Limits...
A Fatty Acid Based Bayesian Approach for Inferring Diet in Aquatic Consumers.

Galloway AW, Brett MT, Holtgrieve GW, Ward EJ, Ballantyne AP, Burns CW, Kainz MJ, Müller-Navarra DC, Persson J, Ravet JL, Strandberg U, Taipale SJ, Alhgren G - PLoS ONE (2015)

Bottom Line: To test the model, we simulated hypothetical Daphnia comprised of 80% diatoms, 10% green algae, and 10% cryptophytes and compared the FA signatures of these known pseudo-mixtures to outputs generated by the mixing model.The accuracy and precision of SI based estimates was extremely sensitive to both resource and consumer uncertainty, as well as the trophic fractionation assumption.These results indicate that when using only two tracers with substantial uncertainty for the putative resources, as is often the case in this class of analyses, the underdetermined constraint in consumer-resource SI analyses may be intractable.

View Article: PubMed Central - PubMed

Affiliation: Oregon Institute of Marine Biology, University of Oregon, Charleston, Oregon, 97420, United States of America.

ABSTRACT
We modified the stable isotope mixing model MixSIR to infer primary producer contributions to consumer diets based on their fatty acid composition. To parameterize the algorithm, we generated a 'consumer-resource library' of FA signatures of Daphnia fed different algal diets, using 34 feeding trials representing diverse phytoplankton lineages. This library corresponds to the resource or producer file in classic Bayesian mixing models such as MixSIR or SIAR. Because this library is based on the FA profiles of zooplankton consuming known diets, and not the FA profiles of algae directly, trophic modification of consumer lipids is directly accounted for. To test the model, we simulated hypothetical Daphnia comprised of 80% diatoms, 10% green algae, and 10% cryptophytes and compared the FA signatures of these known pseudo-mixtures to outputs generated by the mixing model. The algorithm inferred these simulated consumers were comprised of 82% (63-92%) [median (2.5th to 97.5th percentile credible interval)] diatoms, 11% (4-22%) green algae, and 6% (0-25%) cryptophytes. We used the same model with published phytoplankton stable isotope (SI) data for δ13C and δ15N to examine how a SI based approach resolved a similar scenario. With SI, the algorithm inferred that the simulated consumer assimilated 52% (4-91%) diatoms, 23% (1-78%) green algae, and 18% (1-73%) cyanobacteria. The accuracy and precision of SI based estimates was extremely sensitive to both resource and consumer uncertainty, as well as the trophic fractionation assumption. These results indicate that when using only two tracers with substantial uncertainty for the putative resources, as is often the case in this class of analyses, the underdetermined constraint in consumer-resource SI analyses may be intractable. The FA based approach alleviated the underdetermined constraint because many more FA biomarkers were utilized (n < 20), different primary producers (e.g., diatoms, green algae, and cryptophytes) have very characteristic FA compositions, and the FA profiles of many aquatic primary consumers are strongly influenced by their diets.

No MeSH data available.


The results of an analysis where three potential food resources were either set to 100% of the food consumed or 0% (e.g., 100% diatoms, 0% green algae, and 0% cyanobacteria).This resulted in 9 outputs for the three SI based analyses (red). Because the outputs for the three SI cases where the subsidy should have been 100% were very similar, as was also true for the six cases where the subsidy should have been 0%, the three 100% responses and the six 0% responses were aggregated in this plot. For the FA based analyses (blue), we simply analyzed the original 10 cases where Daphnia consumed diatom monocultures. We also analyzed the 8 cases where the Daphnia consumed green algae monocultures as well as the 8 cases where they consumed cryptophyte monocultures. In these 26 cases, the correct answer was always obtained to multiple decimal places. The curves represent the 1/10th percentile (n = 1000) density distribution of the model posterior densities grouped into 40 bins.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4482665&req=5

pone.0129723.g002: The results of an analysis where three potential food resources were either set to 100% of the food consumed or 0% (e.g., 100% diatoms, 0% green algae, and 0% cyanobacteria).This resulted in 9 outputs for the three SI based analyses (red). Because the outputs for the three SI cases where the subsidy should have been 100% were very similar, as was also true for the six cases where the subsidy should have been 0%, the three 100% responses and the six 0% responses were aggregated in this plot. For the FA based analyses (blue), we simply analyzed the original 10 cases where Daphnia consumed diatom monocultures. We also analyzed the 8 cases where the Daphnia consumed green algae monocultures as well as the 8 cases where they consumed cryptophyte monocultures. In these 26 cases, the correct answer was always obtained to multiple decimal places. The curves represent the 1/10th percentile (n = 1000) density distribution of the model posterior densities grouped into 40 bins.

Mentions: For the 26 cases where Daphnia were fed algal monoculture diets, i.e., diatoms (n = 10), green algae (n = 8) and cryptophytes (n = 8), the FA profiles for the individual cases had strong statistical associations with the averages for their respective resource libraries (i.e., the r2 = 0.88 ± 0.07). For these 26 cases, the algorithm in all cases correctly classified the diet contributions from the different algal groups to multiple decimal places (Fig 2). The algorithm was much less capable of classifying diet using carbon and nitrogen SI. The three scenarios tested (i.e., 100% diatoms, 100% green algae and 100% cyanobacteria), generated nine outcomes (i.e., three for each scenario). The actual outcomes for the three 100% cases, and the six 0% cases, were very similar within group so the results of these groups were pooled into two sets of responses. For the three cases that should have had a 100% contribution, the algorithm median and 95% credible interval (i.e., the 2.5th to 97.5th percentile range of the model solution posterior density) contribution was 64% (3–95%) (Fig 2). In the six cases that should have had a 0% contribution, the modeled median contribution was 14% (0–80%). In the 100% cases, the posterior distribution was very flat and few outcomes were excluded from the 95% credible interval (Fig 2), and the correct answer was one of the few outcomes that was excluded. In both the 100% and 0% cases, the output values were very similar to the average of the prior expectation (i.e., equal contributions from the three putative resources) and correct values, i.e., 67 and 17%, respectively.


A Fatty Acid Based Bayesian Approach for Inferring Diet in Aquatic Consumers.

Galloway AW, Brett MT, Holtgrieve GW, Ward EJ, Ballantyne AP, Burns CW, Kainz MJ, Müller-Navarra DC, Persson J, Ravet JL, Strandberg U, Taipale SJ, Alhgren G - PLoS ONE (2015)

The results of an analysis where three potential food resources were either set to 100% of the food consumed or 0% (e.g., 100% diatoms, 0% green algae, and 0% cyanobacteria).This resulted in 9 outputs for the three SI based analyses (red). Because the outputs for the three SI cases where the subsidy should have been 100% were very similar, as was also true for the six cases where the subsidy should have been 0%, the three 100% responses and the six 0% responses were aggregated in this plot. For the FA based analyses (blue), we simply analyzed the original 10 cases where Daphnia consumed diatom monocultures. We also analyzed the 8 cases where the Daphnia consumed green algae monocultures as well as the 8 cases where they consumed cryptophyte monocultures. In these 26 cases, the correct answer was always obtained to multiple decimal places. The curves represent the 1/10th percentile (n = 1000) density distribution of the model posterior densities grouped into 40 bins.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4482665&req=5

pone.0129723.g002: The results of an analysis where three potential food resources were either set to 100% of the food consumed or 0% (e.g., 100% diatoms, 0% green algae, and 0% cyanobacteria).This resulted in 9 outputs for the three SI based analyses (red). Because the outputs for the three SI cases where the subsidy should have been 100% were very similar, as was also true for the six cases where the subsidy should have been 0%, the three 100% responses and the six 0% responses were aggregated in this plot. For the FA based analyses (blue), we simply analyzed the original 10 cases where Daphnia consumed diatom monocultures. We also analyzed the 8 cases where the Daphnia consumed green algae monocultures as well as the 8 cases where they consumed cryptophyte monocultures. In these 26 cases, the correct answer was always obtained to multiple decimal places. The curves represent the 1/10th percentile (n = 1000) density distribution of the model posterior densities grouped into 40 bins.
Mentions: For the 26 cases where Daphnia were fed algal monoculture diets, i.e., diatoms (n = 10), green algae (n = 8) and cryptophytes (n = 8), the FA profiles for the individual cases had strong statistical associations with the averages for their respective resource libraries (i.e., the r2 = 0.88 ± 0.07). For these 26 cases, the algorithm in all cases correctly classified the diet contributions from the different algal groups to multiple decimal places (Fig 2). The algorithm was much less capable of classifying diet using carbon and nitrogen SI. The three scenarios tested (i.e., 100% diatoms, 100% green algae and 100% cyanobacteria), generated nine outcomes (i.e., three for each scenario). The actual outcomes for the three 100% cases, and the six 0% cases, were very similar within group so the results of these groups were pooled into two sets of responses. For the three cases that should have had a 100% contribution, the algorithm median and 95% credible interval (i.e., the 2.5th to 97.5th percentile range of the model solution posterior density) contribution was 64% (3–95%) (Fig 2). In the six cases that should have had a 0% contribution, the modeled median contribution was 14% (0–80%). In the 100% cases, the posterior distribution was very flat and few outcomes were excluded from the 95% credible interval (Fig 2), and the correct answer was one of the few outcomes that was excluded. In both the 100% and 0% cases, the output values were very similar to the average of the prior expectation (i.e., equal contributions from the three putative resources) and correct values, i.e., 67 and 17%, respectively.

Bottom Line: To test the model, we simulated hypothetical Daphnia comprised of 80% diatoms, 10% green algae, and 10% cryptophytes and compared the FA signatures of these known pseudo-mixtures to outputs generated by the mixing model.The accuracy and precision of SI based estimates was extremely sensitive to both resource and consumer uncertainty, as well as the trophic fractionation assumption.These results indicate that when using only two tracers with substantial uncertainty for the putative resources, as is often the case in this class of analyses, the underdetermined constraint in consumer-resource SI analyses may be intractable.

View Article: PubMed Central - PubMed

Affiliation: Oregon Institute of Marine Biology, University of Oregon, Charleston, Oregon, 97420, United States of America.

ABSTRACT
We modified the stable isotope mixing model MixSIR to infer primary producer contributions to consumer diets based on their fatty acid composition. To parameterize the algorithm, we generated a 'consumer-resource library' of FA signatures of Daphnia fed different algal diets, using 34 feeding trials representing diverse phytoplankton lineages. This library corresponds to the resource or producer file in classic Bayesian mixing models such as MixSIR or SIAR. Because this library is based on the FA profiles of zooplankton consuming known diets, and not the FA profiles of algae directly, trophic modification of consumer lipids is directly accounted for. To test the model, we simulated hypothetical Daphnia comprised of 80% diatoms, 10% green algae, and 10% cryptophytes and compared the FA signatures of these known pseudo-mixtures to outputs generated by the mixing model. The algorithm inferred these simulated consumers were comprised of 82% (63-92%) [median (2.5th to 97.5th percentile credible interval)] diatoms, 11% (4-22%) green algae, and 6% (0-25%) cryptophytes. We used the same model with published phytoplankton stable isotope (SI) data for δ13C and δ15N to examine how a SI based approach resolved a similar scenario. With SI, the algorithm inferred that the simulated consumer assimilated 52% (4-91%) diatoms, 23% (1-78%) green algae, and 18% (1-73%) cyanobacteria. The accuracy and precision of SI based estimates was extremely sensitive to both resource and consumer uncertainty, as well as the trophic fractionation assumption. These results indicate that when using only two tracers with substantial uncertainty for the putative resources, as is often the case in this class of analyses, the underdetermined constraint in consumer-resource SI analyses may be intractable. The FA based approach alleviated the underdetermined constraint because many more FA biomarkers were utilized (n < 20), different primary producers (e.g., diatoms, green algae, and cryptophytes) have very characteristic FA compositions, and the FA profiles of many aquatic primary consumers are strongly influenced by their diets.

No MeSH data available.