Campylobacter jejuni and Campylobacter coli: Dose Response Models

Campylobacter jejuni and Campylobacter coli

Author: Kyle S. Enger

Overview

Campylobacter are microaerophilic gram-negative curved or spiral rods with a polar flagellum. Gastroenteritides are typically caused by C. jejuni and C. coli . It can cause acute self-limiting diarrhea in healthy humans with an incubation period of 2-3d, and appears very common worldwide. It is mainly a zoonosis, being primarily associated with birds (especially poultry). They do not grow in water and (like Escherichia coli) are an indicator of post-treatment contamination in water distribution systems.

According to feeding studies with chickens, strains of C. jejuni that have been passaged many times in the laboratory tend to have a lower ID₅₀ than strains that are isolated from infected hosts and then used to infect new hosts, with minimal passage . Minimally passaged strains also had more variation in ID₅₀ . Given safety concerns, strains used for human studies may be passaged and studied more, possibly underestimating infectiousness in actual human exposure scenarios .

Campylobacter epidemiology varies greatly between the developed and developing world, probably due to development of immunity early in life. Illness is rare after about 5 years of age (or earlier) in developing countries, but occurs among adults in industrialized countries, probably because they avoided exposure (and therefore immunity) in childhood . However, immunity appears to protect against disease rather than infection, and asymptomatic shedding is common . In a comparison of Mexican children <4y and Swedish patients (ages not given), Swedish patients tended to carry only 1 Campylobacter serotype, while mixed serotypes were common among Mexican children (42%)

---

Summary of data

Blaser et al. (1983) experimentally infected adult female HA-ICR mice intragastrically with 3 different serotypes of C. jejuni (strains T1, T2, and T3).

Chen et al. (2006) describe dose response models fitted to data from feeding studies in chickens with 19 different strains (18 C. jejuni, 1 C. coli). They found that isolates that had been passaged multiple times in the laboratory were more infectious than isolates taken from infected animals and reused with minimal passaging. Also, one passaged strain (11168) was substantially less infectious than the other passaged strains, which all had very similar dose response curves.

Black et al. (1988) fed human volunteers 2 different strains (81-176 and A3249) of C. jejuni suspended in 150 mL nonfat milk. Neither strain showed an increasing trend of illness with dose. The A3249 strain showed an increasing trend of infection with dose; however, all volunteers became infected with the 81-176 strain regardless of dose. The data describing infection with the A3249 strain were fit by Medema et al. (1996). Teunis et al. (1999) pooled all illness data from Black et al. (1988) to fit a model for strains 81-176 and A3249, which remained similar to Medema et al. (1996). This model was later elaborated by Teunis et al. (2005) by including information from 2 outbreaks of C. jejuni in contaminated milk.

Tribble et al. (2009) fed healthy adult human volunteers with the CG8421 strain of C. jejuni, along with bicarbonate. However, only 2 doses were used, and all but 1 volunteer became ill.

Tribble et al. (2010) experimentally infected humans with strain 81-176 of C. jejuni, one of the strains used by Black et al. (1988). Although Black et al. (1988) administered the dose in milk, Tribble et al. (2010) administered the dose in a solution containing 2g of bicarbonate. All volunteers were infected, but a dose response trend was seen in the development of disease.

Experiment serial number Reference Host type Agent strain Route # of doses Dose units Response Best fit model Optimized parameter(s) LD50/ID50 106* human strain A3249 oral (in milk) 6 CFU infection beta-Poisson α= 1.44E-01 , N50 = 8.9E+02 8.9E+02 184 mice type strain for serotype PEN 1 intragastric 5 CFU infection exponential k = 9.01E-07 7.69E+05 185 mice type strain for serotype PEN 2 intragastric 5 CFU infection beta-Poisson α = 3.19E-01 , N50 = 6.68E+04 6.68E+04 186 {{{host}}} type strain for serotype PEN 3 intragastric 5 CFU infection beta-Poisson α = 1.17E-01 , N50 = 3.14E+04 3.14E+04 188 human strain 81-176 oral (w. 2g NaHCO3) 3 CFU campylobacteriosis beta-Poisson α = 1.66E-01 , N50 = 1.23E+05 1.23E+05 *This model is preferred in most circumstances. However, consider all available models to decide which one is most appropriate for your analysis.

---

Optimization Output for experiment 106

Strain A3249 model data Dose Infected Non-infected Total 810 5 5 10 8100 6 4 10 91000 11 2 13 810000 8 3 11 1.1E+06 15 4 19 1.1E+08 5 0 5

Goodness of fit and model selection Model Deviance Δ Degrees of freedom χ20.95,1 p-value χ20.95,m-k p-value Exponential 110 108 5 3.84 0 11.1 0 Beta Poisson 2.43 4 9.49 0.658 Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.

Optimized parameters for the beta-Poisson model, from 10000 bootstrap iterations Parameter MLE estimate Percentiles 0.5% 2.5% 5% 95% 97.5% 99.5% α 1.44E-01 2.05E-02 3.61E-02 4.99E-02 2.66E-01 2.98E-01 3.71E-01 N50 8.9E+02 6.54E-10 1.47E-04 8.11E-02 6.69E+03 8.97E+03 1.53E+04

Parameter scatter plot for beta Poisson model ellipses signify the 0.9, 0.95 and 0.99 confidence of the parameters.

beta Poisson model plot, with confidence bounds around optimized model

Optimization Output for experiment 184

T1 strain for serotype PEN 1 data Dose Infected Non-infected Total 1 0 5 5 100 0 5 5 1E+04 0 5 5 1E+06 3 2 5 1E+08 5 0 5

Goodness of fit and model selection Model Deviance Δ Degrees of freedom χ20.95,1 p-value χ20.95,m-k p-value Exponential 0.0918 -0.000599 4 3.84 1 9.49 0.999 Beta Poisson 0.0924 3 7.81 0.993 Exponential is preferred to beta-Poisson; cannot reject good fit for exponential.

Optimized k parameter for the exponential model, from 10000 bootstrap iterations Parameter MLE estimate Percentiles 0.5% 2.5% 5% 95% 97.5% 99.5% k 9.01E-07 4.61E-08 2.21E-07 2.21E-07 4.61E-06 4.61E-06 4.61E-06 ID50/LD50/ETC* 7.69E+05 1.51E+05 1.51E+05 1.51E+05 3.14E+06 3.14E+06 1.51E+07 *Not a parameter of the exponential model; however, it facilitates comparison with other models.

Parameter histogram for exponential model (uncertainty of the parameter)

Exponential model plot, with confidence bounds around optimized model

Optimization Output for experiment 185

T2 Strain for serotype PEN 2 model data Dose Infected Non-infected Total 1E+04 1 4 5 1E+05 3 2 5 1E+06 4 1 5 1E+07 4 1 5 1E+08 5 0 5

Goodness of fit and model selection Model Deviance Δ Degrees of freedom χ20.95,1 p-value χ20.95,m-k p-value Exponential 25.4 24.5 4 3.84 7.59e-07 9.49 4.13e-05 Beta Poisson 0.969 3 7.81 0.809 Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.

Optimized parameters for the beta-Poisson model, from 10000 bootstrap iterations Parameter MLE estimate Percentiles 0.5% 2.5% 5% 95% 97.5% 99.5% α 3.19E-01 8.29E-02 1.26E-01 1.49E-01 1.05E+01 5.53E+02 1.52E+03 N50 6.68E+04 4.06E+02 5.55E+03 1.04E+04 3.58E+05 4.65E+05 8.48E+05

Parameter scatter plot for beta Poisson model ellipses signify the 0.9, 0.95 and 0.99 confidence of the parameters.

beta Poisson model plot, with confidence bounds around optimized model

Optimization Output for experiment 186

T3 strain for serotype PEN 3 data Dose Infected Non-infected Total 1E+04 2 3 5 1E+05 4 1 5 1E+06 2 3 5 1E+07 3 2 5 1E+08 5 0 5

Goodness of fit and model selection Model Deviance Δ Degrees of freedom χ20.95,1 p-value χ20.95,m-k p-value Exponential 45.6 40.2 4 3.84 2.25e-10 9.49 2.92e-09 Beta Poisson 5.41 3 7.81 0.144 Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.

Optimized parameters for the beta-Poisson model, from 10000 bootstrap iterations Parameter MLE estimate Percentiles 0.5% 2.5% 5% 95% 97.5% 99.5% α 1.17E-01 1.06E-02 1.79E-02 2.48E-02 2.79E-01 3.29E-01 4.72E-01 N50 3.14E+04 3.29E-09 2.55E-05 2.53E-02 4.21E+05 9.69E+05 3.68E+06

Parameter scatter plot for beta Poisson model ellipses signify the 0.9, 0.95 and 0.99 confidence of the parameters.

beta Poisson model plot, with confidence bounds around optimized model

Optimization Output for experiment 188

Strain 81-176 model data Dose Campylobacteriosis Non-campylobacteriosis Total 1E+05 3 2 5 1E+07 2 3 5 1E+09 33 3 36

Goodness of fit and model selection Model Deviance Δ Degrees of freedom χ20.95,1 p-value χ20.95,m-k p-value Exponential 50.1 46.6 2 3.84 8.56e-12 5.99 1.29e-11 Beta Poisson 3.51 1 3.84 0.061 Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.

Optimized parameters for the beta-Poisson model, from 10000 bootstrap iterations Parameter MLE estimate Percentiles 0.5% 2.5% 5% 95% 97.5% 99.5% α 1.66E-01 2.92E-02 4.29E-02 6.44E-02 3.32E-01 4.07E-01 1.16E+00 N50 1.23E+05 6.24E-10 4.60E-05 9.04E-01 2.00E+06 6.12E+06 3.96E+07

Parameter scatter plot for beta Poisson model ellipses signify the 0.9, 0.95 and 0.99 confidence of the parameters.

beta Poisson model plot, with confidence bounds around optimized model