Co-infections determine patterns of mortality in a population exposed to parasite infection

Highly protective effect of co-infections on mortality due to East Coast fever and consequences for disease epidemiology and control.


This PDF file includes:
Fig. S1. Relationship of subclinical, clinical, and fatal infections to clinical variables. Fig. S2. Relationship between clinical outcome and LPT prevalence. Table S1. Conditional logistic regression analyses of clinical predictors. Table S2. Impact of setting LPT prevalence at age a, L(a), on subsequent acute ECF death rate. References (33,34)  Bovine Interleukin 10 ELISA Kit, Cat No. CSB-E12917B for IL10 (sign-reversed optical density as this is an inhibition assay whereas the TGFβ1 test is a capture assay).

Comparison of diagnostic test results
The performance of the serological tests, PCR-based tests and microscopy for detecting Theileria species were compared as follows: RLB vs serology. Of the 105 calves in the case-control study, 100 were T. parva positive by RLB and/or serology (for the remainder ECF was only diagnosed at post mortem). For 82% of these calves, infection was first detected by RLB. Those first detected by serology alone were all subclinical infections.
Seroconversion appeared delayed for calves that were RLB positive at 6 weeks.
Of the case-control study calves 88 were T. mutans positive by RLB and/or serology. T. mutans infection was first detected by RLB in 93% of these. We also explicitly compared T. mutans status at T0 (see Materials and Methods) by RLB and serology in order to evaluate serology as a marker of current infection. For 100 of the calves in the case-control study, there was 79% agreement; however, for all but one of the nine RLB negative/serology positive calves the data were consistent with there being an older infection that had been cleared, and for all but one of the 12 RLB positive/serology negative calves the data were consistent with the infection being too recent for seroconversion to have occurred. This suggests a level of agreement between serology and current T. mutans infection in the range 79-98%. RLB vs p104. No calf was positive by p104 that was never positive by RLB. However, for two calves the age of first infection with T. parva was moved forward by the p104 test (by 15 and 16 days).
RLB vs microscopy. There were 376 samples that were Theileria positive by RLB and/or microscopy. Of these, 311 (83%) were positive by RLB. There were 29 calves that were RLB negative but positive by microscopy on at least one occasion prior to seroconversion to T. parva. These calves were not at altered The above results suggest that the sensitivity of the RLB test is in the range 80-90%. The protective effect reported in the main text is robust to imperfect sensitivity in the observed range: if RLB test sensitivity is assumed to be 90% and the number of LPT positives scaled accordingly we obtain odds ratio=0.18, P=0.004; assuming 80% sensitivity we obtain odds ratio=0.25, P=0.011.

Clinical severity
Analysis of the relationship between outcome and haematological and immunological variables were confined to case-control study calves where T. parva was detected prior to seroconversion, i.e. 19 cases and 49 controls. Included variables were: white blood cell count (WBC), total serum protein (TSP), packed cell volume (PCV), platelet count (PLT) (log 10 transformed), IL10 and TGFβ1 (see above), all measured at T0. Univariate conditional logistic regression analyses were used to explore the relationships between each of these variables and acute ECF death (Table S1). In addition, principal components analysis (PCA) was applied to haematological and immunological data. PCA was implemented using PC-ORD software version 6.03 (MJM software Design, Gleneden Beach, OR). The first three principal components were significant (p<0.05) and explained 41%, 31% and 17% of the total variance respectively (Fig. S1). Four variables were loading (>0.4) on PCA1 (PCV, logPLT, TSP, IL10), with WBC and TGFβ1 loading (>0.4) on PCA2 and PCA3 respectively. As the principal components are not equally important in measuring overall clinical severity, PCA1-3 were combined to create a single composite index. Using the proportion of these percentages as weights on the score coefficients, a nonstandardized index (NSI) was calculated using the following formula: NSI = (40.6/88.3)x(Factor 1 score) + (30.8/88.3)x(Factor 2 score) + (16.9/88.3)x(Factor 3 score).
The non-standardized 3-axis index (NSI3) from the PCA was the single best predictor of acute ECF death (Table S1) and was used as an index of clinical "severity" in subsequent analyses. The fit of a univariate model of parasite load is also reported in Table S1. This indicates that high visible parasite load in the blood is associated with increased mortality risk, as previously reported (16,17). and 80% confidence ellipses (R pgl package scatterplot 3D) illustrating the results of the principal components analysis on a subset of N=68 calves using six variables: platelet count (PLT, log transformed); total serum protein (TSP); packed cell volume (PCV), white blood cell count (WBC); and levels of the cytokines IL10 and TGFβ1. There were three significant principal components: PCA1, PCA2 and PCA3 which explained 40.6%, 30.8% and 16.9% of the total variation in the data respectively. Individual data points and confidence ellipses distinguish calves that died (red), had clinical ECF but survived (orange), and remained healthy throughout (green).

Fig. S2. Relationship between clinical outcome and LPT prevalence.
All T. parva seronegative calves from the case-control study are grouped into those that died (N=19), experienced clinical illness (N=26) and those that remained healthy (N=37). Observed LPT prevalence in surviving calves (with 95% CIs) is compared with model-predicted LPT prevalence (see Fig. 3C). Overall, surviving calves have close to expected LPT prevalence, but the presence or absence of LPT partly sorts the calves into healthy or sick respectively (general linear model, overall model: p<0.001, post hoc test for differences between clinically ill and dead: p=0.32).  Table 1). Odds ratios give the effect of the predictor on the odds of death by ECF. AICc is the corrected Akaike Information Criterion. Bold type indicates models with relative probabilities ≥ 0.05.
Statistics derived from principal components analysis used as predictors are: the first principal component (PCA1); and the non-standardized 2-and 3-axis indices (NSI2 and NSI3 respectively). NSI3 is the best predictor but packed cell volume, total serum protein and PCA1 all have relative probabilities >0.05. A univariate model with parasite load is compared.

Predictor
Odds ratio (95% CIs) P value AICc Predicted acute ECF death rates are compared for different ages and three different scenarios: all infected, L(a)=1; none infected, L(a)=0, and L(a) as expected using baseline model parameters (Table   3). Odds ratios ( The data used in the analyses reported here are openly accessible at http://datashare.is.ed.ac.uk/. The data are presented in a single file, IDEALSciAdvdata.csv. The data comprise records of observations made at every visit to all 548 calves included in the IDEAL study cohort and identifies the calves and visits that contribute to each of the major analysis described in the main text. In addition, we have placed biological samples (of sera and, where appropriate, other tissues) collected from the calves in a repository. The samples can be linked to the data files (via the calf and visit ID numbers) and are available to researchers on request (subject to international shipping regulations).