Deep sequencing analyses of low density microbial communities: working at the boundary of accurate microbiota detection

Abstract

Introduction: Accurate analyses of microbiota composition of low-density communities (10(3)-10(4) bacteria/sample) can be challenging. Background DNA from chemicals and consumables, extraction biases as well as differences in PCR efficiency can significantly interfere with microbiota assessment. This study was aiming to establish protocols for accurate microbiota analysis at low microbial density.

Methods: To examine possible effects of bacterial density on microbiota analyses we compared microbiota profiles of serial diluted saliva and low (nares, nasopharynx) and high-density (oropharynx) upper airway communities in four healthy individuals. DNA was extracted with four different extraction methods (Epicentre Masterpure, Qiagen DNeasy, Mobio Powersoil and a phenol bead-beating protocol combined with Agowa-Mag-mini). Bacterial DNA recovery was analysed by 16S qPCR and microbiota profiles through GS-FLX-Titanium-Sequencing of 16S rRNA gene amplicons spanning the V5-V7 regions.

Results: Lower template concentrations significantly impacted microbiota profiling results. With higher dilutions, low abundant species were overrepresented. In samples of <10(5) bacteria per ml, e.g. DNA <1 pg/µl, microbiota profiling deviated from the original sample and other dilutions showing a significant increase in the taxa Proteobacteria and decrease in Bacteroidetes. In similar low density samples, DNA extraction method determined if DNA levels were below or above 1 pg/µl and, together with lysis preferences per method, had profound impact on microbiota analyses in both relative abundance as well as representation of species.

Conclusion: This study aimed to interpret microbiota analyses of low-density communities. Bacterial density seemed to interfere with microbiota analyses at < than 10(6) bacteria per ml or DNA <1 pg/µl. We therefore recommend this threshold for working with low density materials. This study underlines that bias reduction is crucial for adequate profiling of especially low-density bacterial communities.

Figure 1. Microbiota composition and Principal Coordinate of Analyses for serially diluted saliva.

a. Microbiota composition for undiluted and serially diluted saliva of individual 1 isolated with the Agowa method. For undiluted saliva, the dilutions and the PCR blank, relative abundance of the genera expressed in percentages are shown on the y-axes. The legend shows the 30 most abundant taxa and genera found in colors. Microbiota composition starts to deviate from the original sample at dilution 3 (10⁵ bacteria/ml). In dilutions 1 and 2 low abundant genera seemed to increase in abundance, while high abundant genera decrease in abundance. b. Unweighted UniFrac Principal Coordinate Analyses plot of undiluted and serially diluted saliva isolated with four DNA extraction methods. Great overlap in sequence representation was seen between undiluted samples and samples diluted up to dilution 2 (10⁶ bacteria per ml) for all DNA extraction methods except for Mobio. Dilution 3 samples (10⁵ bacteria per ml) are deviating from the original sample. DNA extraction methods are depicted in characters (E = Epicentre, M = Mobio, Q = Qiagen and A = Agowa). The dilutions are depicted in symbols as shown in the legend. Undiluted = undiluted saliva, 1 = dilution 1, 10⁷ bacteria/ml, 2 = dilution 2, 10⁶ bacteria/ml and 3 = dilution 3, 10⁵ bacteria/ml.

Figure 2. Average relative abundance of the 6 main taxa in undiluted saliva and dilutions 1 to 3 (10⁷ to 10⁵ bacteria per ml, respectively).

Shown in error bars is the standard deviation per dilution indicative of the variation between DNA extraction methods. We used ANOVA statistics to test for significant differences. Dilution 3 shows a significant increase in Proteobacteria (p<0.001, mean 26,11% and 3.6%, SD 17.5 and 2.5% respectively) and a significant decrease in Bacteroidetes (p<0.001, mean 18.99% and 43.1% respectively) compared to the undiluted saliva samples. By diluting the sample up to 10⁵ bacteria per ml an increase in Firmicutes, mostly Veilonella, was observed, and a decrease in Bacteroidetes, mostly Prevotella.

Figure 3. Principal component analyses (PCA) of the microbiota profiles and Principal Coordinate of Analyses (PcoA) plot of the weighted and unweighted UniFrac average distance per site and DNA extraction method.

a. Principal component analyses (PCA) of the microbiota profiles of the nares and nasopharynx depicted per dilution and DNA extraction method. Depicted in colors are 16S DNA levels (blue = ≥1 pg/µl, green = <1 pg/µl). Depicted in characters are the DNA extraction methods (A = Agowa, E = Epicentre, Q = qiagen, M = Mobio). Clustering of the samples is according to DNA level and DNA extraction method. Differences in template concentration were due to differences in DNA extraction efficiency between used methods and effects of template concentration on microbiota analyses were therefore fully tied to DNA extraction effects. b.PcoA plot of the weighted UniFrac. Shown in colored circles are the DNA extraction methods (yellow = Epicentre, red = Mobio, blue = Qiagen and green = Agowa). The abbreviations represent the site of sampling (NP = nasopharynx, N = nares, OP = oropharynx, SA = saliva). Clear clustering per site of sampling was observed with saliva and oropharynx distant from nares and nasopharynx samples. For the oropharynx and saliva clusters significant sub-clustering per DNA extraction method was seen with clusters of Epicentre and Mobio, distant from Agowa and Qiagen clusters. DNA extraction method in these high density sites even introduced a larger distance in microbiota profile than origin of the sample (saliva or oropharynx). c. PCoA plot of the unweighted UniFrac as described above. Clear clustering per site of sampling was observed with saliva and oropharynx distant from nares and nasopharynx samples, and also between saliva and oropharynx. For the nares and nasopharynx clusters significant sub-clustering per DNA extraction method was seen with clusters of Agowa, Qiagen and Epicentre distant from Mobio. Both weighted and unweighted UniFrac analysis of sequence data revealed distinct clustering of saliva and oropharyngeal separate from nares and nasopharynx samples, reflecting unique differences in microbiota composition between these sites (Amova, p<0.001, Figure S4 and S5).

Figure 4. Phyla composition of individual 1 per sample site and DNA extraction method.

Samples extracted with Agowa and Qiagen showed a significant higher proportion of Actinobacteria and Firmicutes and a lower proportion of Bacteriodetes, compared to Epicentre and Mobio, especially in the oropharyngeal and saliva samples for all 4 individuals. Note that nasopharyngeal samples from all four individuals isolated with Epicentre failed to give results.