Metabolic theory and taxonomic identity predict nutrient recycling in a diverse food web

Excretion rates of N or P or both and body mass were measured for 667 individual fish and 233 individual macroinvertebrates. Of these individuals, at least one of three additional traits were measured: (i) somatic nitrogen (N), (ii) somatic phosphorus (P), or (iii) δ¹⁵N (as a proxy for trophic position). Vertebrate species were diverse in taxonomy, morphology, and physiology (28 families, 43 genera, and 72 species, including boxfish, moray eels, and pipefish), as were invertebrates (22 families, 28 genera, and 30 species, including crabs, shrimp, jellyfish, gastropods, and sea cucumbers). Variation in body nutrient content was substantial [%N: 2.32 (pitho crab) to 14.2 (parrotfish); %P: 0.19 (sea cucumber) to 8.7 (purple mouth moray eel); N:P: 0.26–146; %C: 11.1 (pitho crab) to 60.1 (green moray eel) of dry mass]. Mass-specific excretion rates were variable among individuals (N: 0.002–351.2 μg/g per hour; P: 0.0009–35.9 μg/g per hour; N:P: 0.038–4,156 molar ratio) and span a broader range than previous studies (e.g., ref. 16: N excretion: ∼9–30 μg/g per hour; P excretion: ∼2–13 μg/g per hour; NP: ∼1–7.5; see Table S1 for further information).

Relationships between the logarithm of nutrient excretion and the logarithm of body wet mass (hereafter, mass) were positive and significant for N, P, and N:P (Fig. 1). Evidence for allometric scaling of nutrient excretion at the predicted 3/4 power rule was found for P excretion (slope = 0.73, 95% CI = 0.66–0.80) but not for N excretion, which did not significantly differ from 1 (slope = 1.0, 95% CI = 0.95–1.04), suggesting isometry (8, 19). The slope for N:P excretion was significantly positive (slope = 0.16, 95% CI = 0.16–0.22; Fig. 1), in contrast to predictions.

Linear regression was used to explore the variation in E_N, E_P, and E_NP as explained by mass, body nutrient content, and δ¹⁵N. In each case, model residuals (from mass-excretion, body nutrient-excretion, and δ¹⁵N-excretion regression models for E_N, E_P, and E_NP) were used to explore how much additional variation in the data was explained by each subsequent variable (e.g., the amount of variation explained by body nutrient content and δ¹⁵N after first accounting for the mass; Fig. 2). In all cases, body mass explained the largest proportion of variation in the data. Nutrient content of the consumer explained a relatively minor fraction of the total variation in E_N, E_P, and E_NP (r² = 0.44, 0.11, and <0.1, respectively; all positive relationships, in contrast to expectations from EST; Figs. S1–S3) and less of the residual error from the mass-excretion models for E_N, E_P, and E_NP (Fig. 2). δ¹⁵N explained a substantial proportion of the variance for E_N and E_P (r² = 0.65 and 0.35, respectively, both positive relationships; Figs. 1 and 2), suggesting the potential importance of trophic position for the recycling of nutrients. However, these relationships are confounded by the positive δ¹⁵N-body mass relationship (r² = 0.32, P < 0.001), which is highlighted by the relatively weak relationships between δ¹⁵N and E_N (and in particular E_P) after accounting for body mass (Fig. 2), suggesting effects of trophic position were almost entirely accounted for by body mass.

Taxonomic identity has previously been reported to have important implications for metabolic scaling (22–24). We tested for the importance of taxonomic identity for metabolic scaling coefficients using mixed effects models (22, 23). Two types of models were run. First we fit nested models that included all levels of taxonomic identity (class, order, family, genus, and species), and then we tested models that included only a single taxonomic level (22, 23). To test for potential differences in metabolic scaling of excretion rates by vertebrates and invertebrates, we included a vertebrate/invertebrate level of classification (in addition to all other taxonomic levels) into the second class of models. In all cases, we tested for random effects on the slope, the intercept, or both the slope and the intercept (22) (Table 1). Models that included all levels of taxonomic classifications were within the top two models (as determined by AIC) for E_N (slope and intercept), E_P (intercept only), and E_NP (slope and intercept) (Table 1; note that class was excluded as a random effect in the model for E_N because it appeared to be nonidentifiable, resulting in an overparameterized model). Only in the case of E_NP was the top model a single-taxon model; in this case, the best model had random effects for slope and intercept at the species level.

For E_N, E_P, and E_NP, we then tested the degree to which the allometric scaling coefficient varied within the taxonomic group that produced the best model; i.e., family (slope and intercept), species (intercept only), and species (slope and intercept), respectively (Figs. 3 and 4) (25). Variation around the slope was highly constrained for E_N such that the 95% predictive intervals of the allometric coefficient for all families with the exception of one (Strombidae; queen conch, an invertebrate family) overlapped with the predicted 0.75 scaling coefficient (Fig. 3). In contrast, the best fit model for E_P did not include a random slope, indicating species-level differences in the mean excretion rate (Fig. 3), not the scaling of this rate, which was consistent (∼0.75) among species. Two of the 42 species had positive allometric coefficients with 95% predictive intervals that did not overlap with a zero slope for E_NP: the only two negative coefficients did not overlap with zero (Fig. 4).

The best models (lowest AIC) for E_N, E_P, and E_NP were used to further test the relative importance of body mass, body nutrient content, and δ¹⁵N. For the E_N and E_P models, strong support was found for body mass as the best predictor. It was the only parameter included in the top four competing models, and in the case of E_P, it had the largest influence on model performance (the model explained >25% less variation in the data when comparing models with body mass and without). We found support for body nutrient content as a predictor of E_N only. It was retained in the top two models, the 95% CIs did not overlap zero, and the model was not as good when this predictor was removed (ΔAIC decreased by 7.4; Table 2). Little support was found for effects of δ¹⁵N for all responses. In all models in which it was retained, the 95% CIs overlapped zero (Table 2).

Animals are critical components of nutrient cycles in most ecosystems (26–28). Using a unique dataset from a diverse marine community, our analysis showed that individual level nutrient recycling was highly predictable based on body mass alone, but models were improved substantially when accounting for variation due to taxonomic identity. Surprisingly, neither body nutrient content nor relative trophic position substantially improved model performance, in contrast to the predictions of EST. These results suggest that, among individuals in a diverse community, variation in excretion rates is best explained simply by a combination of allometry and the taxonomic identity of the organism. Our results support theory that seeks to integrate MTE with EST to improve the predictability of material flow through ecosystems (21) but highlights the need for further specificity regarding the unique effects certain species can have on ecological processes.

The slope of the E_P relationship with body mass was consistent with that predicted by allometric scaling laws (∼0.75), whereas E_N scaled isometrically (slope ∼ 1.0) without accounting for taxonomy. When accounting for variability in this relationship among taxonomic levels, both E_N and E_P were found to scale with a 3/4 coefficient because larger fish, which tended to belong to a few families, excreted higher levels of N than would be predicted by body size alone (Fig. 1). Previous research using bioenergetics models showed that freshwater temperate fishes deviate from the allometric scaling coefficient for E_P, but not E_N, and that accounting for P nutrient content of dietary resources (i.e., stoichiometric imbalances) adjusted the slope for the P models in agreement with the 3/4 scaling coefficient (19). Our findings, although we did not explicitly test for the importance of body nutrient-dietary imbalances, emphasize the importance of evolutionary factors, e.g., taxonomic identity, for differences among individual nutrient recycling rates in a diverse community.

Taxonomic identity proved to be a particularly important predictor in all models of nutrient excretion. E_P were most variable among species (as opposed to among other levels of taxonomy) in terms of absolute differences in the mean excretion rate (i.e., random intercept model). However, the scaling of E_P with body size was not dependent on taxonomic differences (i.e., the random slope effect was negligible) and was roughly consistent with the predicted 0.75 in all of the top models regardless of the taxonomic level of the random effects. In contrast, both mean rates (intercept) and the degree to which E_N scaled with body size (slope) varied among families such that only when accounting for these differences did the global scaling coefficient align with MTE predictions (i.e., 95% CIs of the E_N slope overlapped with 0.75). However, almost all families were still found to have 95% prediction intervals that overlapped with 0.75 for E_N (Fig. 3). Interestingly, there were little differences among invertebrates and vertebrates in these scaling coefficients, as may have been predicted given basic physiological differences among these groups (2, 29). Beyond a few exceptions (discussed below), the relative homogeneity in scaling coefficients found in our study contrasted with previous work demonstrating that, whereas scaling coefficients converged at roughly 3/4 scaling across broad taxonomic grouping values, they were highly heterogeneous among these groups when estimated independently (22, 23).

As hypothesized, E_NP was invariant with body mass (slope = 0), but only when accounting for taxonomy (species level). An alternative hypothesis, based on previous evidence that larger individuals (possibly due to ontogenetic shifts in diet) tend to have lower E_NP and feed on prey with lower E_NP (often because they are higher on the food chain) (19, 27), predicted a negative relationship with E_NP and body size. When accounting for differences in this relationship in both the mean (intercept) and slope among species, only 4 of the 43 species were found to have 95% CIs of slope that did not overlap with zero. Consistent with the alternative hypothesis, two of these species were fish species, Sparisoma chrysopterum (redtail parrotfish) and Lutjanus apodus (schoolmaster snapper), both of which have strong ontogenetic dietary shifts and had negative E_NP-body size relationships, demonstrating support for the role of ecological interactions for influencing E_NP. In contrast, the other two species were invertebrates with positive E_NP-body size relationships. This finding, however, opposes expectations based on the growth rate hypothesis (30), which states that at smaller sizes invertebrates with higher growth rate will have an increased demand for P (due to allocation in RNA) and thus would be expected to have a negative E_NP-body size relationship. Two aspects must be considered when evaluating these findings (1): these exceptions constitute only a limited subsample of species (4 of 43) (2); in contrast to E_N and E_P, E_NP models generally described the data poorly, highlighting the difficulty of predicting E_NP with these predictors used herein alone.

Surprisingly little support for the role of additional predictors of ecological interactions (e.g., body nutrient content and δ¹⁵N) for individual nutrient recycling was found in our data. Only with E_N did body nutrient content improve model performance, and even in this case, the effect of body nutrient content explained minimal variation in the data. The lack of importance of body nutrient content for E_P is also surprising for two reasons: (i) the extensive range of P nutrient content found within our dataset (0.19–8.7% dry mass) and (ii) the previous support for this relationship in the literature (16). Although we did not test explicitly for nutritional imbalances in our analysis, the fact that accounting for the trophic position of the consumer did not improve predictive power of individual nutrient excretion raises questions of the importance of dietary nutrient content, a key predictor of EST, as a useful factor in explaining nutrient excretion rates across a diverse group of animals.

Interestingly, the slope of the relationship between body nutrient content and excretion rate was positive for E_N, contrasting theoretical expectations that an animal with lower requirements for a given nutrient should release more of that nutrient back to its environment (2). To some extent this finding can be explained by the natural history and physiology of these animals. For example, E_N by fishes is directly associated with heart rate and thus metabolism (31). Atlantic silversides (Menidia menidia) provide an example of an animal that would be expected to have low excretion rates based on its relatively high demand for N (%N ∼ 11) and only modestly nutrient-rich diet (zooplankton). However, their family (Atherinopsidae) ranks among the highest in E_N, presumably due to the fact that these are small active pelagic fishes with high metabolic rates. This example highlights the importance of additional factors, namely metabolism, beyond nutritional imbalances for explaining the basic physiological process of nutrient excretion.

The simple, yet highly predictable, nature of allometric relationships and MTE provide a useful framework through which to explore ecological interactions at multiple scales of organization, ranging from potential effects of climate change on ecosystems (14) to scaling C flux from individual to global scales (11). Here, using an unprecedented dataset, we show the applicability of MTE to additionally explain individual-level nutrient recycling, a complex physiological process that encompasses ecological interactions commonly described by EST. Additional predictors of EST may improve predictability, but our analyses suggest that these improvements are negligible without substantially increasing model complexity. Our findings, however, underscore the importance of taxonomic identity in fundamentally regulating these processes, thus confirming a basic expectation of EST and challenging general theory with an additional layer of biological complexity.

Individual organisms were collected using hook and line, traps, cast nets, and dip nets between 2008 and 2011, within the same large embayment (the Bight of Old Robinson) on Abaco, The Bahamas. Habitat types within the embayment consisted of seagrass, mangrove, and patch reef complexes (32–34).

Excretion rates, for nitrogen-NH₄⁺ and phosphorus-soluble reactive phosphorus (SRP), were measured in situ following the methodologies of ref. 35, as modified by ref. 36. In each case, animals were placed into clean plastic zippered bags (1- to ∼37-L bag sizes) for either 30 (vertebrates) or 60 (invertebrates) min. Experiments were conducted in situ, and bags were placed in bins of ambient seawater. As such, temperature was highly regulated for all experiments (∼22–23 °C) and was similar to the temperature of the environment from which the subject animals were extracted. The volume of water per experiment varied according to the size of the individual (0.15–22 L), and net excretion rates were corrected for water volume to achieve rate of excretion in grams of nutrient per unit time. Values were control corrected through the use of multiple (typically n = 6) identical control incubation bags without fish. Each individual used for excretion experiments [n = 667 individual fish; size range: 0.14–2597 g (range, 2–107 cm), n = 233 invertebrates, size range: 0.04–487 g, 900 total individuals] was weighed for wet mass and measured to standard length. Fish were dissected to remove stomach contents, and invertebrates were incubated in seawater (for at least 6 h but typically 10–24 h) to allow for gut evacuation, and after identification, all animals were frozen for transport to Odum School of Ecology and processed for elemental content (see below). The University of Georgia’s (UGA) Institutional Animal Care and Use Committee approved protocols for the capture and handling of fish (AUP A2009-10003-0). Water samples (filtered with 0.45-μm Whatman nylon membrane filters) were immediately placed on ice and, within 10 h, analyzed for NH₄⁺ using the methodologies of ref. 37 or frozen for transport to the Odum School of Ecology (UGA) for SRP analysis using the ascorbic acid method and colorimetric analyses (American Public Health Association 1995) (38).

Individuals used for somatic nutrient content analyses (n = 494 vertebrates, n = 164 invertebrates, and n = 658 individuals total) were weighed for wet mass and measured to standard length. Samples were lyophilized to a consistent dry weight and then ground to a powder with a ball mill grinder. Larger individuals required blending to homogeneity before mill grinding. Ground samples were analyzed for %C and N content and δ¹⁵N with a CHN Carlo-Erba elemental analyzer (NA1500) CN Analyzer and for %P using dry oxidation-acid hydrolysis extraction followed by a colorometric analysis (Aplkem RF300). Elemental content was calculated on a dry weight basis. δ¹⁵N values were used as a proxy measure for the trophic position of an individual following the rationale that the higher the δ¹⁵N value, the higher on the food chain an individual is feeding (39–41). We did not correct for potential temporal variation in isotopic baselines for three reasons: (i) because the myriad basal resources within these systems makes correcting for specific isotopic baselines impossible (39, 41), (ii) because we are not attempting to use these data to calculate trophic position sensu stricto (42), and (iii) because previous research using the same species from the same study area found clear gradients of δ¹⁵N values that accurately depict relative trophic levels in this system with relatively little interannual variation in isotopic values among consumers (40). We acknowledge some potential error associated with this approach, but suggest the continuous nature of δ¹⁵N renders it preferable to simply assigning species to exact trophic levels based on presumed natural history or literature sources from studies conducted in other systems.

Simple linear regression was used to first explore the role of body mass in predicting excretion rates for all individuals for which we had excretion rate and body mass (n = 868, 471, and 461 for N, P, and N:P, respectively). To understand the degree to which body nutrient content and a species’ trophic position (δ¹⁵N) complement allometric relationships, we consolidated data to individuals for which all predictors of interest were available (n = 625, 371, and 365 for N, P, and N:P, respectively). Simple linear regression was used to test allometric relationships, and the model residuals were then regressed against the additional predictors independently. Model r² and significance (α < 0.05) were reported. This process was repeated for N, P, and N:P excretion.

To explore the importance of taxonomy variation in the relationship between body mass and nutrient recycling, we used mixed effects models testing for variation in slope or intercept or both across (i) all taxonomic levels (class, order, family, genus, or species) using a nested model structure and (ii) each taxonomic level independently; models were fitted by restricted maximum likelihood. Because we were interested in testing for species-level random effects, we used a subset of our data that included only species with three or more individuals (E_N = 56 species, 789 individuals, 14.8 individuals per species; E_P = 44 species, 430 individuals, 9.8 individuals per species; E_NP = 43 species, 420 individuals, 9.8 individuals per species). In the case of one model for E_N containing random slope and intercept for all nested taxonomic levels, the random effect for class appeared to be unidentifiable and thus was removed from the model.

The model with the lowest AICc values was then used to further explore how other covariates (body nutrient content, δ¹⁵N, i.e., fixed effects only) may explain additional variation in nutrient excretion rates using mixed effects models fitted by maximum likelihood using the dredge function in the R package MuMIn. Comparisons among competing models were made using information theory (AIC; note AICc was used in our analysis, but yielded the same ranking as AIC) (43, 44) and the r² statistic (conditional; which includes fixed and random effects, as well as for fixed effects alone). Mixed effects models were run using the lme4 package in R (25, 45). All other data analysis was conducted using R software (44).

We thank D. Haynes, F. DeRosa, R. Appaldo, K. Rennirt, J. Richard, and Friends of the Environment for logistic and field support. M. Vanni and P. McIntyre were helpful in discussions about excretion methods, and T. Maddox provided invaluable support for nutrient analyses. The manuscript was greatly improved by comments from three reviewers and the editor. Funding was provided by an Environmental Protection Agency Science to Achieve Research Fellowship, National Science Foundation (NSF) Doctoral Dissertation Improvement grant, and NSF Postdoctoral Research Fellowship in Biology (to J.E.A.) and NSF Biological Oceanography Grants 0746164 and 1405198 (to C.A.L.).