These weak correlations, combined with the rarefaction curve, mean that we needed to rarefy samples to make proper comparisons without jeopardizing the accuracy of sample diversity or losing information on OTUs, particularly rarer ones, in the data set. Hence, we chose the depth of 30,000 based on the rarefaction curve so that the number of OTUs in most samples could be accurately represented.After rarefaction, we compared the number of OTUs to that before rarefaction. Just like the preliminary experiments, rarefying to an even depth reduced the absolute numbers of OTUs across all samples while retaining the relative trends . The similarity between the OTU distributions before and those after rarefaction speaks to the efficacy of this technique in terms of standardizing sample sizes while avoiding loss of useful comparative information about the community structure. To look at diversity more critically from a different angle, we also examined the inverse Simpson’s index vales before and after rarefaction . In this case, rarefying to 30,000 led to somewhat more pronounced effects on this index than on the number of OTUs. For all samples except for pure E. coli, rarefaction shrank the range of the inverse Simpson’s index values, thus making samples look more similar to one another. For instance, blueberry plant pot the culture with a value of more than 12 before rarefaction , compared to the rest of the cultures with values less than 5, saw a reduction to approximately 4, compared to the rest of the cultures that had inverse Simpson’s values of 3 or lower.
Nevertheless, the relative diversity rankings across samples were retained, and the difference between that particularly diverse culture sample and the rest of the cultures was still pronounced enough after rarefaction that this statistical procedure remained valid for temporal cultures. Principal Coordinate Analysis of all samples with spike-ins shows distinct clusters of cultures, controls, and plaque . As in the preliminary experiments, rarefaction does not visibly change the clustering patterns except for a few cultures of longer incubation times . Bar plots of read counts for negative controls show that the Escherichia-Shigella OTU, the major OTU in the E. coli spike-ins, dominates the negative controls for Host 3 until the 168-hour incubation time. For Host 1 and 2 controls, this OTU is not consistently dominant as incubation time increases, and the lack of consistent dominance is shown clearly in bar plots of relative abundances of controls . In this figure, we see that Escherichia-Shigella OTU0001 occupies more than 90% of the read counts in Host 1 for only the 12-hour controls and in Host 2 for the 12-hour controls, one 24-hour control, and one 48-hour control. After 48 hours, the relative abundance of the Escherichia-Shigella OTU does not consistently decrease as incubation times increase, and the relative abundances of the spike-in OTU differ across the two wells from the same incubation time, especially for the 168-hour controls in all hosts. These results are somewhat difficult to interpret because of the uneven sequencing depths of the controls , as well as the experimental setup in which we sampled from distinct sections of the well plates instead of the same wells.
However, one observation is clear – the contamination is still entirely internal, i.e. sourced from the cultures from the same plate. Contaminated controls in all hosts are dominated by Streptococcus OTUs until Veillonella OTUs take over at 168 hours, and as we will observe later, this succession occurs in the cultures as well. Together, these results validate that our methodology prevents external contamination. The relative abundances of cultures and plaque, with spike-ins, are shown in Fig-ure 21. The plaque samples contain very little biomass, as expected while cultures from all hosts contained high biomass from oral bacteria, as evidenced by the dominance of non-Escherichia-Shigella OTUs. Dominance of the oral bacteria, however, is not always consistent between the 2 wells with the same incubation time. This is especially visible in the 48- and 96-hour cultures from Hosts 2 and 3, where Wells 1 and 2 differ in terms of relative abundances of oral bacterial OTUs. Oral bacterial dominance also does not always increase with increasing incubation time. In fact, there seemed to have been a decrease in oral bacteria biomass, relative to the E. coli spike-in, from 48 hours to 96 hours for Hosts 3. Once again, these results are difficult to interpret because the wells being sampled at these time intervals were different wells. The inconsistency in the relative abundances of the oral bacterial OTUs could have easily arisen from differential growth rates of different wells. Such a possibility seems especially likely in light of the substantial differences in the read depths of the cultures . However, a definitively consistent trend did surface from this relative abundance data: Cultures from all 3 hosts show a predominance of Streptococcus OTUs for both the 12- and 24-hour cultures, followed by the rise of Veillonella OTUs , which is in turn followed by a shift toward Prevotella OTUs .
Host 1 relative abundances show the earliest visible appearance of the Prevotella OTUs at the 96-hour mark while Hosts 2 and 3 only show the presence of Prevotella OTUs in the 168-hour cultures. Such changes in these community compositions indicate a shared succession of colonization, much like the sequential colonization of the human oral cavity in vivo, where members from the Streptococcus genus lay the ground work, and those of the Veillonella and Prevotella genera follow as either middle or late colonizers. It is also interesting to note that regardless of initial plaque composition, even without continuous inoculation of the in vitro cultures by plaque, the communities in these experiments evolved to include OTUs from the Streptococcus, Veillonella, and then the Prevotella taxa. It is possible, then, that members of the Veillonella and Prevotella genera stay dormant and/or protected until they can proliferate in the environment created by Streptococcus OTUs, though this succession would be best tested by co-culturing known strains. We can observe an interesting similarity between the relative abundances of negative controls and those of the cultures/plaque samples. The community composition changes in the contaminated controls track the changes in the cultures, albeit at a slower rate; controls in these experiments did not reach a mature enough stage to include the Prevotella OTUs. It is likely that because the controls received the inoculum later during the incubation period – at some point between 12 and 24 hours rather than at hour 0 – their development seemed delayed compared to the cultures. After we examined the biomass of the controls and cultures, using the E. coli spike-in as a qualitative standard, we removed the major spike-in OTUs in order to assess sample similarities. We performed PCoA on the cultures and plaque samples without spike-in OTUs, and the results clearly show three clusters of samples regardless of rarefaction status . Plaque samples cluster loosely in a group separate from the cultures , as expected because of the inherently selective nature of in vitro culturing procedures. The large spread in the plaque samples is also expected because these samples originated from different hosts, though this cluster did not show three finer subclusters. Unlike the plaque samples, cultures cluster into two distinct groups, separated by the length of incubation time. The 12- and 24-hour cultures cluster somewhat tightly together while the 96- and 168-hour cultures cluster tightly together , with some spread of the 48-hour cultures in between . The two clusters are also situated somewhat symmetrically across the loosely vertical line formed by the plaque samples. The divide between cultures of different incubation times suggests a compositional shift in the cultures for all three hosts starting around 48 hours, as we observed in the relative abundances of the cultures, albeit with less certainty. Another feature of some interest in the PCoA plot is the amount of variation accounted for by the two coordinates. In this case, the differences in the cultures account for 55 to 57% of the total variation in the samples, plastic gardening pots while differences between cultures and plaque account for approximately 22 – 23% of the variation. The difference between Streptococcus dominance and Veillonella/Prevotella dominance is clearly the largest contributing to inter-sample variation. However, the two coordinates in PCoA only account for about 80% of the total variation, which puts into question what constitutes the other 20%.
To help answer this questions, we constructed a dot plot of OTUs that make up at least 1% of the relative abundance in rarefied samples , and performed Principal Component Analysis on the relative abundances . From the dot plot, we found that the most abundant OTUs in plaque samples came from 7 genus-level taxa , two of which include members known for early and middle colonization of the plaque community and one of which contains members that have been shown to co-aggregate with organisms involved in all stages of colonization. In cultures, we observe the trends present in the relative abundance bar plots and the PCoA plots: compositions of 12- and 24-hour cultures are dominated by Streptococcus OTUs , then transition to Veillonella OTUs in 48-, 96-, and 168-hour cultures, with the simultaneous rise of Prevotella OTUs at 96 and 168 hours. Of the Streptococcus and Veillonella OTUs, a single OTU from each genus dominates at certain points in time while the other OTUs tend to persist at lower abundances. On the other hand, only one Prevotella OTU plays a role in the temporal cultures. Interestingly, neither this Prevotella OTU nor a Fusobacterium OTU appears much in the cultures until 168 hours of incubation. In addition, a Megasphaera OTU also begins appearing between 48 and 96 hours of incubation. Other OTUs with somewhat substantial presence in plaque samples Pseudomonas OTU0027, Corynebacterium OTU0017, Actinobacillus OTU0019, and Acinetobacter OTU0030 – seemed to have been selected out of the temporal cultures, as they do not appear in abundances higher than 1%, and all except for the Actinobacillus OTU disappears between 96 and 168 hours of incubation. Based previous research on the order of succession in human oral microbiome formation, we see that these temporalcultures were potentially transitioning into later or late colonization stages at 168 hours, with the rise in relative abundance of the Fusobacterium OTU. Here, extending the time incubated beyond 168 hours and/or regular re-inoculation with host-specific plaque would help us greatly in probing whether such a transition is occurring in vitro. For the Principal Component Analysis , we first performed it on untransformed relative abundances of plaque and culture samples. The resultss show distinct clusters much like in PCoA , with plaque samples situated between cultures of different incubation times. The underlying factors that contribute toward differential clustering seem to originate from a division between Streptococcus OTU0002 and Veillonella OTU0003. As expected based on results in the dot plot, Pseudomonas OTU0027 contributes to the difference between plaque samples and culture samples , though the other prominent plaque OTUs from the Corynebacterium, Actinobacillus, and Acinetobacter genera surprisingly do not seem to contribute as much as the Pseudomonas OTU does. When colored by host, the cultures display no visibly distinct grouping, though duplicate plaque samples of individual hosts stay quite close to each other . We then performed centered-log-ratio transformation on the relative abundances of plaque and culture samples, and performed PCA again on the transformed data. CLR is commonly used to take the simplex space of the relative abundance data of the sample – the nature of any data that sum to a value of 1, or 100, for any individual sample – into real Euclidean space, hence making valid any distance metrics and statistical method that can be applied to data in Euclidean space . Because PCA typically needs to operate in a real Euclidean space to avoid artifacts and spurious patterns, CLR allows us to perform PCA on the data set in a much more statistically valid manner. Mathematically, this transformation is done by using the log function on a ratio between a sample and the geometric mean of the sample, as shown in Eq. 4 . Results of PCA on the CLR-transformed relative abundances show that plaque and cultures cluster in groups similar to those in PCoA and PCA of untransformed relative abundances . As shown in the analyses above, the differences across hosts do not play a large role in accounting for the variation in the samples, but incubation time does.