Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Testing for effects of growth rate on isotope trophic discrimination factors and evaluating the performance of Bayesian stable isotope mixing models experimentally: A moment of truth?

  • Kirsty E. B. Gurney ,

    Contributed equally to this work with: Kirsty E. B. Gurney, Robert G. Clark

    Roles Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Writing – original draft, Writing – review & editing

    kirsty.gurney@ec.gc.ca

    Affiliation Science and Technology Branch, Environment and Climate Change Canada, Saskatoon, Saskatchewan, Canada

    ⨯
  • Henry L. Classen,

    Roles Conceptualization, Investigation, Methodology, Project administration, Writing – review & editing

    Affiliation College of Agriculture and Bioresources, University of Saskatchewan, Saskatoon, Saskatchewan, Canada

    ⨯
  • Robert G. Clark

    Contributed equally to this work with: Kirsty E. B. Gurney, Robert G. Clark

    Roles Formal analysis, Investigation, Methodology, Project administration, Writing – original draft, Writing – review & editing

    Affiliation Science and Technology Branch, Environment and Climate Change Canada, Saskatoon, Saskatchewan, Canada

    ⨯

Abstract

Discerning assimilated diets of wild animals using stable isotopes is well established where potential dietary items in food webs are isotopically distinct. With the advent of mixing models, and Bayesian extensions of such models (Bayesian Stable Isotope Mixing Models, BSIMMs), statistical techniques available for these efforts have been rapidly increasing. The accuracy with which BSIMMs quantify diet, however, depends on several factors including uncertainty in tissue discrimination factors (TDFs; Δ) and identification of appropriate error structures. Whereas performance of BSIMMs has mostly been evaluated with simulations, here we test the efficacy of BSIMMs by raising domestic broiler chicks (Gallus gallus domesticus) on four isotopically distinct diets under controlled environmental conditions, ideal for evaluating factors that affect TDFs and testing how BSIMMs allocate individual birds to diets that vary in isotopic similarity. For both liver and feather tissues, δ13C and δ 15N values differed among dietary groups. Δ13C of liver, but not feather, was negatively related to the rate at which individuals gained body mass. For Δ15N, we identified effects of dietary group, sex, and tissue type, as well as an interaction between sex and tissue type, with females having higher liver Δ15N relative to males. For both tissues, BSIMMs allocated most chicks to correct dietary groups, especially for models using combined TDFs rather than diet-specific TDFs, and those applying a multiplicative error structure. These findings provide new information on how biological processes affect TDFs and confirm that adequately accounting for variability in consumer isotopes is necessary to optimize performance of BSIMMs. Moreover, results demonstrate experimentally that these models reliably characterize consumed diets when appropriately parameterized.

Citation: Gurney KEB, Classen HL, Clark RG (2024) Testing for effects of growth rate on isotope trophic discrimination factors and evaluating the performance of Bayesian stable isotope mixing models experimentally: A moment of truth? PLoS ONE 19(6): e0304495. https://doi.org/10.1371/journal.pone.0304495

Editor: Lee W. Cooper, University of Maryland Center for Environmental Science, UNITED STATES

Received: February 9, 2024; Accepted: May 13, 2024; Published: June 14, 2024

Copyright: © 2024 Gurney et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Our entire data set is freely accessible on the Zenodo general-purpose open repository (https://doi.org/10.5281/zenodo.10927544).

Funding: Funding was provided by Environment and Climate Change Canada. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

In applications related to trophic ecology, stable isotopes (i.e., the naturally occurring ratio of heavy to light isotopes, primarily for carbon, δ13C, and nitrogen, δ15N) are common chemical tracers [1, 2]. Stable isotope analyses are particularly well established for diet reconstruction and expand the scope of inference that can be obtained from other dietary methods, particularly in systems where direct observations are not possible and potential food items are isotopically distinct [3–5]. When metabolically inert tissues such as feathers, hair, or nails are sampled, such methods offer the additional advantage of being non-lethal and less invasive than sampling of other tissues, like blood or muscle [6, 7]. To estimate the proportional contributions of different food resources to consumer tissues, stable isotope analyses primarily use Bayesian stable isotope mixing models (BSIMMs) [8, 9].

Broadly, by using a variety of biological and chemical tracers, statistical models of mixtures (i.e., mixing models) quantify the relative contributions of different source materials to composite, end-member samples [10–12]. Given mathematical limitations of simple linear mixing models, however, BSIMMs, which are flexible and based on a likelihood structure, are most commonly used across the natural sciences to identify proportional inputs of sources to a mixture [13, 14]. Such models incorporate variability into model inputs, allow for inputs from multiple possible sources, and generate potential end-mixture solutions as true probability distributions [15]. Importantly, the Bayesian approach improves upon frequentist methods because processes known to affect the variability in the mixture data are accounted for in model error structures, so that probability distributions reflect associated uncertainties and allow inference about which solutions are more likely than others [16, 17].

The accuracy with which BSIMMs identify source contributions to consumer tissues is strongly influenced by model specifications, especially error assumptions and variation in consumer and resource isotope values related to a suite of factors that can often not be controlled or accounted for [16, 18]. In addition, stable isotope ratios of food items change when incorporated into the tissues of consumers–these changes are described by a trophic discrimination factor (TDF), a key input to BSIMMs that further affects model outputs [19, 20]. Most commonly derived from feeding trials of captive animals, TDFs are highly variable and generally depend on species, diet, and tissue type, although our understanding of how different factors affect TDFs, particularly for carbon, remains limited [21, 22].

To date, most evaluations of how accurately BSIMMs predict diet have been conducted through simulations [16], or controlled studies of captive [19, 23] or wild [5, 24] animals, and reach different conclusions about the relative importance of TDFs on bias, precision, and accuracy of posterior estimates, as well as the factors that affect TDFs. For example, some studies suggest that species-specific TDFs are of secondary importance to the use of informative priors (i.e., existing knowledge about dietary composition) [25], whereas others find that use of an appropriate TDF has a stronger effect on performance of BSIMMs [19, 23, 24]. Further, some data indicate that TDFs vary not only based on nutrient composition of consumed foods and consumer type (i.e., omnivorous, carnivorous), but also on interactions of these factors–and that the factors affecting TDFs vary for different isotopes [22, 26]. An experimental study with tilapia (Oreochromis mossambicus) showed the importance of diet type and tissue type in determining TDFs, with simulated data providing additional evidence that the use of experimentally-derived TDFs, compared to those from the literature, improved accuracy of dietary estimates [19]. This is similar to Stephens et al.’s (2022) findings that using existing data and phylogenetic relatedness (i.e., Stable Isotope Discrimination Estimation in R; SIDER) [27] to derive TDFs can lead to misleading conclusions about diet, but is in contrast to other studies that suggest such methods improve BSIMMs performance, or that TDFs affect model results less than other factors [28, 29]. Reliable information about factors affecting TDFs is needed to resolve these apparently contrasting results and to more clearly understand the true, best-case accuracy of BSIMMs outputs in trophic ecology, as well as in other applications [30, 31].

To identify the most critical factors to consider in applying BSIMMs and thus improve inference related estimates of animal diet, we aim to test for factors that affect TDFs and to evaluate how different sources of variation in isotope values drive uncertainty in BSIMM outputs. We test the performance of BSIMMs experimentally, by feeding domestic broiler chicks (Gallus gallus domesticus) isotopically distinct diets under controlled environmental conditions, asking three key research questions: (i) how does variation among individual growth rates affect TDFs for carbon (Δ13C) and nitrogen (Δ15N) in an omnivorous bird; (ii) how do variable estimates of TDFs and different estimates of uncertainty in BSIMMs influence model performance; and (iii) what is the best-case accuracy of BSIMM outputs when environmental and resource-related properties are primarily controlled for? Although broiler chickens have undergone extensive artificial selection for growth in captivity, variation in growth rate remains [32]. To our knowledge, although the effects of individual growth rates on TDFs have been demonstrated in fish species, similar evaluations have not been conducted in endotherms with higher basal metabolic rates, like birds [33, 34], but see [35]. Because variation in tissue isotopic values related to environmental factors and dietary composition (isotopic values and source of macronutrients) is minimized in our captive setting, the broiler chicks are an ideal system to test how variation in growth and other organism-dependent factors, such as tissue type, affect TDFs and how accurately BSIMMs allocate individual birds to assigned dietary groups when model inputs vary.

We anticipated that both individual growth rate and tissue type would be important predictors of Δ13C values, with Δ13C decreasing as growth rate increases and routing of dietary constituents (especially amino acids) to tissues is prioritized over the production of energy (the “anabolism hypothesis” [36, 37]. We also expected the growth rate effect to be tissue dependent, being more evident in liver, a tissue that is vital for bodily function and has more rapid turnover, relative to feathers, which are metabolically inert [38]. For Δ15N, we expected that tissue type would have a similar effect as Δ13C (liver having lower values compared to feathers), but that effects of metabolic routing would be reduced, resulting in a lesser effect of growth on Δ15N, relative to Δ13C [39]. Model outputs from BSIMMs should provide the most accurate assessments of assimilated diets when observed variation in TDFs (i.e., diet-specific Δ13C and Δ15N values) are applied [19, 40]. In terms of uncertainty estimates, we presume that birds in our study could not preferentially consume dietary components, so that process error (i.e., individual specialization) will have a minimal influence on isotopic variation of tissues. We therefore predict that models specifying a residual error-only formulation will be most supported by the data [15, 16].

Methods

Diet composition and experimental design

Using commercially available feed ingredients, we designed two isotopically distinct, nutritionally balanced diets to meet or exceed nutrient specifications for as-hatched Ross-308 broiler chickens [41]. Detailed summaries of dietary constituents are available as Supporting Information (S1 Table). The two primary diet formulations (diet 1 and diet 2) were manufactured in mash form, with two secondary formulations based on mixtures: diet 3 (67% diet 1; 33% diet 2) and diet 4 (33% diet 1; 67% diet 2). All four diets were then pelleted (< 80°C in a conditioner) and crumbled, with each dietary treatment designated as an experimental group. Day of hatch, mixed-sex Ross-308 broiler chicks were purchased from a commercial hatchery (Sofina Foods Inc., Wynyard, Saskatchewan, Canada), weighed, individually banded, and randomly assigned to one of 24 cages (six replicate cages per group). Each experimental unit (i.e., cage) contained four chicks, and to replace any birds that died during the first week of the trial, we also maintained extra chicks for each group (n = 10 per group) in separate cages.

Housing procedures followed standard production guidelines for caged broiler chickens, and we obtained written consent and approval for all protocols from the Animal Care Committee (Animal Research Ethics Board), University of Saskatchewan, on behalf of the Canadian Council on Animal Care (Protocol 19940248, Poultry Centre File Number CTR 1702). We placed individual birds in the top compartments (51 cm W x 51 cm L x 46 cm H; 650 cm2 per bird) of two-tier battery cages and gave birds ad libitum access to feed (front-mounted feed trough) and water (nipple, automatic, adjustable height). Extra feed was provided in supplemental trays for the first four days, and water was provided in ice cube trays for the first week. Room temperature was gradually reduced (3°C weekly) from 34°C to 22°C. The light:dark cycle began at 23:1 for day-old chicks and was gradually adjusted to 19:5 by day six, where it remained for the remainder of the trial (total of 29 days). The light period included 15-minute dawn and dusk periods, and light intensity at the feeder was 20 lux. To eliminate the possibility of dietary contamination across groups, chicks were placed only in the top tier of cages, and an empty cage was left between experimental replicates.

To index feed intake and growth rates, changes in food mass were estimated over a 24-hour period throughout the trial and on days one, 14, and 29, remaining chicks (n = 93; one bird censored due to a missing mass value at day 29) were weighed on a top-loading balance (± 1 g). Dietary samples from each group were collected throughout the trial (weekly) to test isotope values of feed. After 29 days, we euthanized all birds via cervical dislocation. We immediately excised the entire liver, collected flank-breast feathers, and sexed birds via visual confirmation of testes or ovaries. Liver and feather samples were stored in sterile polyethylene bags at -80°C (liver) or in paper envelopes at -18°C (feathers).

Stable isotope analyses

Feather samples were rinsed with a 2:1 chloroform-methanol solution, allowed to air dry for 24 hours, and finely cut into pieces (calamus removed); liver samples were freeze-dried and ground to a homogenous powder; and diet samples were finely powdered in a homogenizer. To remove lipids, liver and diet samples were rinsed with a 2:1 chloroform-methanol solution and dried to constant mass. All tissue and diet samples were packed (approximately 1 mg) into tin capsules and combusted in a Costech (Milan, Italy) ECS4010 elemental analyzer interfaced with a Delta V continuous-flow isotope-ratio mass spectrometer at the Department of Soil Science, University of Saskatchewan. By convention, isotope values are reported in delta notation as δ13C and δ15N according to: where X = 13C or 15N, R = 13C:12C or 15N:14N, and international standards (std) are Vienna PeeDee Belemnite (VPDB) for 13C and atmospheric N2 (AIR) for 15N, and where measured delta values were converted to international isotope reference scales using two-point linear normalization [42].

Two secondary isotopic reference materials (egg albumen and lyophilized bowhead whale baleen) that were expected to span the range of experimental values were selected by the analytical lab, as standards to allow estimation of measurement precision [43]. Standards–provided by the lab (in the same mass range as test samples)–were calibrated by running them against USGS 40 and USGS 41a L-glutamic acids, which have internationally accepted δ13C and δ15N values [44, 45]. For further quality assurance / quality control, standards were processed before and between every five tissue samples within each analytical run (for albumen) and at the start, midpoint, and end of each run (for baleen). We used replicate standard measurements within a run to estimate analytical precision: mean δ13C (± 1 standard deviation, SD) within runs = –22.4 ‰ ± 0.0 ‰, and mean δ15N (± 1 SD) within runs = 6.8 ‰ ± 0.1 ‰ for egg albumen; mean δ13C (± 1 SD) within runs = –18.5‰ ± 0.1%, and mean δ15N (± 1 SD) within runs = 14.8 ‰ ± 0.2%, for bowhead whale baleen. Measurement error estimates (± 1 SD) were ± 0.1 ‰ for δ13C and ± 0.3 ‰ for δ15N, based on measurements of laboratory standards across multiple runs [43].

Statistical approach

To evaluate the relative influence of organism-related factors (growth rate and tissue type) on TDFs, analyses were implemented in SAS® Version 9.4 (SAS® Inc., Cary, NC). We calculated Δ13C as δ13CTissue− δ13CDiet and Δ15N as δ15NTissue− δ15NDiet for both liver and feathers, estimated growth parameters for each individual, based on three-parameter Gompertz growth curves [46, 47], and used general linear mixed models for our analyses. These models allowed us to account for dependence among birds at the cage level (random effects), while evaluating the specific and separate influences of our predictor variables of interest [48]. For each TDF, model covariance structure was determined based on data structure and convergence criteria [49]. Marginal log-likelihood values for each model were computed using restricted maximum likelihood methods, and a Kenward-Roger correction was applied in computing denominator degrees of freedom (PROC MIXED) [50]. After developing an a priori set of candidate models (one each for Δ13C and Δ15N), we compared models based on an information-theoretic approach, with models ranked according to 2nd-order Akaike’s Information Criterion corrected for sample size (AICc) and relative likelihoods for each model estimated by Akaike weights (wi) [51].

Key predictor variables (fixed effects) in our models included individual growth rate based on mass (MassB), tissue type, and their interaction (MassB*Tissue). To account for additional factors known to affect TDFs, we also included diet type (Group) and sex in our model sets [40, 52] and evaluated potential interactions between diet and tissue (Group*Tissue) and sex and tissue (Sex*Tissue). Inference concerning fixed effects was based on precision (85% confidence interval, CI) of the regression coefficient (β), as this interval is more compatible with the information-theoretic approach. When wide confidence intervals indicated imprecise slope estimates, we considered respective covariates as uninformative [53].

To compare the effects of different TDFs and error structures on stable isotope dietary inference and test how accurately BSIMMs estimated known proportions of the diets that were consumed, we used hierarchical BSIMMs, implemented in the open-source R package, MixSIAR Version 3.1.10 [54]. Within MixSIAR, each model was first run under test settings to confirm models were correctly specified–cage as a random effect, nested within the fixed effect of group (Cage(Group)). For final analyses, Gibbs sampling parameters were set to 3 parallel chains of 300,000 vectors (burn-in = 200,000, thinning interval = 100). To determine whether mixing models converged, we examined trace plots for relatively constant mean and variance, as well as for small fluctuations in the estimated parameters–results from the Gelman-Rubin diagnostic test were also considered [54]. TDFs were specified as diet-specific (mean ± standard deviation)(Diet 1: Δ13C = 0.7 ± 0.2, Δ15N = 0.7 ± 0.2; Diet 2: Δ13C = 2.3 ± 0.2, Δ15N = 1.2 ± 0.4) or combined (Δ13C = 1.4 ± 0.8, Δ15N = 0.9 ± 0.3). Different specifications of BSIMMs that converged were ranked using the Deviance Information Criterion, DIC [55], and to estimate the relative accuracy of model outputs across groups, we compared the actual diet proportions to the model-based 95% credible intervals–when intervals included the known proportion, we considered the output accurate.

Results

Diet composition and chicken growth

Feed isotopic values (δ13C, δ15N) were significantly different among the experimental diet groups (MANOVA, Wilks’ λ = 0.007, F6, 54 = 97.06, p < 0.0001); as expected, our formulations resulted in a high δ13C / δ15N diet (Group 1), a low δ13C / δ15N diet (Group 2) and 2 intermediates (Groups 3 and 4) (Fig 1).

thumbnail
Download:
Fig 1. Dietary δ13C (upper panel) and δ15N (lower panel) by experimental dietary group.

Isotope values differed among diets fed to growing Ross-308 strain chicks under controlled environmental conditions. Shown are the median (horizontal line within shaded boxes), lower and upper quantiles (lower and upper edge of box, respectively) and range (vertical line) for each dietary group (n = 8 samples per group).

https://doi.org/10.1371/journal.pone.0304495.g001

On day one, chick body mass averaged 47.1 g (± 1.0 g standard error, SE), and no effects of sex or dietary grouping were detected. Group 3 had three more females than males and Group 1 had six more males than females, but sex ratio did not differ among groups (G-test, G3 = 1.91, p = 0.59); overall, there were 46 females and 48 males (Table 1). By day 29, after accounting for cage effects, additional effects of sex (p < 0.0001) and diet group (p = 0.01) on body mass were detected, with no interaction. Across all birds, coefficients for daily rate of mass gain (MassB) ranged from 0.0403 (a bird in Group 3) to 0.0907 (a bird in Group 3), solving the Gompertz equations for days 15 and 16, estimated values were equivalent to 34 to 83 g of body mass gained per day. Although we detected no effects of sex or diet group (or their interaction) on MassB, by the end of the diet trial, female chickens (least squares mean = 1621 g ± 28 SE) averaged ~144 g less than males (1765 g ± 28 SE). Adjusting for sex effects, least square mean body masses (SE) of chickens were 1732 g (46), 1737 g (46), 1539 g (47), 1736 g (46) for diet Groups 1–4, respectively.

thumbnail
Download:
Table 1. Descriptors of growth for Ross-308 broiler chicks.

https://doi.org/10.1371/journal.pone.0304495.t001

Stable isotope values & factors affecting TDFs

For both liver and feather tissue, δ13C and δ 15N values differed among dietary groups, reflecting patterns in feed isotope values; compared to feather, δ13C was generally more negative for liver, with the opposite pattern for δ15N (Fig 2). Similarly, as expected, Δ13C was affected by dietary group, tissue type, and their interaction, but models including an effect of sex on Δ13C were not supported. Consistent with our predictions under the anabolism hypothesis, we detected a negative effect of MassB on Δ13C, for liver tissue only (βMassB*Feather = 13.7, 85% CI = 7.0 to 20.5), with some minor differences among groups (Table 2, Fig 3). For Δ15N, the most-supported model included effects of dietary group, sex, and tissue type, as well as an interaction between sex and tissue type (wi = 0.63; βFemale*Feather = -1.30, 85% CI = -0.19 to -0.06), although a model without the interaction also received moderate support (wi = 0.36, Table 3). The sex effect (females having higher Δ15N relative to males) was observed only for liver, and contrary to our expectation, in general, Δ15N values were greater for liver than feather (Fig 4).

thumbnail
Download:
Fig 2. Tissue δ13C (upper panel) and δ15N (lower panel) for liver and feather.

Values varied across dietary groups. Shown are the median (horizontal line within boxes), lower and upper quantiles (lower and upper edge of box, respectively) and range (vertical line) for each dietary group, with values above or below the interquartile range (upper quartile–lower quartile) indicated by closed circles.

https://doi.org/10.1371/journal.pone.0304495.g002

thumbnail
Download:
Fig 3. Δ13C in captive broiler chicks was affected by individual growth rate.

The effect varied across dietary groups and between tissue types. All plots show model-based estimates (± 85% confidence intervals) as indicated in the legend. Raw data (circle symbols) are plotted for visual comparison.

https://doi.org/10.1371/journal.pone.0304495.g003

thumbnail
Download:
Fig 4. Dietary group, sex, and tissue type influenced Δ15N in captive broiler chicks.

For liver, Δ15N was higher in females (triangles), but for feathers, values were similar between females and males (circles). All plots show model-based estimates (± 1 standard error). Raw data (grey boxplots) are plotted for visual comparison.

https://doi.org/10.1371/journal.pone.0304495.g004

thumbnail
Download:
Table 2. Model ranking for Δ13C.

https://doi.org/10.1371/journal.pone.0304495.t002

thumbnail
Download:
Table 3. Model ranking for Δ15N.

https://doi.org/10.1371/journal.pone.0304495.t003

Influences of TDFs & error structure on performance of BSIMMs

For liver tissue, our top-ranked BSIMM (based on DIC) showed that combining the variance in TDFs across diet types and applying a multiplicative error structure would improve model accuracy compared to the residual error-only model (ΔDIC = 6.6; Table 4). Models with diet-specific TDF values for Δ13C and Δ15N either did not converge (multiplicative error) or were poorly supported by the data (residual error). Results for feather models were similar in that increasing variance estimates for TDF values–by combining diet-specific values–and using a multiplicative error structure resulted in the most-supported model, although there was some indication that a residual error-only model (with combined TDF variance) performed well (ΔDIC = 2.5; Table 5). Outputs from our top-ranked BSIMMs were generally very similar to known dietary contributions, with some variation in accuracy, depending on group and tissue type. For liver, estimated dietary proportions based on 95% credible intervals were accurate (i.e., included known proportions) for dietary groups 3 and 4, whereas for feather–best model outputs were accurate for group 4 only (Tables 4 and 5). Elsewise, credible intervals generally were within 10% of known proportions (Fig 5).

thumbnail
Download:
Fig 5. Estimated contributions (Bayesian credible intervals for scaled posterior densities) of each formula (= ‘form’, as legend), for both liver and feather.

Estimates based on top-ranked Bayesian stable isotope mixing models were similar to known dietary contributions (indicated by vertical dotted lines), with some variation in accuracy related to dietary group and tissue type.

https://doi.org/10.1371/journal.pone.0304495.g005

thumbnail
Download:
Table 4. Bayesian stable isotope mixing model outputs for liver tissue.

https://doi.org/10.1371/journal.pone.0304495.t004

thumbnail
Download:
Table 5. Bayesian stable isotope mixing model outputs for feather tissue.

https://doi.org/10.1371/journal.pone.0304495.t005

Discussion

Using captive, domestic birds being fed specifically formulated, isotopically distinct diets, we have directly measured how several factors, including diet, tissue type, sex, and individual growth rates, affect trophic discrimination factors and, further, how variation in the specifications of Bayesian stable isotope mixing models affect the accuracy of their outputs. Although TDFs varied considerably across dietary groups and tissue types (as expected), our work highlights two other factors that influence TDFs in birds–sex and individual growth rate. Importantly, we also show that adequate estimates of uncertainty in BSIMMs improve performance of these models, and that when variation in the isotope values of animals due to environment and food resources is controlled (i.e., secondary mechanistic factors; 18), BSIMM-derived estimates of consumed diet can closely reflect known proportions.

Factors influencing trophic discrimination factors

As expected, our analyses indicated a negative relationship between individual growth rate and Δ13C –for liver tissue only. Although the relationship between growth rates and TDFs has rarely been investigated directly, this finding is generally aligned with other studies of vertebrates that suggest an important role for metabolic processes in determining trophic discrimination factors [56–58]. For example, faster growing juvenile fish tend to have higher metabolic rates, relative to slower-growing conspecifics, which in turn is linked to decreases in Δ13C [36, 59]. It is unclear if adaptations to endothermy would result in similar relationships for birds. Intraspecific variation in percentage mass change did not affect turnover rates of 13C in dunlin (Calidris alpina), and Δ13C did not differ predictably between slower and faster-growing captive American crows (Corvus brachyrhynchos) [38, 60]. Yet Δ13C in domestic chickens ranges from approximately -4‰ to 2‰, and observed variation among captive-raised Caipirinha broilers suggests a growth effect in this species [40, 61], as do the lower TDFs for younger age classes that have been observed in other species [62]. Our results thus add to a growing body of evidence that differences in physiology among individuals have an important influence on TDFs, which we speculate is linked to increased routing of dietary protein to body tissues during rapid growth (i.e., anabolism). That we observed an inverse relationship between growth rate and Δ13C only in liver, a highly metabolically active tissue (relative to feathers), is consistent with this idea [63, 64]. However, differences in the specific amino acid composition of livers and feather (i.e., keratin) may be an additional factor affecting differences in fractionation values between tissue types and the interaction with growth rate [65, 66].

For Δ15N, our finding that diet and tissue type are reliable predictors of fractionation values is consistent with expectations, which were based on other studies that consistently highlight the importance of these factors, especially dietary composition [21, 22, 67]. The sex dependence of the tissue effect on Δ15N, however, was not anticipated. In liver only, fractionation values for nitrogen were marginally higher for female chicks than for males. Although few studies have specifically evaluated sex effects on TDFs, higher Δ15N in females have previously been noted in some mammalian species [68, 69]. Although sex-based variation in growth is a proposed mechanistic explanation in both cases, this hypothesis is unsupported by our study, where there was no evidence of a growth rate effect on Δ15N. Instead, physiological differences related to oogenesis may play an important role. If female broiler chicks increase routing of nitrogen to liver for production of eggs–even prior to sexual maturity–enrichment of 15N would be expected [70].

We did not design our feeding trial to specifically evaluate the influence of dietary composition (isotopic values and sources of macronutrients) on trophic discrimination factors for carbon or nitrogen, but our findings that dietary group affected both Δ13C and Δ15N were not unexpected. Accounting for other factors, in our study, Δ13C and Δ15N were generally highest for dietary groups 2 and 4 (Figs 3 and 4), which had reduced δ13C values (Fig 1), a greater percentage of C3-based carbon (i.e., wheat and soy), and less fish meal, relative to Groups 1 and 3. Although evidence of an influence of δ13C values on Δ13C in birds is lacking [71], origin of carbon sources (i.e., C3-based, C4-based, marine-based, or a mixture) and dietary protein content can explain substantial variation in Δ13C for avian species [39, 40]. Given similar estimated protein content across the different diets in our study, our data support a carbon source hypothesis, but we note that differences in digestibility of different protein sources may also be important, not only for Δ13C, but also for Δ15N [67, 72]. In particular, for Δ15N, our results are consistent with an important influence of protein quality; our diet groups with higher levels of easily-digestible fish protein also had lower Δ15N values. Additional studies–especially those using specifically formulated food sources and compound-specific isotope analysis of amino acids–will provide improved insight about how diet composition affects Δ13C [73, 74].

Performance of Bayesian stable isotope mixing models

Varying TDF estimates and error structures in our BSIMM formulations affected performance of the models (based on DIC and estimated dietary contributions), with variance in TDFs having the greater effect, both for liver and feather tissue. Although we anticipated that model performance would be improved by specifying diet-specific Δ13C and Δ15N values–we observed the opposite. When we included estimates of TDFs that were combined across dietary sources, model performance and accuracy of diet assignments were improved, relative to models with diet-specific estimates. Although this is in apparent contrast to the idea that diet-specific estimates of TDFs are more valuable for isotopic dietary assignments than literature-based values [5, 19], our findings align with work suggesting that larger uncertainties in TDFs can improve fit and accuracy of BSIMMs [24, 75].

Regarding the accuracy of BSIMM outputs under best-case conditions, we note that under both TDF specifications, relative dietary contributions estimated by the Bayesian credible intervals were similar to known proportions (Tables 4 and 5), being within about 10% of the known diet (or better) in all cases we looked at. This is encouraging for use of BSIMMs to identify diets in field studies, albeit not perfect. Importantly, our results, particularly for dietary groups 3 and 4, suggest that dominant dietary components would be clearly identified by BSIMM outputs, even though accuracy of estimates might vary. Ecologists can therefore expect predicted diets based on these models will give solid insights into the dominant assimilated dietary components, at least in food webs with isotopically distinct dietary options. We recommend, however, that results not be considered as indicative of strictly true diets, as revealed in our “moment of truth”.

Because our results indicate that accuracy, more than inference, is affected by TDF variance estimates when isotopic differences between source proportions are high, we recommend that future studies should consistently consider overall sensitivities of BSIMM outputs to TDF variance, particularly because assuming single, fixed TDFs can artificially reduce estimates of sampling error and the importance of ecological processes such as individual specialization [16]. That models with process-related error terms (for both tissue types) were more supported by our data than residual error only models is consistent with the idea that variability in consumer isotope values due to chance–as well as due to distinct dietary choices of individual animals–are important elements of mixing models. This was an unexpected finding for our captive setting, where we expected that the relatively homogenous nature of the pelleted feed would prevent individuals from preferentially consuming one formula (i.e., source) over another. We also assumed that by using the same lipid extraction methods for all diet and liver samples (feathers do not contain lipids), possible effects of lipid extraction on BSIMM outputs would be negligible [76, 77]; indeed, model-predicted dietary assignments for lipid-rich liver and lipid-free feather tissues were similar (Fig 5).

Although the importance of process-related error in our models implies that source selectivity among individual consumers may be underestimated in many natural settings, we caution that factors leading to individual specialization are likely dependent on variation in an individual’s state, as well as competitive pressure for resources [78, 79]. We suggest that additional studies assessing intraspecific dietary variation are needed to fully understand the extent of this phenomenon and the conditions under which it is expected to arise. We also further echo the suggestions of others that understanding the underlying ecological assumptions of different BSIMM formulations, particularly error structures, is key to improving inference in dietary studies [13, 80, 81].

Supporting information

S1 Table. Commercially available feed ingredients were used to formulate experimental diets.

Percentages of each constituent are shown.

https://doi.org/10.1371/journal.pone.0304495.s001

(PDF)

Acknowledgments

We thank Jamille McLeod, Robert Gonda, Steve Leach, Lisha Berzins, and Katelyn Luff for assistance with sample collection and processing, Stuart Bearhop for encouragement, anonymous reviewers for their constructive comments, and the University of Saskatchewan’s Poultry Centre and Department of Soil Science for logistics support.

References

  1. 1. Pinzone M, Damseaux F, Michel LN, Das K. Stable isotope ratios of carbon, nitrogen and sulphur and mercury concentrations as descriptors of trophic ecology and contamination sources of Mediterranean whales. Chemosphere. 2019;237:124448. pmid:31398606
  2. 2. Bonin M, Dussault C, Taillon J, Lecomte N, Côté SD. Combining stable isotopes, morphological, and molecular analyses to reconstruct the diet of free-ranging consumers. Ecology and Evolution. 2020;10(13):6664–76. pmid:32724540
  3. 3. Gurney K, Clark R, Slattery S, Ross L. Connecting the trophic dots: responses of an aquatic bird species to variable abundance of macroinvertebrates in northern boreal wetlands. Hydrobiologia. 2017;785(1):1–17.
  4. 4. Potapov AM, Tiunov AV, Scheu S. Uncovering trophic positions and food resources of soil animals using bulk natural stable isotope composition. Biological Reviews. 2019;94(1):37–59. pmid:29920907
  5. 5. Johnson DL, Henderson MT, Anderson DL, Booms TL, Williams CT. Bayesian stable isotope mixing models effectively characterize the diet of an Arctic raptor. Journal of Animal Ecology. 2020;89(12):2972–85. pmid:33020919
  6. 6. Cherel Y, Jaquemet S, Maglio A, Jaeger A. Differences in δ 13 C and δ 15 N values between feathers and blood of seabird chicks: implications for non-invasive isotopic investigations. Marine Biology. 2014;161:229–37.
  7. 7. Steenweg RJ, Crossin GT, Kyser TK, Merkel FR, Gilchrist HG, Hennin HL, et al. Stable isotopes can be used to infer the overwintering locations of prebreeding marine birds in the Canadian Arctic. Ecology and Evolution. 2017;7(21):8742–52. pmid:29177032
  8. 8. Graves SD, Liber K, Palace V, Hecker M, Doig LE, Janz DM. Trophic dynamics of selenium in a boreal lake food web. Environmental Pollution. 2021;280:116956. pmid:33799129
  9. 9. Phillips DL, Inger R, Bearhop S, Jackson AL, Moore JW, Parnell AC, et al. Best practices for use of stable isotope mixing models in food-web studies. Canadian Journal of Zoology. 2014;92(10):823–35.
  10. 10. Fry B. Stable isotope ecology: Springer; 2006. 308 p.
  11. 11. Wefing AM, Casacuberta N, Christl M, Gruber N, Smith JN. Circulation timescales of Atlantic Water in the Arctic Ocean determined from anthropogenic radionuclides. Ocean Sci. 2021;17(1):111–29.
  12. 12. Smith HG, Karam DS, Lennard AT. Evaluating tracer selection for catchment sediment fingerprinting. Journal of Soils and Sediments. 2018;18(9):3005–19.
  13. 13. Stock BC, Jackson AL, Ward EJ, Parnell AC, Phillips DL, Semmens BX. Analyzing mixing systems using a new generation of Bayesian tracer mixing models. PeerJ. 2018;6(e5096):e5096. pmid:29942712
  14. 14. Liu J, Han G. Tracing Riverine particulate black carbon sources in Xijiang river Basin: insight from stable isotopic composition and Bayesian mixing model. Water Research. 2021;194:116932. pmid:33640754
  15. 15. Parnell AC, Inger R, Bearhop S, Jackson AL. Source partitioning using stable isotopes: coping with too much variation. PLoS One. 2010;5(3):Article. pmid:20300637
  16. 16. Stock BC, Semmens BX. Unifying error structures in commonly used biotracer mixing models. Ecology. 2016;97(10):2562–9. pmid:27859126
  17. 17. Cooper RJ, Krueger T, Hiscock KM, Rawlins BG. Sensitivity of fluvial sediment source apportionment to mixing model assumptions: AB ayesian model comparison. Water Resources Research. 2014;50(11):9031–47. pmid:26612962
  18. 18. Boecklen WJ, Yarnes CT, Cook BA, James AC. On the use of stable isotopes in trophic ecology. Annual review of ecology, evolution, and systematics. 2011;42:411–40.
  19. 19. Kadye WT, Redelinghuys S, Parnell AC, Booth AJ. Exploring source differences on diet-tissue discrimination factors in the analysis of stable isotope mixing models. Scientific reports. 2020;10(1):15816. pmid:32978550
  20. 20. Bond AL, Diamond AW. Recent Bayesian stable-isotope mixing models are highly sensitive to variation in discrimination factors. Ecological Applications. 2011;21(4):1017–23. pmid:21774408
  21. 21. Bastos RF, Corrêa F, Winemiller KO, Garcia AM. Are you what you eat? Effects of trophic discrimination factors on estimates of food assimilation and trophic position with a new estimation method. Ecological Indicators. 2017;75:234–41.
  22. 22. Stephens RB, Ouimette AP, Hobbie EA, Rowe RJ. Reevaluating trophic discrimination factors (Δδ13C and Δδ15N) for diet reconstruction. Ecological Monographs. 2022;92(3):e1525.
  23. 23. Derbridge JJ, Merkle JA, Bucci ME, Callahan P, Koprowski JL, Polfus JL, et al. Experimentally derived δ13C and δ15N discrimination factors for gray wolves and the impact of prior information in Bayesian mixing models. PloS one. 2015;10(3):e0119940.
  24. 24. Swan GJ, Bearhop S, Redpath SM, Silk MJ, Goodwin CE, Inger R, et al. Evaluating Bayesian stable isotope mixing models of wild animal diet and the effects of trophic discrimination factors and informative priors. Methods in Ecology and Evolution. 2020;11(1):139–49.
  25. 25. Moore JW, Semmens BX. Incorporating uncertainty and prior information into stable isotope mixing models. Ecology letters. 2008;11(5):470–80. pmid:18294213
  26. 26. Hopkins JBI, Ferguson JM. Estimating the diets of animals using stable isotopes and a comprehensive Bayesian mixing model. PLoS One. 2012;7(1):e28478. pmid:22235246
  27. 27. Healy K, Guillerme T, Kelly SB, Inger R, Bearhop S, Jackson AL. SIDER: An R package for predicting trophic discrimination factors of consumers based on their ecology and phylogenetic relatedness. Ecography. 2018;41(8):1393–400.
  28. 28. Morgenthaler A, Millones A, Gandini P, Frere E. Which trophic discrimination factors fit the best? A combined dietary study of a coastal seabird. Journal of Ornithology. 2021;162:179–90.
  29. 29. Sugden S, Murray M, Edwards MA, St. Clair CC. Inter-population differences in coyote diet and niche width along an urban–suburban–rural gradient. Journal of Urban Ecology. 2021;7(1).
  30. 30. Fathabadi A, Jansen JD. Quantifying uncertainty of sediment fingerprinting mixing models using frequentist and Bayesian methods: A case study from the Iranian loess Plateau. Catena. 2022;217:106474.
  31. 31. Thieltges DW, Goedknegt MA, O’Dwyer K, Senior AM, Kamiya T. Parasites and stable isotopes: a comparative analysis of isotopic discrimination in parasitic trophic interactions. Oikos. 2019;128(9):1329–39.
  32. 32. Zuidhof M, Schneider B, Carney V, Korver D, Robinson F. Growth, efficiency, and yield of commercial broilers from 1957, 1978, and 2005. Poultry science. 2014;93(12):2970–82. pmid:25260522
  33. 33. Reich KJ, Bjorndal KA, Martínez del Rio C. Effects of growth and tissue type on the kinetics of 13C and 15N incorporation in a rapidly growing ectotherm. Oecologia. 2008;155(4):651–63. pmid:18188602
  34. 34. Buchheister A, Latour RJ. Turnover and fractionation of carbon and nitrogen stable isotopes in tissues of a migratory coastal predator, summer flounder (Paralichthys dentatus). Canadian Journal of Fisheries and Aquatic Sciences. 2010;67(3):445–61.
  35. 35. Campos D, Macari M, Fernadez-Alarcon M, Nogueira W, de Souza F, Hada F, et al. Performance and carbon turnover in fast-and slow-growing broilers submitted to cyclic heat stress and fed on high-protein diets. British poultry science. 2016;57(1):84–92. pmid:26634283
  36. 36. Scharnweber K, Andersson ML, Chaguaceda F, Eklöv P. Intraspecific differences in metabolic rates shape carbon stable isotope trophic discrimination factors of muscle tissue in the common teleost Eurasian perch (Perca fluviatilis). Ecology and Evolution. 2021;11(14):9804–14. pmid:34306663
  37. 37. Gorokhova E. Individual growth as a non-dietary determinant of the isotopic niche metrics. Methods in Ecology and Evolution. 2018;9(2):269–77.
  38. 38. Ogden LJE, Hobson KA, Lank DB. Blood isotopic (δ13C and δ15N) turnover and diet-tissue fractionation factors in captive dunlin (Calidris alpina pacifica). The Auk. 2004;121(1):170–7.
  39. 39. Oortwijn T, de Monte LG, Varley DP, van der Meer MT, van Gils JA. Tissue-and diet-dependent stable carbon and nitrogen isotope discrimination: a calibration study in a captive shorebird species. Journal of Avian Biology. 2023:e03094.
  40. 40. Stephens RB, Shipley ON, Moll RJ. Meta-analysis and critical review of trophic discrimination factors (Δ13C and Δ15N): Importance of tissue, trophic level and diet source. Functional Ecology. 2023;n/a(n/a).
  41. 41. Aviagen. Ross 308 Nutrient Specifications 2014 [http://en.aviagen.com/assets/Tech_Center/Ross_Broiler/Ross308BroilerNutritionSpecs2014-EN.pdf.
    • 42. Paul D, Skrzypek G, Fórizs I. Normalization of measured stable isotopic compositions to isotope reference scales–a review. Rapid Communications in Mass Spectrometry. 2007;21(18):3006–14. pmid:17705258
    • 43. Bond AL, Hobson KA. Reporting stable-isotope ratios in ecology: recommended terminology, guidelines and best practices. Waterbirds. 2012;35(2):324–31.
    • 44. Dunn PJHC J.F. Good practice guide for isotope raito mass spectrometry. Second ed: Forensic Inotope Ratio Mass Spectrometry (FIRMS) Network; 2018 2018. 94 p.
    • 45. Qi H, Coplen TB, Mroczkowski SJ, Brand WA, Brandes L, Geilmann H, et al. A new organic reference material, l-glutamic acid, USGS41a, for δ13C and δ15N measurements− a replacement for USGS41. Rapid Communications in Mass Spectrometry. 2016;30(7):859–66.
    • 46. Song C, Kuznetsova O, editors. Fitting Gompertz Nonlinear Mixed Model to Infancy Growth Data with SAS-« version 8 Procedure NLMIXED2001 2001.
      • 47. Sakomura N, Longo F, Oviedo-Rondon E, Boa-Viagem C, Ferraudo A. Modeling energy utilization and growth parameter description for broiler chickens. Poultry Science. 2005;84(9):1363–9. pmid:16206556
      • 48. Littell RC, Milliken GA, Stroup WW, Wolfinger RD, Schabenberger O. SAS system for Mixed Models, Second Edition. Second ed. Cary, NC: SAS Institute Inc.; 2006 2006.
        • 49. Kiernan K, Tao J, Gibbs P, editors. Tips and strategies for mixed modeling with SAS/STAT® procedures. SAS Global Forum; 2012: SAS Institute Inc Cary, NC.
          • 50. SAS Institute Inc. SAS/STAT® 9.3 User’s Guide Cary, NC: SAS Institute Inc; 2011.
            • 51. Burnham KR, Anderson DR. Model selection and inference: a practical information-theoretic approach. 2nd ed. New York, New York: Springer-Verlag; 2002 2002.
              • 52. He M, Liu F, Wang F. Determination of the stable isotope discrimination factor of wild organisms: a case study of the red swamp crayfish in integrated rice–crayfish (Procambarus clarkii) culture without artificial diets. Isotopes in Environmental and Health Studies. 2022;58(1):81–98. pmid:34890270
              • 53. Arnold TW. Uninformative parameters and model selection using Akaike’s information criterion. Journal of Wildlife Management. 2010;74(6):1175–8.
              • 54. Stock BC, Semmens BX. MixSIAR GUI user manual. Version 3.1. 2016.
                • 55. Spiegelhalter DJ, Best NG, Carlin BP, Linde A. The deviance information criterion: 12 years on. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 2014;76(3):485–93.
                • 56. Hughes KL, Whiteman JP, Newsome SD. The relationship between dietary protein content, body condition, and Δ 15 N in a mammalian omnivore. Oecologia. 2018;186:357–67.
                • 57. Roth JD, Hobson KA. Stable carbon and nitrogen isotopic fractionation between diet and tissue of captive red fox: implications for dietary reconstruction. Canadian Journal of Zoology. 2000;78(5):848–52.
                • 58. Whiteman JP, Kim SL, McMahon KW, Koch PL, Newsome SD. Amino acid isotope discrimination factors for a carnivore: physiological insights from leopard sharks and their diet. Oecologia. 2018;188(4):977–89.
                • 59. Rosenfeld J, Van Leeuwen T, Richards J, Allen D. Relationship between growth and standard metabolic rate: measurement artefacts and implications for habitat use and life‐history adaptation in salmonids. Journal of Animal Ecology. 2015;84(1):4–20. pmid:24930825
                • 60. Hobson KA, Clark RG. Assessing avian diets using stable isotopes.2. Factors influencing diet-tissue fractionation. Condor. 1992;94(1):189–97.
                • 61. Coletta L, Pereira A, Coelho A, Savino VJM, Menten JFM, Correr E, et al. Barn vs. free-range chickens: Differences in their diets determined by stable isotopes. Food Chemistry. 2012;131(1):155–60.
                • 62. Micklem IA, Connan M, Stander N, McQuaid CD. Influence of ontogeny on stable isotope ratios and trophic discrimination factors of African penguin (Spheniscus demersus) tissues. Marine Biology. 2021;168(7):96.
                • 63. Cruz V, Ducatti C, Pezzato AC, Pinheiro D, Sartori JR, Goncalves J, et al. Influence of diet on assimilation and turnover of 13C in the tissues of broiler chickens. British poultry science. 2005;46(3):382–9. pmid:16050194
                • 64. Pelícia V, Araujo P, Luiggi F, Stradiotti A, Denadai J, Sartori J, et al. Estimation of the metabolic rate by assessing carbon-13 turnover in broiler tissues using the stable isotope technique. Livestock Science. 2018;210:8–14.
                • 65. Wolf N, Newsome SD, Peters J, Fogel ML. Variability in the routing of dietary proteins and lipids to consumer tissues influences tissue-specific isotopic discrimination. Rapid Communications in Mass Spectrometry. 2015;29(15):1448–56. pmid:26147485
                • 66. McMahon KW, Polito MJ, Abel S, McCarthy MD, Thorrold SR. Carbon and nitrogen isotope fractionation of amino acids in an avian marine predator, the gentoo penguin (Pygoscelis papua). Ecology and Evolution. 2015;5(6):1278–90. pmid:25859333
                • 67. Whiteman JP, Rodriguez Curras M, Feeser KL, Newsome SD. Dietary protein content and digestibility influences discrimination of amino acid nitrogen isotope values in a terrestrial omnivorous mammal. Rapid Communications in Mass Spectrometry. 2021;35(11):e9073. pmid:33634533
                • 68. Kurle CM, Koch PL, Tershy BR, Croll DA. The effects of sex, tissue type, and dietary components on stable isotope discrimination factors (Δ13C and Δ15N) in mammalian omnivores. Isotopes in Environmental and Health Studies. 2014;50(3):307–21.
                • 69. Lecomte N, Ahlstrom O, Ehrich D, Fuglei E, Ims RA, Yoccoz NG. Intrapopulation variability shaping isotope discrimination and turnover: experimental evidence in Arctic foxes. PLoS One. 2011;6(6). pmid:21731715
                • 70. Johnston TA, Casselman JM, Leggett WC. Sex-based divergence in isotopic compositions of north temperate freshwater fishes. Hydrobiologia. 2021;848(4):873–84.
                • 71. Caut S, Angulo E, Courchamp F, Figuerola J. Trophic experiments to estimate isotope discrimination factors. Journal of Applied Ecology. 2010;47(4):948–54.
                • 72. McMahon KW, Fogel ML, Elsdon TS, Thorrold SR. Carbon isotope fractionation of amino acids in fish muscle reflects biosynthesis and isotopic routing from dietary protein. Journal of Animal Ecology. 2010;79(5):1132–41. pmid:20629794
                • 73. Ishikawa NF. Use of compound-specific nitrogen isotope analysis of amino acids in trophic ecology: assumptions, applications, and implications. Ecological Research. 2018;33(5):825–37.
                • 74. Petta JC, Shipley ON, Wintner SP, Cliff G, Dicken ML, Hussey NE. Are you really what you eat? Stomach content analysis and stable isotope ratios do not uniformly estimate dietary niche characteristics in three marine predators. Oecologia. 2020;192(4):1111–26. pmid:32179976
                • 75. Davis P, Syme J, Heikoop J, Fessenden-Rahn J, Perkins G, Newman B, et al. Quantifying uncertainty in stable isotope mixing models. Journal of Geophysical Research: Biogeosciences. 2015;120(5):903–23.
                • 76. Tarroux A, Ehrich D, Lecomte N, Jardine TD, Bêty J, Berteaux D. Sensitivity of stable isotope mixing models to variation in isotopic ratios: evaluating consequences of lipid extraction. Methods in Ecology and Evolution. 2010;1(3):231–41.
                • 77. Cloyed CS, DaCosta KP, Hodanbosi MR, Carmichael RH. The effects of lipid extraction on δ13C and δ15N values and use of lipid-correction models across tissues, taxa and trophic groups. Methods in Ecology and Evolution. 2020;11(6):751–62.
                • 78. Pagani-Núñez E, Barnett CA, Gu H, Goodale E. The need for new categorizations of dietary specialism incorporating spatio-temporal variability of individual diet specialization. Journal of Zoology. 2016;300(1):1–7.
                • 79. Maldonado K, Bozinovic F, Newsome SD, Sabat P. Testing the niche variation hypothesis in a community of passerine birds. Ecology. 2017;98(4):903–8. pmid:28170088
                • 80. Schulting RJ, MacDonald R, Richards MP. FRUITS of the sea? A cautionary tale regarding Bayesian modelling of palaeodiets using stable isotope data. Quaternary International. 2023;650:52–61.
                • 81. Litmanen JJ, Perälä TA, Taipale SJ. Comparison of Bayesian and numerical optimization-based diet estimation on herbivorous zooplankton. Philosophical Transactions of the Royal Society B: Biological Sciences. 2020;375(1804):20190651. pmid:32536310