The division of cultures into those with shorter incubation times and longer ones persists across the CLR transformation, though the magnitude of the differences decreases . Notably, the results here, which should mathematically represent trends that are valid in Euclidean space, also correspond to trends from the PCA analysis of the relative abundances still in simplex space. The agreement between the PCA results from the untransformed and those from the transformed datasets indicates that these results are very unlikely to be mathematical artifacts. In fact, they seem to carry enough biological significance to persist through extensive mathematical and statistical manipulation, including rarefaction, relative abundance calculations, and logratio transformation. Results from PCoA, dot plot of abundances, and PCA together serve as exceptional evidence of the compositional changes in the temporal cultures as well as the compositional “neutrality” of plaque. They also point to the lack of host driven differences, implicating the possibility of a baseline in vitro culture that can be developed by manipulating incubation times. A clear difference between the untransformed and CLR-transformed PCA that is worth some exposition is the number of factors shown graphically and accounted for mathematically.
In the untransformed PCA, blueberry pot size the first two components account for almost 90% of the variation in the samples while in the CLR-transformed PCA, the first two components only account for 45% of the total variation. Untransformed PCA on the temporal samples attributes variations to mostly the Streptococcus, Veillonella, and Pseudomonas OTUs while CLR-transformed PCA includes contributions from the Actinobacillus and Acinetobacter OTUs, along with a number of other genera. Even in the most prevalent and abundant genera, the PCA on the untransformed dataonly accounts for two of the three more prominent Streptococcus OTUs and one of the two Veillonella OTUs. On the other hand, results from the CLR-transformed PCA are much more consistent with the dot plot. Regardless of whether PCA on relative abundance data led to spurious trends in the temporal cultures, it is clear that performing PCA on simplex spaces can obscure the observation of important factors. For these reasons, we elected to keep performing PCA on CLR-transformed relative abundances in subsequent experiments, in order to gain a more complete understanding of what contributes to compositional differences.The results from the analyses of the temporal cultures confirm that our culturing procedures helped avoid contamination.
We discovered that controls in the temporal cultures were much more difficult to keep clean than those in the preliminary cultures. This may be due to the physical proximity of the control wells to the culture wells, which enables accidental splashing during feeding and moving of the plates, though we took great care to avoid doing so. As in the preliminary experiments, OTUs in the controls fell entirely within the taxa expected from either the E. coli spike-in or from the cultures, with no sign of external contamination. In terms of diversity, the temporal samples seemed slightly higher than the preliminary samples , though by what quantitative extent we do not know. We may estimate the differences between the preliminary cultures and the temporal cultures by observing the differences between the 24-hour cultures and the more aged cultures from the temporal samples. This particular comparison shows that the more aged cultures contained different distribution of OTUs and more OTUs in some cases. The increased diversity with increased incubation times was very much expected, and the procedures we used in to generate the temporal cultures were capable of capturing early colonizers as well as some of the middle colonizers from the oral bacterial community. In terms of procedural differences between the preliminary and temporal experiments, we used rarefaction to standardize and make samples comparable to one another, as rarefaction was once again shown to be a defensible approach in the temporal experiments.
We also opted not to separate the liquid from the sedimentary cells in the temporal experiment because of the lack of biological significance of the liquid samples. During the preliminary experiments, we confirmed that the liquid samples did not contain aberrant OTUs, i.e. OTUs belonging to neither the human oral bacterial community nor to the spike-in E. coli, and that the liquid samples contained low biomass. These observations led us to believe that the liquid was simply an extension of the sedimented cells, not a saliva-like substance that helps inoculate the cultures, especially given the lack of liquid movement in the cultures at all. As for the composition of the temporal cultures, we observed a shift from Streptococcus OTUs to Veillonella OTUs followed by the rise of Prevotella OTUs in all three hosts. This observation corroborates previous research on the succession of microbes in the colonization of the human dental surface, where Streptococcus spp. serve as the earliest colonizers of dental enamel, followed by microbes from several other genera that include Veillonella and Prevotella. This order of succession of species is in large part explained by the attachment capabilities of streptococcal bacteria. The two most abundant streptococcal OTUs in this study have been shown to produce surface proteins that bind to salivary amylase, human extra parotid glycoprotein, salivary surface lectins, human salivary antigen C fibrillar glycoprotein, and Veillonella fibrillar antigen B. Clearly, organisms in these streptococcal OTUs can act as intermediaries between the salivary pellicle deposited on dental enamel and other bacteria, and our methods provided appropriate conditions for this temporal succession observed in humans as well as in other in vivo and in vitro models of the human oral microbiome. We may be able to observe finer details of temporal development if we increase the sampling frequency, i.e. add more time points to the 5 existing ones during the 168-hour interval. However, it is not clear how the increased aerobic exposure would affect the purity and the development of the cultures. We will likely have to introduce additional measures to avoid contamination should we choose to sample the cultures more frequently. The E. coli spike-in in the temporal experiments served as a qualitative approach for estimating the biomass of the controls. For controls not contaminated by oral bacterial OTUs, the Escherichia-Shigella OTU dominated the relative abundance as in preliminary experiments . Even in controls that had been internally contaminated, the Escherichia-Shigella OTU still occupied 50% or more of the total abundance until the 168-hour incubation. At this time, we have confirmed that up to 168 hours of incubation time, the culturing procedure does not introduce microbes that interfere greatly with the in vitro compositions; both the OTUs and the estimated biomass in the controls, relative to the E. coli spike-in, imply that the OTUs and their distributions in the cultures were likely minimally affected by any potential cross-contamination between the wells. In future experiments, plant raspberry in container we may quantify the biomass directed by weighing the cells from the controls and cultures to gain an understanding of how much cross-contamination in a single plate affects the culture compositions in that plate, though we may not need to do so with the E. coli spike-in. Since the original purpose of the E. coli spike-in – developing a quantitative internal standard that would allow us to estimate the number of cells of the OTUs in the in vitro bacterial community – has been shown to be infeasible, we will eliminate this step to improve procedural efficiency as well as increase the sequencing depths of the cultures. Interestingly, a somewhat recent paper details a quantitative approach that links the number of cells per gram to the differences in gut microbiota.
The authors use flow cytometry cell-counting and 16S copy number corrections to bypass issues stemming from the fixed depth of Illumina MiSeq 16S rRNA sequencing process. The results from this quantitative profiling experiment showed that microbial load is significantly associated with changes in the microbiota, underscoring the importance of the ability to quantify sample biomass. It may be useful to incorporate this particular quantitation approach into future experiments, though the differences in microbial load are likely to be much higher in direct fecal samples across different hosts than the differences generated by cellular growth in in vitro cultures inoculated by similar amount of host plaque. Another potential improvement, should we reinstitute the spike-in procedure, may be using a species or type of bacteria other than E. coli. Ideally, the spike-in should be absent from the human oral microbiome while simultaneously demonstrating extraction and amplification efficiencies similar to members of the oral microbiome.
Because E. coli, a Gram-negative bacterium, has been shown to extract more efficiently than Grampositive bacteria such as S. aureus , E. coli K12 ER2738 may be a non-ideal choice for spike-ins. In this respect, we may be able to use the phi X 174 DNA sequences employed by the UC Davis Host-Microbe Systems Biology Core, as the sequences from that bacteriophage have been shown not to interfere much with the sequencing processes of microbes. Another key quantification step for reinstituting a spike-in would be to compare extraction efficiencies across different Gram-positive species in the oral bacterial community so that we understand potential extraction bias. With that information, we may be able to devise corrective measures to estimate the number of cells in each OTU more accurately. For the temporal cultures, we did not perform an in-depth analysis of host plaque because the goal for this phase of the project was not to generate in vitro cultures with composition that exactly matched that of host plaque. The purpose of this phase was to establish a baseline community that contained organisms common to oral and dental plaque microbiome, with some mimicry of the organismal succession observed in hosts. In this case, though the composition of plaque is useful for making comparisons, the absolute plaque composition from any single host and the precise replication of host specific oral communities were not of central importance. Furthermore, we have been prepared for the fact that unless able to perfectly reproduce the natural conditions, anyin vitro culturing process inevitably selects for a subset of organisms from the microbial community in nature. The short incubation time of one week and an environment absent of host nutrient sources and immune systems in this model make it extremely unlikely that we would be able to replicate the formation of a mature oral or dental plaque community. However, should we look to cultivate communities with higher diversity, we can now build upon this simple yet robust methodology by varying and/or adding nutrients, and regularly re-inoculating the cultures to increase the presence of middle and later colonizers. Here, re-inoculation may be the key to mimicking host variation in the “base” in vitro cultures. In any case, our methodology seemed capable of establishing communities with compositional developments that mimicked in vivo bacterial successions as well as capturing host differences in a qualitative, limited manner. As a last comment on the bio-informatics analyses: As we continued to augment the statistical methods from the preliminary experiments, we adopted the centered-logratio transformation to make PCA a mathematically sound process. This transformation helped confirm the biological significance of the patterns observed in the temporal cultures. The comparison between Figures 24 and 25 shows that CLR helps correct some potential artifacts created by working with relative abundance data, i.e. working in simplex space. For instance, the differences between clusters of samples in the PCA plot before transformation were much more pronounced than those after transformation . Given that CLR has been shown to reduce false discovery rates substantially, the decrease in the effect size in the temporal experiments as a result of the CLR transformation is simultaneously expected and highly desirable. Hence, we will use this method as a tool and a check on patterns observed in PCoA of relative abundance data going forward.In the temporal experiments, we expanded upon the previously established culturing procedures by extending the incubation time and implementing a feeding regimen. Analysis wise, we were able to characterize the community composition of the temporal cultures by using the established, validated bio-informatics pipeline. We also gained the capability to scrutinize and correct the results from Principal Component Analysis on relative abundance datasets, with the addition of the centered-log-ratio transformation.