Experimental tissue mixture samples
Across the portfolio of four top-performing markers, sequencing read depth did not provide a consistent quantitative proxy for the relative amount of tissue added to the fishmeal mixture. While there was a tendency for species with lower input amounts to receive fewer reads, the proportion of reads varied substantially among species that were added in the same amounts (Fig. 4, Fig. 6A). The mean proportion of reads for the five taxa added in the same amount showed a positive relationship for tissue inputs <1%, with the lowest tissue input (0.01%) receiving an order-of-magnitude smaller proportion of reads across all four markers (1.7 x 10-4 of reads) than the next lowest tissue input (0.1%; 5.3 x 10-3of reads; SI, Fig. S6). However, the relationship between read proportion and tissue input was inconsistent because the 1.91% tissue input level received, on average, a smaller proportion of reads than taxa that comprised 1.0% of the tissue input (5.6 x 10-3 and 2.7 x 10-2 of reads, respectively) and for three of the four markers, the highest tissue input level (13.32% of fishmeal) received a smaller proportion of reads than taxa comprising 3.65% of the fishmeal mixture (Fig. 4 and SI, Fig. S6).
The ratio of fishmeal-to-filler in the experimental feeds had little effect on the proportion of reads per taxon (Fig. 4). Taxa that comprised 0.01% and 0.1% of the fishmeal showed more variation than taxa present in higher proportions, with inconsistent read depth across fishmeal-to-filler ratios (2%, 10% or 25% fishmeal) and across the four markers (Fig. 4). Although read depths were inconsistent in terms of presence/absence detection, fewer of the low-input taxa were recovered, and this pattern of taxon drop-out increased as the ratio of fishmeal-to-filler decreased (SI, Fig. S7).
Matrix composition (filler) of experimental feeds did not impact the recovery of reference taxa or the proportion of reads attributed to those taxa for constituents that make up >1% of fishmeal. Patterns in proportion of reads were dominated by variation among taxa rather than between filler types, and the proportion of reads did not change between the soy filler and the animal/plant filler for taxa added at >1% fishmeal (Fig. 5). However, low-input taxa, those that comprised 0.01% or 0.1% of fishmeal, displayed more variable read depth between fillers, but not in a consistent way within or across markers. At the lowest tissue input (0.01%), two markers (12S; MiFish and one COI primer set; nsCOIFo) consistently showed a higher proportion of taxon drop-out (SI, Fig. S7).
Although two feeds consisted entirely of filler with no added fishmeal, sequencing results showed that 80% of fishmeal taxa were present in these filler-only feeds, indicating that simultaneous preparation of filler-only feeds and fishmeal feeds resulted in contamination that was sufficient to be detected by our PCR assay (SI, Fig. S8). Sequencing read counts for filler 1 (soy flour) were lower than for filler 2, which contained bloodmeal (from Sus sp. ) and feathermeal (fromGallus sp. ). Both taxa were detected by all four of the markers designed to amplify fishes, and in the absence of added fishmeal tissue,Sus sp. and Gallus sp. received an average of ~25% of the sequencing reads per replicate, whereas in the experimental feeds that contained 75% filler 2 and 25% fishmeal,Sus sp. and Gallus sp. received only ~1% of sequencing reads, highlighting the influence of template competition during library preparation.
Comparing taxon recovery from the 100% fishmeal and pooled DNA extracts (MV) showed that reference taxa were detected (presence/absence) equally well regardless of relative representation when tissue input/DNA concentration was >1% (Fig. 6A, 6B). For input categories below 1% (0.01% and 0.1%), more taxa were identified from the 100% fishmeal sample than from the MV (that contained DNA extracted separately from each experimental feed fish constituent and pooled in proportions equivalent to relative tissue inputs to the feeds). In comparison, when the DNA extracts from each experimental feed constituent was pooled in equal concentration (the ME pool), we recovered all taxa with at least one locus with just one exception (Ophisternon sp. ), indicating that taxon drop-out in the MV pool was not a result of primer mismatch but rather a result of low input concentration (all species except for Loliolus sp. were consistently identified at concentrations above 2.24 x 10-4 ng/ul; Fig. 6B, 6C).