Adélie penguin scats were collected from the areas near colonies of breeding penguins and preserved in 70–80% ethanol. Scats were collected from snow or rock between the colony and the sea. The penguins tend to follow set paths when entering or leaving the colony. It is common for them to defecate immediately after coming ashore from a foraging trip and before entering the colony. These areas were sampled frequently and all of each scat was collected, or the remainder buried to preclude repeat sampling. Sampling locations are shown in Figure 1. All scats were collected regardless of colour or form to avoid sampling bias. Samples were collected over several years and the penguin breeding season starts late in one year and finishes towards the end of the Austral summer, generally in February so samples referred to as ‘2011’ are taken from the breeding season starting in 2010 and finishing in 2011. The sample sets we analysed are shown in Table 1.

Preserved samples were stored at −20°C during transport from Antarctica. Once returned to Australia, DNA was extracted from animal scats by a Promega ‘Maxwell 16’ DNA extraction robot with a ‘Tissue DNA Purification’ kit. PCR inhibitor concentrations were reduced in the DNA extracts by a Zymogen ‘One Step™ PCR Inhibitor Removal’ kit.

A PCR primer set for amplifying a ∼140–170 bp region of the 3′ end of the SSU rDNA of eukaryotes was designed manually on an alignment of this region that incorporated representatives of all major eukaryotic lineages. Two regions that had complete conservation in this alignment were then tested against all representatives of animal, plant and fungal phyla by BLAST searches of GenBank and minimal variation was found in the primer binding site. This primer set (SSU3′F and SSU3′R) was tested on a wide phylogenetic range of target species and was found to amplify from them as predicted. A blocking primer for suppressing amplification of penguin DNA (actually all tetrapods) was designed to bind in the region overlapping the 3′ end of SSU3′R and an adjacent region in the amplicon that is present in tetrapods, but not in other animal groups [30]. We tested this empirically as previously described [22]and found it to be successful in suppressing penguin SSU DNA amplification. This blocking primer oligo did also apparently have complete complementarity to some fish SSU rDNA, such as genbank accession JX282337for Sarpa salpa; and GenBank accession AF278682 for Raja schmidti. However, this was not the case for any of the fish groups in the Southern Ocean that are know or likely prey of Adélie penguins.

All subsequent amplifications were performed with fusion primers that incorporated the primer set described above as well as sequences for the IonTorrent sequencing system and short ‘tag’ sequences for identifying individual samples from a pooled set of sequences [31] using previously published sequence tags [32]. Each PCR contained template DNA purified from one scat, one unique combination of one of six forward and one of eight reverse primers and a ‘blocking oligo’ for suppressing amplification of penguin DNA. The primers used are shown in Table 2.

PCR reactions were performed in lots of 48. Each reaction of 10 uL contained 5 uL 2 × Phusion HF (NEB), 1 × Bovine Serum Albumin (NEB), 1 uM of each amplification primer and 10 uM of blocking primer ‘TetrapodBlockC3’ (see Table 2) with the remainder of the reaction volume being scat DNA extract. PCR thermal cycling conditions were 98°C, 5 min; then 40 cycles of 98°C, 5 s; 57°C, 20 s; and 72°C, 20 s. A final extension occurred at 72°C for one minute. PCR reactions from each set were pooled and purified from unincorporated reaction components by washing utilising reversible binding to Ampure (Agencourt) magnetic beads following the manufacturer’s protocol. A negative control reaction with no template DNA was included with each batch of 48 samples and this control reaction was sequenced as well. Batches were discarded if the negative control reaction contained any sequences from diet items.

Sequencing of PCR products was performed with an Ion Torrent next-generation sequencer and OneTouch semi-automated library preparation platform (Life Technology). The 200 bp sequencing kit v1 was used and 314 chips. Primary sequence estimation was done by Torrent Server software version 2.2 with the ‘Beverly read filter’ turned off to ensure the maximum number of sequences were included in the primary data. FASTQ files were transferred from the Torrent Server after checking run success based on the run reports generated by the Torrent Suite software. Samples from each location were split among different runs to avoid run-specific biases.

Sequence data was processed with scripts written in Python 2.7.2 (www.python.org) and R 2.13.1 (www.cran.r-project.org) by the authors and are available in the Dryad database archive for this paper. The Python scripts called the software package USEARCH for OTU clustering [33]. The scripts also used BLAST 2.2.6 for assigning OTUs to taxonomic categories [34]. A standalone BLAST database was created from the SILVA SSURef database release 108. This database was formatted with the full EMBL-format taxonomy used as the name for each sequence and the RNA sequences from the original database converted to DNA. It contained 5978 sequences representing most known eukaryotic lineages to genus or family level. This database is carefully curated, containing only high quality sequences derived from organisms identified by taxonomic experts [35].

Data was processed in two phases. In phase one, no aggregation of sequences to higher taxa was performed. This ‘training’ phase allowed the identification of broad prey groups that were at a taxonomic level higher than the SSU amplicon could resolve. For example, the first time a Scyphozoa sequence was identified, the closest match in our local SSU BLAST database had the full taxonomic description:

Metazoa;Cnidaria;Scyphozoa;Coronatae;Nausithoidae;Nausithoe;Nausithoe_rubra representing SSU sequence from the Scyphozoa species Nausithoe_rubra. At this point, we added ‘Scyphozoa’ to the list for aggregation. In any future sequence processing runs, sequences belonging to Scyphozoa were aggregated into one group in the final summaries. The software preserves each step of the processing as text files so that identifications can be tracked. All files resulting from the processing of the primary sequence data into final aggregated groups belonging to higher taxa were archived in the Dryad database entry for this paper.

After all of the samples were analysed in this training phase, a list of taxa to aggregate sequences to was generated as shown in Table 3. All of the IonTorrent runs were then re-analysed with aggregation to these higher taxa in the following steps, which follow a very similar scheme to other environmental DNA barcoding classification software such as QIIME [36]:

The assignment of sequences to categories started with a list of known Adélie penguin prey groups such as Euphausiidae and Actinopterygii for the ‘food category’; and lists of known contaminating taxa such as Primates (because human DNA is an inevitable contaminant), Diptera (flies) and Pezizomycotina (including many mold fungi) that represent laboratory contamination rather than penguin diet items for the ‘contaminant’ category. These lists were primarily derived from the ‘training’ analysis run of the sequence processing scripts. The ‘unicell’ category was defined by taxonomy and included all basal eukaryotic lineages that do not include metazoans or plants.

Once sequences for each scat were classified by taxonomy, the proportion of sequences belonging to each taxon in each population was calculated. Population average proportions were calculated from the mean of proportions in individual scats. Comparisons between populations were made by creating two Matrices of pairwise differences among the proportions for each food group within each population. Tests for correlation of the two matrices and their significance were made with a Mantel test [37] implemented with the R package ‘ade4’ [38]. This produced an estimated correlation, r, with a range from complete negative correlation at −1 to complete correlation at 1. An estimate of a p value for this r based on the distribution of r was calculated from 9999 simulated matrices. Comparisons were only made among proportions of food items that were found in both populations at 0.1% or more, which represented multiple sequence counts within each population, but excluded comparisons were no sequence was present in one of the two populations. This procedure tested for changes in ratios of prey items between two populations.

A qPCR melt curve assay for identifying penguin sex from scat samples was created by modifying existing assays targeting the Chromodomain helicase DNA binding 1 (CHD1) gene that has size polymorphisms between the Z and W sex chromosomes of most birds. This region has been used in similar assays for many years [39], [40] and our assay targeted a very small region of this, which made it especially successful in amplifying the fragmented CHD1 in penguin scat DNA. A full description of the validation and range of species that this works on has been submitted for publication separately.

Reactions of 10 uL took place on a Roche LC480 thermal cycler. Reactions contained 1 x Light Cycler Probes Master mix (Roche), 1 x EvaGreen, 1 x Bovine Serum Albumin (NEB) and 1 uM of each primer (see Table 2). Amplification conditions were 95°C, 5 min; then 40 cycles of 95°C, 10 s 60°C, 30 s with signal acquisition and 72°C, 10 s. Melt curve conditions were 55°C to 95°C at a ramp rate of 2.2°C per s with 5 signal acquisitions per degree.

Penguin scat samples were collected in the Australian Antarctic Territory with permission of the Australian Commonwealth Government under Australian Antarctic Science permits 2926, 4014, 2722 and 4088. No penguins were handled or approached closely as part of this work as the study material was scat samples collected outside breeding colonies. The collection procedure was approved by the Antarctic Animal Ethics Committee under project permits 2926, 4014, 2722 and 4088.

All data generated in this study, including the raw reads from the IonTorrent sequencer; the scripts used to process them, including example files; and spreadsheets containing the data are available from the Dryad database (http://datadryad.org/) with the doi: 10.5061/dryad.1rf7d.

A total of 534 scats were collected in four different years from four locations in East Antarctica as shown in Figure 1. The numbers of samples taken at each site and year are given in Table 1. After DNA was purified from these scats, PCR was performed with a ‘universal’ primer set which amplifies a taxonomically informative amplicon from the 3′ end of the small subunit ribosomal rDNA (SSU). A blocking oligonucleotide was included in PCR to limit amplification of penguin DNA and resultant amplicons. 389 samples (73% of those collected) produced amplicons that were sequenced with the IonTorrent sequencer.

A total of 1.27×10⁶ sequences passed quality checks on the Ion Torrent software and then Q score filtering and forward and reverse primer tag recognition in our scripts. These were derived from 16 runs of 314 chips and 6.5×10⁶ raw reads. Based on classification of sequences to broad taxonomic groups with our purpose written software, we recovered 650,520 sequences representing food items (∼51%); 136,056 sequences from parasites; 164,913 sequences from unicellular eukaryotes and 189,462 sequences that were classified as ‘contaminants,’ which had a sequence database match and includes penguin sequences. 133,943 sequences were unassignable, meaning that they did not have a match in the SSU database. These sequences include chimaeric PCR products, amplicons from organisms without SSU reference sequences and sequencing reads that although they had good overall Q scores they had low quality regions within them. Our assignment of sequences to individual samples involved recognition of tags on both the forward and reverse primers. This strategy of primer tagging to recover amplicons belonging to individual scats is highly effective in other situations where sequencing is accurate in the range that includes the reverse primer tag [20], [21], [31], [32]. However, the 200 bp IonTorrent sequencing kit produced sequences with a high error rate at their 3′ end, which caused large numbers of reads to not be assignable to the individual scat SSU amplicon pools that they were derived from. This strategy was still valuable for allowing multiple samples to be included in each run, but was not especially compatible with IonTorrent sequencing technology at that stage of its development.

The proportions of food sequences for all scats in this study that belong to each broad prey group are shown in Figure 2. Also shown is the overall frequency of occurrence (FOC) of each food group and co-occurrence of pairs of food groups. These samples were collected at different phases of the breeding season at all locations shown in Figure 1 and Table 1 so grouping them is intended to increase knowledge of the overall food diversity present in Adélie penguin diet. It is important to emphasise that the DNA proportions shown here are not intended to be a direct representation of biomass, energy consumed, or number of individual prey items consumed. There will be some correlations with quantity of sequences and these diet parameters, but biases are certain to exist as with any diet analysis method.

Krill (Euphausiidae) were the most important item in Adélie penguin diet in East Antarctica, making up about 40% of the proportion of the averaged DNA proportions and occurring in about 70% of samples and being the only diet item in 20% of all samples. Bony fish (Actinopterygii) and calanoid copepods are the next most significant diet items and the importance of these three groups correlates with other studies [16], [27]. Interestingly, Calanoida occur with fish (28%) more often than they do with Euphausiidae (23%) despite Euphausiidae occurring more often than fish in the overall diet results. This indicates that some of the detected copepods likely represent secondary predation (i.e. from consumption of fish that have eaten copepods). Amphipod shrimps (Amphipoda), have been identified in stomach contents of Adélie penguins previously [16], but have not been detected at the high relative abundance that we found them in at Mawson in 2010 and 2011. These also occur more frequently along with fish (12.7%) than with krill (8.7%) indicating that some of the penguin amphipod consumption may also be secondary.

The next most important groups are ‘true’ jellyfish (Scyphozoa) and comb jellies (Ctenophora), which have not previously been thought to be common in Adélie penguin diet, although gelatinous species have been recognised as a significant part of the diet of other seabirds [41], [42]. Other groups that have not to our knowledge previously been identified in Adélie diet and are unequivocally diet items include Siphonophora, Anthomedusae, Canalipalpata, Harpacticoida, Siphonostomatoida, Mysida, Asellota and Pleocyemata. The frequency of co-occurrence of prey groups in all scats in this study in Figure 2 shows that some of these rare items occur both with predatory food items such as fish or jellyfish and also with non-predatory groups such as pteropods or copepods. For example, ostracods co-occur with Pteropoda, Pleocyemata, Harpacticoida and Siphonostomatoida, none of which are predatory on ostracods, as well as fish. This indicates that they are likely to be mostly consumed by penguins directly.

Some items are not certain to be food. We took a conservative approach of including items like Springtails (Collembola) and terrestrial plants (Caryophyllales) in the ‘food’ category of the analysis as it is possible that these are consumed, although equally likely that they contaminate the scats before collection. Collembola are common in Antarctic soils and Caryophyllales are the only flowering plants in this region of the Antarctic, so their pollen or other parts may contaminate the scats. These are only minor diet items, so their inclusion doesn’t greatly alter the overall analysis.

No squid DNA was identified in this study. Squid beaks were identified previously in the stomach contents of 19 of 105 dead penguin chicks from Adélie Land [8]. We were surprised by this and wondered whether it was an artefact of the PCR method or the subsequent data processing. To test this, we spiked a sample with squid DNA and added extra squid sequences from GenBank to the local SSU BLAST database. When we ran the samples again, we correctly identified the spiked-in squid DNA. We have also identified squid in albatross diet using the same methodology, so we have some confidence that their absence here represents reality and squid were unimportant in the diet of these penguin populations.

The scat DNAs that produced SSU food amplicon results were used as template for PCR amplification of penguin CHD1alleles, which resulted in sex identification of 348 (89%) of the samples. Analysis of all dietary results that also had a sex assigned to them allowed an overall comparison of diet between male and female penguins as shown in Figure 3. This shows a remarkably similar overall diet among male and female Adélie penguins (Mantel r = 0.995, p = 0.0001).

Changes in relative proportions of DNA sequences were used to infer differences in prey consumption among different sets of samples. Overall differences in prey consumption were inferred by comparing matrices of pairwise ratios among all prey species present in each population. Significant differences as recognised by a Mantel test indicated an overall change in relative proportions of prey items. A summary of all the comparisons made in this way is given in Figure 4. DNA sequence proportions for individual scats can be seen in the population summaries included in the Dryad database entry for this study.

Two spatial scales of diet comparison were made. Comparisons were made among three widely separated sites in East Antarctica at ranges of 700–2100 km separation as shown in Figure 1. In 2012, collections were made at Mawson, Casey and Davis during the Crèche phase of the breeding cycle. The population diets estimated from these locations were substantially different as shown in Figure 4A. Notable differences include a higher proportion of fish (Actinopterygii) consumed at Mawson than at Casey or Davis; and a higher proportion of jellyfish (Scyphozoa) consumed at Davis than at Casey or Mawson. In contrast, population diets were estimated to be very similar from samples collected at two colonies in the Casey region at Whitney Point and Blakeney Point that are only 3.5 km apart as shown in Figure 4D.

Temporal comparisons of population diet were made at two scales. Interannual comparisons were made for samples collected from Mawson at the same phase of the breeding cycle in three consecutive years. This site produced the majority of the amphipod sequences found in this study. Small numbers were also found at Casey but Figure 4B shows that during 2010 and 2011 these were a major diet item during both Guard and Crèche breeding cycle phases at Mawson. However, in 2012 they are almost absent, only being found as a minor component of food DNA in one scat collected during Guard phase.

Comparisons among different phases of the breeding cycle were made at Davis as shown in Figure 4C. The diet during the Incubation and Guard phases of the breeding cycle was similar but changed with the transition to Crèche phase. This apparently represents a decrease in the amount of krill (Euphausiidae) in the diet and an increase in jellyfish (Scyphozoa) and calanoid copepods (Copepoda).

The SSU 3′ primer set amplifies an SSU amplicon from a wide phylogenetic range of eukaryotes. Many of these are not food items, but some of the groups are of interest for other reasons. Figure 5 shows metazoan parasites and unicellular eukaryotes identified in all scats analysed in this study. Tapeworm (Cestoda) DNA is present in almost 93% of scats that contained amplifiable DNA and forms a large proportion of the parasite DNA. Monogenea are ectoparasitic flatworms generally known as fish parasites and these may have been ingested with fish, or they may be penguin parasites that are consumed in grooming. Mites (Oribatida) are likely to be penguin mites that either are consumed in grooming or contaminate scats from nests or the body of penguins. The unicellular eukaryote diversity is dominated by Saccharomycotina, which are apparently a large part of the gut microflora.

Ecosystem monitoring utilising DNA based diet analysis of top level predator diet is a new approach that has not been widely tested. Our results indicate that a broad-scale DNA based analysis can identify changes in Adélie penguin diet at regional, inter-annual and inter-season scales. The diet of this species is far more diverse than has previously been revealed by gut content and stable isotope analysis. It is important to be able to detect all significant components of Adélie penguin diet as fisheries for krill and bony fish are likely to cause a shift to some of these groups such as jellyfish and comb jellies. Scat collection is a simple and efficient field collection procedure that allows a wide range of questions on diet to be addressed. Analysis of food DNA with the broad-scale PCR and sequencing approach presented here is efficient once established in a laboratory and can be supplemented with other tests for penguin gender, or for species-level food identification. We anticipate that the sort of approach we present here could make a significant contribution to ecosystem monitoring programs and we show how it might be useful for CEMP here. The methodology used here is directly applicable to monitoring the diet of any bird or mammal and could be used to address a variety of ecological questions.