Methods
Study sites and sample collection
Samples were collected at two sites: in Maryland, at the Chester River Field Research Station, Washington College, Chestertown, MD, and in Michigan at a Michigan State University field station, Camp WaWaSum, Grayling, MI. In Maryland, cowbirds and blackbirds were trapped by mist-net during January – February 2011, and cloacal samples collected. In Michigan, cowbirds and blackbirds were trapped in decoy traps in May 2007, by US Fish and Wildlife Service to reduce cowbird parasitism on the endangered Kirtland’s Warbler (Walkinshaw 1983). In Michigan, cloacal samples were taken from cowbirds and blackbirds. In addition, pharyngeal and gut tissue samples were taken from each cowbird.
Cloacal samples were taken using sterile urethral swabs manufactured by Medical Wire and Instrument, Durham, NC, following the method of (Lombardo 1998). In brief, the swab was gently inserted into the cloacal opening and slowly twirled, then the tip was inserted into a sterile tube filled with buffer, the metal handle cut, and the tubes were frozen and stored at -90 ready for genetic processing. Cloacal samples are used to an estimate of the gut microbe community, since the reproductive and elimination channels in birds both use the cloaca (Lombardo 1998) and since cloacal sample collection does not require euthanizing the bird. Gut tissue samples were taken from each cowbird immediately after euthanasia by cervical location. The cowbirds were necropsied, a 1 mm2 section of gut tissue was excised from the ileo-cecal region and placed into a sterile tube with buffer and frozen at -90.
Laboratory Methods and Statistical Analyses
PCR, DNA sequencing
We analyzed 35 samples (16 from Michigan and 19 from Maryland, Table 1) that were sampled in order to address each of our three objectives. Samples were amplified for pyrosequencing using a forward and reverse fusion primer. The forward primer was constructed with (5’-3’) the Roche A linker, an 8-10bp barcode, and the forward gene specific primer sequence. The reverse fusion primer was constructed with (5’-3’) a biotin molecule, the Roche B linker and the reverse gene specific primer sequence. The gene specific primer pair for bacterial SSU rRNA genes was 27F/519R (Lane 1991).
Amplifications were performed in 25 ul reactions with Qiagen HotStar Taq master mix (Qiagen Inc, Valencia, California), 1ul of each 5uM primer, and 1ul of template. Reactions were performed on ABI Veriti thermocyclers (Applied Biosytems, Carlsbad, California) under the following thermal profile: 95°C for 5 min, then 35 cycles of 94°C for 30 sec, 54°C for 40 sec, 72°C for 1 min, followed by one cycle of 72°C for 10 min and 4°C hold. Amplification products were visualized with eGels (Life Technologies, Grand Island, New York). Products were then pooled equimolar and each pool was cleaned with Diffinity RapidTip (Diffinity Genomics, West Henrietta, New York), and size selected using Agencourt AMPure XP (BeckmanCoulter, Indianapolis, Indiana) following Roche 454 protocols (454 Life Sciences, Branford, Connecticut). Size selected pools were then quantified and 150 ng of DNA were hybridized to Dynabeads M-270 (Life Technologies) to create single stranded DNA following Roche 454 protocols (454 Life Sciences). Single stranded DNA was diluted and used in emPCR reactions, which were performed and subsequently enriched. Sequencing followed established manufacture protocols (454 Life Sciences).
DNA sequence quality control, OTU binning and Statistical Analyses
The 16S rRNA gene sequence collections were demultiplexed and sequences with sample barcodes not matching expected barcodes were discarded. We used the maximum expected error metric (Edgar, 2013) calculated from sequence quality scores to cull poor quality sequences from the dataset. Specifically, we discarded any sequence with a maximum expected error count greater than 1 after truncating to 200 nt. The forward primer and barcode was trimmed from the remaining reads. Reads were taxonomically annotated (below) and all Chloroplast, Eukaryal, Archaeal, "Unassigned" and mitochondrial sequneces were discarded. We checked that all primer trimmed, error/taxonomy screened and truncated sequences were derived from the same region of the SSU rRNA gene by aligning the reads to Silva SSU rRNA gene alignment with the Mothur (Schloss 2009) NAST-algorithm (DeSantis TZ, Hugenholtz et al. 2006, Schloss 2009), aligner (using the Mothur version of the Silva alignment) and inspecting the alignment coordinates. Reads falling outside the expected alignment coordinates were culled from the dataset. Remaining reads were trimmed to consistent alignment coordinates such that all reads began and ended at the same position in the SSU rRNA gene.
Sequences were taxonomically classified using the UClust (Edgar 2010) based classifier in the QIIME package (Caporaso et al., 2010) with the Greengenes database and taxonomic nomenclature (version “gg_13_5” provided by QIIME developers, 97% OTU representative sequences and corresponding taxonomic annotations (McDonald, McGehee et al. 2012). We used the default parameters for the algorithm (i.e. minimum consensus of 51% at any rank, minimum sequence identity for hits at 90% and the maximum accepted hits value was set to 3).
Reads were clustered into OTUs following the UParse pipeline. Specifically USearch (version 7.0.1001) was used to establish cluster centroids at a 97% sequence identity level from the quality controlled data and map quality controlled reads to the centroids. The initial centroid establishment algorithm incorporates a quality control step wherein potentially chimeric reads are not allowed to become cluster seeds. Additionally, we discarded singleton reads because it is difficult to asses the quality of singleton reads and this quality control parameter in addition to maximum expected error screening has proven to be similarly useful if not superior for reducing 454 sequencing error as denoising (Edgar 2013). Moreover, two popular denoising algorithms have been shown to add sequencing errors while correcting others sometimes in a nearly equal ratio (Bragg L, Stone et al. 2012). Ninety-nine percent of quality controlled reads could be mapped back to the cluster seeds.
Alignment of SSU rRNA genes was done with SSU-Align, which is based on Infernal (Nawrocki, Kolbe et al. 2009, Nawrocki and Eddy 2013). Columns in the alignment that were not included in the SSU-Align covariance models or were aligned with poor confidence (less than 95% of characters in a position had posterior probability alignment scores of at least 95%) were masked for phylogenetic reconstruction. Additionally, the alignment was trimmed to coordinates such that all sequences in the alignment began and ended at the same positions. FastTree (Price, P.S. Dehal et al. 2010) was used to build the tree. Samples were compared using the weighted normalized Unifrac metric (Lozupone and Knight 2005). Ordination was done by Non-Metric Multi-Dimensional Scaling (NMDS) with Phyloseq and Vegan (R packages). Multivariate dispersion for sample groups were calculated using the "betadisper" function in Vegan and differences in group dispersions were assessed with ANOVA.