Ethics Statement

This study was approved by the University of New Brunswick’s Animal Care Committee (Animal Care Permit No. 09012) as well as the Canadian Wildlife Service (Scientific Take Permit No. ST2642 for handling puffins and MBS/MSI 09-6 for disturbing birds in a protected, federally-owned Migratory Bird Sanctuary).

Sampling

To assess the effectiveness of conventional field observations as an appropriate method to study puffin diet, we compared diet as estimated from 68 hours of field observations and through DNA within fecal samples of puffin chicks in the 2009 breeding season (May-August) on Machias Seal Island (44°30’N, 67°6’W), New Brunswick, Canada. Field observations were conducted in three hour stints with binoculars from observation blinds according to Machias Seal Island’s research protocols [23,31] where the number, length, and species or category (e.g. “krill”) of all prey were recorded in each bill load delivery of a provisioning adult. A total of 91 chick fecal samples were collected opportunistically during regular research activities. To test the similarity between the diets of puffin adults and chicks we collected 10 adult fecal samples per week (total 146) throughout the season. Fecal samples were stored in 70% ethanol. To assess herring diet and to investigate the potential effect of secondary consumption, we dissected the stomachs from 77 collected juvenile herring (~5-15cm) that been dropped accidentally by provisioning seabird adults. Other prey found within the colony, as well as several invertebrate and fish species obtained from the island’s intertidal zone and from refuse buckets donated by passing fishermen, were stored in Whirl-pak® bags (Nasco) at -20°C to generate a sequence database of local fauna from which fecal and stomach DNA could be identified. All fecal samples and stomach contents were stored in 5-10ml 70% ethanol at 4°C for three months, -20°C for four months, and then -80°C for two years.

Primer Design

Short target amplicons are preferred for PCR-based molecular diet analysis because fecal DNA is often highly degraded [22,32]. We used universal primer pairs to target small (~130-300bp) fragments of the mitochondrial genes 16S and cytochrome c oxidase subunit 1 (CO1). The degenerate primers 16S1F and 16S2R amplify a ~180-270bp region of 16S and have been shown to amplify prey DNA from the feces of penguins (e.g., fish, euphausiids, squid [12]). To complement these data, we employed a second set of universal primers [33] that successfully amplifies a 130bp region of the CO1 in over 600 species of mammals, fishes, birds, and insects, making it a good candidate to detect species not amplified by 16S (Table 1).

We used a pooled massively parallel sequencing (MPS) approach following the protocol of Puritz et al. [34]. To allow recovery of sample identification from sequence data we included a 10bp multiplex identifier (MID) tag between the Lib-L 454 sequencing adapter (26bp plus a 4bp signal calibration key) and the universal primer (16S or CO1) in our custom engineered forward and reverse primers (Table 2).

Stomach and fecal samples can contain significant amounts of DNA from the host species due to sloughing of cells in the digestive tract [19]. Host DNA may represent a huge proportion of the total number of sequences obtained through sequencing, reducing the detection of prey with little or difficult-to-amplify DNA [19]. Preliminary sequencing of cloned16S PCR products indicated a high frequency of host DNA sequences recovered from both stomach and fecal samples. Thus, we designed herring and puffin blocking primers for 16S1F and 16S2R (respectively) using the C3 spacer method described by Vestheim and Jarman [35] (Table 1). Blocking primer efficiency was tested with puffin and herring DNA in isolation, and under competitive conditions in PCR reactions consisting of mixtures of prey (our reference samples) and predator (puffin or herring) DNA. Although we did not identify interference between the blocking primers and any of our reference samples, it is possible that the blocking primer (overlapping a highly conserved region of DNA) prevented ideal amplification of prey types that we were unable to asses. Blocking primers were used with 16S amplifications at a blocking to universal primer ratio of 5:1. The use of a blocking primer to enrich the genomic DNA samples for rare prey templates reduced the proportion of herring sequences by 86% and doubled the number of taxa identified (from 10 to 20 taxa) in a test of eight 16S-amplified herring stomach contents (data not shown).

DNA Extraction and Amplification

DNA from stomach contents and reference samples was isolated with a CTAB protocol [36]. DNA was resuspended in 20-100 ul TE (Ambion pH 8.0) and concentration and purity were evaluated with NanovueTM (General Electric Life Sciences). Fecal samples were centrifuged for 30 minutes at 4°C and storage ethanol was poured off. DNA was extracted with QiaAmp DNA Stool Mini Kit (Qiagen). Samples with small amounts of fecal material were eluted with 75-100ul of buffer AE instead of the recommended 200ul. DNA was stored in 2ml microcentrifuge tubes at -20°C.

Amplification of fecal DNA with 16S MID-tagged sequencing primers was achieved in 30ul reactions containing 6ul undiluted template, 0.2mM dNTP, 1X bovine serum albumin (BSA; New England Biolabs), 5mM MgSO4, 0.25uM of each primer, and 1.25 of blocking primer, 1X High Fidelity Buffer, and 1.2 units Platinum® Taq DNA Polymerase High Fidelity (Life Technologies). DNA from fish stomach contents was amplified with 16S MID-tagged sequencing primers in 30ul reactions containing 3ul undiluted template, 0.2mM dNTP, 5mM MgSO4, 0.25 uM of each primer, 1.25uM of blocking primer, 1X High Fidelity Buffer, and 1.2 units Platinum® Taq DNA Polymerase High Fidelity. DNA from samples intended for the reference database of potential prey was amplified with 10ng of template, 0.2mM dNTP, 0.5mM MgSO4, 0.25uM of each forward and reverse 16S primer, 0.04 units of Taq DNA polymerase, and 1X ThermoPol Buffer (New England Biolabs). Thermocycling protocol for 16S began at 94°C for 10 minutes followed by 35 cycles of 94°C for 15s, 55°C for 15s, and 68°C for 30s, with a final extension of 68°C for 5 minutes (BioRad C-1000).

Amplification with CO1 for both sample types followed similar component and cycling conditions as the 16S with the following modifications: 3ul undiluted template was used, BSA was omitted, there was no blocking primer, and annealing temperature was set at 53°C. All amplicons were visualized under UV light in 2% agarose using SYBR Safe (Life Technologies), and those that amplified were cleaned with Agencourt AMPure XP (Beckman Coulter) at a 0.9 beads to 1 PCR product ratio. DNA concentration was determined with dsDNA BR assays on a Qubit 2.0 Fluorometer (Life Technologies). PCR was repeated for samples with less than 20ng of 16S-amplified DNA then pooled after cleaning with QIAEX II Gel Extraction Kit following the manufacturer’s instructions.

All pipetting was done with barrier tips, negative controls were used in every PCR, small aliquots of reagents were used, and lab benches were cleaned between uses. Adult, chick, and herring diet samples were DNA-extracted on different days, several weeks in advance of the PCR component of this project.

We made considerable efforts to obtain a sufficient number and an even temporal distribution of samples throughout the sampling period. However, a limiting step in our protocol was the extraction of high quality DNA. Amplification success of herring stomach contents and puffin fecal samples ranged from 47% to 71%. Insufficient homogenization during DNA extraction and PCR inhibitors are two potential explanations for the large number of samples that could not be successfully PCR amplified. There was also considerable variation in sample coverage. Although sequencing success of submitted samples was over 94%, only 61-84% of these samples produced a sufficient number of sequences to be included in calculations of frequency of occurrence (see next section). This variation may be a result of MID tag-induced amplification and sequencing bias [37], imprecision in equimolar pooling of amplicons, or variability in sample quality (as both fecal and stomach samples are likely to contain many PCR-inhibitors).

Sequencing, Bioinformatics and Analysis

Samples were quantified and pooled into a single equimolar library and sequenced by Génome Québec Innovation Centre at McGill University in Montreal, Quebec, Canada. The library was sequenced unidirectionally on half a pico titre plate using the Roche GS-FLX (454) platform. Bases were recalled using Pyrobayes [38] and sequences were filtered for length (40-400bp) and quality (mean Q20 across fragment) using Prinseq [39]. Sequences were then demultiplexed based on their MID tag combinations (exact matches only) and divided by amplicon type (16S or CO1, exact matches only) with jMHC [40]. The program jMOTU [41] was used to assemble sequences into molecular operational taxonomic units (MOTUs) at various levels of base pair differences (cutoffs). Sequences were clustered into MOTUs with cutoff ranges from 1-30 base pairs (and suggested gathering parameters of 95% low BLAST identity filter and default settings for sequence alignment overlap). At increasing cutoff values (number of base pairs different between sequences), the number of MOTU produced by jMOTU decreases and eventually reaches an asymptote. Cutoffs were chosen at the beginning of this asymptote to ensure a high degree of taxonomic diversity at the expense of producing multiple MOTUs belonging to a single taxon. MOTUs at the chosen cutoff were exported to an sql file, which was uploaded to the relational database management software, PostgreSQL (hwww.postgresql.org). Representative sequences were chosen as the first sequence appearing in jMHC data output belonging to a unique MOTU. All CO1 representative MOTUs were queried using Barcode of Life Database (BOLD) online identification tool [42]. All representative MOTUs that were not identified to species by BOLD’s online tool were assigned to a taxanomic level that incorporated all potential matches to the queried sequence at 100% identity. MEGAN [43] was used to corroborate BOLD-based taxon assignments as well as to assign taxonomy to representative MOTUs with no match to BOLD. BLAST files (BLASTn, queried 22 March, 2013 with default algorithm parameters) were imported to MEGAN (min score=75, min support=2, top percent=10, min complexity=0) and the MEGAN taxon assignments were accepted or rejected based on the % identities of each MOTU’s top matches. MOTUs were assigned to the top match if it was 96% similar or higher and was a possible component of this study system, otherwise MOTUs were assigned to the most highly resolved taxonomic node common to all significant matches. Similar criteria were used for taxon assignment for 16S MOTUs except we also included the sequences generated from our reference samples in our consideration of taxon assignment.

We used PRIMER 6 (PRIMER-E Ltd) [44] for multivariate statistical analyses. We used an Analysis of Similarity (ANOSIM) to compare chick diet with method type (field observation or DNA) as a fixed factor. We also conducted an ANOSIM to test the difference between puffin diet with age (chick or adult) as a fixed factor. Data were converted to binary (presence/absence for fecal samples or for field observation provisioning deliveries). Resemblance matrices were built from Bray-Curtis similarity, from which the ANOSIM was run with 999 permutations. We used non-metric multidimensional scaling (MDS, 100 restarts) with overlaid vectors representing the correlations (Pearson correlation coefficients, >0.35 for chick/adult comparisons, all for diet by method (DNA/field)) to visualize differences between diets or chick and adults and between diet studying methods. All MDS graphs had stress values below 0.2 [45].

We used the frequency with which a prey was detected across samples, the frequency of occurrence (FOO), to quantify diet because the proportion of prey sequences found within a sample does not necessarily correlate quantitatively with proportions of prey consumed [14]. We limited FOO calculations to samples containing at least 50 sequences per marker for each of the three sample types obtained within the common sampling period (the time frame where all sample types were available for collection, 13 June to 29 July). When a taxon was identified with both 16S and CO1, we used all samples (that met above criteria) to estimate FOO, otherwise FOO was calculated based on number of samples amplified by the respective gene. Table S4 outlines gene-specific sample size and sequencing success as well as which data are used in the creation of various figures and tables.

Comparison of Diet Methods

The diet of chicks when assessed by molecular methods is more highly resolved taxonomically than when assessed by conventional field observations (Figure 6, Table S2). Gulf of Maine puffin chick diet is known from years of observing adults carrying prey crosswise in the bill and delivering them to chicks in burrows [23]. Prey types are often reported in broad categories, such as ‘polychaete’ or ‘sculpin’. Through fecal analysis we can attribute ‘polychaetes’ to include the Ragworm and the Slender ragworm, ‘sculpin’ includes the Sea raven (Hemitripterus americanus) and Grubby (Myoxocephalus aenaeus), and add Thysanoessa sp. to the ‘krill’ category, which was previously thought to comprise only Northern krill.

We saw a similar improvement on the taxonomic resolution of herring prey using a DNA-based approach compared to stomach content analyses. In published studies of juvenile herring diet (Table S3), prey items are also classified in broad categories such as: cladocerans; eggs from fish, crustaceans, or decapod crustaceans; barnacle or decapod larvae; a class of tunicate; a protozoan family; and krill, with the exception of one krill species (Thysanoessa raschii) and 12 copepod species and genera. Through DNA sequencing of stomach contents we identified three cladoceran species, eight fish species/genera, three crab species/genera, one shrimp species, and a species and genus each of both barnacle and krill.

Our molecular method identified prey not detected or not accurately identified by conventional methods. DNA from insects, echinoderms, amphipods, bivalves, gastropods, polychaetes, and cnidarians was present in herring stomach contents. None of these groups had been observed in previous studies, which may either be due in part to a change in herring diet in the 30 years since the latest published study was conducted, or reflect the increased resolution of a DNA-derived diet. Some taxa, however, would have never been detected in herring diet, such as the Common jellyfish (Aurelia aurita), as this prey would be digested completely, leaving nothing for identification. We also detected several taxa that were either visually misidentified as common prey or absent from the diet in 15 years of field observations of puffins (Giant wrymouth, Threadfin rockling, Atlantic cod, Hippolytidae (shrimp), Scallop snailfish, Haddock, and Acadian redfish). As many of the fish caught by puffins are small (~5-15cm), accurate identification can be difficult in the field. Shrimp from the Hippolytidae family, for instance can be easily mistaken for krill, a common prey item. While shrimp were not recorded in chick diet during 2009 field observations, one species (Pandalus montagui, Pandalidae) was found as a dropped prey item in the colony that year (personal observation), offering further evidence that field observations suffer from misidentification error.

Failure to detect or misidentification of prey in predator diet can be a substantial hindrance to our understanding of how components of an ecosystem interact. For example, identifying commercially fished species consumed by puffins (Cod, Haddock, and Redfish) is important for the effective management of these stocks, as the impact of non-human predators on fish has historically been severely underestimated [3]. The natural mortality rate of herring used in stock assessment models, for example, was less than 25% of the estimated consumption by mammals, piscivorous fish, and seabirds [3]. Explicit consideration of the links between exploited species and the rest of the ecosystem, termed ecosystem-based management, has superseded historical stock-based resource management [1,2,4,5,6,7]. As predators, puffins can play a part in the population dynamics of their prey and therefore diet studies that do not suffer from misidentification of prey are preferred and their role needs to be incorporated into models of ecosystem function.

Effect of Secondary Consumption

While field observations did not accurately identify or quantify chick diet, they did provide a necessary framework from which to interpret our molecular diet data. The fish and shrimp species found in chick diet but not observed in the field were most likely misidentified as one of the more common fish prey or as krill, in the case of shrimp. However, the DNA-identified taxa without visual equivalents that had not been identified through field observations are either a result of secondary consumption, or are newly-identified puffin prey. Adult puffins catch and hold prey in their bills as they propel themselves underwater and upon return to the colony they drop the intact prey in their nest burrow, making very unlikely any transfer of non-targeted prey to their chicks. The copepod A. longiremis as well as larval stages of the Crevice brittlestar (Ophiopholis aculeata) and Paguridae crabs are components of plankton that are too small to be target prey for provisioning puffins. The adult forms of these species have never been seen in chick diet, nor do they look sufficiently similar to taxa that have been identified in the field that they could have been be mistaken (“visual equivalents”). Further, these three taxa were also detected in herring samples, implicating secondary consumption as the explanation for these records.

With evidence for secondary consumption in the prey shared between chicks and herring only, we can consider removing barnacles (Balanus sp., Semibalanus balanoides), copepods (Calanoida, Temora longicornis, Acartia longiremis), cladocerans (Evadne nordmanni, E. spinifera), crabs (Great spider crab, Paguridae), and a brittlestar (Crevice brittlestar) from our picture of the diet of chicks (Figure 2). The remaining prey types of chicks (fish, krill, polychaetes, and shrimp) are known prey or have visual equivalents and can therefore be proposed as the true shared resources between herring and puffins (Figure 3). If the prey items found in chick feces were considered in isolation from those in herring diet, we would have made erroneous conclusions about foraging biology and diet of puffin chicks. The inclusion of herring diet in our study was critical to understanding puffin chick diet due to the confounding effect of secondary consumption.

Comparison of Chick and Adult Diet

Fecal DNA analysis has shown that the community of taxa in adult puffin samples is indistinguishable from that of chicks (Figure 4, Figure 5), supporting results of a 2006 stable isotope analysis study of the same puffin population [24]. Central-place foraging theory leads to the expectation that adult and chick diets should differ, a phenomenon observed in other seabirds [48,49,50,51,52]. Though supported by stable isotope data, we question whether our data lacked the power to test this idea conclusively. Future studies should account for the difficulty in amplifying fecal DNA and collect many samples in appropriate time periods.

Adult puffins forage at sea and do not leave identifiable components of prey in feces or in the form of a pellet. While stomach content analyses of dead or sacrificed birds have been conducted in European populations, no study has yet identified the prey of Gulf of Maine adult puffins [53]. As a result, chick diet has been used as a best estimate of adult diet. Our results provide the first ever assessment of puffin adults in the Gulf of Maine and, in addition to stable isotope data, support the use of chick diet as an acceptable proxy for adult diet.

Foraging Habitat

This DNA-based method identified many herring prey to species or genus level, allowing us to draw conclusions about herring foraging habitat and behaviour based on the biology and selection of prey on which they feed, which in turn can be used to understand how changes at the base of the food chain may affect puffins.

Juvenile herring are considered as “plankton feeders” (Table S3 and references therein) and components of the pelagic food web. In the summer months, schools of juvenile herring occupy discrete small home ranges in and around inlets and bays on the Gulf of Maine coast [29]. Our data support previous observations of juvenile herring occupying near-shore habitats as we identified the coastal and ‘shallow water’ copepods Acartia longiremis [54] and T. longicornis [55], and the fresh- or brackish water cladoceran Pleopsis polyphemoides [56] in our herring diet samples. We also identified pelagic prey, as both species of krill (Northern krill and Thysanoessa inermis) detected in herring diet have been shown to be absent from areas where bottom depth was less than 45 metres [57]. However, our data show herring to consume non-planktonic prey from non-pelagic habitats. For instance, Corophium volutator (Figure 2), is a mud flat-living amphipod [58], Lacuna sp. has one regional epifaunal representative [47], and the Ragworm is a shallow-sediment dweller [46].

Juvenile herring are believed to capture prey items actively as opposed to passively ingesting whatever floats into their open mouth [29]. It has been shown that year-class size is unrelated to primary production [59] and the amount of food in herring stomachs is not correlated with the abundance of zooplankton in the water [30]. The composition of herring diet is therefore a matter of choice and not a random sample of species available in the environment. Further, juvenile herring remain in the same general locality during the principal growing period (June to August) and the body condition of these distinct schools of herring is consistent across years [29]. We offer two potential explanations for how we might observe herring prey from a range of habitats (near shore and benthic to planktonic and pelagic) while herring themselves occupy small and discrete home ranges. The full diversity of prey taxa might be available at each location if bottom topography is sufficiently variable within such a short distance to support food chains based on both benthic and pelagic production. Alternatively, if each location used by juvenile herring provides a different range of prey, foraging puffins must be exploiting different patches of juvenile herring to produce the varied prey community we have detected in their stomach contents. These alternative interpretations provide competing predictions about both spatial ecology of juvenile herring and foraging behaviour of breeding puffins that should be amenable to testing in the field.

Taxa identified in adult puffin diet may also provide evidence for foraging habitat. Several taxa present in adult diet occupy freshwater (a rotifer (Bdelloidea), a diatom (Pinnularia sp.), an oligochaete (Nais elinguis)) and intertidal (an amphipod (Gammarellus angulosus) and Rockweed (Ascophyllum nodosum)) habitats. The presence of these taxa may reflect a coastal foraging habitat of adult puffins, or secondary consumption that was not confirmed due to insufficient overlap between adult puffin and herring diet samples. We recommend future DNA-based diet studies aim to control for confounding secondary consumption through the investigation of prey diet but also by ensuring sufficient number and overlap of predator diet samples across the sampling period so that temporal differences in prey availability are not confused with secondary consumption.

Methods

We wished to encompass as much taxonomic diversity as possible in this food chain study so we chose two loci with published degenerate primers which produce small amplicons. The combination of universal primers (chosen in regions of non-variable DNA) and short amplicons (necessary to function effectively on degraded DNA) did pose some problems resolving a few taxa. For both 16S and CO1, the fragment produced from the universal primers was not sufficiently polymorphic to identify which of the two Gulf of Maine Ammodytes species (dubius or americanus) were present in our diet samples. Additionally, there were many instances where prey could not be resolved to species or genus level because those species had not been previously sequenced at our loci of interest. In several cases the taxon assignment was so general that it was not useful in describing diet at all (e.g.: the first entries in Table S1) or no assignment could be made, which may be a due to either considerable deficiencies in the reference database (GenBank) or sequencing error. There were stark differences in the type, number, and coverage of taxa that each marker identified. Marker choice has a considerable impact on how diet is assessed as only 20% of all animal taxa were detected by both markers. Dietary data, and the interpretation of how an organism relates to the greater food web, are skewed by the effectiveness of a marker to amplify various groups of organisms. The variability in prey taxa coverage between loci (Figure 1) is a methodological bias that strongly supports a multi-locus approach.

Although our method identified many more taxa than field observations, an estimated 6 species are present but undetected in our samples (PRIMER Choa2 S extrapolator, Figure 8). Our data provide overwhelming support for the use of multiple barcoding markers, particularly when no a priori knowledge of diet is available, to increase the likelihood of capturing the true assemblage of prey present in predator diet and improve the chance of accurately describing the trophic interactions within a food web.

Download:

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

Figure 8. Species accumulation curve for DNA-based chick diet.

Only samples with >50 sequences from kingdom Animalia were included. Crosses with standard deviation error bars represent Chao2 S extrapolator for prediction of the true total number of taxa present in samples.

https://doi.org/10.1371/journal.pone.0083152.g008

There was sufficient overlap in chick and herring samples to confidently identify secondary consumption in puffin chicks, however this is not true for adult puffin diet. We were unable to detect the full complement of prey for all components of this food chain (Figure 8). It is also important to note that puffins feed on fish species other than herring, which may contribute to the incidence of secondary consumption. However, we predict the effect of secondary consumption from non-herring prey is limited as both field observations and molecular data (Figure 6) show herring as the predominant food of puffins in this sampling time period.