …

https://doi.org/10.7554/eLife.28298.024
Figure 7—figure supplement 1
Correlation between the number of P. rettgeri in head and abdomen of the same fly (Spearman correlation test, S = 633.31, r = 0.84, p=8.8e-09).
https://doi.org/10.7554/eLife.28298.025

Tables

Table 2
Effect of early growth (tlag, µ and σb) and control (c, Vc and ntip) parameters on the capacity of the model to predict survival.

Log-likelihood is computed either on bacterial load data (i.e. on the data set we used to fit the model) or on survival data. In this latter case, we used the probability of control predicted by the …

https://doi.org/10.7554/eLife.28298.028
Complete data setIntermediate survival
Bacterial loadSurvivalBacterial loadSurvival
dflogLiklogLikp-valuelogLiklogLikp-value
Full−1243.119−17.85860−800.0603−11.48770
Control24−1629.965−21.684080.999−839.5523−17.183190.496
c8−1660.396−19.189410.954−839.5791−16.825340.030
Vc8−1643.985−17.376631−826.4296−14.677600.173
ntip8−1634.382−25.233970.064−832.8896−12.855960.603
Growth24−1798.630−81.133556.512e-16−854.1100−11.377391
tlag8−1692.743−16.490761−835.3833−11.303971
µ8−1705.175−61.560151.564e-15−836.3794−11.754060.970
σb8−1704.156−21.952700.415−848.2682−12.284870.810
Table 1
List of parameters of the mixture model with their signification.
https://doi.org/10.7554/eLife.28298.029
Baranyi model
n0Bacterial load upon injection
nmaxMaximum bacterial load
tlagLag time
µEarly bacterial growth rate
σbStandard deviation of loads in the absence of control
Exponential model
ncIntercept of the exponential decrease model
δDecrease rate in bacterial load when infection is controlled
σcStandard deviation of loads in controlled infections
Control
cAverage time to control
VcVariance in time to control
ntipBacterial load above which the host cannot control infection

Additional files

Download links