Benchmarking DNA isolation kits used in analyses of the urinary microbiome

The urinary microbiome has been increasingly characterized using next-generation sequencing. However, many of the technical methods have not yet been specifically optimized for urine. We sought to compare the performance of several DNA isolation kits used in urinary microbiome studies. A total of 11 voided urine samples and one buffer control were divided into 5 equal aliquots and processed in parallel using five commercial DNA isolation kits. DNA was quantified and the V4 segment of the 16S rRNA gene was sequenced. Data were processed to identify the microbial composition and to assess alpha and beta diversity of the samples. Tested DNA isolation kits result in significantly different DNA yields from urine samples. DNA extracted with the Qiagen Biostic Bacteremia and DNeasy Blood & Tissue kits showed the fewest technical issues in downstream analyses, with the DNeasy Blood & Tissue kit also demonstrating the highest DNA yield. Nevertheless, all five kits provided good quality DNA for high throughput sequencing with non-significant differences in the number of reads recovered, alpha, or beta diversity.

www.nature.com/scientificreports/ In addition to sample collection, handling and storage, the DNA isolation methods used are another important step for microbiome analysis where bias could be introduced prior to sequencing. At this step, human and microbial DNA are extracted from the proteins, salts, and other components of the physiologic sample. This requires lysis of human cells and bacterial cell walls in order to isolate the DNA contained within. When performing marker gene sequencing, the isolated DNA is later subjected to PCR, where the marker gene is amplified and uniquely tagged for sequencing. The bacterial 16S rRNA gene is one of the most commonly used marker genes used for bacterial identification. To date, a range of amplicons encompassing multiple different variable regions of the 16S rRNA gene including V2, V3-V4, V4, V4-V5, and V6 8,15,16 have been applied to urinary microbiome samples.
Regardless of which segment of DNA is used as the marker gene, reliably isolating all of the DNA in a sample is an important step prior to PCR and sequencing. Many commercial kits and custom protocols for DNA isolation were developed for microbiome analyses specifically for microbe-rich or microbe-poor environments 19,22,25,26 . This tailoring of DNA isolation methods was required to achieve more representative identification and quantification of the microbial composition for each respective environment. Nevertheless, different methods for DNA isolation show variable efficiencies of DNA recovery and quality. A large number of studies report significant differences in microbial composition identified with the use of different DNA isolation protocols [17][18][19][20][21][22][23][24] . Biases introduced by the DNA isolation methods to microbial composition persist both in microbe-rich communities such as gut, soil, sewage 19,25 and in microbe-poor communities such as water, meconium, and animal larvae 22,26 . However, there are occasionally studies that do not show notable differences among DNA extraction methods in other microbial niches 23 .
Most studies that examine differences among DNA extraction protocols note that the main hurdles are incomplete cellular lysis and presence of PCR inhibitors that could interfere with downstream sequencing. Incomplete cellular lysis for some microbes biases the compositional analyses towards more easily lysed taxa. These differences in lysis were repeatedly recorded for Gram positive bacteria and fungi that both have more robust cell walls in comparison with Gram negative bacteria 20,26 . For the urinary microbiome, researchers have previously found a significant bias in the ability to detect fungi due to the inability to efficiently lyse hardy fungal cells 13,27 . However, many urinary microbiome studies employ a variety of DNA isolation techniques without considering these potential sources of bias. Furthermore, microbe-poor communities, such as urine, are especially sensitive to potential influence from contaminants, which should be taken into account during analysis 6,28 . Table S1 summarizes multiple studies to highlight the diversity of the methods of DNA isolation used to date. These studies use both custom and commercially available DNA isolation methods. As there are no studies directly comparing the results obtained when urine samples are subjected to different commercial DNA isolation kits, our primary objective was to assess whether recovered microbes identified by 16S rRNA sequencing differ based on the DNA isolation protocol.

Results
DNA recovery and performance in high throughput sequencing. A total of 11 urine samples and one negative control containing phosphate buffered saline (PBS) were equally divided and subjected to parallel DNA isolation procedures with five DNA isolation kits ( Table 1). The total DNA concentration recovered from each DNA isolation kit was highly variable (Fig. 1A, Table S3) with the Qiagen DNeasy Blood and Tissue kit resulting in the highest concentrations and the Promega kit with the lowest (Kruskal-Wallis p = 0.0007 overall, pairwise Wilcoxon rank sum < 0.05). Since each aliquot from urine samples contained the same starting material per kit, and each kit elutes DNA into a 50 µL volume, higher concentrations would reflect a higher total amount of DNA isolated. Of the 60 samples (55 from urine and 5 controls), a total of 7 (11.6%) did not produce identifiable bands on gel electrophoresis after PCR amplification of the V4 region of the 16S rRNA gene (Table S3, Fig.  S1A). The majority of these samples were derived from one urine specimen with the lowest quantity of recovered DNA (Sample 11) and negative controls, suggesting truly low quantity DNA in these samples. However, in one instance (Sample 4), no gel band was detected after PCR when DNA was extracted with the Qiagen DNeasy Ultraclean kit though bands were identified when DNA was isolated with all of the other four kits.
Despite the differences in DNA concentrations between isolation kits, DNA isolated from all kits appeared to perform similarly in high throughput sequencing, possibly due to the general normalization of starting DNA amounts in the PCR amplification step prior to library preparation and sequencing.
We did not identify significant differences in the total number of recovered reads based on the DNA isolation kit (Kruskal-Wallis p = 0.806, Fig. 1B). Notably, sequencing reads were obtained even in negative controls and in samples without gel bands after PCR that might have originally been presumed to be devoid of DNA. www.nature.com/scientificreports/ Microbial composition. Alpha diversity measures summarize the composition of bacteria in a sample in terms of the numbers of different taxa present (richness) and their distribution (evenness). We did not identify significant differences in alpha diversity measured as the number of observed genera, the Shannon index, or the inverse Simpson index based on DNA isolation kit (Kruskal-Wallis p = 0.292, 0.363, and 0.436, respectively; Fig. 2).
To evaluate the differences in the overall composition of taxa between DNA isolation kits, we estimated beta diversity using the Bray-Curtis distance and nonmetric multi-dimensional scaling (NMDS, Fig. 3A), and evaluated the relative abundance of recovered bacteria in each sample (Fig. 3B). For most of the samples, the composition appears to be consistent despite the DNA isolation kit that was used. As such, the overall microbial composition was not significantly different based on the DNA isolation protocol (PERMANOVA p = 0.87, permutations = 999), with the exception of Sample 7, which displays high variability in both the relative abundance and NMDS plots. As expected, recovered microbes differed significantly between the 11 urine samples (PER-MANOVA p = 0.001, permutations = 999).

Recovery of Gram positive versus Gram negative bacteria.
Prior studies comparing methods of DNA isolation from non-urine microbiome samples strongly indicated that the envelope structure of Gram positive organisms represents an impediment for uniform cell lysis [17][18][19][20][21][22][23][24] . Therefore, we analyzed whether DNA isolation kits biased the identified microbial composition towards Gram negative species. We compared relative abundances among all genera with known Gram staining of representatives ( Fig. 4 and Fig. S2 for individual sample results). Four out of five DNA isolation kits yielded comparable overall relative abundances of Gram positive bacteria. The Promega kit resulted in fewer Gram positive bacteria, though this was not statistically significant (Kruskal-Wallis, p = 0.197), likely due to the small sample size and highly variable data.
Microbiome studies of different niches reveal that overall composition is important for health and disease. However, infectious disease studies also show that specific microbes may be important for an underlying condition. Therefore, we further analyzed the presence of eight specific genera relevant for the urinary niche (Fig. 5). These include genera containing three known urinary pathogens (Escherichia, Klebsiella, Enterococcus) and five genera typically considered as commensals (Lactobacillus, Corynebacterium, Prevotella, Staphylococcus, Gardnerella). This analysis shows a non-significant reduction of Enterococcus, Corynebacterium, and Staphylococcus genera for the Promega kit. It also appears that the PowerSoil kit could recover more 'easy-to-lyse' Gram negative organisms such as Klebsiella and Escherichia compared to the other kits. In our study, this phenomenon was driven by approximately half of the samples (see Fig. 3B and Fig. S2, Samples 2,4,7,9,and 11).
To assess the trend towards bias in recovery of specific types of bacteria, we further tested DNA isolation results in a subset of kits using a well characterized mock microbial community (ZymoBIOMICS Microbial Community Standard, Zymo Research, Irvine, CA). This community contains known quantities of 8 bacteria, including both Gram positive and Gram negative microbes. In this secondary experiment, we did not confirm differences among kits based on Gram staining status, though we were unable to include the Promega kit due www.nature.com/scientificreports/ to a loss of proprietary instruments in our facility. Rather, we observed that results were highly variable by kit in extremely low microbial biomass samples. DNA isolation kits performed similarly with undiluted or slightly diluted bacteria. However, when bacteria were substantially diluted, replicating the bacterial cell content of urine, only DNA extracted with the DNeasy Blood & Tissue kit resulted in all of the expected organisms after sequencing (Fig. S3A). When substantially diluted, other kits showed varying amounts of expected organisms with a substantial amount of additional contaminant bacterial DNA (Fig. S3B, Fig. S4).

Discussion
The field of urinary microbiome research is still relatively new. As such, studies benchmarking DNA isolation kits and their performance in recovering urinary microbial composition data are lacking. This study aimed to compare several methods of isolating microbial DNA from human urine. In particular, we compared five commercially available DNA isolation kits and estimated not only the quantity of DNA, but also the quality of DNA when utilized in downstream compositional analyses. It has previously been shown that biases are introduced to microbial composition analysis based on the DNA isolation technique in both high biomass and low biomass communities of microbes 19,22,26 . Our results echo those found in oral microbial communities, where the DNA isolation method may result in significantly different DNA yield, though overall non-significant differences in downstream sequencing 23 . Though many of our downstream assessments showed non-significant differences, our data do not support the assumption that all DNA isolation kits perform equally in urinary microbiome studies, as we identified some qualitative differences in recovery of Gram positive versus Gram negative organisms, and some differences in overall performance with low biomass samples. Since microbiome data are typically presented in terms of relative abundance, if one type of microbe is absent due to a technical bias, it will artificially make other microbes appear more abundant. This is evident when viewing graphs in Figure S2, where relative abundances of Gram positive and Gram negative bacteria are inversely proportional to each other.
In our study, after initial PCR amplification, four samples and three controls derived from extremely low quantities of DNA failed to show a band on electrophoresis. The lack of amplified DNA after beginning with extremely low quantities of DNA could be expected. However, in one instance a sample extracted with the UltraClean kit had normal quantities of starting DNA with no evident PCR product on electrophoresis. This one result could have been spurious or possibly indicative of the presence of PCR inhibitors in the sample, as have been identified in other studies 12,13 .
Our findings are strengthened by the multiple ways in which we assessed quality of DNA after isolation. This included evaluation of PCR products, assessment of the number of sequencing reads after high-throughput sequencing, as well as detailed compositional analyses of microbial data. We utilized an updated and rigorous bioinformatics pipeline to identify the genera corresponding to recovered sequences. We then utilized this information to assess the quality of sequencing information, which revealed some initial differences based on www.nature.com/scientificreports/ Gram staining characteristics and in urogenital genera that are highly relevant to the urinary microbiome field. However, these initial differences were not confirmed in a secondary experiment with a mock microbial community. Rather, we established that high variability and potential contaminants can be observed in dilute, low biomass environments. These results confirm the importance of including a dilution series of positive controls when performing sequencing, as has been previously described, to control for potential contaminants during bioinformatic processing of low biomass sequencing data 28 .
Our study certainly has multiple limitations, which are mainly related to technical factors. After assessing recovered DNA quantity using Qubit, we did not perform additional testing to assess the proportion of microbial versus human DNA contained in each sample. Thus, it is unclear if differences identified in total DNA recovery actually translate to differences in microbial DNA within different samples, which is the component of interest

Figure 5.
Comparison of relative abundances of genera with biologically-significant representatives. We compared relative abundances of bacteria recovered from eight genera with high biologic relevance including urinary pathogens (Escherichia, Klebsiella, Enterococcus) and commensals (Lactobacillus, Corynebacterium, Prevotella, Staphylococcus, Gardnerella). The Promega kit tended to recover fewer Gram positive Corynebacteria, Enterococci, and Staphylococci compared to other kits while the PowerSoil kit recovered more Escherichia and Klebsiella compared to other kits. In follow up experiments, we did not confirm a bias in favor of Gram negative organisms for the PowerSoil kit. Rather, the PowerSoil kit was substantially influenced by contaminants in low biomass environments (See Fig. S3 and S4). www.nature.com/scientificreports/ in microbiome studies. Another limitation was inherent in the need to divide urine samples. Though one urine sample was produced, we needed to ensure that it was equally divided prior to performing parallel testing with five kits. Since the biomass (e.g. cellular material containing DNA) may not be evenly distributed within the fluid of a urine sample, we addressed this issue by first centrifuging whole urine to produce a cell pellet containing the biomass. This cell pellet was then reconstituted in a smaller volume, thoroughly mixed, and then divided into five aliquots. However, it is still possible that due to pipetting or mixing errors, slightly different amounts of starting material were present in aliquots, which could have contributed to some of the variability seen in our results. However, we believe this factor is less important since urine volume did not correlate with biomass. For example, as shown in Tables S2 & S3, a 50 mL sample (Sample 3) had the highest amount of recovered DNA while another 100 mL sample had the lowest amount of recovered DNA. We utilized a negative control (PBS buffer) that was processed and sequenced in parallel to the urine samples. Though there was no starting DNA added to this sample, we recovered a small number of sequences (Fig. S1) suggesting presence of low level contaminants. Unfortunately, we did not use separate controls at each analytic step and thus we are unable to distinguish the sources of the observed contamination, which could come from plastics in the laboratory, reagents within the DNA isolation kits, or during multiple technical steps prior to sequencing. This study utilized voided urine, which is more reflective of the urogenital microbiome than the bladder microbiome. Since we are not attempting to characterize a niche, the method of urine sample acquisition is less important. However, microbes from the vagina are found in higher abundance in voided compared to catheterized urinary samples, and thus may have higher representation in the compositional data presented here. Since vaginal and urinary microbes are highly related in terms of the genera and species represented, vaginal contamination theoretically should not negatively impact the results of this benchmarking study 29,30 . Nevertheless, studies such as this one would ideally be replicated numerous times to confirm the findings.

Conclusions
When considering the totality of our findings, DNA extracted with the Qiagen Biostic Bacteremia and DNeasy Blood & Tissue kits showed the fewest technical issues in downstream analyses, with the DNeasy Blood & Tissue kit also demonstrating the highest DNA yield. Nevertheless, all five kits provided good quality DNA for high throughput sequencing with non-significant differences in the number of reads recovered, alpha, or beta diversity. In qualitatively assessing the types of bacteria, the Promega and DNeasy PowerSoil kits appear to have some biases towards over-representing certain Gram negative bacteria of biologic relevance within the urinary microbiome. This bias in the DNeasy PowerSoil kit was not confirmed in follow up analysis of a mock microbial community. Rather, analyses in a mock microbial community confirmed that kits may perform differently when applied to high versus low biomass environments, and that we should anticipate and control for potential contaminants in low biomass samples. These findings have implications for research teams wishing to maximize utility of low biomass samples, particularly for sequencing strategies where more DNA is required. Furthermore, these findings are relevant for interpretation of microbiome studies. The results presented here are certainly in line with other microbiome niches suggesting that the DNA isolation methods used could potentially bias downstream results. As such, we urge caution to investigators when selecting which DNA isolation method is used in future urinary microbiome studies, caution to the scientific community when assessing findings from studies where isolation methods with known bias were used, and further urge a high level of caution in general when trying to compare or extrapolate results from studies where different DNA isolation methodologies were used.

Materials and methods
Sample collection and processing. This study was deemed exempt by the Duke University Institutional Review Board (Pro00085111). Following all relevant guidelines, de-identified voided urine samples were collected in sterile cups from the Duke Urogynecology clinic, refrigerated (4 °C), and processed within 4-10 h (Table S2). As the study was deemed exempt by IRB no consent was obtained.
During processing, samples were handled aseptically, transferred to 50 mL conical tubes and spun without any buffer pretreatment to collect all of the biomass, including human and microbial cells (4 °C, Eppendorf 5810R centrifuge, 15 min, 3220 rcf) represented in the "cell pellet". Supernatants were decanted and the remaining cell pellets with residual urine were transferred into sterile 1.5 mL tubes, then spun again at 10,000 rcf in the Eppendorf 5340R centrifuge for 5 min at 4 °C. The total cell pellet per sample was resuspended in sterile filtered phosphate buffered saline (PBS) on ice. Re-suspended pellets were divided into 5 identical aliquots, and stored at -80 °C until DNA isolation. DNA isolation procedures. This step started with the five identical aliquots and thus the same starting material was processed in parallel with five commercially available DNA isolation kits. Each kit had differing levels of chemical, mechanical, and enzymatic cell lysis, as summarized in Table 1. PBS buffer was used as a negative control sample with each DNA isolation kit. For the Qiagen DNeasy Blood & Tissue kit we performed the optional steps as recommended in the protocol for optimizing recovery of Gram positive bacteria. All samples were assessed using the Agilent 2100 Bioanalyzer, Promega GlowMax spectrophotometer and ThermoFisher Qubit HR reagents to determine the quality and quantity of recovered DNA. Recovered DNA concentrations are provided in Table S3.
Bacterial ribosomal DNA amplification and sequencing. DNA samples and negative control were subjected to PCR in order to amplify the V4 variable region of the 16S rRNA gene. For PCR, forward primer 515 and reverse primer 806 were used following the Earth Microbiome Project protocol (http://www.earth micro biome .org/). These primers (515F and 806R) carry unique barcodes allowing for construction of a library of