1The use of stable isotopic techniques to study animal diets and trophic levels requires a priori estimates of discrimination factors (Δ13C and Δ15N, also called fractionation factors), which are the differences in isotopic composition between an animal and its diet. Previous studies have shown that these parameters depend on several sources of variation (e.g. taxon, environment, tissue) but diet as a source of variation still needs assessment.
2We conducted an extensive review of the literature (66 publications) concerning estimates of animal-diet Δ13C (n = 290) and Δ15N (n = 268). We analysed this data set to test the effect of diet isotopic ratio on the discrimination factor, taking into account taxa, tissues, environments and lipid extraction treatments. Our results showed differences among taxonomic classes for Δ13C, but not for Δ15N, and significant differences among tissues for both Δ13C and Δ15N. We found a significant negative relationship between both, Δ13C and Δ15N, with their corresponding diet isotopic ratios. This relationship was found also within taxonomic classes for mammals (Δ13C and Δ15N), birds (Δ13C), fishes (Δ13C and Δ15N) and invertebrates (Δ13C and Δ15N). From these relationships, we propose a method to calculate discrimination factors based on data on diet isotope ratios (termed the ‘Diet-Dependent Discrimination Factor’, DDDF).
3To investigate current practice in the use of discrimination factors, we reviewed studies that used multi-resource isotopic models. More than 60% of models used a discrimination factor coming from a different species or tissues, and in more than 70% of models, only one Δ13C or Δ15N was used for all resources, even if resources had very different isotopic ratios. Also, we estimated DDDFs for the studies that used isotopic models. More than 40% used Δ15N values and more than 33% used Δ13C values differing > 2‰ from estimated DDDFs.
4Synthesis and applications. Over the last decade, applied ecologists have discovered the potential of stable isotopes for animal diet reconstruction, but the successful adoption of the method relies on a good estimation of discrimination factors. We draw attention to the high variability in discrimination factors, advise caution in the use of single discrimination factors in isotopic models, and point to a method for obtaining adequate values for this parameter when discrimination factors cannot be measured experimentally. Future studies should focus on understanding why discrimination factors vary as a function of the isotopic value of the diet.
Stable isotopic analyses are becoming widespread as a tool for studies of community structure and ecosystem function (Post 2002). In trophic studies, heavier isotopes of any given element increase in abundance compared with lighter isotopes, through the process of isotope discrimination. Early laboratory studies showed that for carbon (C) the isotopic ratio values of consumers are usually similar to those of their diets (DeNiro & Epstein 1978a,b). Since the ratio of carbon isotopes changes little (about +1‰; DeNiro & Epstein 1981; Peterson & Fry 1987; France & Peters 1997) as carbon moves through food webs, this ratio is commonly used to evaluate the source of the carbon, typically to distinguish carbon fixed by terrestrial C3 plants from that fixed by C4 plants or marine C3 plants (Peterson & Fry 1987). In contrast, consumers are typically enriched by 3–4‰ (DeNiro & Epstein 1981; Minagawa & Wada 1984; Peterson & Fry 1987) in isotopic ratios of nitrogen (N) relative to their diets. The isotopic ratio of nitrogen is thus commonly used to estimate trophic positions.
Based on the assumption that ‘you are what you eat’, two major types of studies have used differences in isotopic ratios between consumers and their resources: (i) trophic relationship studies and (ii) animal diet reconstruction studies. Each type uses the difference between isotopic ratios of an animal and its diet, called discrimination factor or trophic enrichment (Δ13C and Δ15N for carbon and nitrogen, respectively). Recent approaches in the use of isotopic mixing models to derive quantitative estimates of dietary contributions from isotopically distinct components specifically require precise estimates of diet discrimination factors (Phillips & Gregg 2001). Small variations in the values used for the discrimination factor may lead to important differences in the output of isotopic-mixing models (Ben-David & Schell 2001).
In this study, we assess the importance of diet isotopic values on discrimination factors. To this end, we examined some sources of variation potentially affecting discrimination factors: consumer class, environment, type of tissue, presence or absence of lipid extraction treatment, and finally diet isotopic ratios. This was done through analysis of estimates of Δ15N and Δ13C values in the literature. Based on the extensive set of studies analysed, we propose a method (the Diet-Dependent Discrimination Factor method) for obtaining a baseline for appropriate isotope discrimination factors calculated from the diet isotope values, controlling for other sources of variation. This provides suitable discrimination factors for each consumer class and tissue, and can be used to infer discrimination factor values in cases where data are lacking. In addition, to investigate current pratice in the use of discrimination factors, we reviewed 32 multi-resource isotope model studies. Finally, we make recommendations for the use of discrimination factors in isotope models.
We searched the ISI Web of Knowledge electronic data base (1983–2007, http://portal.isiknowledge.com) for literature involving stable carbon and nitrogen isotopic discrimination factors for any species, using the keywords stable isotope, stable carbon, stable nitrogen, carbon-13, nitrogen-15, discrimination factor, isotopic fractionation, isotopic enrichment and trophic enrichment. References cited in each of the resulting studies were reviewed for the presence of any additional studies (especially prior to 1983) that could have been missed in the previous search step. Because the focus of this review was the relationship between discrimination factor and diet isotopic ratio, estimates were not included where the diet was a mixture or was not controlled for (i.e. wild studies). If a study provided multiple discrimination factors estimated for one diet fed to one species of consumer, we pooled the data. If a study involved the same diet fed to different consumer species in the same taxon, the result for each species was considered one estimate. If a study involved different diets fed to the same species, the result for each diet was considered one estimate (see Supporting Information, Table S1). Estimates were not included if consumer diets were not specified, or if the authors explicitly stated that the duration of laboratory experiments was insufficient to allow for isotopic equilibrium between the consumer and its diet (see the ‘time’ column in Supporting Information, Table S1). Data available only in a graphical form were converted to a numerical form following fourfold enlargement of the graphs involved (estimated error = 0·05‰).
The literature search identified 66 references involving discrimination factors concerning 86 different species (Supporting Information, Table S1). To examine the use of discrimination factors in isotope models, we included only the 32 studies involving the four most used isotope models (Supporting Information, Table S2): (i) the geometric dual-isotope mixing model (Kline et al. 1993; Ben-David, Flynn & Schell 1997a; Ben-David et al. 1997b); (ii) the linear mixing model (Phillips 2001); (iii) the concentration-weighted linear mixing model (Phillips & Koch 2002); and (iv) the IsoSource partitioning mixing model (Phillips & Gregg 2003).
To examine variables affecting discrimination factors of carbon and nitrogen, we applied general linear mixed models (GLMM) in which the dependent variable was the carbon or the nitrogen discrimination factor. The normality of dependent variables was confirmed prior to analysis. We used mixed models because data from the same literature reference were correlated, and this covariance structure was handled by introducing the reference as a random effect into the GLMM.
To determine whether discrimination factors of carbon and nitrogen differed among consumer taxonomic classes, environments in which the species live, lipid extraction treatment of samples, or the analysed tissues, we performed independent GLMMs for each one of these independent variables. We distinguished four consumer classes (fishes, birds, mammals and invertebrates), three environments (terrestrial, marine and freshwater), and nine tissues (blood, collagen, feather, hair, liver, muscle, plasma, red blood cells and whole body). We performed a similar analysis to determine whether there was any relationship between discrimination factor and diet isotopic ratio for both carbon and nitrogen.
In analyses of the effect of tissues, we also investigated whether discrimination factors of carbon and nitrogen for each tissue differed among consumer classes by performing independent GLMMs for each type of tissue, using the consumer class as the independent variable.
After testing the general effects of consumer taxonomic classes, environment, lipid extraction treatment, tissue and diet isotopic ratios on the discrimination factors, we tested which of these variables significantly affected discrimination factors within each consumer class. To this end, we ran independent GLMMs for each consumer class in which the dependent variable was the carbon or the nitrogen discrimination factor, and the main independent variable was the diet isotopic ratio (carbon or nitrogen). Depending on the data available, three more independent categorical variables were added to the models: the environment of the species (as noted above), the tissue analysed (as noted above), and lipid extraction treatment (yes or no). We used these categorical variables when more than one category were available and valid (each category had more than five data points and/or the sample sizes of different categories were equilibrated). We ran a full model to identify the significant variables. However, the full model is unsatisfactory for prediction because it includes variables that are nonsignificant (Whittingham et al. 2006); and because one of the goals of this study was to develop a method for estimating discrimination factors when diet isotopic values are available, we searched for significant regression equations between discrimination factors and diet isotopic ratio, through simple general linear models (GLMs) within consumer classes (and within tissues, lipid extraction categories or environments, when these variables were significant in the full model). As these regressions do not consider random effects, GLM results could be slightly different from GLMM results with respect to the significance of the diet isotopic ratio. Discrimination factors (δY–δX) and diet isotopic values (δX) are not totally independent variables as one is partially derived from the other. Although the use of related variables (such as ratios) in regression is controversial as it could lead to spurious relationships, differences seem to not be so problematic (Hills 1978). Differences are frequently used, for example, in sexual size dimorphism (Szekely, Freckleton & Reynolds 2004) or community ecology (Cerda, Retana & Cros 1998) studies. Indeed, using the residuals of the relationship between diet and tissues should yield the same results. Computations were performed with statistica 6·0 (StatSoft Inc. 2001) and sas package (procedure MIXED, version 8·2, SAS Institute Inc. 2004).
Variation in discrimination factors
The literature review yielded 290 animal-diet discrimination factor estimates of carbon and 268 animal-diet discrimination factor estimates of nitrogen from 66 publications (Supporting Information, Table S1) distributed (for Δ13C and Δ15N, respectively) as follows: mammals (95 and 89), birds (61 and 52), fishes (41 and 47), reptiles (3 and 3), and invertebrates (90 and 77). The overall mean estimates for Δ13C and Δ15N were 0·75‰ (SE = 0·11) and 2·75‰ (SE = 0·10), respectively.
effect of lipid extraction treatment
Samples are sometimes treated to extract lipids before isotope analysis, and this can affect the values obtained (Murry et al. 2006; Sweeting, Polunin & Jennings 2006; Bodin, LeLoch & Hily 2007). We separated discrimination factor values into two categories according to whether consumer and diet samples were subject to lipid extraction. If lipids were extracted from one but not the other sample type, data were not included in the analysis. Discrimination factors of carbon and nitrogen did not differ with lipid extraction (F1,196 = 1·22, P = 0·271 and F1,187 = 0·89, P = 0·348, for Δ13C and Δ15N, respectively).
effect of environment
We found significant differences among environments for the carbon discrimination factor but not the nitrogen discrimination factor (F2,229 = 3·10, P = 0·047 and F2,209 = 0·09, P = 0·912, for Δ13C and Δ15N, respectively). Higher mean estimates of Δ13C were obtained for organisms inhabiting freshwater environments (1·33‰, SE = 0·07, n = 42) than for those inhabiting marine (0·96‰, SE = 0·18, n = 87) or terrestrial (0·32‰, SE = 0·17, n = 158) environments.
effect of taxon
Discrimination factors differed among the consumer classes for carbon (F3,228 = 2·96, P = 0·033) but not for nitrogen (F3,207 = 1·51, P = 0·214, Fig. 1). However, interpretation of the effect of taxon difference was somewhat confounded by other sources of variation that may have influenced the discrimination factor values, in particular that the data represented different tissues, environments, diets, or treatment of samples (i.e. lipid extraction or not) before isotopic analysis.
effect of tissue
We initially analysed all taxa combined. Discrimination factors of carbon and nitrogen differed among tissues (F9,220 = 1·93, P = 0·049, and F8,198 = 2·71, P = 0·007, respectively). Consequently, we analysed differences among the consumer classes for muscle, plasma, liver, blood and the whole body (Fig. 2); other tissues did not have sufficient data to carry out this analysis. The carbon discrimination factor for muscle was significantly different among birds, fishes and mammals (F2,21 = 9·15, P = 0·001), but the differences were not significant for the nitrogen discrimination factor (F2,18 = 1·86, P = 0·184). For plasma, both the carbon and the nitrogen discrimination factors were not significantly different between birds and mammals (F1,16 = 3·52, P = 0·079 and F1,16 = 2·53, P = 0·131, respectively). The carbon discrimination factor for liver did not differ significantly among birds, fishes and mammals (F2,16 = 2·92, P = 0·083), but did differ significantly for the nitrogen discrimination factor (F2,16 = 6·67, P = 0·008, Fig. 2). For blood, the carbon discrimination factor was no different between birds and mammals (F1,13 = 0·05, P = 0·834), but the difference was significant for the nitrogen discrimination factor (F1,11 = 5·52, P = 0·039). For the whole body, the carbon discrimination factor was significantly different between invertebrates and fishes (F1,80 = 9·14, P = 0·003), but no significant differences were found for the nitrogen discrimination factor (F1,68 = 0·13, P = 0·721).
effect of diet
Our initial analysis of all taxa combined showed significant negative relationships between discrimination factors and their diet isotopic ratios (F1,230 = 51·58, P < 0·001, R2 = 0·19 and F1,210 = 50·54, P < 0·001, R2 = 0·16, for ΔC and ΔN, respectively). We then performed independent GLMMs for these relationships within each consumer class (birds, mammals, fishes and invertebrate) and, taking into account where possible the type of tissue, the environment and the lipid extraction treatment (we used one of these variables when more than one category was available and valid, as described in the Statistical analysis section). In general, we found the same trend as in the initial analysis (all combined taxa) of significant negative relationships between discrimination factors and their corresponding isotopic ratios (Table 1). However, in some consumer classes, the relationship was not significant, as described below.
Table 1. Factors affecting the carbon and nitrogen discrimination factors in general linear mixed models. The analysed variables are presented for each consumer class and in italics are the ones significant in the full model
For mammals, only two categorical independent variables were added to the GLMM: the lipid extraction treatment and the type of tissue (six categories: blood, red blood cells, hair, liver, muscle and plasma). The carbon discrimination factor was negatively correlated with the diet carbon isotopic ratio, but none of the categorical variables was significant in the full model (Table 1). The nitrogen discrimination factor showed differences among tissues and a significant negative correlation with the diet nitrogen isotopic ratio (Table 1). The GLM on the relationships between discrimination factors and their corresponding diet isotopic ratios confirmed these results, showing significant relationships for both carbon and nitrogen (Δ13C: F1,88 = 91·44, P < 0·001, R2 = 0·51; and Δ15N: F1,78 = 14·25, P < 0·001, R2 = 0·15; Fig. 3a,b). For the nitrogen discrimination factor and within tissues, the GLM between Δ15N and diet isotopic values were only significant for muscle, liver and plasma (F1,13 = 17·74, P = 0·001, R2 = 0·58; F1,14 = 8·17, P = 0·013, R2 = 0·37; and F1,17 = 19·17, P < 0·001, R2 = 0·53, respectively).
For birds, three categorical independent variables were added to the GLMM: the lipid extraction treatment, the environment of the bird (three categories: terrestrial, marine and freshwater), and the type of tissue (five categories: blood, feather, liver, muscle and plasma). The carbon discrimination factor was positively correlated with the diet carbon isotopic ratio, and there were significant differences among tissues (Table 1). The nitrogen discrimination factor showed differences among tissues, environments and lipid extraction treatments, but there was no significant relationship with the diet nitrogen isotopic ratio (Table 1). The GLM showed no significant relationships between the carbon discrimination factor and carbon diet isotopic ratio (Δ13C: F1,53 = 0·64, P = 0·425, R2 = 0·012; Fig. 3c); within tissues the GLM between Δ13C and the carbon diet isotopic value was only significant negative for blood (F1,14 = 8·92, P = 0·010, R2 = 0·39). For nitrogen the GLM showed no significant relationships between the discrimination factor and diet isotopic ratio (Δ15N: F1,46 = 0·009, P = 0·924, R2 = 0·00; Fig. 3d). Within tissues and lipid extraction categories, the GLMs between Δ15N and nitrogen diet isotopic value were not significant, and for environment, the relationships were only significant for marine and terrestrial environments (F1,16 = 18·91, P < 0·001, R2 = 0·54 and F1,15 = 18·88, P < 0·001, R2 = 0·56, respectively).
For fishes, three categorical independent variables were added to the GLMM: the lipid extraction treatment, the type of tissue (three different tissues: liver, muscle or whole body) and the environment of the fish (two categories: marine and freshwater). The GLMM for the carbon and nitrogen discrimination factors showed negative relationships with the diet isotopic ratios, and there were also differences among tissues (Table 1). The GLM analysis showed significant relationships between both discrimination factors and the corresponding diet isotopic ratios (Δ13C: F1,39 = 10·69, P = 0·002, R2 = 0·22; and Δ15N: F1,45 = 19·28, P < 0·001, R2 = 0·30; Fig. 3e,f). For the carbon and nitrogen discrimination factors and within tissues, the GLM between discrimination factors and diet isotopic values were only significant for whole body and muscle (carbon: F1,14 = 8·34, P = 0·012, R2 = 0·37, and F1,16 = 7·01, P = 0·018, R2 = 0·30, respectively; nitrogen: F1,15 = 16·65, P < 0·001, R2 = 0·53, and F1,17 = 7·85, P = 0·012, R2 = 0·32, respectively).
For invertebrates, one categorical variable was added to the GLMM model that was not significant in the full model: the environment of the invertebrate (three categories: terrestrial, marine and freshwater). The discrimination factors for nitrogen and carbon were negatively correlated with their corresponding diet isotopic ratios (Table 1). The same trend was evident in the GLM analysis (Fig. 3g,h), which was significant for carbon (F1,84 = 7·97, P = 0·006, R2 = 0·09) and nitrogen (F1,71 = 39·50, P < 0·001, R2 = 0·36) .
Use of discrimination factors for dietary reconstruction
The literature review identified 32 publications that used one of the four main model types (Supporting Information, Table S2): 10 involving the geometric dual-isotope mixing model, 11 involving the linear mixing model, 3 involving the concentration-weighted linear mixing model, and 9 involving the IsoSource source partitioning mixing model. All taxa were represented: mammals (8 publications), birds (8 publications), fishes (6 publications), invertebrates (11 publications) and reptiles (1 publication).
We first investigated the source of Δ13C and Δ15N used in the isotopic models in the reviewed papers (i.e. the study cited in each paper), and examined the concordance between the consumer classes and tissues of the discrimination factor in the publication source, and the discrimination factor used in the isotopic models. In most cases (20 of 35 publications for carbon and 23 of 38 publications for nitrogen), the discrimination factor for the consumer class and tissue in the publication source differed from that of the model (references marked with D in Supporting Information, Table S2). In addition, in 16 cases for carbon and 22 cases for nitrogen, the discrimination factors used in the isotopic models came from published reviews (references marked with an asterisk in Supporting Information, Table S2). These reviews included both field and laboratory studies. Therefore, it is likely that the values of discrimination factors coming from these reviews (i.e. the values used in 19 of the 32 studies) strongly biased the results of the isotopic models. However, these were probably the best values available at the time of publication.
To assess the potential inappropriate use of discrimination factor values in isotopic models for diet reconstruction, we calculated new nitrogen and carbon discrimination factors for all the reviewed papers using a Diet-Dependent Discrimination Factor method (DDDF, see Fig. 4). We constructed a decision diagram (Fig. 4) using the quantifications of the effects of diet isotopic values and type of tissues within different consumer classes. We used the regression equations for each consumer class when only the relationships between discrimination factors and their corresponding diet isotopic ratios were significant (e.g. invertebrates, see Table 1). When both the diet isotopic ratio and the tissue significantly affected discrimination factors, we used the regression equations between discrimination factors and diet isotopic ratios for each tissue (e.g. muscle for fishes), or the mean for the tissue when this regression was not significant (e.g. liver for fishes). In the case of birds, the diet isotopic ratios were not significantly related to their corresponding discrimination factors, and hence, we took the mean for each tissue (except for blood and Δ13C, where we used the regression equation). In the case of mammals, the tissue was not a significant variable affecting carbon discrimination factor but we were interested to search for regression equations among tissues (although main equation for all tissues together was also included).
For both carbon and nitrogen, we calculated two DDDFs corresponding to the minimum and maximum values of the isotopic ratios of the diet (except where there was only one diet isotopic value). We found that 18 of 39 Δ13C, and 29 of 42 Δ15N used in the isotopic models were different from estimated DDDFs (e.g. outside of our interval). Moreover, some of the used discrimination factors were included in the estimated DDDF range but this range was very wide.
To illustrate the possible deviation between the range of the estimated DDDFs and the discrimination factors used in the isotopic models, we calculated a parameter called the deviation coefficient (CD). This parameter was the maximum deviation between the discrimination factor value used in the isotopic study and the discrimination factor of DDDFs for each diet item. Thus, CD represents the deviation (in ‰) of discrimination factors used in the isotopic models from DDDFs. Based on these criteria, we found 11 used Δ13C and 15 used Δ15N within [0–1‰CD] (‰ of deviation from DDDFs), 16 Δ13C and 11 Δ15N within [1–2‰ CD], 6 Δ13C and 13 Δ15N within [2–3‰CD], 3 Δ13C and 3 Δ15N within [3–4‰CD], 3 Δ13C within [4–5‰CD] and 1 Δ13C and 1 Δ15N > 5‰CD, for carbon and nitrogen respectively (see Supporting Information Table 2A). Averaged CD of used values were 1·78‰ (SD: 1·36) for Δ13C and 1·69‰ (SD: 1·09) for Δ15N. In summary, 40% used Δ15N and 33%Δ13C differed more than 2‰ from estimated DDDFs. This means that more than 35% of the values used in isotopic model studies could have an error ≥ 2‰ leading to incorrect results, and that an appropriate value should be used instead.
Two key points emerge from the thorough review of the literature reported here. First, there is high variability in both Δ13C and Δ15N values, and this is mainly dependent on the consumer class, the tissue and the diet isotopic ratio. Indeed, we show a significant negative relationship between both Δ13C and Δ15N and their corresponding diet isotopic ratios. Secondly, most studies have used inappropriate discrimination factor values, or averages from data that should not have been combined, both of which could generate inaccuracies in isotope model outputs, and hence, in interpretation in diet reconstruction studies. Although these were probably the best values available at the time the studies involved were published, we stress that using inappropriate values for discrimination factors potentially leads to large errors or meaningless results. We propose the DDDF method to generate values when the data for use in isotope models are lacking.
In this study, we have shown that discrimination factors are significantly affected by the consumer taxonomic group and the consumer tissue. Indeed, we found differences between taxonomic classes in Δ13C but not in Δ15N, although these results could be partly confounded by the use of different tissues. Previous studies have noted that Δ15N may vary among species (e.g. DeNiro & Epstein 1981; Vanderklift & Ponsard 2003). Differences may be partly due to the excretion mode of each class (Vanderklift & Ponsard 2003). The results also revealed consistent differences in nitrogen discrimination factor among tissues that could be explained by different metabolic properties characterizing organs and tissues within the body that are similar across taxa (e.g. turnover rates, biochemical composition). In this aspect, several studies have found contrasting results (e.g. DeNiro & Epstein 1981; Hobson & Clark 1992a; Hilderbrand et al. 1996).
Stable isotope models are used to quantify the contributions of multiple sources to a mixture, such as the proportions of different types of food sources in an animal's diet. The use of isotope models in ecology has dramatically increased in the last several decades (see references in Supporting Information, Table S2). Mean signature values are calculated for each of the sources, and based on this, the fractional contribution of each source to the mixture is calculated. However, the most important parameter in the isotope model is the discrimination factor, which can considerably modify the model output with respect to the difference in isotopic composition between an animal and its diet (Ben-David & Schell 2001). Despite the large variability in nitrogen and carbon discrimination factors highlighted here, most isotope model studies have used a single discrimination factor for carbon and nitrogen, often obtained from a published review. Although these were probably the best values available at the time the studies involved were published, this has created two problems. Firstly, the estimates in the reviews were derived from inappropriate combinations of laboratory studies (which represent discrimination factors) and field studies (which represent trophic enrichment), and included different consumer classes, tissues and other variables. Secondly, a common assumption has been that discrimination factors are independent of the diet isotope value. We have shown that diet-dependent discrimination factors calculated using the method we have proposed differ markedly from those used in the reviewed studies (see Supporting Information, Table S2), with the consequence that most models generate incorrect results. In fact, as isotope models are very sensitive to changes (> 1‰) in discrimination factors (Ben-David & Schell 2001), the use of diet-dependent discrimination factors could significantly change the results and hence the interpretation (Caut et al. 2008b).
Based on an extensive literature review, the trend between discrimination factors and diet isotopic ratios is consistent among consumer classes and explains between 9% and 51% of the variation in discrimination factors. Caut et al. (2008a) showed higher percentages in a controlled experiment: diet isotopic values explained 60–98% of the variation of discrimination factors in different tissues in the rat (Rattus sp.). Similarly, Felicetti et al. (2003) showed that diet isotopic values explained 88% of Δ13C and 98% of Δ15N of bear plasma. Here, we have provided a decision diagram (Fig. 4) for estimating the discrimination factors for different animal consumer classes (rows) and two isotopic elements (columns) when using isotope models. For each class and isotopic element, ecologists can easily estimate a discrimination factor relevant to the particular field situation, according to the significant variables in each case. The decision diagram includes the three most significant variables (taxon, tissue and diet isotopic value) from the five studied in this review, but other variables not considered here could also play a role (e.g. quality of diet, type of excretion, trophic level). Moreover, the data in Supporting Information, Table S1 could be used to construct and test regression equations for specific situations (e.g. for a researcher working with insects, where only data for this consumer class can be used).
We do not have a functional explanation for this relationship, and future effort should be focused on laboratory-based studies to quantify the involvement of factors including C : N ratio, amino acid composition and nutritional state. However, knowledge of this relationship provides researchers with a tool to reduce model output errors created by use of inappropriate discrimination factors. To date, the lack of an easy method to quantify species-specific discrimination factors for different diets and tissues has led ecologists to use an idiosyncratic collection of procedures mainly relying on fixed discrimination factors, particularly in studies of elusive, rare or endangered species. Although we concur with the note of caution expressed by Ben-David & Schell (2001), that regressions of this type cannot be used as surrogates for mixing models, such regressions should be very useful in providing estimates of discrimination factors when no field data are available. We strongly recommend using the decision diagram when the diet sources used in the isotope model have significantly different isotope ratios (e.g. in omnivore species). Finally, field ecologists should recognize that discrimination factor values are estimated with error and that this error propagates in the use of mixing models. For example, some of the relationships we present are noisy and this noise is ignored when running computer programs to estimate sources in animal diets. Future studies should devise models that incorporate errors in discrimination factors.
In summary, understanding and estimating discrimination factors remains problematic. Stable isotope methods are currently among the most powerful tools for the study of trophic relationships and animal diets. However, the assumptions underpinning isotope models, such as the potential sources of variation in discrimination factors, should not be overlooked. In particular, researchers using isotope models should consider the diet-dependent discrimination factor as a tool for obtaining more accurate results.
We wish to thank V. Corvest for help in the search for manuscripts for inclusion in the review. We would like to thank Carlos Martínez del Rio for valuable and encouraging comments on the ms. This study was financed by a grant from INSU: ACI ECCO-PNBC. E.A. was supported by a postdoctoral fellowship from the Spanish Ministry of Education and Science (SEEU-FSE), and F.C. was supported by l’Agence Nationale de la Recherche.