The microbiome concept has altered the way we perceive the relationship between microbes and their hosts.
We have moved from a view that all microbes are bad to a recognition that all organisms on the planet interact with a wide variety of microbes, some harmful, others benign and some positively helpful to their host. This shifting view has had a major impact on the science of farmed animals, where clinical and subclinical infections can have a major impact on productivity and sustainability. The prevailing view in the dairy industry was that the ideal state of the dairy cow mammary gland was for it to be sterile and that any bacteria present in the gland were bad for the animal and bad for productivity. However, a number of studies in cows, sheep and humans have consistently detected microbes in milk samples and inside the glands of slaughtered animals. The fact that bacteria are naturally present in animals with healthy udders, as well as those with acutely infected udders, raises the possibility that there are specific bacteria that are consistently found in the mammary gland microbiome, which if true, might protect an animal from mastitis. Therefore we ask whether there are probiotic bacteria that could protect against mastitis.
There is now a well-established idea that there is a core of microbes that are almost always found in sites like the gut and vagina, where they play a positive role in host health and wellbeing. Therefore, it would seem reasonable to expect a similar situation in the mammary gland, although not all agree. It has been argued, based on the physiology and immunology of the bovine mammary gland that it would not maintain a consistent microbiome. These two contrasting views have not been resolved but the prospect of a cheap and easy prophylactic treatment for mastitis would make an enormous difference to dairy farm productivity, reduce antibiotic use and improve sustainability.
At Warwick, we have studied mammary gland health and the bacteria in milk samples taken from the end of the lactation period (drying off), through the birth of a calf and into the first month of lactation. This longitudinal study enabled us to determine whether there were patterns in the disease status of individual mammary glands over time. A latent class analysis, which finds groups in datasets, produced eight groups with consistent patterns in the changing innate immune status of the mammary gland. These ranged from a class that were consistently healthy to one that was subclinically mastitic. These classes were exciting as they were potentially disease status markers that might highlight udders that needed a probiotic anti-mastitis bacterial inoculation.
Therefore, we analysed the detected bacterial communities in each sample using a sophisticated clustering method, which showed distinct separation in latent classes, especially of the extreme healthy and subclinically diseased classes. This was very exciting; it seemed different latent classes contained different bacteria – evidence for a protective microbial community, maybe? However, we then analysed individual cows within a healthy latent class and found that each cow was distinct, as were individual mammary glands within a class. This surprised us as we expected at least some of the cows and glands to be somewhat consistent in the clustering if there was a core bacterial community in healthy glands. Also, the species that were important in the analysis were mostly rare in the total dataset. This is a classic sign that the data is being ‘over-fitted’, meaning patterns are produced using data that amplify differences excessively. This can be checked, as in a good analysis a similar pattern should be produced when using only part of the data, a method called cross-validation. Our analysis failed this test! In essence, we had found ‘patterns’ in our data but could not reproduce them, meaning we could not trust this analysis. Moreover, we had found clusters with all the data, then more clusters within a latent class and within each mammary gland – layers of complexity on complexity, a fractal-like pattern. Therefore, we need to reassess and reanalyse our data, using methods that protect against over-fitting; only then will we be able to assess whether there are patterns in the data that would support the idea of a core microbiome in the dairy cow mammary gland.
The alternative, that there is no consistent microbial community in the mammary gland, could challenge the idea that all microbiomes are beneficial to their host.