The end of Koch's postulates

How do you know whether disease X is the result of infection by pathogen Y? 

In 1890 Robert Koch published a list of 'postulates' which would form the foundation of ascribing disease causation from that day - until now. Though Koch established his list based upon his studies of bacteria, more specifically Mycobacterium tuberculosis and Bacillus anthracis, (the causative agents of tuberculosis and anthrax respectively), his postulates could be applied more widely. No more would infectious diseases be merely regarded as mysterious happenings. Scientists could now pursue and nail down the cause of a disease by fulfilling Koch's postulates. There are many ways of expressing the same thing, but Wikipaedia describes them as:

1) The microorganism must be found in abundance in all organisms suffering from the disease, but should not be found in healthy organisms. 

2) The microorganism must be isolated from a diseased organism and grown in pure culture.

3) The cultured microorganism should cause disease when introduced into a healthy organism.

4) The microorganism must be reisolated from the inoculated, diseased experimental host and identified as being identical to the original specific causative agent.

Association vs. causation will always be a key issue when it comes to diagnosis and Koch's postulates, as important as they are, soon start to fall apart under closer scrutiny. For instance, what happens in circumstances where the organism can't be isolated in culture - a not unusual occurrence in the world of virology? 

The most significant problems occur though simply due to words such as 'must'. Infectious disease is never so clear cut. Koch himself acknowledged as much during his studies of cholera; whilst the causative agent, Vibrio cholerae, could be isolated from people with cholera, it could also be isolated from healthy people. This fails the first postulate (depending on how loosely you apply 'should'). 

What about the same agent in different species? The dogma with Bluetongue virus is that it is asymptomatic in cattle, but can be lethal in sheep. Even with sheep though, only a few of those which become infected actually show bluetongue disease. The fact that the term 'case fatality' exists is an acknowledgement itself that only a fraction of animals which become infected fall ill. You could argue that this is an irrelevance as, even if it doesn't cause disease in every animal that it infects, it is still the causative agent of those which do become sick. Even though there haven't been enormous numbers of cases reported (presumably because there's no evidence of disease), I would put a lot of money on the fact that a large proportion of cows in southern England are currently seropositive for Schmallenberg virus.

In turn, when attempts are made to fulfill postulate 3, it would be perfectly normal, expected even, that an animal doesn't become sick.   

Discussions of these limitations are not new. The biggest development that has prompted the questioning of Koch's postulates has been molecular biology. The extreme sensitivity of nucleic acid-based methods means that evidence of pathogens can be detected at extremely low levels. Now it is possible to go searching for viruses with relative ease, increasing the speed and efficiency of virus detection and discovery. There are many approaches; PCR is arguably the most common, and metagenomics the most recent. I compare the methods to fishing: firstly, it’s possible to go fishing for specific viruses using PCR. Metagenomics on the other hand is more like a deep sea trawler – sequencing everything within a sample and looking within the ‘catch’ for virus sequences. Using the latter approach has allowed the identification of numerous viruses, and this is reflected in the number of publications, as revealed in a paper by Mokili et al. (2012).

Viral metagenomic studies between 2002 and 2011. From Molili et al 2012.

The paper by Mokili et al. (2012) also discusses the fact that molecular approaches remove the need for the growth of an agent in pure culture. Significantly, there is also the acknowledgement that determining causality merely by a pathogen’s presence is difficult to achieve, particularly with metagenomics approaches which are capable of finding a diverse array of agents within the same sample. The authors go on to describe a ‘metagenomic Koch’s postulates' approach whereby the metagenomes of individuals are compared. The postulates are:

1) The diseased metagenome must be significantly different from the healthy control and contain a greater abundance of the suspected metagenomic traits.

2) Inoculation of a healthy individual with a sample from the diseased individual must result in disease state.

3) Selected, specific samples containing the suspected traits from the individual infected for step 2 must cause disease when injected into another healthy individual.

Metagenomic Koch's postulates as described by Mokili et al 2012. Comparison between a diseased and healthy control animal shows a significant difference between the metagenomic libraries (depicted by the histograms of relative abundance reads). In order to fulfill the metagenomic Koch's postulates: (1) The metagenomic traits in diseased subject must be significantly different from healthy subject. For example traits A, D, E and J found in the disease animal that are not present in the healthy control; (2) Inoculation of samples from the disease animal into the healthy control must lead to the induction of the disease state. Comparison of the metagenomes before and after inoculation should suggest the acquisition or increase of new metagenomic traits (A, E and P). New traits can be purified by methods such as serial dilution or time-point sampling of specimens from a disease animal. (3) Inoculation of the suspected purified traits into a healthy animal will induce disease if the traits form the etiology of the disease

Following this procedure allows the sequence associated with disease, i.e. a biomarker of the etiological agent, to be discovered by the process of elimination.

Metagenomics will, rightly, more than likely become established as the method of choice for diagnosis. The technology is still developing, but not too far in the future I suspect metagenomics will follow a similar path to that of real-time PCR into the molecular diagnostics setting. Instead of testing a samples against some 'likely suspects' using PCR approaches, it will be possible to get a complete picture of the complex 'virome' associated with that sample.

Although the 'metagenomics Koch's postulates' are a step towards linking metagenomics to disease, there is still the issue of ‘must cause disease’; if this ‘disease’ by definition involves clinical signs, then similarly to Koch’s original postulates this may fail. However, the approach does allow the picking apart of causality in complex scenarios where multiple pathogens are present. There are still issues to be ironed out as to how molecular data is interpreted, but this is going to become increasingly important as metagenomic approaches become even more widespread. From this perspective, it appears that Koch’s postulates, landmarks and revolutionary as they may have been, may be nearing the end of their life.

Mokili, J., Rohwer, F., & Dutilh, B. (2012). Metagenomics and future perspectives in virus discovery Current Opinion in Virology, 2 (1), 63-77 DOI: 10.1016/j.coviro.2011.12.004