The role of a healthy cohort has gained significance owing to the general concept of high differences between individuals. The empirically designed sampling intervals were based on a monthly and quarterly follow-up schedule to closely mimic the common research scenario, comprising a “healthy” reference cohort. The findings from the present study show that the “healthy cohort” has significant variations in the microbiome profile of periodically collected samples, implying a difference in the state of microbial stability.
According to the population richness and evenness results, the sampled cohort behaved similarly. However, the significant Shannon index changes indicated species level bio-interaction. These findings of microbial interaction were further evident in the participant-wise index pattern, and the sharp contrast to the phenotypic (clinical) selection criteria was markedly observed on visualizing the changes across three-time points. Moreover, intraindividual variations were evidenced over a broader range of variability, particularly in the Shannon diversity metric traced about T0. The observations mentioned above were mainly in accordance with pioneering results from the human microbiome project . Similarly, in the findings of Sato et al., a significant difference was reported in a consecutive day and intraday samples of healthy adults . These results are collectively conclusive of varying stability as a factor of time within the microbiome classified as health-associated.
The observed time-dependent variations become particularly concerning when interpreted in light of the Anna Karenina principle (AKP), which hypothesizes that increased within-participant variability in the microbiome as a good marker for dysbiosis [26, 27]. In the present study, both PERMANOVA and ANOSIM displayed statistically significant differences in participant-based comparisons, thus indicating a significant variation within the cohort, i.e., a higher predilection for dysbiosis. Therefore, the observed within-participant variability of the diversity metric challenges the “healthy cohort” concept, synonymous with a state of eubiosis.
To reveal the pattern of variations observed from T0 to T2, within the study cohort, we evaluated a correlation to stability metric from ecology indices [16, 28]. Independent of the net abundance, in a fluctuating community, unstable species (species with high turnover) have been reported to maintain community states . Analytically, such states present with a negative covariance or asynchrony pattern against stability when observed over time . A moderate to strong negative correlation pattern observed in the present study also indicated an inverse relation to stability. Therefore, our findings would also imply the presence of an alternative state of balance amidst participants of the cohort, suggesting the presence of unstable species.
Enterotypes have been defined as the state where samples get assigned when binning different individuals into classes that share some similarity in microbiota composition.[17, 30] We adopted a cluster model that bins taxa into somatotypes to explore the distribution similarity. As a characteristic of a stable community, clustering of taxa would result in a reproducible pattern. [15, 25]. This characteristic response to clustering is attributed to low inherent variation within an individual. However, our findings challenged the concept with the observation of marked differences in the aspects of: (i) clustering pattern, (ii) biomarkers, and (iii) metabolic indicators. The lack of cluster reproducibility contrasts the outcomes for a healthy cohort from the conclusions of Sato et al. . However, such differences can be attributed to comparatively longer sampling intervals analyzed in this study, thus highlighting a notable effect of temporal dynamics on the oral microbiome.
We performed the PICRUSt analysis to estimate the metagenomic function. Our findings were in contrast to earlier research, which predicted greater stability of the metagenome across time . However, the interpretations were consistent with the finding that single estimates (cross-sectional) do not effectively describe the equilibrium abundance . These findings become increasingly relevant while considering diagnostic or intervention-related changes. Consequently, identifying stable somatotypes within a cohort across time points can markedly contribute to identifying bioindicators .
Eventually, we nominated two individuals to illustrate differences due to microbiome stability. The turnover metric revealed an inverse trend upon considering the number of taxonomic units gained (appearance) and lost (disappearance). Abundance profiling of the groups resonated the differences with a distinct group of dominant genera between the two participants at all time points. Comprehensively, they presented 49.11% (Table S2) dissimilarity of significant taxonomic units (Fig. 5c-i). Subsequently, we could identify differences at the KEGG metabolism level with significant variations in secondary metabolite synthesis, carbohydrate, and nucleotide metabolism (Fig. 5c-ii), further indicative of a metabolically different microbial community composition due to a difference in stable states.
Despite fluctuations over a period, it is believed that net stability can be achieved when an increase in one species compensates for a decreased abundance in another . Previous microbiome studies have described temporal fluctuation as a ubiquitous and vital factor in the stability of the aggregate community [33–37]. In other words, temporal fluctuations and asynchronous patterns observed over a time series are natural processes toward the stability of the microbial community. However, substantial distress can elicit a switch of the stomatotype (clustering pattern), thereby resulting in an alternative state . While this step may not express significantly between niches, researchers cannot overlook the impact of time on stability.
The aforementioned points are further validated under the multi-stable ecosystem concept, which outlines the need to consider the prevalence of more than one stable system within a particular condition . The stable states can experience trigger events (favorable or unfavorable) that can lead to an abrupt change. Therefore, only a single sampling instance for a study may have a transient state, potentially wrongfully estimating eubiosis or dysbiosis. With context to the present study, a shift in the baseline point to T1 instance from T0 may lead to a significantly different inference, as vindicated by stomatotyping variations. Figure 6 schematically depicts the multi-stable state of microbial community ecology while also considering the effect of AKP.[26, 39, 40] Changes in conditions can affect the microbial community, where a multi-stable community adopts different stable states within similar environmental conditions. However, the system states vary across a tipping point (black stars), and changes between stable states are not entirely reversible. Such effect is referred to as hysteresis . Therefore, a novel stable state or reversal between states can occur only with a stimulus beyond the trigger threshold. Studying samples in a transitional state, when regarded as standard “healthy,” may result in bias seeding owing to the locational effect (exampled as points 1, 2, and 3 as points of study commencement) .
To summarize, the prestudy findings suggest that phenotypic screening alone may fall short as the confounding factors extend beyond basic subject features while planning an oral microbiome study . A genomic screening step will facilitate understanding the cohorts’ homeostatic range and the sub-selection of objectively relevant groups. Taken together, they will enhance the quality of oral microbiome studies to identify the diagnostic and treatment markers. Another point of emphasis for intervention and follow-up studies is to include a “false start,” and allow an adaptational equilibrium (with time), and help in finding a true baseline .
The present study provided a comprehensive analysis of time-and participant-based data; however, the use of web-based tools for analysis may be an inherent limitation. Furthermore, the sample sourcing from biobank was a drawback, in terms of the assessment period being limited to a quarterly interval. Third, we could not compare associative phenotype patterns owing to limited host metadata for the available samples.