• inflammation
• B cells
• T cells

Rheumatic diseases are a diverse group of conditions in which the host and the environment interact to drive inflammation and autoreactivity. In this dyadic model, the strongest environmental influence is infection, with bacteria and viruses likely to trigger disease in a host predisposed by genetics; infection can either non-specifically stimulate the immune system to heighten the propensity for autoreactivity or specifically stimulate B and T cell autoreactivity by molecular mimicry. For many years, this venerable model has powerfully influenced the design of experiments in both the human and animal systems as well as their interpretation.

While the ‘one organism, one disease’ mechanism may pertain to at least some rheumatic diseases (eg, rheumatic fever following Streptococcus infection), for most other conditions, pathogenesis is more complicated, at least in part because one of the major sources of foreign organisms—bacteria, viruses, fungi and others—resides in or around the host as the microbiome. The microbiome is a huge biomass, with the mammalian host containing as many prokaryotic as eukaryotic cells; the number is in the trillions. Many studies have therefore addressed whether the microbiome can influence the occurrence of not only rheumatic diseases but also metabolic, cardiovascular and neuropsychiatric conditions, among others.1 2

As shown recently in a provocative study by Rosshart et al 3, the host and the environment are not truly distinct, with data presented suggesting that the microbiome and the host have coevolved to produce an optimal balance of positive and negative effects. In their drive to study disease in a more reductionist way, investigators have likely disrupted this balance by breeding genetically inbred mice in very clean environments. Although a clean environment would seemingly reduce potential ‘contaminating’ influences of infection, it is very much unnatural and leads to a microbiome that itself can alter host responses and skew the mechanisms on pathogenesis in large and unpredictable ways.

The study by Rosshart and colleagues3 represents an important step in understanding how the microbiome can influence disease by exploring a very novel model system called ‘wildling’ mice. The opposite of germ-free mice, wildling mice are the offspring of a pseudopregnant wild mouse dam which has received embryos from a C57BL/6 mouse by surgical transfer. This approach allows the development of a laboratory strain mouse with the microbiome of a wild mouse exposed to a natural environment. In this case, the wild mice were trapped in barns in the Washington DC area. Like laboratory mice, the wild mice were of the Mus musculus domesticus species.

In a series of elegant experiments, the investigators compared the immunological responses of conventional C57BL/6 mice, the wildling mice and wild mice as well as their microbiomes. Since gut is the largest and most accessible of the microbiomes, the investigators analysed it in the most detail. In addition to bacteria, the gut microbiome also contains viruses, fungi and, in some cases, multicellular organisms such as helminths, but the enumeration of bacteria is the most straightforward by sequencing ribosomal genes.4 As the studies of Rosshart et al 3 showed, wildling mice significantly differ from laboratory mice in their microbiomes (gut, skin and vagina) and more closely resemble those of the wild mice in their size and diversity.

By the nature of this system, the relative contribution of genetics and the microbiome to shaping the immune system can be studied via a comparison between the immune cells of the laboratory, wildling and wild mice. Using mass spectroscopic techniques, Rosshart et al showed that the phenotypic properties of spleen immune cells appear to reflect primarily the microbiome, whereas the genome has a greater influence on the lymphocyte populations at sites such as the gut, skin and vagina. For peripheral blood, the pattern of gene expression in terms of transcriptional profile showed that wildling and wild mice are very similar despite genetic differences.

Wildling mice are not the only system that can help explore the impact of the microbiome on immune responses as well as disease. Studies of this kind go back many decades and began with the use of germ-free mice to elucidate the impact of infection (including colonisation) on various diseases. Other versions of this approach include treatment of mice with antibiotics; repopulation of germ-free mice with single or multiple organisms, including pathogens; and transfer of the microbiome of wild mice to conventional laboratory mice.5–9

While the microbiomes that develop in these different models vary, nevertheless, studies have clearly established that the microbiome can have a profound effect on immune responses. These effects are demonstrable in the response to infection, vaccination and development of autoimmunity in models such as the New Zealand mice.1 2 In this regard, the perspective in evaluating these effects is important since the microbiome can promote host defence against infection as well as predispose to the occurrence of autoimmune and inflammatory diseases.

Other evidence for the impact of the microbiome on immune responses concerns aspects for the use of animal models to study diseases that have been troubling to investigators. The first concerns reproducibility of findings from different laboratories, which is now a major concern for both mechanistic and translational research. Unfortunately, studies addressing the same question can sometimes arrive at quite different conclusions about the role of particular cell types, for example, in a phenomenon. Among causes of irreproducibility or inconsistency of experimental results, issues of animal husbandry are always considered possible, with the nature of the microbiome high on the list.4

Another troubling aspect of animal research concerns the difficulty in translating results of studies with mice to patients. The greatest difficulties with translation have occurred in the setting of sepsis. While studies using agents such as tumour necrosis factor (TNF)-α blockers to treat sepsis appeared very promising in mice, clinical trials failed.10 11 A failure to translate TNF-α blockers for sepsis is perhaps not surprising since, as shown in studies on gene transcription of immune cells, human and murine responses to challenges such as sepsis, trauma and burns show very little similarity. Of note, the responses of patients with these conditions can show common patterns of gene transcription.12

The divergent findings on the effects of TNF-α blockade in human and murine sepsis highlight a well-established fact: despite sharing of many genes, humans and mice have many differences in both innate and adaptive immune systems, the nature of their antibodies and even the composition of the blood in terms of different cells.13 Humans and mice diverged between 65 and 75 million years ago in evolution and inhabit very different environments. In an obvious case, mice live close to the ground and have intimate contact with soil as its microbial components. In humans, on the other hand, the breathing apparatus is safely in the air, reducing a constant barrage of dirt and its abundance of foreign organisms.

For the wildling model to have utility as a model for translation, it must display certain properties and most importantly predict responses of human subjects more accurately than those of conventional mice. The investigative team, therefore, conducted studies to explore the characteristics of the wildling mice as a model for translation, testing basic properties of the transferred wild microbiome, including its stability over time and its resiliency. These studies demonstrated that at least through the F5 generation, the gut microbiome of the wildling mice is stable, an important consideration in the intended use for translational research since their production takes time and skill. Other studies showed that a natural microbiome from wild mice is resilient and apparently better adapted than the microbiota of a conventional laboratory mouse.3 In its own way, the microbiome of the laboratory mouse is dysbiotic, since the transition from the wild to the laboratory has represented a huge jolt to the system, upsetting millions of years of evolution.

With data demonstrating that wildling mice have favourable properties as laboratory models, the next experiments concerned the ability of these mice to predict responses to therapeutic interventions. For this purpose, the investigators explored two systems in which studies in mice did not predict eventual results with human subjects. The first system tested involved the CD28 superagonist antibody (CD28SA). In studies with mice, the CD28SA dramatically boosted the number of Treg cells and showed impressive efficacy in a variety of disease models, promising a new approach to control unwanted or deleterious immune responses.14 The results in human trials, however, were totally divergent, leading to catastrophe.

In a phase I study, the CD28SA induced a cytokine storm that was near fatal in treated subjects.15 Importantly, in the study by Rosshart et al, while the conventional laboratory mice showed the expected Treg cell induction with CD28SA treatment at day 4, the wildling mice did not increase this population but rather showed high levels of proinflammatory cytokines as well as interleukin (IL)-10 at various times after treatment depending on the cytokine. Figure 1 shows the differences in responses of these two types of mice. Importantly, for these responses, wildling mice showed much better prediction of the cytokine storm observed in the human subjects. Had the wildlings been used to screen for the effects of CD28SA in preclinical studies, the translation to humans would not have been contemplated and certainly no lives would have been placed in jeopardy.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The response of laboratory and wildling mice to the administration of CD28SA. The figure illustrates the outcome of CD28SA treatment of conventional and laboratory mice. As the figure shows, while conventional mice showed a dramatic upregulation of Treg cell numbers at day 4, the wildling mice showed an increase in proinflammatory cytokine production at 2 hours (IL-1β, TNF-α, IL-2, IFN-γ), 6 hours (IL-6), 24 hours (IL-4) and 96 hours (IL-10). In contrast to the situation with conventional mice, the changes in Treg numbers of the wildling mice were not significant. The differences between cytokine levels of conventional and wildling mice were all significant at the time of sampling, which differed depending on the cytokine. Since CD28SA provoked cytokine storm in human subjects, the results of these experiments indicate that wildling mice provide much better prediction of the response of humans than conventional laboratory mice. CD28SA, CD28 superagonist antibody; IFN-γ, interferon gamma; IL, interleukin; TNF-α, tumour necrosis factor-alpha.

The other model tested concerned the effects of TNF-α blockade on sepsis. Following the discovery of TNF-α as an important proinflammatory mediator, seminal experiments demonstrated that passive immunisation of mice with a polyclonal anti-TNF antiserum can prevent the shock syndrome induced by lipopolysaccharide (LPS) from Escherichia coli. Not surprisingly, clinical trials soon thereafter evaluated TNF-α blockade in humans as a treatment for sepsis (a condition with high morbidity and mortality), but these efforts were unsuccessful and the development programme curtailed.

The failure of TNF-α blockade in humans is often cited as a prime example of the limitations of animal models as a step towards translation. It is therefore important that, in the wildling mice, like humans and unlike conventional mice, TNF-α inhibition with either a monoclonal anti-TNF antibody or a TNFR:Fc fusion protein did not block shock induced by LPS administration. Taken in concert with the results of the study on the effects of CD28SA, these findings on sepsis suggest that wildling mice, because of the microbiome of the wild mice, show patterns of immune responsiveness more analogous to those of humans, pointing to an enhanced utility in translation studies.

Along with other studies on the microbiome, the paper by Rosshart et al shows the remarkable effects of the microbiome on the immune system and suggests that the mammalian organism is itself an environment or an ecosystem in which the host genome as well as the microbiome contribute to both physiology and pathophysiology. All of these ‘omes’ create the metagenome which has components of both eukaryotic and prokaryotic genomes.4 Conceptually, the findings of the interaction of the host and the microbiome are fascinating since they reveal hitherto unappreciated facets of biology and should stimulate rethinking of fundamental questions on the nature of humans as mammals.

While the study of the microbiome promises a wealth of exciting new knowledge, operationally, the prospect of incorporating consideration of the role of the microbiome in all animal work is also intimidating and even terrifying for the experimentalist. As studies show, the composition of the microbiome (and hence its effects on the immune system) of mice is susceptible to a wide variety of factors, ranging from the diet to the temperature of the room to the nature of the bedding.4 16 Determining the microbiome of a mouse or colony also requires special expertise as well as knowledge of many species whose names are unfamiliar to most investigators. Studies will have to determine which of the many differences between the microbiomes of conventional and wild mice are most relevant. How many bacterial, viral or fungal species are determining the differences in functional outcomes? Can mice with a standardised microbiome be created and, if so, what should its composition be?

The microbiome impacts on many physiological responses, and studies will need to determine whether changes in the microbiome important for immunological disease impact positively or negatively on cardiovascular or neuropsychiatric disease, for example. Genetics is also at play since genotype can affect the microbiome, as demonstrated in studies on the effects of human leukocyte antigen (HLA) molecules associated with rheumatoid arthritis and spondyloarthropathy on the composition of the microbiomes by unaffected subjects.17 The current studies assessed C57BL/6 strain mice. Whether other strains commonly used by immunologists (eg, BALB/c) would behave similarly as those of C57BL/6 background is unknown.

Even as the studies are elucidating the impact of the microbiome on immune function, the microbiome is also becoming a target of therapy, with faecal transplants for Clostridium difficile infection just the beginning. Making the microbiome a ‘drug’ will be a huge undertaking, with issues such as composition, pharmacokinetics and pharmacodynamics important considerations as this work goes forward.18 While altering the microbiome could be contemplated to prevent or treat disease, the side effects are entirely unknown, and creating a microbiome to resist rheumatoid arthritis, for example, could predispose to other conditions in an unfavourable way.

In a more speculative vein, it is interesting to consider what would have happened to the field of rheumatology had mice with properties similar to those of the wildling mice been used to test the effects of anti-TNF on murine sepsis. Would investigators have concluded that monoclonal antibodies for inhibiting cytokine action in vivo are simply ineffective and the project dropped? Or would they have gone on to try TNF-α blockade in collagen-induced arthritis (CIA), undeterred by the failure with sepsis? Future experiments will have to address this and many other important questions as the most appropriate models for translation are developed and refined.

While many responses in mice show a striking effect of the microbiome, it is not unlikely that others are sufficiently hardwired and robust so that the effects of genetics, epigenetics and the metagenome are not determinative of outcome. The results with biological disease-modifying antirheumatic drugs to treat rheumatoid arthritis, spondyloarthropathy and psoriatic arthritis indicate that anticytokine agents can work widely among patients despite differences in genetics and no doubt the composition of their microbiomes.

For conditions such as rheumatoid arthritis, current animal models have been very informative and have been essential to the development pathway of many agents. Perhaps a model such as the CIA does not need much tinkering with respect to its microbiomes although, despite many successes, CIA has not always predicted results in humans. Thus, while antibodies to IL-17 performed well in CIA, the main benefits of anti-IL-17 therapy has occurred with psoriasis, psoriatic arthritis and ankylosing spondylitis and not rheumatoid arthritis.19 20 Pending refinement of models for various facets of arthritis including synovitis as well as bone and cartilage destruction, it will be very important that scientific publications include information on animal care and husbandry (eg, light cycle, chow, bedding) so that, if differences between studies do occur, papers have sufficient detail in the methods section to identify possible contributory factors.4 21

John Donne, the poet, created some of the most memorable lines in the literature in his mediation that starts with ‘No man is an island entire of itself’. As the elegant and exciting studies on the microbiome have shown, no person—man, woman or child—is an organism unto itself. Rather, each person progresses through life with trillions of companions in the microbiome that have participated in the same evolutionary pathway and seem to live in harmony most of the time. Future studies will determine whether the current relationship between humans and microbes can be made more harmonious and whether new animal systems can provide better prediction to promote translational research and develop new therapies.