Evaluating the performance of the Bayesian mixing tool MixSIAR with fatty acid data for quantitative estimation of diet
Case 1: spectacled eiders, from Wang et al.42
The experiment
This case is based on captive feeding trials conducted on 8 adult spectacled eiders, Somateria fischeri, which were maintained on an initial diet containing 1% Atlantic surf clam, 3% Antarctic krill, 88% Mazuri sea duck formula, 4% blue mussel and 4% Atlantic silverside, for 69 days prior to the start of the feeding experiment. After this, on day 0, a biopsy sample of the synsacral adipose tissue was obtained from each eider. With the FA data of the adipose tissue the authors calculated CCs. Feeding trials started on Day 0, and spectacled eiders were switched to diet A, consisting of 56% krill and 44% Mazuri sea duck formula for 21 days. On Day 21, eiders were biopsied again and switched to diet B consisting of 48% Mazuri formula and 52% silverside. On Day 50, a final biopsy sample was collected (Fig. 2A). FA turnover was considered near complete by 69 days.
Figure 2
Spectacled eiders case. (A) Feeding experiment: spectacled eiders (n = 8) spent 69 days on the initial diet described in the figure; after this, on day 0 eiders were biopsied and switched to diet A. On day 21 they were biopsied again and switched to diet B. After 29 days on diet B, eiders were biopsied on day 50. (B) Plots for diet estimations of spectacled eiders fed different combined diets. The true diet is indicated in each plot by the blue asterisks. CCs were calculated using FA data of day 0. Images by Alicia Guerrero.
Full size image
The model
This analysis is based on FA data from day 0, 21 and 50. We used the CCs calculated in this study after eiders were maintained on the same initial diet for 69 days. All sources were significantly different from each other (PERMANOVA, F4 = 9936.9, P = 0.001). ‘Day’ was set as fixed factor in the model.
Diet predictions
Based on FA data of day 0, MixSIAR estimated a contribution of 0.1% clam, 4% krill, 88% Mazuri, 6% mussel, and 1% silverside (Fig. 2B). After the first shift of diet, MixSIAR estimations changed to 28% krill and 62% Mazuri for FA data obtained on day 21. The final biopsy sample on day 50 produced estimations of 62% Mazuri, and 27% silverside.
Case 2: Steller’s eiders, from Wang et al.42
The experiment
This case corresponds to a feeding trial conducted simultaneously to the previous case by Wang et al.42, although the diet of Steller’s eiders, Polysticta stelleri, differed slightly. For 69 days prior to the start of the feeding trial, 8 adult Steller’s eiders were maintained on an initial diet containing 1% clam, 1% Antarctic krill, 88% Mazuri sea duck formula, 7% mussel, and 3% silverside. CCs were calculated after a biopsy was extracted to each eider on day 0. At the start of the feeding experiment, on day 0, Steller’s eiders were switched to diet A, containing 66% krill and 34% Mazuri formula. Then, on day 21, they were switched to diet B, consisting of 34% Mazuri formula and 66% silverside. Biopsy samples were collected on days 0, 21 and 50 (Fig. 3A). FA turnover was considered near complete by 69 days.
Figure 3
Steller’s eiders case. (A) Feeding experiment: eiders (n = 8) were maintained on the initial diet described in the figure for 69 days after which biopsy samples were collected (day 0) and eiders were switched to diet A. After 21 days, eiders were biopsied again and switched to diet B. On day 50, after 29 days on diet B, eiders were biopsied one last time. (B) Plots of diet estimation for Steller’s eiders fed different combined diets. Diet estimates are based on biopsy samples collected on days 0, 21 and 50. The true diet is indicated in each plot by red asterisks. CCs were calculated using FA data of day 0. Images by Alicia Guerrero.
Full size image
The model
We used the CCs calculated in the same study after Steller’s eiders were maintained on the same diet for 69 days. All sources were significantly different from each other (PERMANOVA, F4 = 10,079, P = 0.001). ‘Day’ was set as fixed factor in the model.
Diet predictions
Based on FA data of Day 0, MixSIAR estimated a contribution of 0.2% clam, 2% krill, 90% Mazuri, 4% mussel, and 3% silverside (Fig. 3B). After the first diet switch, MixSIAR estimations changed to 46% krill and 27% Mazuri. The final biopsy sample on day 50 produced estimations of 9% krill, 28% Mazuri, and 46% silverside.
Case 3: Atlantic salmon, from Budge et al.43
The experiment
For 22 weeks, tank-reared juvenile Atlantic salmon, Salmo salar (n = 132), were fed one of four different formulated feeds based on two marine oils: 100% krill oil, 100% herring oil, a mixture of 70:30 herring to krill oil, or a mixture of 30:70 herring to krill oil. Muscle samples were analysed for FAs after the 22-week experiment, which allowed the calculation of CCs. In this experiment, two sets of CCs were calculated: one derived from salmon fed the diet based on 100% herring oil and another from salmon fed the diet based on 100% krill oil (Fig. 4A). Unlike the previous two cases, where CCs were derived from consumers eating a mixed diet, here CCs were obtained from consumers feeding on a single type of source: either herring or krill oil. This allowed us to evaluate whether the source used to calculate the CCs affected dietary predictions. Additionally, we calculated a combined CC (an average value between CCs derived from krill and herring diets) and run a separate model. FA turnover was considered complete after 22 weeks on the same diet.
Figure 4
Atlantic salmon case. (A) Feeding experiment: Atlantic salmon fed formulated feeds based on either solely herring (n = 36) or krill oil (n = 28), or in proportions of 70:30 or 30:70 herring to krill oil (n = 34 each), for 22 weeks. (B) Plots of diet proportions estimated using MixSIAR for models using CCs derived from salmon fed a herring-oil diet (HO-CC), a krill-oil diet (KO-CC) or CCs averaged from these two treatments (Combined-CC). The true diet is indicated in each plot by black asterisks. Salmon image designed by Creazilla (https://creazilla.com).
Full size image
The model
We ran three independent models to estimate the diet of salmon fed each of the four diets: one using the set of CCs derived from herring oil, another using the CCs derived from krill oil, and one using the combined CCs. For each model, we used krill oil, herring oil and initial diet (commercial feed given to salmon prior to the experiment) as sources, and set ‘diet group’ as fixed factor. Significantly different FA compositions were found for the three sets of sources: those multiplied by herring oil CCs (PERMANOVA, F2 = 4008.7, P = 0.003), those multiplied by krill oil CCs (PERMANOVA, F2 = 3354.7, P = 0.002), and those multiplied by the combined CCs (PERMANOVA, F2 = 3677.2, P = 0.003).
Diet predictions
When we used CCs derived from salmon fed on a diet supplemented with herring oil only, MixSIAR correctly estimated the contribution of herring oil in the consumers (salmon) diet (98%). However, the contribution of herring was slightly overestimated where the consumers had been fed a mixture of herring and krill oil (Fig. 4B), and where the salmon’s diet was based on krill oil the contribution of krill oil was slightly underestimated (89%).
We found the opposite trend when we used CCs derived from salmon that had been fed a diet where krill oil had been the lipid source. Here, MixSIAR correctly estimated the contribution of krill oil in the salmon’s diet when they had been fed a diet based on krill oil (98%) or a mixture of 70:30 herring to krill oil (33% krill) or 30:70 herring to krill oil (70% krill); however, when herring oil had been the only dietary source this dietary contribution was underestimated (81%) (Fig. 4B).
Our dietary estimates were less biased when we used CC values that had been derived from the average between the herring- and krill-oil treatments (Combined-CCs). For example, we estimated herring contribution to be 90% of the diet when the actual diet was supplemented only with herring oil, and an estimate of 95% krill contribution when the actual diet was supplemented with krill oil only, and when the actual diet was a combination of herring (70%) and krill (30%), the estimations were 71% and 27%, and where the actual contribution of herring was 30% and 70% of krill, the estimated diets were 34% herring and 65% krill (Fig. 4B).
Case 4: tufted puffin nestlings, from Williams et al. 44
The experiment
Tufted puffin, Fratercula cirrhata, nestlings (n = 6) underwent an experimental feeding trial in their own burrows. Chicks were fed by their parents for approximately 10 days since hatching. When they were estimated to be 10-days old, the access to the burrows was blocked, so adults could not continue feeding their chicks. Through another access hole excavated by the researchers, chicks began being fed Pacific herring once a day, for 27 days. To infer the diet of free-living puffin nestlings during the first 10 days after hatching, wire screens were placed at burrow entrances to collect whole fish dropped by the parents. The species identified, in descending order (by mass), were Pacific sandlance, capelin, Pacific sandfish, salmonid and Pacific cod. An adipose tissue sample was collected on days 10 (start of the feeding trial), 19, 28 and 37. On day 37, assuming complete FA turnover after 27 days on a single prey diet (herring), the researchers calculated CCs (Fig. 5A). The data used to run this model included day 10, which represents the unknown diet provided by the parents, days 19, 28, and 37 which represent the herring diet at different extents. FA turnover was considered “close to, but not entirely complete” after 27 days44.
Figure 5
source even though it was not part of the nestlings’ diet. The red asterisks in each plot represent the potential diet fed by the parents (and used as priors in the first model). (D) Plots of diet estimations for tufted puffins fed herring, based on their FA profiles of days 19, 28 and 37. The true diet is indicated in each plot by red asterisks. CCs were calculated from tufted puffins fed herring, using FAs from day 37. Images by Alicia Guerrero.
Tufted puffins case. (A) Nestlings (n = 6) were fed by their parents for approximately 10 days since hatching. After this, they were fed herring for another 27 days as the entrance to their burrows was blocked and parents could not feed their chicks. (B) Non-metric multidimensional scaling plots for FAs obtained from chicks at different stages of the experiment and their sources. When FAs of sources were multiplied by their respective CCs, source (herring) and consumer (chicks, day 37) overlap in the plot. (C) Plots of the three models run to estimate the diet of nestlings on day 10. From left to right: Model using informative priors based on meals brought by the parents after the burrows were blocked; the same model without informative priors; and a third model without informative priors but including herring as
Full size image
The model
We used the CCs derived from these chicks feeding on Pacific herring for 27 days. For biopsy samples collected on day 10, we conducted three separate analyses: the first model excluded herring as potential prey, and incorporated informative priors based on the amount of different fish (% by mass) dropped by the parents at the burrow entrances; the second model was exactly the same but without prior information; and the third model was run without priors, and included herring as potential prey in order to determine whether this prey could be identified as absent.
For days 19, 28 and 37, we ran another analysis and included herring as source and no informative priors. ‘Day’ was set as a fixed factor in this model. All sources had significantly different FA compositions (PERMANOVA, F5 = 430.9, P = 0.001).
Diet predictions
In this example we included a non-metric dimensional scaling plot (Fig. 5B) to evaluate the effect of applying CCs to sources. Day 0 biopsies show greater variation than those of successive days, in both plots. When using original sources and consumer FA values, chicks are segregated from all the sources, although the similarity of the FAs increases toward the FA of herring as days pass, but they do not match. When CCs were applied to sources, the FA values of herring and chicks from day 37 overlap, indicating that they have the same FA compositions.
For day 0 (Fig. 5C), the estimated diet contributions were similar to meals brought by the parents when the model included informative priors, where the main dietary sources were sandlance (72%) and capelin (15%). Whereas estimates from the model without informative priors misrepresented the diet, as capelin was wrongly identified as the main dietary source (61%), cod the second most important prey (26%), and sandlance was incorrectly estimated to be only 6% of the diet. The third model including herring again identified capelin and cod as the main contributors (56% and 23%, respectively) whereas herring was identified as the least important prey, with 0.9% of contribution.
For day 19 (Fig. 5D), herring was identified as the main source, with 60% of contribution, followed by capelin and sandlance although with greater variation. From day 19 to 37, the contribution of herring increases from 60 to 97%, respectively.
Case 5. Harp seals, from Kirsch et al.45
The experiment
This study evaluated the effect of a low-fat diet on blubber FAs of harp seals, Pagophilus groenlandicus. Only for this experiment, the fat content of the different sources was available. Juvenile harp seals (n = 5) had been maintained on a diet of Atlantic herring (≥ 9% fat) for approximately 1 year prior to the feeding trial. On day 0, a full-depth blubber sample was collected from the posterior flank of each animal. For 30 days, seals were kept on a diet consisting solely of Atlantic pollock (1.7% fat). Blubber biopsies were taken again on days 14 and 30 (Fig. 6A). FA turnover was not considered complete after 30 days on the same diet, and authors suggest that a longer period on the diet, or higher intakes of fat, would be needed to accomplish it.
Figure 6
Diet estimates for juvenile harp seals fed a low-fat prey. (A) Feeding experiment: For a year prior to the feeding experiment, harp seals (n = 5) had been eating only Atlantic herring, a prey with a high-fat content. During the feeding experiment, seals were fed Atlantic pollock, a low-fat prey, for 30 days. (B) Plots of estimates derived from MixSIAR models based on FA data of whole blubber cores from days 0, 14 and 30. The black asterisks in the plots indicate the true diet. CCs correspond to harp seals fed herring, calculated using FAs from day 0. Images: fish by Lukas Guerrero Zambra, harp seal by Alicia Guerrero.
Full size image
The model
Since seals had been feeding on the same source for a year, we calculated CCs using FA data from day 0. Thus, consumer values were divided by Atlantic herring values producing CCs that were applied to both herring and pollock FA data. Sources were significantly different (PERMANOVA, F1 = 3680.4, P = 0.001) and ‘day’ was set as fixed factor in the model.
Diet predictions
For all three data sets (day 0, 14 and 30) the main contributor to the diet was Atlantic herring. The predicted proportion of Atlantic herring decreased only slightly from 99% on day 0 to 95% on day 30. Consequently, the contribution of pollock increased from 1% on day 0, to 5% on day 30 (Fig. 6B).
Case 6: Harbour seals, from Nordstrom et al.46
The experiment
Estimations are based on a feeding experiment by Nordstrom et al.46. Prior to the feeding study, juvenile harbour seals, Phoca vitulina, were fed a homogenate of 2:1 Pacific herring to salmon oil for approximately three weeks, and then fed only Pacific herring for four to six days. The feeding experiment consisted of three diets: one group of seals was fed only herring for 42 days (n = 3); the second group was fed only surf smelt for the same period (n = 6); and a third group (n = 7) was fed smelt for 21 days and then only herring for 21 days (Fig, 7A). Whole blubber core samples were collected on days 0, 21 and 42 for each group. Complete FA turnover was estimated to occur after at least 55 days on the same diet, although it could extend well beyond if turnover rate slowed with time.
The model
Since prey FA data was not provided in the same study, we used Pacific herring, surf smelt, and salmon FA values from Huynh and Kitts47, which had significantly different FA composition (PERMANOVA, F2 = 87.7, P = 0.001). CCs were derived from other harbour seals on a Pacific herring diet for over a year, in the same study46. We estimated the diet of the three groups of harbour seals, based on samples collected on day 42, setting ‘diet group’ as fixed factor in our model.
Diet predictions
Overall, diet differences were evident among groups, and the direction of the change was consistent with the shifts in diet. Estimates for harbour seals fed exclusively Pacific herring for 42 days, correctly showed that diet was predominantly based on herring (94.7%). For seals fed surf smelt for 42 days; however, estimates showed that surf smelt only accounted for 26.6% of the diet whereas herring remained to be the main component. For seals fed surf smelt for 21 days and then herring for another 21 days, MixSIAR again identified herring as the main component, with 90.9%, whereas surf smelt was only 3% (Fig. 7B).
Figure 7
Diet estimates for harbour seals. (A) Feeding experiment: For 42 days, seals were fed the following diets: solely herring (n = 3), solely surf smelt (n = 6), or surf smelt for the first 21 days and then herring for the remaining 21 days (n = 7). Prior to the feeding experiments they had been fed a mixture of herring and salmon. (B) Plots for MixSIAR diet estimations, based on blubber FAs obtained on day 42. The red asterisks indicate the true diet. CCs were obtained from harbour seals (other individuals) fed herring for over a year. Image by Alicia Guerrero.
Full size image More
