The study of the human microbiome has advanced considerably due to experimental and computational approaches, yet its clinical applications remain challenging due to the lack of standardised protocols for sample collection, processing and analysis of results. Given the complexity of microbiome-host interactions and the influence of numerous confounding factors, strict regulatory frameworks are necessary to ensure reliable and reproducible findings1. Moreover, a multidisciplinary approach is crucial, requiring collaboration between microbiologists, bioinformaticians and clinicians2. While Wensel et al.’s attempt to apply Koch’s postulates to microbial signatures provides us with a theoretical guide to look for potentially pathological microbial signatures3; most of the available studies on microbial signatures in pathologic conditions are isolated snapshots of reality that may result from several confounding factors. Unlike traditional pathogens, microbial communities exist in dynamic equilibrium with the host, making it difficult to identify causal relationships between specific microbial alterations and disease.

Several studies have focused on understanding the broader role of microbial ecosystems in disease aetiology4. While this may be of great intellectual value, due to limitations in current methodologies it might be worth taking a step back and focusing on the application of new methodologies to identifying disease-associated biomarkers at the top of the disease cascade.

A potential conceptual framework for understanding microbial ecosystem dynamics is the “theory of broken windows” introduced in criminology by James Q. Wilson and George L. Kelling in 1982 and later revisited by Kees Keizer et al. in 20085 based on a social experiment conducted in 1969.

In the experiment, two identical cars were abandoned in neighbourhoods with vastly different social conditions; one in the Bronx (a crime-ridden area of New York) and another in Palo Alto (a wealthy and peaceful city in California), with a team of social psychologists observing people’s behaviour in each location. The car left in the Bronx was stripped within hours: anything of value was taken, and what was left was destroyed. Meanwhile, the car in Palo Alto remained untouched. While it would have been easy to stop the experiment and to attribute crime to poverty, the researchers continued the experiment by breaking one of the windows of the vehicle left in Palo Alto. Following this, the same pattern of destruction seen in the Bronx emerged (Fig. 1).

Fig. 1
Fig. 1
Full size image

Schematic representation of the “theory of broken windows” and its counterpart applied to microbiota’s studies.

This theory suggests that visible signs of disorder in urban environments, such as broken windows, can escalate into widespread antisocial behaviour and crime. Similarly, small disturbances in microbial communities may trigger a cascade of dysbiosis, leading to disease. Most of the studies available on the microbiome compare a disease status (the Bronx) to healthy controls (Palo Alto)6. This perspective suggests that stringent inclusion and exclusion criteria alone may be insufficient to establish a causal link between the microbiome and disease.

Numerous confounding factors complicate the interpretation of microbiome data (the absence of an “identical car”) and biomarkers (such as the increased relative abundance of a specific species or its pulsed presence - the “broken window”) remain elusive.

Available microbiome analysis methods cover a vastly wide. The lack of a standardised framework for analysis means that technologies used by different groups for analysis can vary from classical PCR to whole genome sequencing. Sample collection and transport should be standardised; sample collection kits including containers and slips for safe collection should be provided to study participants to reduce the risk of sample contamination. Samples should be stored at the same temperature after collection. These indications are the tip of the iceberg in microbiome investigations.

The international consensus statement by Porcari S, Mullish BH, Asnicar F, et al. on microbiome testing in clinical practice is one of the first attempts in providing a tangible guidance to regulate the provision of microbiome testing. As with all laboratory procedures, an essential statement should involve the bioinformatics guidelines, which are currently well partially established for group comparisons (for example, describing normalization methods7) but are not clearly defined and standardized for application in clinical practice.

The microbiome of two individuals may be significantly different due to geography8, diet9,10, medical comorbidities and medication intake11. To minimise bias, participants of different study groups should be accurately matched and ideally differ only in the state of the disease being studied. Importantly, the concept of the “identical car” represents a normalisation step upstream of sample collection, sample handling, sequencing and bioinformatic/statistical analyses. Ways to ‘break a window’ in-vivo could vary from the use of a specific diet, probiotics, or antibiotics of the investigator’s choice for a specified period of time.

Correlation between microbial signatures and disease symptoms could be investigated through the use of e-diaries collecting patient reported outcomes (PROMs). PROMs collected using a Likert scale creates a dataset which can be analysed using statistical techniques including multivariate analysis to assess the relationship between symptom variation (i.e. minimal clinically important differences) and a change in microbial signature.

The introduction of a normalisation strategy, the identical car, before characterising microbial signatures may allow us to identify three different scenarios: a) microbiome perpetuity without host improvement (a disturbed microbiome that remains unchanged and/or imbalanced, even after an intervention, failing to restore host health: i.e. pathogenic microbiomes); b) microbiome variations without host improvement (observed microbial changes occur independently of disease progression, indicating a non-causal association) and c) microbiome variations with homoeostasis recovery (microbial shifts contribute to restoring host health, pathobiont-microbiomes).

As previously postulated by the adaptation of Anna Karenina’s principle by Zaneveld et al.: “all healthy microbiomes are similar, each dysbiotic microbiome is dysbiotic in its own way”12. Therefore, the use of metagenomic techniques for the identification of markers of disease and the normalisation of upstream factors may aid us in taking a step towards reconciling metagenomics and traditional microbiology, thereby advancing its role in precision medicine. We must bridge the gap between metagenomic data and basic research in order to extend beyond the limits of microbiome observational studies.