C.perfringens incidence and clinical factors associated with its colonisation
Faecal samples and complete clinical notes were available for 333 infants. The demographics of these babies are shown in Table 1. C. perfringens was isolated in faecal samples from 98 of the infants (29.4%).
Colonisation data were used to predict the risk of colonisation of the infant gut by C. perfringens over time through a survival analysis (Fig 1). By the day of life of the median stay on the neonatal unit for the cohort (29 days), a predicted 36% of infants would be colonised (95% confidence band 25%, 43%).
We next repeated the survival analysis with the aim of determining clinical variables (shown in Table 1) that were associated with C. perfringens colonisation. A series of univariate models were created and significantly associated variables (after multiple hypothesis correction) are shown in Table 2.
For each significant factor, values are provided at each of its quartiles to illustrate the spread of data. The exponentiated coefficients provide the change in risk of colonisation per unit of each factor. Risk change indicates the relative change in risk of colonisation between the minimum value (0%) and the 75% quartile for a given clinical factor. Abbreviations: CI, confidence interval.
Given the potential for correlation between these variables, a multivariate survival analysis was used to identify a minimal set of clinical factors to best predict colonisation. Four factors were found to remain significant in this model, with associations between increased probability of C. perfringens colonisation and fewer days of CPAP with supplemental oxygen (CPAP oxygen), fewer days of maternal milk feeds (via feeding tube), fewer days of breast feeding and fewer days of antibiotics over the course of the infant’s admission. The variation of colonisation probabilities when the infant cohort is divided into quartiles for each of these factors is shown in Fig 2.
As the risk of colonisation for each infant is associated with the combined effects of each of these factors, we performed a multivariate analysis where each infant in the cohort was annotated either “low” (<= median) or “high” (>median) for each of the four factors. Multiple sets of analyses were run, illustrating the effects of higher than average measure of a single factor, or combinations thereof, on the probability of colonisation. The results are shown in Fig 3.
The dominant variable of the four appeared to be feeds with breast milk, with increased feeds being associated with the greatest shift towards low probability of C. perfringens colonisation. We theorised two modes of action for this association, with breast milk acting either directly (inhibiting the growth of C. perfringens) or indirectly (promoting the growth of other components of the gastro-intestinal microbiota which outcompete C. perfringens). We explored these possibilities in culture experiments as follows.
Growth of C. perfringens in breast milk
A C. perfringens isolate was grown in either nutrient rich medium (supplemented WCB broth) or breast milk, either as a monoculture or in co-culture with Bifidobacterium infantis which was chosen to represent a typical competing gut species, one that thrives on breast milk oligosaccharides (19-21). Both species grew in each substrate and under both culture conditions, and each grew significantly better in rich medium under monoculture than in breast milk (B. infantis, p = 0.008, C. perfringens, p < 0.0001). There was no significant difference in growth rate between the two species in the rich media. However, in breast milk C. perfringens grew at a significantly lower rate than B. infantis in monoculture (p = 0.003) and co-culture (p = 0.0001). Growth of C. perfringens appeared lower in co-culture than monoculture when grown in breast milk, but the reduction was not significant (p = 0.11) (Fig 4).
Toxic potential of C. perfringens isolates
The harmful effects of C. perfringens arise in large part through the production of toxins. Of particular interest in the neonatal field is the potential for C. perfringens toxin to play a part in the pathogenesis of necrotising enterocolitis (NEC). We surveyed our cohort for the presence of toxin genes through targeted PCR and found that a range of toxin genes were present in C. perfringens isolates. The presence of toxin genes in C. perfringens isolates during any point in their admission was scored for each neonate, allowing comparison of the toxic potential of the isolates prior to either discharge from the neonatal intensive care unit (“Control infants”) or NEC incidence (“NEC Infants”) (see Fig 5).
Statistical analyses (Barnard's test and survival analysis) found no significant associations between the presence of toxin genes and the development of NEC. Given the infant numbers available in our cohort, Barnard's test would detect a proportional increase of 0.19 or greater in the occurrence of toxin in infants developing NEC compared to controls with 95% confidence (assuming a one-sided test and 80% power). These results would therefore do not support the hypothesis of NEC being associated with the prevalence of a particular toxin gene across our infant cohort.