Strategies to Facilitate Translational Advances from Microbiome Surveys

Abstract

Manipulating the microbiome has enormous potential to treat important human diseases. Microbiome surveys are often used to identify potential therapeutic targets by finding associations between microbial elements and disease status. We argue that many reported associations between the microbiome and disease are incompatible with translational research because they are insufficiently specific. We encourage the clear specification of manipulable microbial elements that can be tested in follow-up randomized experiments, and we provide multiple examples of specific and nonspecific microbial elements.

Microbiome researchers have made significant progress over the past decade characterizing the microbes which inhabit the human body, and have begun to identify mechanisms which can directly impact human health [1]. The next frontier for translational microbiome science is to develop diagnostics and therapeutics that leverage those connections between the microbiome and human health. By identifying therapeutic targets, we have the opportunity to make great strides in preventing and treating important human illnesses such as inflammatory bowel disease, colorectal cancer, and metabolic disorders. However, microbiome researchers need to carefully consider their scientific approach to achieve high-impact translational advances.

To make the transition from descriptive studies of the microbiome to microbiome therapies it is necessary to identify which biological elements to target. Is it the relative or absolute abundance of a particular strain that drives disease? Is it the presence of a specific collection of genera? Is it the absence of a metabolic pathway? Or the presence of a metabolic pathway within a particular species? In many of the disease settings which have a published microbiome connection it is unclear which specific microbial elements are associated with disease.

An excellent tool for identifying potential therapeutic targets is performing microbiome surveys. Broadly speaking, microbiome surveys are conducted by collecting microbiome samples from a cohort of participants, making a measurement of some aspect of host health or disease, while also measuring the abundance of microbial elements in the microbiome sample (e.g., with amplicon or shotgun sequencing). While the experimental design and measurement strategy may vary widely, the outcome of these microbiome surveys typically is some association metric linking each microbial feature to a measure of health or disease. The benefit of a survey-driven approach is that it generates data from natural populations, instead of contrived laboratory-based models. Unfortunately, strong assumptions are needed to establish causality from survey data, which are inherently observational.

The highest standard for assessing causality is a randomized experiment. Therefore, for an association result observed in a microbial survey to demonstrate translational potential, validation experiments must be performed. Generalizable validation approaches include the controlled administration of specific microbial isolates, prebiotics, or small molecules. There are examples of such controlled manipulations demonstrating a causal role for specific microbial species [2], strains [3], and genes [4] on host health. We argue that microbiome researchers who wish to demonstrate the translational potential of a finding should perform a randomized validation experiment, or clearly state and acknowledge the assumptions needed to make causal inference from observational data [5].

We believe that there are many types of observation which are regularly made from microbiome surveys which lack translational potential because they are not specific enough to be tested in a randomized experiment. For example, observing significantly lower alpha diversity in subjects with disease compared to nondiseased controls is unlikely to advance disease treatments on its own because there are an infinite number of ways for a microbiome to have increased diversity. For example, lower alpha diversity is associated with Clostridium difficile colonization [6,7], but there is no finite collection of cross-sectional surveys which can falsify the hypothesis that such lower alpha diversity causes the associated diarrheal disease, and is not its consequence. Similarly, observations about decreased or increased relative abundance of broad taxonomic groupings (e.g., Firmicutes) in microbiomes associated with diseased versus nondiseased cohorts is unlikely to advance treatment because of the immense diversity of strains under this grouping. We are observing a growing body of literature which reports associations of nonspecific microbial elements with human disease, including (but not limited to) high-level metabolic pathways, vague organismal groupings, phylum- or class-level taxonomy, alpha- or beta- diversity, or undefined community-level attributes (e.g., ‘dysbiosis’). This lack of biological specificity inhibits experimental validation and slows the progress of translational work.

To accelerate the pace of translational microbiome research we encourage researchers to identify specific strains, specific metabolic pathways, or specific genes that they observe to be associated with health or disease. We also encourage researchers to state the nature of the association of the biological element with health status. For example, is it the presence of a strain which is most strongly predictive of a disease, or does the relative abundance need to be above a threshold? We encourage colleagues who report observations from microbiome surveys to carefully consider whether the associations that they are reporting can be validated in a randomized experiment.

Hypotheses about the presence or abundance level (whether relative or absolute) of microbial strains are specific enough to be testable using preclinical models, cell cultures, and human clinical trials, which make them possible to falsify. Similarly, hypotheses about the presence of specific metabolic pathways in specific organisms are also falsifiable using existing laboratory-based technology. Even as our techniques and technology for testing preclinical models evolve, we ask researchers to emphasize and articulate candidate hypotheses that are specific enough to be investigated further.

Science advances through the proposal of hypotheses that can be falsified through multiple experimental techniques. We argue that if a result about the microbiome cannot be evaluated in a randomized experiment then it is not falsifiable and does not provide an avenue towards microbial therapeutics. We are optimistic that the microbiome field can have great translational impact, and we believe its advance can be accelerated by an increased focus on biologically specific and experimentally falsifiable hypotheses.