Introduction

Livestock grazing grass-legume mixed pastures can construct their own diets by selecting different forage species and fulfilling their nutritional requirements with varying proportions of each1. Diet selection in grazing animals is a function of preference modified by opportunity2, thus, when multiple species are present in a pasture, the accurate representation of the animal’s diet is difficult to achieve through pasture sampling alone. Knowing the true diet composition of a grazing animal is an important tool for linking pasture canopy composition and animal diet selection, and diet selection to animal performance. However, accurate representation of grazing animal diet composition has been a challenge in grazing research, with methods often relying on invasive techniques (e.g., collection of extrusa through oesophageal fistulae) which, among others, have certain limitations, including number of animals required, surgical costs, and difficulty in handling extrusa3. Other techniques utilizing plant wax components (e.g. n-alkanes, or long-chain alcohols) have also been employed for estimating diet botanical composition but remain challenging to employ under grazing due to intensive labor requirements related to dosing n-alkanes or long-chain alcohols, and elevated-costs related to sample processing4.

Stable isotope techniques provide additional options for inferring diet botanical composition across several animal species5,6, and thus far, C and N techniques have been implemented for this purpose. Stable isotope composition is expressed in delta (δ) notation, which is the difference between the ratio of heavy isotope (e.g. 13C or 15N) to light isotope (e.g. 12C or 14N) in a given sample, to the same ratio of the standard (Pee Dee Belemnite for C, or air for N)5. The C isotope techniques are contingent upon differences in the accumulation of C isotope in plants of contrasting photosynthetic pathways 6. Legumes and temperate, cool-season grasses are C3 (Calvin pathway) plants, while warm season, tropical and sub-tropical grasses are C4 (dicarboxylic acid pathway) plants. The C3 pathway discriminates against 13C in favor of 12C to a greater extent than C4, resulting in a more depleted (more negative) δ13C composition for C3 plants compared to that of C4 plants 5,7. The range of δ13C of C3 plants is −35 to −20‰, and C4 plants from −16 to −9‰ 7.

Stable N isotope techniques have been largely utilized in ecological studies to establish trophic schemes since consumers generally increase their 15N (the heavy N isotope) composition by a factor of 3‰ compared with their diet source 5. This occurs, to an extent, because of faster loss of 14N than 15N in metabolism and excretion leaves consumers with enriched (more positive) δ15N composition5. However, in agronomic studies, δ15N is often used to quantify biological N2-fixation8. Plant δ15N composition generally ranges between −10 and 5‰, with depleted δ15N being indicative of biological N2-fixation processes9. Plants that undergo biological N2-fixation have δ15N resembling that of the atmosphere, 0‰, where plants reliant upon soil mineral N, exhibit δ15N resembling that of the soil, oftentimes above 2‰8,10. Legumes are well known for their symbiotic relationship with N2-fixing bacteria, but grasses have also demonstrated potential to associate with N2-fixing microorganisms11,12,13.

Thus far, the use of fecal δ13C has been an accurate and rapid method for estimating the diet botanical composition of livestock grazing C3–C4 binary pastures6. This technique is advantageous in providing an inference on effects of dietary proportion of C3 or C4 species on short-term intake with ease of sample collection, and minimal sample processing requirements 6,14; however, a comparison across various body samples and their response to both 15N and 13C has yet to be made for beef cattle. Understanding the differences in response across body samples, in relation to C and N stable isotope composition is important for evaluating methodologies that aid in understanding grazing animal diet composition. This study was designed to compare the use of 15N and 13C composition in blood plasma, red blood cells (RBC), hair, and feces to estimate the diet botanical composition of beef steers fed C3–C4 binary diets. These body samples were selected provided their universal ease of collection across grazing studies. The two hypotheses in the study were: (1) δ13C of feces, plasma, red blood cells, and hair will be useful to back-calculate the diet of steers consuming C3–C4 binary diets; and (2) that δ13C is a more accurate predictor than δ15N of diet composition from steers consuming C3–C4 binary diets.

Results

Chemical composition of the diet

Concentrations of dry matter (DM) were similar across the five diets, and the concentrations of organic matter (OM) appeared to decrease with increasing proportion of RP inclusion (Table 1). Crude protein concentration was least in 0% RP, which contained bahiagrass only, and greatest in 100% RP. The in vitro digestible OM (IVDOM) concentrations were also least in 0% RP, at 46.4%, and increased to 67.8% in 100% RP. As the proportion of RP increased in the diet, the δ13C became more depleted, ranging from −16.04 to −29.41‰, in 0% RP and 100% RP, respectively. The reverse was observed for δ15N, where 0% RP was more depleted (−0.24‰) than 100% RP (1.71‰), and the treatment diets became more enriched with increasing proportion of RP.

Table 1 Chemical composition of the five treatment diets offered to beef steers, each with increasing proportion of rhizoma peanut (RP) hay into bahiagrass hay.
Full size table

Stable isotope composition across sample type

Feces

Throughout the experimental period, there was a treatment × collection day interaction on fecal δ13C (P = 0.002; Fig. 1). Feces from 0%RP remained at -18‰ throughout the 32-days experimental period, while 100%RP became the most depleted with time, reaching −31.8‰ by 32 days. The feces were depleted in their δ13C on the first day of collection, given the animals were fasted on 0 day, and therefore did not produce any fecal material upon collection. Despite depleted δ13C on 1 day, there were no treatment differences within this evaluation, whereas treatments within each evaluation day differed thereafter. The fecal δ15N composition also showed a treatment × collection day interaction (P = 0.005). Initial δ15N was 4.41‰ across all treatments, and by 32 days, 0%RP, 25%RP, and 50%RP were the most depleted, while 75%RP and 100%RP were the most enriched (Fig. 1). Across the five collection dates, there was a linear effect (P < 0.001) on the fecal δ15N composition, where the feces were more enriched with increasing proportions of RP, which might be indicative of discrimination against the lighter isotope, resulting in fractionation. Across all treatments, the feces became enriched in δ15N compared with the diets, and by 32 days, Δ (the fractionation factor, comparing diet and fecal δ15N) averaged 2.09‰ for each treatment.

Figure 1
figure 1

δ13C (a) and δ15N (b) feces of beef cattle consuming increasing proportions of rhizoma peanut (RP) in bahiagrass hay. Day of collection represents the number of days after dietary change. Asterisk indicates treatment differences (P ≤ 0.05) exist within the day of collection.

Full size image

Plasma

Plasma δ13C composition became enriched as the proportion of RP increased in the diet. Furthermore, there was a treatment × collection day interaction (P < 0.001), in plasma δ13C. All treatments averaged −14.02‰ at the beginning (0 day) of the experimental period (Fig. 4). The 0%RP remained the most enriched, and was −14.32‰ by 32 days. In general, as the proportion of RP increased, the δ13C in the plasma became more depleted. The largest magnitude in difference from 0 to 32 days was in 100%RP, where the δ13C decreased from −14.09 to −19.73‰. By 16 days, plasma appeared to begin reaching equilibrium, since changes in δ13C were minimal in each of the treatments after 16 days. Estimating RP proportions based on plasma δ13C showed adjusted R2 of 0.82 and 0.74, for 8 and 32 days, respectively (Fig. 2). Similarly, the δ15N composition in the plasma also showed a treatment × collection day interaction (P = 0.049). Overall, there was a trend for plasma δ15N to become depleted throughout the duration of the experiment, where the plasma δ15N across all treatments was 7.60‰ at 0 day and was 5.99‰ by 32 days (Fig. 4).

Figure 2
figure 2

δ13C (a) and δ15N (b) plasma of beef cattle consuming increasing proportions of rhizoma peanut (RP) hay in bahiagrass hay. Day of collection represents the number of days after dietary change. Asterisk indicates treatment differences (P ≤ 0.05) exist within the day of collection.

Full size image

Red blood cells

Despite an observed treatment × collection day interaction on the δ13C composition of RBCs (P = 0.004; Fig. 3), there was no observable pattern to the response in terms of δ13C composition, nor was there an appearance of equilibrium by 32 days. Across all treatments, there was a difference of − 0.29‰ on δ13C of RBC, from 0 to 32 days, indicating small magnitude in change. Discrimination (Δ) was 13‰ in 100%RP, which was the largest among all treatments (P < 0.001). In addition, while there was a treatment effect (P = 0.04) on RBC δ15N composition (Table 2), a collection day effect was not observed (P = 0.55). The RBC δ15N decreased linearly (P = 0.004), as RP proportion increased in the diet (Table 2).

Figure 3
figure 3

δ13C of red blood cells (RBC) of beef cattle consuming increasing proportions of rhizoma peanut (RP) in bahiagrass hay. Day of collection represents the number of days after dietary change.

Full size image
Table 2 δ13C and δ15N composition of red blood cells from beef steers consuming increasing proportions of rhizoma peanut (RP) hay in bahiagrass hay.
Full size table

Hair

Hair δ13C composition showed a treatment × collection day interaction (P < 0.001; Fig. 4). During the 0-day and 8-days day of collection, which implies the number of days after diet implementation, there were no treatment differences within each day. However, the collection days thereafter showed treatment differences within each collection day. By 32 days, 100%RP showed the most depleted δ13C (−22.67‰). The 0%RP was enriched from −17.18 to −14.63‰ from 0 to 32 days, while 25%RP remained at −16.53‰ during the same period. Lastly, there were no differences in treatments on δ15N composition of the hair (P = 0.58; Table 2) and averaged 7.15‰ across treatments.

Figure 4
figure 4

δ13C of hair from beef cattle consuming increasing proportions of rhizoma peanut (RP) in bahiagrass hay. Day of collection represents the number of days after dietary change. Asterisk indicates treatment differences (P ≤ 0.05) exist within the day of collection.

Full size image

Prediction of diet composition

Three subgroups (13C fecal 8 days, 13C fecal 32 days, and 15N fecal 32 days) had concordance correlation coefficients (CCC) ≥ 0.80 (Table 3). They were also characterized by regression model intercepts and slopes relatively close to zero and 1.0, and adjusted R2 ≥ 0.70. These subgroups had comparatively favorable profiles with respect to the MSE partitioning (smaller MSE and higher proportion of the MSE due to random error). Several subgroups (13C plasma 8 days, 13C plasma 32 days, and 13C hair 32 days) had Pearson correlation coefficients ≥ 0.80 and regression model adjusted R2 ≥ 0.70, but also exhibited low concordance correlation coefficients, model intercepts that deviated substantially from zero, and large MSE with a high proportion attributable to the MC component. Although the predictions for these latter subgroups exhibited strong correlation with the actual dietary percentages, the magnitude of the predictions was meaningfully different from the actual percentages. This translates to not being able to predict dietary percentages directly using Eq. (3).

Table 3 Analytic regression analysis results for 16 combinations of isotope (13C or 15N), day (8 or 32), and sample type (feces, plasma, red blood cell (RBC), or hair).
Full size table

The two one-sided test (TOST) with an equivalence region of (−15%,15%) were conducted for the three sets of fecal samples identified above. In each case the mean actual minus predicted difference was < 0 indicating that on average the predicted dietary percentages were greater than the actual percentages (Table 3). The P-values for the 13C fecal 8 days, 13C fecal 32 days, and 15N fecal 32 days samples were 0.1035, 0.1039, and 0.0298, respectively. The corresponding minimum equivalence regions for these subgroups were (−16.31, −5.48), (−16.37, −5.11), and (−13.95, −0.51). These results indicate that it would be reasonable to predict dietary percentages directly from Eq. (3) for each of these subgroups with a tolerance of ± 15%.

The analyses where predicted percentages outside of the valid range were assigned the appropriate boundary value prior to analysis resulted in five (one < 0% and four > 100%), three (three > 100%), and two (one < 0% and one > 100%) percentages being reassigned for the 13C fecal 8 days, 13C fecal 32 days, and 15N fecal 32 days subgroups. The results for 15N fecal 32 days were essentially unchanged however meaningful improvements were realized for the two 13C fecal subgroups with respect to the proximity of the regression slopes to 1.0 (both slopes were very close to the benchmark, although the corresponding intercepts were somewhat farther away from zero than in the primary analyses), the MSE partitioning (smaller MSE with 80% attributable to random error), and the TOST equivalence tests (P < 0.01) (Table 4).

Table 4 Analytic regression analysis results of predicted vs. actual proportions of rhizoma peanut into bahiagrass hay diets consumed by beef cattle.
Full size table

Discussion

The various samples collected in this study differed in their δ13C and δ15N compositions throughout the 32-days period, and therefore their utility for use as predictor of diet composition varies. The use of δ13C across the samples collected proved a more reliable predictor, than δ15N, of C3–C4 diet composition. However, the δ15N of feces indicated a probable method for estimating C3:C4 diet composition, which was unexpected. Additionally, the response to δ13C composition was also affected by the collection day, implying the time after diet change affects these results and reliability of these samples as C3–C4 diet predictors.

Altogether, the fecal δ13C results corroborate data by Norman, et al. 6, where feces from sheep fed various proportions of C3–C4 forages remained at δ13C equilibrium within 18-days after dietary change and were useful predictors of diet composition. Rates of passage will have an impact on fecal composition, and therefore the isotopic composition of the fecal material is reflective of the ingested diet after the retention period along the digestive tract. Bahiagrass has a mean retention time of 42 h in rumen15, suggesting that feces will likely reflect dietary changes beyond this time period, especially as the animal integrates the diet into their system16. In addition, considering total tract retention of diet feedstuff is important for ensuring accuracy of sample collection. Therefore, the fecal δ13C obtained 8-days after diet change, does indeed represent the diet consumed within this 8-days time frame, and given that fecal sample collection does not require specialized equipment, additional personnel training, or elaborate sample processing, this methodology will likely continue to be favored over the collection and analysis of other samples.

Regarding fecal δ15N, enrichment (in relation to the diet) has also been observed in other studies, and it is likely a result of 15N discrimination during the digestive process17,18,19. Sutoh, et al. 17 also showed rumen bacteria and protozoa differ in their δ15N composition, which might also provide insights into the δ15N metabolic fates. In meerkat (Suricata suricatta) feces, δ15N enrichment occurred during transit through the digestive system, indicating discrimination is not restricted to ruminant species alone19. Discrimination factors (Δ) are likely affected by the diet composition19. For instance, Sare, et al. 20 established that δ15N composition in stomach contents, feces, collagen, and hair collected from red-backed voles (Clethrionomys gapperi) varied in each of the sample types according to the protein concentration in the diets. Similar studies are warranted in ruminants, especially to establish accurate discrimination factors since models have been developed that incorporate diet-sample discrimination6.

In this study, fecal δ15N was useful for predicting diet proportions at 32-days, owed to the strong CCC and adequate adjusted R2 from regression of actual vs. observed proportions of RP in diet (Table 3). These results are novel since previous uses of 15N have not resulted in capability of predicting dietary proportions, especially that of C3–C4 diets. Possible explanation may be attributed to the magnitude in difference of the feed δ15N in 0%RP and 100%RP diets, which was 1.92‰, was large enough such that differences in species composition could be detected, similar to the mechanisms observed using δ13C. However, Cantalapiedra-Hijar et al. 21 indicate that differences in animal 15N discrimination are highly affected by the animal’s metabolism. While fecal δ15N may be used as a diet predictor in C3–C4 binary mixtures, further studies are still warranted to corroborate these results and evaluate if differences may occur due to an animal’s age or stage of production. If corroborated, these results may provide options for potentially reconstruction diet components in diets consisting only of either C3 or C4 forages, so long as the feed δ15N differences are large in magnitude.

The time required to reach equilibrium is directly influenced by the range in isotopic compositions of the diets consumed, and is also affected by the magnitude in change of diet isotopic composition6. In plasma there is a high correlation at both 8 and 32-d for predicted vs. observed RP diet proportions, however the intercept of the model falls below any realistic value, along with poor CCC, which results in predictions well below 0% across all treatments (Table 2). It may be speculated that a period beyond 32-d may be required to establish plasma δ13C equilibrium. Similar results are reported to other species, including American Black Ducks (Anas rubripes), where diet changes were detected using δ13C of plasma within 1 week after diet change, but required at least 67 days to reach δ13C equilibrium 22.

Isotopic fractionation in plasma N has been reported in other studies and is attributed to liver urea synthesis and splanchnic amino acid metabolism 17,23. In addition, fractionation in plasma N can also be affected by both diet type and nutrient intake level23. Plasma δ15N might be a better indicator of the N status of the animal by encompassing variability in the conversion of metabolizable N into animal proteins 23. Halley et al. 24 show that 15N fractionation in male reindeer (Rangifer tarandus) varies by individuals, tissue/sample collected, sex, and even physiological status of the animal, thus establishing accurate fractionation values is important for increasing their value in diet-reconstruction models. In the present study, the observed plasma δ15N did not appear to reach equilibrium, since each of the collection days differed in their δ15N composition and the overall values appeared to become depleted with time. The inherent variability in body tissues and within animal species, as well as differences in turnover rates may affect the predictive power across different body samples collected25. Results should be interpreted with caution, especially when adequate equilibrium periods have not been established for animals on a specific diet 23,26.

The utility of RBC as diet component indicator was limited for the duration of this study. Bovine RBC lifespan is relatively long, between 130 and 160 d, indicating that the isotopic composition in RBC may not reach an equilibrium until this time period 27,28,29, therefore an experiment extending beyond 160 d may be more representative of RBCs produced from nutrients absorbed from the diets provided. In essence, RBC δ13C composition is more greatly influenced by what the animal has consumed throughout a longer time frame 30. Other potential uses of RBC δ13C or δ15N compositions include the capability of estimating foraging and migratory behavior of waterfowl22 or other wildlife24.

Stable isotope analyses of keratinized tissues, such as hair, have been useful for dietary reconstruction of ruminant diets 31, as well as other species32. Zazzo, et al. 31 also detected dietary changes through analysis of hair in beef steers consuming C3 or C4 forage-based diets. The use of keratinized tissues offers several advantages since keratin generally contains all the major light elements (C,H, N, O, and S) and can be sampled non-invasively 31. Hair growth rates and length will pose an inherent challenge to the accurate representation of the animal’s diet, especially since these vary between breed and individuals 31. For the purposes of this study, new growth was always collected since all animals were shaved from the same location at each evaluation period. Even though by 32 d there was greater correlation between observed and predicted RP proportions, the CCC remained low, indicating the values predicted were still less than the observed. This might indicate the dietary history of the steers through a wider period of time, rather than being indicative of more recent dietary composition. Hair samples may contain dietary information for up to 15 mo., though further studies are required to establish adequate equilibration periods, as well as to establish differences among breeds 31,33.

In conclusion, the accurate establishment of isotopic equilibrium across body samples is still required to assure accuracy of utilizing stable isotopes for predicting diet composition of cattle consuming C3–C4 binary mixed diets. Each sample type has contrasting equilibrium times, which in turn, may reflect relative diet composition across different time scales (e.g. days, weeks, or months etc.). Estimating diet intake proportions in the short term (e.g. 8 days after diet change) was best utilizing fecal δ13C, compared to all other sample types evaluated. In the longer term (e.g. 32-days after diet change), feces δ13C and δ15N were both considered adequate predictors of diet composition. Despite plasma and hair showing high correlation between predicted and actual values, the estimates were not adequate enough to provide accurate predictions of diet composition in C3–C4 binary mixtures at 32-days after diet change. The nature of protein turnover and deposition in plasma and hair requires a larger window of time before they represent the animal’s diet. Provided that fecal sample collection is non-invasive, does not require additional processing equipment, or trained personnel, it is thus concluded that fecal δ13C continues to be the optimal method to estimate diet proportions in C3–C4 binary diets. Further evaluation into the usefulness of fecal δ15N for diet prediction is warranted, as this has not been yet evaluated in beef cattle.

Materials and methods

Animals, housing, and treatments

All procedures involving animals were approved by the University of Florida Institutional Animal Care and Use Committee (Protocol #201709925). All methods were performance in accordance with the relevant guidelines and regulations, and permission and informed consent was obtained from the University of Florida (owners) for the use of the steers in this experiment.

The experiment was carried out during July and August of 2017 at the Feed Efficiency Facility of University of Florida, North Florida Research and Education Center, located in Marianna, Florida (30°52′N, 85°11″W, 35 m asl). Both ‘Argentine’ bahiagrass and ‘Florigraze’ rhizoma peanut hays were obtained from commercial producers. The hay bales were stored in enclosed barns throughout the duration of the experiment.

Twenty-five Brahman × Angus crossbred steers (Bos sp.) were utilized (average BW = 341 ± 17 kg, approx. 16 months of age). The steers were grazing bermudagrass (Cynodon dactylon) pastures, a C4 grass, prior to the start of the study. The day prior to the start of the experiment (e.g. day-1), steers brought to working facilities, where they remained 16 h off feed and water, in order to obtain shrunk bodyweights. On day 0 of the experiment, steers were weighed, blocked by bodyweight, and allocated to five treatments (5 steers per treatment) and housed in grouped pens. Hay intake was recorded utilizing GrowSafe© systems (GrowSafe Systems Ltd., Calgary, AB, Canada), which utilize radio frequency identification to record feed intake by weight change measured to the nearest gram. Water was available ad libitum. Forage treatments were offered ad libitum by providing sufficient hay to maintain full feed troughs throughout each day of the experiment. Treatments were five proportions of ‘Florigraze’ rhizoma peanut hay in ‘Argentine’ bahiagrass hay: (1) 100% bahiagrass hay (0% RP); (2) 25% rhizoma peanut hay + 75% bahiagrass hay (25% RP); (3) 50% rhizoma peanut hay + 50% bahiagrass hay (50% RP); (4) 75% rhizoma peanut hay + 25% bahiagrass hay (75% RP); (5) 100% rhizoma peanut hay (100% RP). Diet chemical composition is presented in Table 1. All treatment proportions were weighed and mixed on as-fed basis. Mixing of diets was done manually; no hay mixers or choppers were used, to minimize leaf shatter.

Sample collection

Steers were housed for 32 days and sampling occurred on 0, 8, 16, 24, and 32 days after initiation of treatment diets; exception was for feces, which were collected on d-1 given steers were fasted on d-0 of the experiment. The hay mixtures offered to the steers were collected (10 samples of each diet) and analyzed for nutritive value (Table 1), at the start of the experiment. All sampling occurred between 0700 and 1000 h on each of the sampling days.

Fecal samples were collected directly from the rectum and placed in quart-sized plastic bags to avoid contamination. The feces were frozen at −20 °C. All fecal samples were thawed, dried at 55 °C for 72 h, and ground to pass a 2-mm stainless steel screen using a Wiley Mill (Model 4, Thomas-Wiley Laboratory Mill, Thomas Scientific, Swedesboro, NJ, USA). Samples were then ball milled using a Mixer Mill MM400 (Retsch GmbH, Haan, Germany) at 25 Hz for 9 min.

Blood was obtained through jugular venipuncture using 10-mL K2 EDTA vials (Becton Dickinson and Company, Franklin Lakes, NJ, USA), and stored in ice until centrifugation. All tubes were centrifuged at 714 G for 20 min using an Allegra X-22R centrifuge (Beckman Coulter, Brea, CA, USA). A 10-mL sample of plasma was collected and placed in a separate glass vial, the remaining plasma, white blood cell, and platelet fractions were discarded. The remaining RBC pellet was re-suspended with 9 vol. 0.9% NaCl solution and mixed at room temperature for 15 min at 2 Hz orbital shaker. The tubes were then centrifuged at 714 G for 20 min. The saline solution from the centrifuged tubes was discarded after centrifugation. The rinse procedure was repeated two more times for a total of three rinses. After the third rinse procedure, a 500-µL sample was removed, frozen at −20 °C, and subsequently freeze-dried for isotopic analyses.

Hair clippings were obtained from an area of 20 × 20 cm on the left hindquarter, utilizing veterinary hair clippers (Sunbeam-Oster Inc., Boca Raton, FL, USA). Hair clippings were collected, placed in nylon bags (Ankom Technology, Macedon, NY, USA), and frozen for subsequent analysis. Clippings were always collected in the same location from each animal in order to ensure new hair growth would be analyzed. All hair samples were cleaned using soapy water and defatted following protocols for other keratin-based tissues 31,34. Each sample was sonicated twice for 30 min in a methanol and chloroform solution (2:1, v/v), rinsed with distilled water, and oven dried overnight at 60 °C. Each hair sample was ball milled using a Mixer Mill MM400 (Retsch GmbH, Haan, Germany) at 25 Hz for 9 min.

Calculations

After processing, all samples were analyzed for total C and N using a CHNS analyzer through the Dumas dry combustion method (Vario MicroCube, Elementar Americas Inc., Ronkonkoma, NY, USA) coupled to an isotope ratio mass spectrometer (IsoPrime 100, Elementar, Elementar Americas Inc., Ronkonkoma, NY, USA). Certified standards of L-glutamic acid (USGS40, USGS41; United States Geological Survey) were used for isotope ratio mass spectrometer calibration. Isotope ratios were as follows: δ13C of −26.39, + 37.63‰, and δ15N of −4.52, 47.57‰ for USGS40 and USGS41, respectively. Calibration of the IRMS was conducted according to Cook, et al. 35, with an accuracy of ≤ 0.06 ‰ for 15N and ≤ 0.08 ‰ for 13C.

The isotope ratio for 13C/12C was calculated as:

$$\delta^{{{13}}} {\text{C}} = \, \left( {^{{{13}}} {\text{C}}/^{{{12}}} {\text{C}}_{{{\text{sample}}}} {-}^{{{13}}} {\text{C}}/^{{{12}}} {\text{C}}_{{{\text{reference}}}} } \right)/ \, \left( {^{{{13}}} {\text{C}}/^{{{12}}} {\text{C}}_{{{\text{reference}}}} \times { 1}000} \right)$$
(1)

where δ13C is the C isotope ratio of the sample relative to Pee Dee Belemnite (PDB) standard (‰), 13C/12Csample is the C isotope ratio of the sample, and 13C/12Creference is the C isotope ratio of PDB standard 5. The isotope ratio for 15N/14N was calculated as:

$$\delta^{{{15}}} {\text{N}} = \, \left( {^{{{15}}} {\text{N}}/^{{{14}}} {\text{N}}_{{{\text{sample}}}} -^{{{15}}} {\text{N}}/^{{{14}}} {\text{N}}_{{{\text{reference}}}} } \right)/\left( {^{{{15}}} {\text{N}}/^{{{14}}} {\text{N}}_{{{\text{reference}}}} \times { 1}000} \right)$$
(2)

where δ15N is the N isotope ratio of the sample relative to atmospheric nitrogen (‰), 15N/14Nsample is the N isotope ratio of the sample, and 15N/14Nreference is the N isotope ratio of atmospheric N (standard) 5. The fraction factor (Δ) in this study was considered to be the difference between the diet isotopic composition (δ13C or δ15N) and that of the given sample 5.

The dietary proportion of rhizoma peanut hay was back-calculated using δ13C and δ15N of each plant on a DM basis 3. This method is advantageous in that it does not require further tissue processing and facilitates implementation at the field scale. The proportion of rhizoma peanut was estimated using Eq. (3), described by Jones et al. 3:

$$\%RP=100-\left\{100 \times \frac{A-C}{B-C}\right\}$$
(3)

where %RP is the proportion of RP in the diet, A is the δ13C or δ15N of the sample obtained, B is the δ13C or δ15N of bahiagrass, and C is the δ13C or δ15N of RP.

Statistical analysis

All response variables were analyzed using linear mixed model procedures as implemented in SAS PROC GLIMMIX (SAS/STAT 15.1, SAS Institute). Individual animals were considered the experimental unit. Treatment, collection day, and their interaction were considered fixed effects, and block was considered a random effect in the model. The data were analyzed as repeated measures, considering collection day as the repeated measure. The best covariance matrix was selected according to the lowest AICC fit statistic. Least squares treatment means were compared through pairwise t test using the PDIFF option of the LSMEAN statement in the aforementioned procedure. Based on the recommendations by Milliken and Johnson 36 and Saville 37, no adjustment for multiple comparisons was made. Orthogonal polynomial contrasts (linear and quadratic effects) were used to test effects of RP inclusion on isotopic responses. The slice option was used when the treatment × collection day interaction was significant (P ≤ 0.05), using collection day as the factor, to test treatment effects across collection days. Significance was declared at P ≤ 0.05. The contrast statement was used to test for linear or quadratic effects. Regression analyses for the dietary predictions were conducted using PROC REG from SAS.

Predictions of dietary proportions based on Eq. (3) were generated for 16 subgroups reflecting combinations of isotope (13C or 15N), day (8 or 32), and sample type (feces, plasm, RBC, or hair). Analyses comparing predicted versus actual diet proportions included several components. First, we computed the concordance correlation coefficient (CCC) following the recommendations from Crawford, et al. 38. The CCC is a measure of agreement that encompasses both precision and accuracy, along with corresponding 95% bias accelerated and corrected (BCa) bootstrap confidence intervals. For comparative purposes we calculated the Pearson correlation coefficient which only reflects precision. Both correlation coefficients range from −1.0 to 1.0 and we interpreted values ≥ 0.80 as indicating strong positive agreement/correlation. Next, we regressed the actual percentages on the predicted percentages using linear regression. Perfect prediction corresponds to the estimated regression line having an intercept of zero and a slope of 1.0. We then partitioned the mean square error (MSE) of the predicted (from Eq. (3), not the above linear regression) versus actual percentages as described in Rice and Cochran 39. This partitioning entails calculating the proportion of MSE attributable to three sources of error: the difference in mean predicted and actual values (mean component, denoted “MC”), the error resulting from the slope of the above linear regression deviating from 1.0 (slope component, denoted “SC”), and random error (random component, denoted “RC”). The results from the above analyses were examined to identify subgroups whose predictions were sufficiently good to warrant hypothesis testing. In this context “good” means that the predicted percentages were strongly correlated with the actual percentages and the magnitudes of the predicted percentages were similar to the actual percentages. The objective of the hypothesis testing was to formally evaluate whether dietary proportions could be directly predicted from Eq. (3) (in contrast to generating predictions using the equation from regressing actual dietary percentages on the predicted percentages from Eq. (3)). Paired two one-sided test (TOST) equivalence tests were conducted for the selected subgroups with α = 0.0540. These tests are formulated such that the null hypothesis is “non-equivalence” and the alternative hypothesis is “equivalence”. An input parameter to the test is the equivalence region, a range for which we consider the mean actual minus predicted difference to be unimportant (“equivalent”) from a practical standpoint. For each equivalence test we also computed the 90% confidence interval for the mean actual minus predicted difference which we refer to as the “minimum equivalence region”. Based on the structure of TOST equivalence tests, to reject the null hypothesis at the 0.05 level, the equivalence region specified for the test must completely contain the minimum equivalence region. For example, if the pre-specified equivalence region is (−15%, 15%) and the computed minimum equivalence region is (−16%, −6%) the null hypothesis would not be rejected. Finally, we re-ran all of the analyses described above for the selected subgroups where, prior to analysis, predicted percentages outside of the valid range were assigned the appropriate boundary value (i.e., predicted percentages < 0% were assigned a value of 0% and those > 100% were assigned a value of 100%).