Assessing the safety of microbiome perturbations

Abstract

Everyday actions such as eating, tooth brushing or applying cosmetics inherently modulate our microbiome. Advances in sequencing technologies now facilitate detailed microbial profiling, driving intentional microbiome-targeted product development. Inspired by an academic-industry workshop held in January 2024, this review explores the oral, skin and gut microbiomes, focussing on the potential long-term implications of perturbations. Key challenges in microbiome safety assessment include confounding factors (ecological variability, host influences and external conditions like geography and diet) and biases from experimental measurements and bioinformatics analyses. The taxonomic composition of the microbiome has been associated with both health and disease, and perturbations like regular disruption of the dental biofilm are essential for preventing caries and inflammatory gum disease. However, further research is required to understand the potential long-term impacts of microbiome disturbances, particularly in vulnerable populations including infants. We propose that emerging technologies, such as omics technologies to characterize microbiome functions rather than taxa, leveraging artificial intelligence to interpret clinical study data and in vitro models to characterize and measure host-microbiome interaction endpoints, could all enhance the risk assessments. The workshop emphasized the importance of detailed documentation, transparency and openness in computational models to reduce uncertainties. Harmonisation of methods could help bridge regulatory gaps and streamline safety assessments but should remain flexible enough to allow innovation and technological advancements. Continued scientific collaboration and public engagement are critical for long-term microbiome monitoring, which is essential to advancing safety assessments of microbiome perturbations.

Keywords: artificial intelligence (AI)/ML; clinical studies; gut; in vitro models; microbiome; oral; safety assessment; skin.

Fig. 1.. Summary of the workshop outputs. Endpoints defining host–microbiome interactions in health and disease remain poorly defined. Effective risk assessment is challenging due to uncertainties introduced by confounding factors such as lifestyle and environmental influences (green circle), as well as biases inherent in measurement techniques and data analysis methods (blue circle). To address these challenges, we propose integrating standardized in vitro testing and longitudinal monitoring of vulnerable populations to better assess potential risks. Additionally, uncertainties can be reduced by capturing extensive host metadata, utilizing advanced digital processing techniques, including emerging AI approaches and employing open access computational models.

Fig. 2.. The distribution of genomes found in the four MGnify human microbiome catalogues: gut, skin, vaginal and oral. While each of these catalogues contains different numbers of genomes and is not directly comparable, the maps highlight the geographic skew of the samples contributing to the genomes found within these catalogues, with human microbiomes from North America, Europe and China representing the bulk of the data [181].