Optimisation of methods for bacterial skin microbiome investigation: primer selection and comparison of the 454 versus MiSeq platform

The existence of commensals is well established, with the term ‘human microbiome’ being used to describe the sum of all the micro-organisms including bacteria, fungi, viruses, archae and eukaryotes that live in or on a human host at a given time [1]. In recent years, as a result of initiatives such as the NIH Human microbiome project, the scientific community has made great progress in cataloguing these micro-organisms [2–5]. One of the most common sites for investigation is skin, due to the ease of sampling and the potential role of the skin microbiome in the aetiology of skin diseases, such as atopic eczema and psoriasis [6–9]. A challenge in studying the skin microbiome is that healthy skin generally harbours a low microbial biomass; several amplification steps are, therefore, required before sufficient starting material is available for the commonly used DNA sequencing approaches and this increases the risk of introducing sequence artifacts (e.g. chimeric DNA fragments) or detecting contamination from the environment or reagents used [10].

Early studies used amplification, sub-cloning and Sanger sequencing of the highly conserved 16S rRNA gene, which can provide sequence information over the entire length of the 16S rRNA gene in a single reaction. Although this is still the most comprehensive method of bacterial identification, it is expensive and time consuming. Next generation sequencing (NGS) technologies make it possible to carry out identification of members of the microbial community at a much lower cost and with a higher with a higher throughput. Although some regions in the 16S rRNA gene sequence are highly conserved within bacterial species, nine hypervariable regions can be used to determine the identity of different species or genera of bacteria; amplification of short read lengths across these hypervariable regions are sufficient for taxonomic assignment of bacteria [11] to the genus level. Hence, NGS platforms provide an ideal technology for studies of the microbiome despite the fact that the current NGS platforms are only able to generate relatively short read lengths (modal read lengths of 150 to 518 base pairs (bp)) [12, 13]. However, it is important to select the most appropriate hypervariable region for each sample type being analysed, as ‘primer bias’ [14] (selective loss of bacterial groups due to the inability to accurately sequence/identify certain bacteria), may limit the diversity of the bacterial species that can be identified in a sample. As bacterial diversity is variable across different niches, careful selection of the primers for 16S amplicon generation is essential so that the particular hypervariable region sequenced results in the most comprehensive capture of the microbial diversity for the niche under investigation [15].

Until recently, much of the NGS work on the human microbiome was performed using the Roche 454 NGS platform, which offers up to 1000 bp reads. However, increasingly, microbiome studies are utilising newer NGS platforms, such as the Illumina MiSeq, which have been reported to be more cost effective and accurate, but have shorter read lengths [13, 16]. The V4 hypervariable region is traditionally selected for work on the MiSeq as it provides adequate information for taxonomic classification of microbial communities and has demonstrated a lower error rate on the Illumina platform [17]. However, to accommodate the current MiSeq protocol (250 bp paired end sequencing), a longer region (~500 bp) of the 16S rRNA gene was considered for the current work. Our aims were, first, to determine the hypervariable region which will provide the most comprehensive capture of the microbial diversity in human skin communities; second, to confirm that the primer set for amplification of that hypervariable region is specific and will not amplify human DNA contaminants and finally, to compare the performance of the MiSeq with the Roche 454 platform, which has been the established technology. A low biomass mock bacterial community was used to test, optimise and develop the protocol. Skin swabs from healthy volunteers were also tested to confirm the method is applicable to the study of the human skin microbiome.

In this paper we have described the development of a methodology for investigation of a very low biomass bacterial community using the MiSeq platform and compared data to that generated using the Roche 454 Junior system.

Comprehensive identification of any bacterial community is dependent on several steps including a robust bacterial DNA extraction method. A major concern with current protocols is their efficiency for extracting DNA from Gram positive bacteria, which have cell walls more resilient to lysis when compared to Gram negative bacteria [22, 23]. Given that Gram positive bacteria are commonly found on skin [24], namely Staphylococcus, a strength of the current study is that the bacterial DNA extraction method was first modified to ensure robust and reproducible capture of these species.

Previous studies [12, 13, 25] have compared the MiSeq and the 454 platforms but have used either mock bacterial communities [12, 13] or a single clinical sample [25]; by contrast, we have used both a mock community to show the accuracy of the V3-4 primer pair on the MiSeq platform when compared to the Roche454 and human samples to demonstrate that the primer pair will amplify sufficiently even in the low biomass environment of the skin.

The Microbiome Quality Control Project [26] recommends the use of a mock community containing taxa typical of site-specific microbial communities as a positive control in studies. Our work used the framework for human microbiome research [3] and included such a mock bacterial community to provide evidence for consistency and accuracy of the protocol. The mock community included a known skin commensal, Staphyloccocus species, which have been implicated in skin diseases such as atopic eczema where there is an increase in their abundance [27] and in psoriasis where reduced abundance has been reported [28]. In addition, the effect of simulated human DNA contamination was assessed. There have been studies on the effect of laboratory and reagent contamination in low biomass samples [10], on human-derived RNA-seq datasets [29] and bacterial contamination in human genome sequencing [30]. However, there is no information on the effect of human DNA contamination on the resulting output which is especially relevant in microbiome work on the skin as the bacterial biomass is low in comparison to the host genetic DNA that may contaminate the skin swab samples. Our work shows that the presence of human DNA does not affect the final output when compared to the results of the sequencing of the DNA from pure bacterial isolates increasing the confidence in the results obtained in human-derived samples. The data in the NIH Human Microbiome project [5] was derived from pyrosequencing using the Roche 454 FLX Titanium technology using amplicons generated from V1-V3 and V3-V5 hypervariable regions. The longer read lengths of 454 technologies cover a larger region of the 16S gene than the MiSeq and, therefore, enable easier taxonomic assignment. Shorter read lengths are sufficient to analyse microbial communities [11, 31], but limits the ability to maximize the capture of taxonomic diversity; therefore, we used the full capacity of the MiSeq with the 2x 250 bp paired end run (available at the time of the work) to achieve the optimal capture of data. Results demonstrated that the MiSeq performance was similar to the Roche 454 Junior platform in identifying community diversity for both the mock bacterial community and the skin samples at the phylum and the genus level.

Studies using the MiSeq platform have commonly used the V4 primer pair [11, 32] or the V3 primer pair [33], which limits the read length to ~250 bp. Our study demonstrates that the longer amplicons across the V3-V4 hypervariable region were identical for the abundance of the bacterial community identified and the beta diversity estimates when analysed on the MiSeq or Roche 454 Junior platforms. However, the MiSeq is more cost effective, as evidenced by the much higher coverage (>400x) obtained when compared to the Roche 454 Junior, using similar library preparation steps.

This work contributes to the evidence [12, 13, 34, 35] that the MiSeq is a suitable sequencing platform for profiling bacterial communities using analysis of 16S rRNA gene fragments. However, some limitations are apparent. We did not sequence negative controls which are considered as a measure of determining erroneous results [10] due to contamination as there was no detectable product in the samples. In our mock community we observed less than 5.5% in total abundance of other bacterial species and Rhizobium and Paracoccus, which have been reported as contaminating species in negative controls previously [10], showed abundances of less than 0.1%, lower than the traditional threshold (>1%) of abundance for inclusion in analysis. Another limitation of the current study reflects the rapidly evolving nature of microbiome research and the availability of advanced NGS with longer reads such as single molecule real time (SMRT) sequencing technologies which have not been explored or cross-validated. However, the results of this study show that the data generated by the MiSeq benchtop sequencer, which is now replacing the Roche454 platform in academic departments, is comparable. Therefore the data generated using the V3-V4 primer pair using the MiSeq can be compared with the existing data in the human microbiome literature.

Four of the eight skin samples showed high levels of chimera (~20%) on data analysis on the MiSeq platform and could not be confidently included in the comparative analysis between platforms. This is likely due to PCR artefacts introduced as a result of incomplete extension of the DNA template, or may be linked to the low starting biomass available, as well as the limitation of the sequence read length of the MiSeq. PCR based limitations are inherent in these analyses due to the need for amplification from low biomass samples and need to be taken into account when preparing libraries. Strategies to suppress the artificial chimera formation during PCR amplification have been reported [36–39] and could be used to minimize chimera generation during PCR. The loss of samples, though not anticipated at the outset of the work, has also been observed in a larger study investigating the skin microbiota in psoriasis [28]. Even in that study, the data from one of the primer pairs which targeted the V3-V5 hypervariable region could not be analysed as ~60% of the samples had to be removed due to amplification and sequencing depth quality issues even before the data analysis could be conducted. In the current work only one paired sample was lost due to sequencing depth issues and there were no losses due to PCR amplification. However, at the data analysis stage, four of the remaining skin swabs had high levels of chimeras detected and were therefore discarded from further analysis. This highlights the limitations that researchers should anticipate and try to mitigate against when studying low bacterial biomass samples. The mock community data could only be assigned to the family level in the taxonomic identification. This was considered to be the result of a combination of factors including the drop in quality of the MiSeq reads beyond 200 bps which resulted in a compromise in the quality of the 50 bp overlap, necessary for joining the paired end reads and the subsequent read/sequence assignment. It is expected that the upgraded version of the MiSeq chemistry, supporting the 300 bp paired end runs, will reduce the bias introduced due to inadequate read overlap at the sequencing stage and allow better taxonomic assignment. However, it was observed that in the HV skin samples, this was not a limiting factor as bacteria could be identified to the genus level despite using the version 2 chemistry on the MiSeq.

Our study reports on the optimization of a protocol that can be used in investigating a low microbial biomass community, such as in skin samples. We have shown that the method is robust to spiking of human DNA into a low bacterial skin biomass mock community sample. Furthermore, we have demonstrated that the community diversity can be captured using the V3-V4 hypervariable region of the bacterial 16S rRNA gene for NGS and that the MiSeq is a suitable next generation sequencing platform, providing comparable data to the Roche 454 GS Junior, with similar capture of bacterial diversity, but with a much improved throughput and cost effectiveness. Therefore, our results provide evidence for validation of the MiSeq platform against the Roche454 using a known mock bacterial community containing typical skin bacteria and also evidence that the optimised method can be used in human clinical samples.