Ethics statement

Samples were selected from a larger study examining the spatio-temporal distribution of A(H1N1) pdm09 across British Columbia during the 2009 pandemic. Identifying data was stripped from the sample record, leaving only patient gender, age, regional health authority, specimen type and STAT/non-STAT designation. The study was approved by the University of British Columbia Clinical Research Ethics Board (Protocols B09-0284 and H09-02695).

Sample collection

A total of 67 clinical samples sent to the British Columbia Public Health Microbiology and Reference Laboratory (BC PHMRL) between April 24, 2009 and March 29, 2010 for influenza A testing were selected. All samples were confirmed A(H1N1) pdm09-positive by RT-PCR and sequencing at the BC PHMRL. Samples were collected by physicians at clinics and hospitals in the Fraser (n=11), Northern (n=13), Vancouver Island (n=8), Vancouver Coastal (n=12) and Interior (n=23) health authorities in British Columbia, Canada. Clinical specimens were chosen for the microbiome study to reflect the type and proportion of samples received for H1N1 testing by the BC PHMRL and included nasal swabs (n=23), nasopharyngeal swabs (n=32), nasopharyngeal washes (n=11) and tracheal aspirates (n=1). Patients included both genders (male=33, female=33, undefined=1) and ranged in age from <1 to 89 years. Samples were collected in hospitals (“STAT” designation, n=35) or non-hospital, community settings (“non-STAT”, n=32) (Table S1).

Nucleic acid extraction

Samples were received at the BC PHMRL through standard clinical sample submission protocols. Total nucleic acid was extracted from clinical material by MagMAX™ Viral RNA Isolation Kit (Life Technologies, Burlington, ON, Canada). Isolated nucleic acid (DNA and RNA) was precipitated for storage and transport. Nucleic acid was re-suspended in 100 µl of TE buffer (10 mM Tris-HCl, 1 mM EDTA; pH 8.0) prior to analysis.

cpn60 universal target (UT) PCR and pyrosequencing

The universal target (UT) region of the cpn60 gene was amplified using a primer cocktail consisting of a 1: 3 molar ratio of primers H279/H280:H1612/H1613 (Table 1) [11–13]. Primer sets were modified at the 5' end with 10-mer multiplexing identification (MID) sequence as per pyrosequencing recommendations to facilitate multiplexing of samples for sequencing (Roche, Brandford CT, USA). PCR reactions consisted of 1 × PCR reaction buffer (20 mM Tris-HCl (pH 8.4), 50 mM KCl), 2.5 mM MgCl₂, 200 µM dNTP, 800 nM of primer cocktail, 2.5 U Platinum Taq DNA Polymerase (Invitrogen, Burlington, ON, Canada) and 2 µl template DNA, in a final volume of 50 µl. Twelve PCR reactions were run for each sample in a thermocycler (Eppendorf Mastercycler) over a temperature gradient: 94°C for 3 min, followed by 40 cycles of 30 sec at 94°C, 1 min at 42-60°C and 1 min at 72°C, followed by a final extension at 72°C for 10 min. PCR reactions from the same sample were pooled and concentrated using the AMPure Purification system (Agencourt Bioscience, Beverly, MA, USA), and further purified by agarose gel separation and extraction (QIAEX II gel extraction kit, Qiagen, Toronto, ON, Canada). Final amplicon was suspended in TE buffer and quantified using a Qubit fluorometer (Invitrogen). cpn60 UT amplicon libraries were pooled in equimolar concentrations and sequenced using the GS FLX Titanium system as per manufacturer’s instructions.

DNA sequence quality control and assembly of OTU

Pyrosequencing data was processed using the default on-rig procedures from 454/Roche. Filter passed reads were assessed for non-cpn60 host DNA by screening against a database comprised of the human genome with annotated cpn60 genes and predicted pseudo-genes removed. MID-partitioned sequences were processed with the microbial Profiling Using Metagenomic Assembly (mPUMA) pipeline (http://mpuma.sourceforge.net [14]) to generate operational taxonomic units (OTU) with gsAssembler (Roche). OTU were screened and filtered for chimeras with Chaban’s Chimera Checker (C3) (http://mpuma.sourceforge.net) and Bellerophon [15]. Remaining OTU were identified by watered-Blast comparison [12] to the cpn60 reference database, cpnDB_nr (downloaded from www.cpndb.ca [16]). OTU abundance was calculated based on mapping of sequence reads to OTU sequences using Bow tie 2 in mPUMA. OTU having the same database reference were collapsed into nearest neighbor “species” and OTU with less than 55% identity to any reference sequence were removed from the dataset as non-cpn60 sequence.

Community composition analysis

A number of parameters and characteristics were calculated for each sample’s microbial community profile. Good’s coverage estimates were determined using mothur [17]. Shannon’s diversity index, Simpson’s diversity index, Chao1 estimate, the number of observed species and Jackknifed beta diversity were calculated with Quantitative Insights Into Microbial Ecology (QIIME) software [18]. Phylogenetic-based distance matrices and clusters were examined using Unifrac [19] and a Bray-Curtis dissimilarity matrix was computed in R using the vegan package [20,21].

Quantitative PCR (qPCR)

Absolute quantification of bacterial species of interest in each sample was done by qPCR using primers and annealing temperatures described in Table 1. Quantification of Streptococcus pneumoniae was achieved by detection of the lytA gene using a Taqman assay [22], while all other species-specific assays were SYBR Green-based, targeting the cpn60 gene. SYBR Green assays were also used for total 16S rRNA gene content estimation based on amplification of the V1-V3 region of the 16S rRNA gene [23], while an estimate of human mitochondrial DNA content was based on quantification of the cytochrome C oxidase subunit 1 (cox1) gene.

For all assays designed in this study (Table 1), the cpn60 sequence for each target species type strain was used to design species-specific primers. Signature regions unique to each target were determined using Signature Oligo software (LifeIntel Inc., Port Moody, BC, Canada), primers were designed with Beacon Designer software (Premier Biosoft International, Palo Alto, CA, USA) and PrimerBLAST [24]. Primers were used to amplify target sequence from patient DNA samples, which were then ligated into pGEM-T-Easy (Invitrogen) to make standards and facilitate confirmation of the target by sequencing. Specificity of all primer pairs was tested by using standard plasmids from other assays as template and no products were generated. Finally, the optimal annealing temperature for each assay was determined via an annealing temperature gradient run and analysis.

All qPCR reactions were run on a plate containing a no template control (NTC) and a standard curve composed of target-containing plasmids at concentrations of 10⁰ to 10⁷ copies/reaction. All reactions were performed in duplicate. Each reaction consisted of 1 × iQ SYBR Green Supermix (BioRad, Mississauga, ON, Canada), 400 nM each primer and 2 µl of template DNA, in a final volume of 25 µl. A MyiQ thermocycler (BioRad) was used for all reactions with the following program: 95°C for 3 min, followed by 40 cycles of 95°C for 15 sec., annealing temperature (Table 1) for 15 sec., 72°C for 15 sec., and a final extension at 72°C for 5 min. A dissociation curve was subsequently run for 81 cycles at 0.5°C increments from 55°C to 95°C for 30 sec at each time point. Fluorescent signals were measured every cycle at the end of the annealing step and continuously during the dissociation curve data collection. The Taqman assay was run under identical conditions, with the substitution of 1× iQ Supermix (BioRad) in place of the SYBR Green Supermix, the addition of 200 nM JH0382 probe and the omission of the dissociation curve. All resulting data was analyzed using iQ5 Optical System Software (BioRad).

Correlations between quantified PCR copies and sequencing read abundance were calculated with SPSS software (SPSS Inc., Chicago, IL, USA).

Pyrosequencing and quality control

cpn60 UT amplicon was generated from all 67 samples. Amplicons were pooled into four libraries and each was sequenced on a half plate region of a Roche 454 GS FLX Titanium pyrosequencer to generate a total of 1,940,006 filter passed sequence reads. Since previous studies of upper respiratory samples report human genomic DNA as an overwhelming proportion of nucleic acid isolated from this environment [10,25], all filter passed sequences were screened against a version of the human genome that had all cpn60 genes and pseudo-genes removed. This screening process filtered out 1,661,090 sequences (85.6% of the data). The remaining sequences were MID-separated and loaded into gsAssembler for de novo assembly. OTU sequences were then filtered to remove chimeras and non-cpn60 sequences (those with <55% identity to any reference database sequence). OTU abundance was calculated by mapping all reads on to the OTU using Bow tie 2 within the mPUMA pipeline. The final dataset contained 270,284 sequences in 490 OTU (Table S2). When each OTU was compared to the reference database (cpnDB_nr), the median percent identify of OTU to reference sequences was 96.4%, with 369/490 (75%) of the OTU having identities of >90% to a reference sequence (Table S3). To further aggregate the data, OTU with the same nearest neighbor (termed nearest neighbor "species") were combined. An average of 4,158 sequence reads per sample (median 1054) was obtained. Samples flu36 and flu48 contained fewer than 100 reads after processing and were removed from the analysis. Raw sequence data files from the 65 samples used in the study have been deposited to the NCBI Short Read Archive, and are associated with the BioProject accession PRJNA200951.

Sequencing depth cut-off

An initial cut-off of at least 100 high-quality cpn60 sequence reads per sample was imposed on the dataset before analysis, which resulted in two samples being removed from the study. To evaluate whether this cut-off was sufficient to allow robust analyses, jackknifed beta diversity was calculated by generating 1000 rarefractions of the data at a depth of 100 reads for each of the remaining 65 samples. A distance matrix was created for each rarefaction and principal coordinates analysis (PCoA) was computed on each rarefied matrix using two different distance measures: Bray-Curtis and Canberra. Each sample was then plotted as a point where the area of each point reflected the interquartile range of the jackknifed PCoA estimates (Figure S1). The relatively compact size of the points for each sample indicated that the distance matrices were fairly consistent over 1000 iterations at the sampling depth of 100 reads, suggesting the sampling depth was adequate for sample comparison. As such, the 65 samples profiled that exceeded this minimum were used for all further analyses.

Diversity and overall composition measures

Since it is well-established that diversity statistics are influenced by different sampling depths [26], values for the Shannon’s diversity index, Simpson’s diversity index, Chao1 estimate and the observed number of species per sample were all derived from the mean value of 1000 iterations of a 100 read subsampling of nearest neighbor “species” values in each sample. The diversity indices were then compared by patient gender, age, health authority, sample type and STAT/non-STAT designation by ANOVA (Table S4). Microbiomes contained an average of 9-14 nearest neighbor “species” when examined at the 100 read subsampling depth, and were predicted by the Chao1 estimate to contain 13-20 “species” if sampled to completion. There was no significant difference in the number of species observed or predicted between any patient characteristic groups. Diversity measurements of richness and evenness determined by Shannon’s and Simpson’s indices also showed no statistically significant difference by patient characteristics. However, there was a trend for both measures to increase as patient age increased (Shannon’s, p = 0.08; Simpson’s, p = 0.11; Table S4), with an observable distinction between patients under and over 20 years of age.

Unifrac clustering, considering both phylogenetic composition of the microbiome and normalized sequence read abundance of the 215 nearest neighbor “species” sequences in each sample, was performed (Figure S2). There were no meaningful groupings of upper respiratory tract microbiomes based on our five patient characteristics. As such, there appeared to be no major phylogenetic differences in the microbiome compositions between patients based on these variables.

Phylum-level comparisons

The phylogenetic composition of the combined upper respiratory tract microbiome dataset consisted of Firmicutes (42.5%), Proteobacteria (27.7%), Actinobacteria (21.7%), Bacteroides (5.5%), Fungi (0.1%), Human (0.2%) and other bacteria (2.3%) (Figure S3). Individual microbiomes were dominated by either Firmicutes, Actinobacteria or Proteobacteria (Figure 1), consistent with previous reports of human upper respiratory tract microbiomes [5–8].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Sample composition at the phylum level.

Phylum level profiles are shown for individual patients (n=65) as well as the average phylum proportions for sample designation (STAT/non-STAT), gender, health authority, patient age (in years) and sample type.

https://doi.org/10.1371/journal.pone.0069559.g001

To identify if a particular microbiome composition was associated with a patient characteristic, the proportion of each phylum in each sample was calculated and samples from the same group were averaged together (Figure 1). No statistically significant differences were observed in the microbiome compositions grouped by any variable investigated.

Species-level microbial comparisons

To investigate the microbiome at a finer taxonomic resolution, the nearest neighbor “species” description of each sample was used to construct a Bray-Curtis dissimilarity matrix. The matrix was calculated by converting the number of sequence reads for each “species” in a sample to its relative abundance within that sample. The jackknife beta diversity analysis results (Figure S1) suggested that the distances between samples were fairly robust, so the actual sequence counts, rather than rarefied numbers, were used for this analysis. Average linkage hierarchical clustering of the dissimilarity matrix generated 11 clusters (Figure 2). There were five clusters consisting of only one or two patient samples (clusters 5, 8, 9, 10 and 11). Cluster 1 was the largest, with 27 patients with Variovorax paradoxus and Enterococcus hirae as the dominant organisms. Clusters 2 and 4 were the next largest, each containing 10 patients, with Corynebacterium pseudodiphtheriticum and Streptococcus pneumoniae as the dominant organisms, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Hierarchical clustering of sample profiles.

(A) Nearest neighbor “species” composing 20% or more of the microbiome in at least one sample are shown. The average percent identities of sequence reads within each nearest neighbor are given in parenthesis. The color scale (black to red) reflects an increasing relative abundance of sequence reads for each nearest neighbor “species” in each sample. (B) Summary of the hierarchical clustering detailing the number of samples and dominant organism(s) in each cluster.

https://doi.org/10.1371/journal.pone.0069559.g002

Absolute quantification of targets

To assess the reliability of sequence read abundance from pyrosequencing of the cpn60 gene, we chose seven species from the dataset to evaluate by quantitative real-time PCR (qPCR). This set was comprised of the three most abundant species in the dataset (Staphylococcus aureus, Streptococcus pneumoniae and Corynebacterium pseudodiphtheriticum), two species representing 1-2% of the dataset (Enterococcus hirae and Propionibacterium acnes) and two species present at less than 1% of the dataset (Streptococcus mitis and Corynebacterium accolens) (Table 2). All seven of these species were also detected in at least 31/65 samples. In addition to the individual species, estimates of total bacterial and human DNA were also obtained for comparison. qPCR was used to quantify the amount of each target of interest in the clinical sample nucleic acid extracts. A well-established Taqman assay targeting the lytA gene was used for quantification of Streptococcus pneumoniae [22]. For the remaining species, specific assays were developed for this study using SYBR Green chemistry and targeting the species type strain cpn60 sequence (whenever available). Assays were also designed to estimate the amount of total bacterial DNA (via quantification of the 16S rRNA gene content with primers designed previously [23]) and human DNA present (via quantification of human mitochondrial cytochrome C oxidase subunit 1 gene (cox1)).

The numbers of copies per sample of all targets were determined (Figure 3). The human cox1 gene had, on average, an abundance 2.6 logs higher that the bacterial 16S rRNA gene (Figure 3, Human compared to 16S). The expected copy number of cox 1 per cell is 50-2000 based on estimates of approximately 42 mitochondria per mammalian epithelial cell [27] and 1-50 mitochondrial genomes per mitochondrion [28]. An average of approximately 1.0×10⁸ copies of cox1 were detected per sample, which translates into 5.0×10⁴ to 2.0×10⁶ epithelial cells per sample (Figure 3). The average 16S rRNA copy number per sample was 2.5×10⁵ (Figure 3), which translates to approximately 7.5×10⁵ bacteria per sample based on a median 16S rRNA copy number of three per bacterial genome [14]. Finally, taking into consideration that an epithelial cell has significantly more DNA than a bacterial cell, it is evident that the standard clinical upper respiratory tract samples used in this study were overwhelmingly dominated by host DNA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Quantification of bacterial and human mitochondrial DNA.

Box and whisker plots showing target copies per sample detected (n = 65). Ca - Corynebacterium accolens; Cp - Corynebacterium pseudodiphtheriticum; Eh - Enterococcus hirae; Pa - Propionibacterium acnes; Sa - Staphylococcus aureus; Sp - Streptococcus pneumoniae; Sm - Streptococcus mitis; 16S -16S rRNA gene; Human - human cytochrome C oxidase gene.

https://doi.org/10.1371/journal.pone.0069559.g003

Comparing sequence read abundance to absolute quantification

To determine how well pyrosequencing read abundance reflected absolute quantities per sample for the species targeted by qPCR, sequence abundance counts were compared to qPCR target copies per sample using a Spearman’s rho correlation (Table 2). With the exception of Streptococcus mitis, all species had a significant positive correlation (p < 0.05) between their sequence read abundance and their absolute quantification values. The S. mitis correlation was just outside the range of significance (p = 0.055) and could be due to qPCR assay biases, such as being too specific (missing natural species variation) or not specific enough (cross-reacting with other organisms). Alternatively, the original cpn60 UT PCR may not have been optimal for this species and S. mitis amplicons may have been underrepresented in the sequencing data.