Details matter: Designing skin microbiome studies

Abstract

The use of genomic sequencing to investigate microbes has expanded, yet has also raised questions regarding optimal approaches to studying the skin microbiome. In this issue, Meisel et al. demonstrate that while whole genome shotgun metagenomic sequencing was most similar to the expected microbial profiles, V1–V3 sequencing of the 16S rRNA gene had greater accuracy than V4 sequencing in determining genus and species level classification of prominent skin bacteria.

COMMENTARY

Advances in genomic sequencing have accelerated investigations into the microbial worlds of humans and the environment. The commonly used method of sequencing the 16S ribosomal RNA (rRNA) gene enables determination of relative compositions of bacterial communities in a sample even if difficult-to-cultivate microbes are present, as reviewed in (Jo et al., 2016). Whole genome shotgun (WGS) metagenomic sequencing is more involved and expensive but it permits taxonomic classification as well as exploration of the full genomic complement, or functional potential, of mixed microbial communities including bacteria, fungi, and viruses (Hannigan et al., 2015; Human Microbiome Project, 2012; Oh et al., 2014). Given the growing interest in understanding complex host-microbial interactions (Cogen et al., 2010), the amount of microbiome research activity has greatly increased (Human Microbiome Project, 2012; Waldor et al., 2015). The quality and outcomes of microbiome studies are substantially influenced by multiple aspects of study design (Goodrich et al., 2014).

The microbial communities in different epithelial sites (gut, oral mucosa, vaginal mucosa, nares, and skin) are distinct; therefore, identifying factors unique to skin-specific investigations is important to advance the field (Costello et al., 2009; Grice et al., 2009). Primer selection for 16S rRNA gene sequencing is a critical factor in skin microbiome studies. Of the nine hypervariable regions in the 16S rRNA gene, targeted V4 sequencing is the most frequently used method because it helps differentiate fecal bacteria in gut microbiome studies. In contrast, prior studies have underscored the importance of targeting the V1–V3 regions of the 16S rRNA gene to optimally identify important bacterial species common to skin, particularly staphylococci (Chakravorty et al., 2007; Conlan et al., 2012; Jonasson et al., 2002; Jumpstart Consortium Human Microbiome Project Data Generation Working, 2012). In this issue, Meisel and colleagues systematically compared the ability of current sequencing strategies to taxonomically classify skin bacteria and confirmed prior analyses recommending V1–V3 targeted sequencing for skin microbiome studies (Meisel et al., 2016).

Using V1–V3, V4, and WGS sequencing methods, Meisel et al. analyzed the bacterial communities identified from skin samples obtained from topographically distinct skin sites and from a standardized mock community with known bacterial DNA content which was developed as a resource of the National Institutes of Health Common Fund’s Human Microbiome Project (Jumpstart Consortium Human Microbiome Project Data Generation Working, 2012; Meisel et al., 2016). The results describing the compositions of skin bacterial communities varied based on sequencing strategy. For the mock community, WGS sequencing results were similar to the expected results. V1–V3 sequencing results more closely resembled expected 16S rRNA gene profiles than the V4 sequencing results, which showed lower relative abundances of S. epidermidis and P. acnes and higher relative abundance of S. aureus. While targeting the V4 region of the 16S rRNA gene more effectively determines gut-associated bacteria, the V4 dataset resulted in inaccurate assessment of the relative abundances of prominent bacteria of human skin. These notable differences based on comparing results with a mock community standard reiterate the importance of selecting V1–V3 primers for investigating these skin bacteria in order to achieve an outcome that more closely resembles the true microbial composition.

The ability to distinguish between different species of bacteria has important clinical and biological implications. For example, the genus Staphylococcus includes known commensal (S. epidermidis) and potentially pathogenic (S. aureus) species, and distinguishing between S. epidermidis and S. aureus is feasible (Conlan et al., 2012; Kong et al., 2012). Using pplacer, Meisel et al. optimally classified staphylococcal sequences with the V1–V3 and WGS sequencing datasets but only classified <1% of staphylococcal sequences from the V4 dataset (Matsen et al., 2010; Meisel et al., 2016). Thus, the ability to draw potential clinically meaningful conclusions regarding particular genera including Staphylococcus strongly endorses V1–V3 sequencing for skin-associated bacteria as compared to V4 sequencing.

Extending beyond the species level, there is greater recognition of the variability among strains from the same bacterial species (Fitz-Gibbon et al., 2013; Oh et al., 2014). Identification of strains and analyses of the functional potential of a given microbial community are possible with WGS sequencing data. However, the cost and complex bioinformatics of WGS sequencing continue to be relatively high in contrast to 16S rRNA gene sequencing. An example of computational tools developed for microbiome analyses, PICRUSt, uses the metagenomic data from available bacterial reference genomes to predict functional gene profiles from 16S datasets (Langille et al., 2013). In comparing PICRUSt’s ability to predict functional pathways reflective of WGS data, Meisel and colleagues identified discrepancies between the WGS data and the predicted metagenomes based on V1–V3 and V4 data. Reverting to less specific KEGG pathways was required to identify correlation trends between predicted functional profiles and WGS data in this study. Predictions of functional profiles based on 16S rRNA gene sequencing data are approximations that rely on a finite number of reference genomes and require validation studies. Recent investigations of bacterial strain-level differences highlight the limitations of 16S rRNA gene-based predictions (Greenblum et al., 2015; Zhu et al., 2015). Improving investigations of functional differences will require significant expansion of biological pathway annotations as well as of publicly available reference genomes and WGS datasets.

Through comparisons of sequencing strategies, Meisel et al.’s work emphasizes the importance of the elements of skin microbiome study design in achieving relevant results. In addition to examining primer selection for skin microbiome studies, the authors incorporated other crucial study design elements including consistency in sampling skin sites based on known topographical heterogeneity of skin microbial communities; parallel sequencing of DNA from the same clinical samples to enable direct comparisons of sequencing methods; incorporation of a standardized mock community to provide a benchmark for assessing quality and comparability of sequencing runs; and use of negative control swabs to monitor for contamination of low biomass skin samples. The findings that targeting the V1–V3 region better differentiates particular skin bacterial sequences at the genus and species level than targeting the V4 region underscore a need for study design standardization based on scientific goals of each investigation. The limitations of predictions of functional profiles also highlight the need for further research into biological pathways and microbial genomes to improve database annotations and our understanding of complex host-microbial interactions. As the microbiome field continues to expand, adoption of standardized methodologies that have been systematically studied will improve our ability to develop a shared language to advance scientific breakthroughs.

CLINICAL IMPLICATIONS.

• Microbiome sequencing provides a more comprehensive determination of skin microbial communities as compared to traditional cultivation methods.

• Of affordable methods, V1–V3 sequencing of the bacterial 16S rRNA gene significantly outperformed V4 sequencing in classifying the genus and species level of prominent skin bacteria.

• Whole genome shotgun metagenomic sequencing most closely represented expected profiles of a standardized mock community, highlighting that results from this expensive method are most reflective of the composition and functional profiles of the skin microbiome.