Evaluation of Oral Cavity DNA Extraction Methods on Bacterial and Fungal Microbiota

The objective of this study was to evaluate the most effective method of DNA extraction of oral mouthwash samples for use in microbiome studies that utilize next generation sequencing (NGS). Eight enzymatic and mechanical DNA extraction methods were tested. Extracted DNA was amplified using barcoded primers targeting the V6 variable region of the bacterial 16S rRNA gene and the ITS1 region of the fungal ribosomal gene cluster and sequenced using the Illumina NGS platform. Sequenced reads were analyzed using QIIME and R. The eight methods yielded significantly different quantities of DNA (p < 0.001), with the phenol-chloroform extraction method producing the highest total yield. There were no significant differences in observed bacterial or fungal Shannon diversity (p = 0.64, p = 0.93 respectively) by extraction method. Bray-Curtis beta-diversity did not demonstrate statistically significant differences between the eight extraction methods based on bacterial (R2 = 0.086, p = 1.00) and fungal (R2 = 0.039, p = 1.00) assays. No differences were seen between methods with or without bead-beating. These data indicate that choice of DNA extraction method affect total DNA recovery without significantly affecting the observed microbiome.

data processing, and statistical analyses 21 . Additionally, each of these steps has associated labor and cost factors that may influence a researcher's decision to use one method over another 22 . Previous research has shown that oral sampling techniques such as saliva, buccal swab, and oral rinse collection may influence overall DNA quantity and spectrum of microbes detected [23][24][25] . It has also been suggested that next-generation sequencing (NGS) may produce variable results particularly when analyzed using different classification algorithms 26 . Given that these processes can influence the understanding of microbial communities, investigating protocols for characterizing the biota of the oral cavity is important to allow inter-study comparisons. Efficient and consistent methods of DNA extraction are central to accurately characterizing these communities. A number of studies have begun to examine the oral microorganisms using NGS with a variety of DNA extraction methods [27][28][29][30][31][32][33] . In addition, a large number of studies have collected and processed Scope mouthwash samples for genomic DNA that might be suitable for microbiome studies. The purpose of this investigation was to compare the most recent techniques to discern the most effective method of DNA extraction utilizing both enzymatic and mechanical lysis techniques across various human oral samples in order to determine the methods with the highest DNA yield and the most consistent results for characterization of both bacterial and fungal communities found in the oral cavity.

Results
For this study, eight DNA extraction methods, utilizing different combinations of enzymatic and mechanical lysis techniques, were compared across six oral samples ( Table 1). The methods were evaluated for DNA yield and variation in the detected oral microbiome. There was a significant difference in DNA quantity among the eight extraction methods (p < 0.001). The phenol-chloroform extraction technique (Method 1) generated the highest DNA yield (Fig. 1) while the UltraClean Microbial DNA Isolation Kit (Method 7) and the UltraClean Microbial DNA Isolation Kit (Method 8) resulted in significantly lower DNA yields (p < 0.01) than the three non-bead-beating methods ( Table 2).
DNA from the 48 DNA samples were amplified using 16 S rRNA V6 barcoded primers and recently described primers for the ITS1 region and submitted for Illumina NGS. Raw sequences were processed for quality control and chimera removal, resulting in a total of 373,840 bacterial reads (average of 7,788 ± 1,837 reads per sample), and 363,881 fungal sequence reads (average of 5,965 ± 1,579 reads per sample). The bacterial community composition and normalized abundances in the oral cavity are displayed in the heat map ( Fig. 2A). Dendrogram clustering based on the top 20 species shows a tendency of samples to cluster by original subject. DNA extraction method did not show clustering. Community clustering based on the top 20 fungi (Fig. 2B) displays a closer distance between samples than seen with the bacterial 16 S data. However, the fungal heatmap also indicated that samples tended to cluster together based on subject and not extraction method.
Seven bacterial phyla were identified; Firmicutes, Bacteroidetes, Actinobacteria, Fusobacteria, Proteobacteria, Spirochaetes, and the candidate phylum TM7 (also known as Saccharibacteria), with the majority of OTUs assigned to Firmicutes and Bacteroidetes. At the genus/species level, Streptococcus dominated the oral cavity, consistent with published studies 27 . Rothia mucilaginosa 34 , an opportunistic pathogen in immunocompromised patients and Prevotella veroralis, a biofilm forming opportunistic pathogen 35 , were the second and third most abundant species, respectively ( Fig. 2A).
The oral mycobiota was dominated by species from Ascomycota, Basidiomycota, an unidentified fungal phyla, and Zygomycota (order based on cumulative dominance across all samples). Constituents of the Candida genus were amongst the top identified OTUs consistent with previous reports on the oral mycobiome 36  of Malassezia were also identified in the oral cavity, including Malassezia restricta (Fig. 2B), a common lipid dependent human pathogen that is usually found on skin 37 . Significant variation in sample evenness, based on the Shannon diversity index, was observed in the bacterial p < 0.001 and fungal p < 0.001 assays (Fig. 3A,B respectively). There was no significant difference in the Shannon diversity index among DNA extraction methods for either the bacterial p = 0.87 or fungal assays p = 0.93 (Fig. 3C,D, respectively). Similarly, β-diversity showed distinct clusters formed on the basis of subject in both the bacterial p < 0.001 and fungal p < 0.001 community analyses, which explained nearly all of the inter-sample community variance, R 2 = 0.80 and R 2 = 0.84, respectively (Fig. 4A,B). β-diversity analyses did not show significant sample clustering based on extracted method for either bacteria R 2 = 0.086, p = 0.996 or fungi R 2 = 0.039, p = 1.00 (Fig. 4C,D, respectively).

Discussion
In the current study, eight methods for DNA extraction from six oral cavity samples were used and DNA quantity and microbial community composition were compared. Our analysis revealed that DNA yield was significantly different among the eight DNA extraction methods with DNA recovery greatest after phenol-chloroform extraction ( Fig. 1). The lower DNA yield of commercially available kits (Table 1) compared to the phenol-chloroform extraction method may be due to DNA loss during silica column purification. DNA yield tended to be greater with enzymatic digestion than using mechanical lysis (bead-beating) approaches. The lower DNA yield among bead-beaten samples is likely due to DNA degradation during mechanical disruption. Thus, for DNA yield, non-bead-beating methods, particularly phenol-chloroform extraction provides the greatest yield of total DNA.
Although DNA for human genetic studies has frequently been obtained using oral mouthwash and/or saliva collection methods 38 , compatibility of the DNA from these studies for future microbiome studies has not been examined in detail. Previous studies found differences in the oral bacterial microbiome based on DNA extraction methods 32,33,39 ; whereas, other studies determined that DNA extraction methods did not seem to influence major differences in the oral microbiome 22,31,40 . Nevertheless, it is hard to do a direct comparison amongst studies in the literature, since many used saliva and/or plaque collection 31,33,39,40 , some compared crude DNA to purified DNA 39,40 , others used mock communities 39 , and one did not include NGS analysis of the microbiome 32 . Only one study examined both bacterial and fungal communities and surprisingly found no differences amongst 4 methods for bacterial communities, but found phenol-chloroform extraction best for fungal community diversity 33 .
Although we found that DNA extraction methods had an influence on DNA yield, we did not find a significant difference in oral microbiome composition across eight DNA extraction methods of oral rinse specimens. Shannon diversity measures for bacterial and fungal communities were similar across the employed extraction methods and did not achieve statistically significant differences. Similarly, PERMANOVA analysis on rank order Bray distances did not demonstrate differences in β-diversity for either assay. Our results instead demonstrated that individual subject differences drove diversity measures across both bacteria and fungi. Taken together, these data suggest that both αand β-diversity measures were consistent for all eight-extraction measures, and that the choice of method does not have a major influence on the observed oral communities. The results of this study might have been influenced by the larger number of samples analyzed compared to previous studies. All eight extraction methods were able to consistently recapitulate the original subject microbiotas as indicated by both alpha and beta diversity measures including Shannon diversity index and Bray-Curtis distances, respectively. These findings are consistent with previous studies that have demonstrated that each person's oral microbiome is unique 41,42 . Additionally, all methods reported here detected hard to lyse gram-positive species,  Table 1 and corresponding to the categories shown on the x-axis. All methods used the same starting quantities of sample and final volumes were equal; concentrations are proportional to total DNA recovered. Statistical analyses of the differences in DNA amounts recovered are shown in Table 2 such as Streptococcus 43 , indicating sufficient lysis of cells. Moreover, the similarity of results for fungal community analyses across all methods is consistent with the one report that found phenol-chloroform extraction yielded the highest fungal diversity in saliva 33 .
In summary, our study compared eight DNA extraction methods tested on oral rinse specimens that are commonly collected in large epidemiological studies and are used or may be used in the future to study the oral microbiome. While the eight methods tested in this study had significantly different DNA recovery, there was no difference in the observed oral microbiotas among methods. This study provides empiric evidence that research studies can select an appropriate DNA extraction method with or without bead-beating for characterization of the oral microbiota without influencing differences between the oral microbiome/mycobiome of individuals. IRB approval for analysis of pilot oral specimens was obtained from the Human Research Protection Program (HRPP) of CUNY. All methods performed in this study were conducted in accordance with Hunter College (CUNY) university integrated IRB approved protocol (PT: 346358-9). Informed consent was obtained from study participants prior to sample collection. Upon receipt all used human specimens received a lab Sample ID and no information regarding, age, race, gender or any other identifying information was used in the presented study. specimen Collection. Consented study participants provided an oral sample by rinsing with 20 mL of Scope mouthwash for 20 seconds. The 20-second oral rinse was broken into two 5-second swish sessions and two 5-second gargle sessions. The oral rinse samples were frozen at −80 °C at the New York State Public Health Laboratory (NYPHL) office and were transported on dry ice to Albert Einstein College of Medicine, where they were immediately stored at −80 °C. DNA extraction. DNA was extracted from the oral rinse samples using eight DNA extraction methods based on physical and/or enzymatic lysis steps and isolation procedures (Table 1). Extraction methods with commercially available kits all used a silica-based column. One extraction method included a non-commercial method using phenol-chloroform. All DNA isolation methods evaluated in this study are either commonly used in DNA extraction or have previously been used in microbial analysis studies. For each method, 1 mL from each oral rinse sample was centrifuged (5,000 × g) for 5 minutes. The cell pellet was re-suspended in 100 μl TE buffer (10 mM Tris. Cl, pH 8.0, 1 mM EDTA) and used for DNA extraction. Upon completion of each extraction method, the purified DNA was eluted in 100 µl of elution buffer (pH 8.0) and DNA concentration was determined using a NanoDrop 2000 (Thermo Scientific, DE).

Method 2 (QIAamp DNA mini kit).
First, 20 μl of proteinase K (20 mg/ml) and 100 μl of Buffer AL were added to 100 μl of pelleted cells in TE. The samples were incubated at 56 °C for 10 minutes. After incubation, 100 μl of 100% ethanol was added to the samples and the DNA was purified following the manufacturer's instructions.

Method 3 (Enzymatic lysis followed by QIAamp DNA mini kit).
The pelleted cells in 100 µl TE were treated with lysozyme (0.84 mg/ml, Sigma Aldrich), mutanolysin (0.25 U/ml, Sigma Aldrich) and lysostaphin (21.10 U/ml, Sigma Aldrich) at 37 °C for 30 minutes. Subsequently, 20 μl proteinase K and 100 μl Buffer AL were added followed by incubation at 56 °C for 10 minutes. DNA was purified using the QIAamp DNA mini kit as described above.

Method 4 (Enzymatic and bead-beating lysis followed by QIAamp DNA mini kit). Pelleted cells
were digested using enzymes as in Method 3. After incubation, the mixture was treated with 15 μl proteinase K (10 mg/ml) and 150 μl Buffer AL (Qiagen) at 56 °C for 10 minutes. The samples were then transferred to a clean screw-cap tube containing 300 mg of 0.1 mm-diameter zirconia/silica beads (BioSpec, Bartlesville, OK) and mechanically lysed using a FastPrep-24 Instrument (MP Biomedicals, Santa Ana, CA) at 6.0 m/s for 40 seconds. Next, the samples were centrifuged (10,000 × g) for 30 seconds and 200 μl of the supernatant was added to a clean microcentrifuge tube containing 100 μl of 100% ethanol. DNA was isolated using the QIAamp DNA Mini Kit (Qiagen) as described above.   16S rRNA gene and ITS1 region amplification and massively parallel sequencing. To amplify the 16SrRNA gene region of bacterial species, an aliquot of 0.5 µl DNA from each sample and DNA isolation method was PCR amplified in a total reaction volume of 25 µl using barcoded primers spanning the V6 variable region of the 16 S rRNA gene as previously described 26  To amplify the ITS1 region of fungal species, 10 µl from each sample and DNA isolation method was PCR amplified in a total reaction volume of 25 µl using barcoded primers specific to the ITS1 region of the fungal ribosomal gene cluster 44   Bray-Curtis Dissimilarity Distances. NMDS plots on rank order Bray-Curtis distances were used to assess significance in bacterial and fungal community composition between individuals (panels A and B) and methods (panels C and D). Plot ellipses represent the 95% confidence regions for group clusters. Clustering by sample is highly significant for bacterial R 2 = 0.80 p < 0.001 (panel A) and fungal communities R 2 = 0.84 p < 0.001 (panel B) communities. DNA isolation method did not exhibit significant clustering in either bacterial R 2 = 0.086 p = 0.996 (panel C) or fungal communities R 2 = 0.039 p = 1.00 (panel D). Significance was determined using PERMANOVA analyses. The 16 S rRNA and ITS1 PCR products each were pooled at approximately equal molar DNA concentrations and purified using the QIAquick Gel Extraction Kit (Qiagen). Following library preparation using TruSeq DNA Sample Prep Kits (Illumina, San Diego, CA), the pooled 16SrRNA DNA was sequenced on an Illumina HiSeq. 2500 using paired-end 150 bp reads, while the pooled ITS1 DNA was sequenced on an Illumina MiSeq using paired-end 300 bp reads, by the Epigenomics and Genomics Core Facility, Albert Einstein College of Medicine (Bronx, NY).

Method 5 (Enzymatic lysis followed by PowerLyzer
Bioinformatics. MiSeq reads were demultiplexed using novocraft's novobarcode 1.00 45 based on sample specific barcodes 46 . Reads were left and right trimmed with PrinSeq. 0.20.4 47 to remove bases that fell below the PHRED score of 25. Paired end reads were merged with PANDASEQ. 1.20 48 using default settings. For 16S rRNA gene reads, OTUs were clustered using closed reference selection with USEARCH using a custom in-house database that contains reference sequences from Green-Genes 13.8 49 . Additionally reference sequences of an oral microbiome specific database, Human Oral Microbiome Database (HOMD) 50 , were retrieved in order to account for bacteria specific to the human oral cavity. Representative sequences were aligned using PyNAST 51 and phylogenic analyses were performed using FastTree 2.0 52 .
For fungal ITS1 reads, open reference OTU picking was employed with QIIME 1.9 53 open-reference OTU picking protocol as previously described 44 . The protocol was modified to use VSEARCH version 1.4.0 54 , which allowed for higher throughput. The OTU clustering threshold was changed from 97% to 99% sequence identity to account for fungal diversity. Sequence dereplication and chimera removal was performed as part of the QIIME's usearch quality control protocol prior to OTU picking with VSEARCH. Representative sequences for each OTU cluster were chosen based on sequence abundance. BLAST was used to assign the taxonomy 55 .
All data were processed in R version 3.2.1 56 . QIIME outputs were imported into R using the phyloseq 57 . package and further processed with vegan 58 , coin 59,60 , and reshape2 60 . Data visualization was performed using ggplot2 61 . General community clustering was performed on the 20 most abundant OTUs (in terms of mean abundance across all samples) collapsed based on shared taxonomy at the species level using ward.D2 hierarchical clustering. β-diversity was assessed using Bray-Curtis distances and significance was calculated with PERMANOVA using the adonis function from the vegan package 58 . Statistical ellipses from the ggplot2 package were used to visualize the sample and method clusters on the NMDS plots. α-diversity was analyzed based on the Shannon's alpha diversity and observed number of OTUs metrics and significance was determined using the Kruskal-Wallis test.