Estimating ethanol correction factors for δ13C and δ15N isotopic signatures of freshwater zooplankton from multiple lakes

In freshwater systems, δ13C and δ15N stable isotopes can be used to differentiate between pelagic and littoral energy sources and to quantify trophic position. In these ecosystems, crustacean zooplankton are frequently used to characterize the pelagic baseline. Zooplankton samples are often preserved prior to processing and analysis, which can affect isotopic signatures. Variability in preservation effects across studies make it difficult to determine if and how to correct for preservation effects. Here, we develop a correction factor for ethanol preservation and present a flexible statistical method that can be updated with additional data to increase its applicability. We collected zooplankton from five lakes in Minnesota, USA encompassing wide isotopic ranges (δ13C from −37.23‰ to −23.96‰; δ15N from 3.07‰ to 14.44‰). Changes in zooplankton δ13C and δ15N signatures were quantified using a Bayesian hierarchical model predicting fresh values from ethanol‐preserved values. Ethanol preservation increased δ13C by a factor of 1.158 (95% CI 0.866–1.441) and had a negligible effect on δ15N (slope = 1.077; 95% CI 0.833–1.359). Lake‐specific values did not differ from the overall relationship. K‐fold and leave‐one‐out cross validation tests verified that both models were accurate; RMSE of predicted δ13C = 0.701 and RMSE of predicted δ15N = 0.590. Our correction factors could be applied to other systems in which baseline δ13C and δ15N values fall within the range of our study, and this approach also enables the inclusion of data from additional lakes to estimate new corrections.

Stable isotope analysis is an important tool for understanding food webs and quantifying the contribution of various energy pathways within an ecosystem.Understanding these pathways is increasingly important as stressors like climate change and aquatic invasive species are rapidly altering aquatic food webs (Havel et al. 2015;Bartley et al. 2019).Stable isotope analysis of aquatic food webs typically relies on δ 13 C and δ 15 N signatures to determine reliance on littoral vs. pelagic energy sources and trophic position (Peterson and Fry 1987;Bearhop et al. 2004).Zooplankton communities primarily consist of primary consumers and are often used to quantify the pelagic baseline for aquatic food webs (Vander Zanden andRasmussen 1999, Matthews andMazumder 2003).As a baseline, δ 13 C and δ 15 N values of zooplankton are critical for calculating other organisms' trophic positions and food web connections (Post 2002;Bearhop et al. 2004).Factors that could potentially bias or decrease the accuracy of baseline isotopic signatures are important to understand and quantify for accuracy in modeling consumer trophic interactions.
Prior to analysis of stable isotope composition, zooplankton samples are often sorted under microscopes to remove predator species, phytoplankton, and detritus to obtain an accurate measurement of primary consumers (Vander Zanden and Rasmussen 1999).However, sorting zooplankton immediately upon collection is not always feasible due to time and gear constraints.Preservation of zooplankton in ethanol allows for temporary storage, but ethanol storage alters isotopic ratios in organic samples when stored for even short periods of time (Feuchtmayr and Grey 2003;Ventura and Jeppesen 2009).Previous studies have analyzed preservation effects of ethanol on a variety of taxa, but only a few have focused on zooplankton specifically (Feuchtmayr and Grey 2003;Syvaranta et al. 2008;Ventura and Jeppesen 2009).Published correction factors for zooplankton fixation in ethanol are variable; results of ethanol exposure on δ 13 C and δ 15 N vary, geographic representation is quite isolated, and experimental design differs between studies (Table 1).Because zooplankton taxa vary by geographic location, coefficients from these existing corrections are accurate but less relevant to the communities studied here (Vanderploeg et al. 1992;Arts 1999).Thus, using correction factors from other regions or taxa may introduce more uncertainty.Most of these studies-both zooplankton focused and otherwise-concluded that although the effects of ethanol fixation may not be statistically significant, they have the potential to impact conclusions of stable isotope analysis research by introducing a new source of error into commonly used stable isotope mixing models (Silberberger et al. 2021).Existing corrections, while accurate, are not as relevant in cases where zooplankton from northern temperate lakes are preserved in ethanol, a common situation in stable isotope studies (e.g., Bethke et al. 2023;Hoffman et al. 2010).
Changes in isotopic signatures associated with preservation are due in part to lipid extraction.While fixation in lipophilic ethanol provides some degree of lipid extraction, it is less polar than methanol, commonly used for lipid extraction, and there is often no filtering or washing of samples to remove the lipids (Folch et al. 1957;Bligh and Dyer 1959;Manirakiza et al. 2001).Because of this, it is important to note that correcting for preservation effects is not a lipid correction and-though the effects may be similar-should not be used in place of one (see Supplement).The use of a separate lipid correction in conjunction with this fixation correction is at the discretion of the researcher.Several frequently cited studies provide lipid corrections that encompass a broad range of species designed to normalize lipid values across food webs.However, there is disagreement over whether enriched δ 13 C signatures in lipids need to be corrected to avoid bias in commonly used stable isotope mixing models, or if doing so would result in inaccurate interpretations of trophic interactions (Mintenbeck et al. 2008;Arostegui et al. 2019).While these lipid corrections do exist, there is a need for similar standardized corrections for ethanol storage as it relates to stable isotope values (Sweeting et al. 2006;Post et al. 2007;Smyntek et al. 2007).In this study, we compare δ 13 C and δ 15 N values of fresh (not preserved) and ethanol-preserved crustacean zooplankton to quantify correction factors that could be applied to zooplankton from other northern temperate lakes.We use a Bayesian hierarchical model that estimates lake-specific and overall correction factors and can be easily adapted to include data from additional lakes.

Sampling
Sampling was conducted during early to mid summer when zooplankton are typically the most productive in northern temperate lakes (De Senerpont Domis et al. 2013).Zooplankton were collected from five lakes in Minnesota, USA, that represent a range of trophic status, morphometrics, and invasion status among other factors (Table 2).Zooplankton were sampled from five locations in each lake which were chosen randomly to account for intra-lake isotopic variation while simultaneously targeting lake depths > 10 m.Samples were collected via vertical tow of an 80 μm mesh zooplankton net to ensure a complete representation of the stratified water column.
Upon collection, samples were placed in 250 mL Nalgene bottles and allowed to sit in lake water for a minimum of 3 h to allow for gut clearance, as failure to do so can result in isotopic errors > 3‰ (Feuchtmayr and Grey 2003).Samples were transported on ice to maintain freshness and viability before being run through 850, 250, and 120 μm fourinch zooplankton sieves once gut clearance was completed, removing the majority of debris and excess algae (Fig. 1).Samples were paired by site, so control ("fresh") subsamples of the filtered material were immediately placed in 2.0 mL VWR microcentrifuge tubes and dried at 60 C for 24 h to be processed, while the remainder was placed in reagent grade ethanol (90.25% ethanol, 4.75% methanol, and 5% isopropanol).Ethanol subsamples remained in solution for $ 30 d before also being dried at 60 C for 24 h.All samples were then ground to a fine powder and encapsulated in 5 Â 9 mm Costech tin capsules at weights of 1 mg (AE 0.2) for combined δ 13 C and δ 15 N stable isotope analysis at the UC-Davis Stable Isotope Facility.Samples were analyzed using an Elementar vario MICRO cube elemental analyzer (Elementar Analysensysteme GmbH, Langenselbold, Germany) interfaced to a Sercon Europa 20-20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK).Analytical precision (standard deviation) was within AE 0.2‰ for δ 13 C and AE 0.3‰ for δ 15 N.

Statistical analysis
Treated and control values for both isotopes (δ 13 C and δ 15 N) were centered around their respective grand mean to remove intercepts from the correction owing to the distance between the data and the axes.Centered data were then analyzed using Bayesian hierarchical models for both δ 13 C and δ 15 N to quantify the change in isotopic values and determine the associated correction factors where β 0 represents the population intercept, β 1 represents the population slope, and b 0 and b 1 are grouplevel intercepts and slopes, respectively (Eq. 1).Individual samples and lakes are represented as the ith sample in the jth group.We used uninformative (flat) slope priors, weakly informative (student t-test) intercept priors, and set the MCMC parameters to: chain length = 4, burn-in = 50,000, thin = 1, and iterations = 100,000.Lakes were included as a group-level effect variable within the model, allowing unique slope and intercept coefficients to be determined for each lake.Ethanol treatments represented the population-level effect across all samples.Population-level intercepts in the models were set to zero given that the data was centered.
K-fold Cross Validation tests with five folds using each individual lake as a test group and leave-one-out cross validation tests were performed to test model accuracy and range.Hypothesis tests were conducted to determine the significance of model output by testing slopes against a value of one which would be representative of no change between fresh and ethanol treatments.Additionally, RMSE (residual mean squared error) values were calculated for the differences between observed values and data fit to both the hierarchical model and a 1 : 1 identity line, representative of no change during treatment.
An additional δ 13 C correction incorporating a common mass balance lipid normalization is provided for cases when both ethanol storage and lipid normalization is desired (see Supporting Information Data S1).The mass balance model shown in Eq. 2 uses the average difference in δ 13 C between proteins and lipids, D, along with atomic C : N ratios to estimate δ 13 C of lipid extracted samples (Δ 13 C ex ) allowing for correction of 13 C depletion in fatty acids, where bulk 13 C and C : N represent fresh, unaltered values (Fry et al. 2003;Smyntek et al. 2007).
Values of D and C : N ex were 6.3 (AE 1.3) and 4.2 (AE 0.4), respectively (Smyntek et al. 2007).Lipid normalized control (fresh) and ethanol fixed δ 13 C values were centered and analyzed using Bayesian hierarchical models identical to those used to develop the combined fixation and lipid normalization correction.Similarly, K-fold cross validation and leaveone-out cross validation tests were performed to test model accuracy and range, and RMSE values were calculated comparing corrected values to values fit on a 1 : 1 line.All statistical analyses were performed using R Statistical Software and the BRMS package (v4.3.0,Bürkner 2017; R Core Team 2023).

Assessment
Population-level ethanol treatments resulted in an estimated increase of 1.158 per unit (95% CI: 0.866-1.441)on δ 13 C and 1.077 per unit (95% CI: 0.833-1.359)on δ 15 N (Fig. 2; Table 3).Hypothesis tests provided strong evidence that both slopes are in fact greater than one, yielding probabilities of 91% and 81% for δ 13 C and δ 15 N, respectively (Table 4).Group-level slope and intercept means were all within the region of practical equivalence, indicating that the coefficients are unlikely to vary by lake (Fig. 3).
Mean fresh sample values were À30.82‰ (SD AE 3.78) for δ 13 C and 7.09‰ (SD AE 3.61) for δ 15 N, and mean treated sample values were À29.13‰ (SD AE 3.23) for δ 13 C and 6.92‰ (SD AE 3.30) for δ 15 N. Slope values determined from the population-level effects of the mixed models and sample means were used to form a correction equation that could be applied to preserved zooplankton samples to establish accurate isotopic baselines (Eqs.3 and 4).
Fresh δ 15 N ¼ 1:077 * EtOH δ 15 N À 6:913 Convergence parameters including R-hat and ESS (effective sample size) indicate both models successfully converged (Table 3).Both models fit extremely well with 98% of the response being explained by the ethanol treatment for both isotopes (Table 3).K-fold cross validation tests verified that both models were able to accurately predict data from new lakes; the δ 13 C model had an RMSE of 0.701‰ and the δ 15 N model had an RMSE of 0.590‰.Comparisons of RMSE values when testing fits between the observed data and either the hierarchical model or a 1 : 1 line indicated that data corrected using the models were more accurate than if data were not corrected at all (0.61 vs. 1.86‰ for δ 13 C, 0.53 vs. 0.61‰ for δ 15 N).Leave-one-out cross validation showed model confidence was high for both isotopes in all lakes except δ 13 C in Lake Alexander.Two samples from Lake Alexander had the highest δ 13 C values observed in our study, so leaving these samples out of model construction in the leave-one-out cross validation resulted in generating predictions outside of the range of data used to construct the model.Both points had Pareto k values > 0.7, so the importance  sampling was not able to provide a useful estimate for those observations as their individual influence on the posterior distributions was extremely high (0.977 and 0.845).This indicates that those isotopic values likely fall outside the accurate range of the model.RMSE for the leave-one-out cross validation was 0.47‰ for both isotopes.The correction including lipid normalization resulted in estimated population level effects of a 1.28 (95% CI: 0.74-1.83)increase per unit on δ 13 C. Similar to the fixation models, the fixation and lipid normalization model showed high model confidence for all but two points from Lake Alexander, which had Pareto k values > 0.7.K-fold and leaveone-out cross validation tests resulted in RMSE of 1.088‰ and 0.634‰, respectively (see Supporting Information Data S1).

Discussion
Our results show that the effect of ethanol fixation on zooplankton stable isotope values is linear in nature and consistent between lakes.In the context of food web analysis, if unaccounted for, these changes would lead to an overestimation of reliance on littoral energy sources by higher level consumers (δ 13 C), and a greater change in trophic position between zooplankton and secondary consumers (δ 15 N).The use of two separate model validation methods offers different insights into its accuracy; low RMSE values for the K-fold cross validation grouped by lake mean the model is accurate when predicting new data from different systems, while the leave-one-out cross validation provides better insight into the useful range of the model.Overall, the model appears to be most accurate within a range of around À26 to À39‰ for δ 13 C and 2 to 14‰ for δ 15 N. Discretion should be used when predicted values fall outside of that range (Fig. 4).
Linear relationships comparing lipid normalized control (fresh) values to both lipid normalized and lipid and fixation corrected samples indicated the presence of a fixation effect not accounted for by the lipid normalization.A correction containing lipid normalization is provided to account for this variation; it is integrated into the storage correction as it is based solely on δ 13 C given that relationships were not as highly correlated for atomic carbon, nitrogen, or C : N ratios.Similar to the results from the storage corrections, K-fold and leave-one-out cross validation tests indicated the model containing the lipid normalization accurately predicted values from new systems and was accurate within a range of roughly À25 to À37‰ for δ 13 C (see Supporting Information Data S1).
Including multiple lakes with varying baseline isotopic values increases the transferability of our study to other systems.Measured δ 15 N values had an overall range of 11.37‰ (3.07 to 14.44), and δ 13 C values had an overall range of 13.27‰ (À37.23 to À23.96) (Fig. 5).The effects of ethanol preservation on δ 13 C were consistent across lakes, while δ 15 N of preserved and unpreserved samples consistently overlap.Even with this much initial isotopic variation between lakes, it is highly unlikely that individual correction factors are necessary.Treated carbon values differed from fresh carbon values by an average of 1.69‰ (AE 0.79) across all sites while nitrogen values only differed by À0.17‰ (AE 0.60).Significant alteration in the δ 13 C values was likely caused by changes in lipid content due to the exposure of lipophilic ethanol (Post et al. 2007).Based on average isotopic shifts between trophic levels shown in literature-3.4‰δ 15 N and 0.4‰ δ 13 C-and the differences in accuracy between the hierarchical models and identity lines, we feel it is necessary to apply the correction factor to δ 13 C given the significant change in isotope ratios after ethanol exposure, while δ 15 N did not show a response of the same magnitude and therefore does not need to be corrected (Post 2002).
Having paired samples from multiple lakes and filtering samples instead of sorting them allowed for a community based zooplankton correction factor as opposed to corrections within individual taxa.While isotopic signatures vary between zooplankton taxa (Feuchtmayr and Grey 2003), we wanted to ensure the samples were representative of the zooplankton community which feature primarily Daphnia spp, Calanoida spp, and Cyclopoida spp.Given the consistency of zooplankton communities present in water bodies throughout the northern temperate region, the treatment should be similar for most lakes (Heiskary et al. 2016).Effects of ethanol storage on individual species could be explored further in future work, but it is unlikely to be necessary given the results on a community scale.Our method of processing samples through a series of filters appears to have been effective.The final filter size of 120 μm was small enough to have removed any large detritus or additional species (e.g., Chaoborus and Leptodora) and large enough to allow phytoplankton to be washed through the filter with deionized water.Filtered samples were still visually inspected for any larger unwanted material that made it through the filtering process.Predatory species like Leptodora usually occur at a low frequency-not reaching maximum population densities until summer months-and appear to have been removed by the filtration process, so they are unlikely to have any effect on results (Herzig and Auer 1990).This method additionally prevents any species selection bias from occurring during the sorting process (Ershova et al. 2021).Samples submitted from Lake Alexander contained minimal zooplankton material and there was some difficulty removing all material from the filters.Therefore, this resulted in the loss of two of the five ethanol treated data points and may explain the increased influence of the remaining three points on the model's posterior predictions as indicated by high Pareto k values in the cross validation.
In our comparison, preserved samples were exposed to ethanol for 1 month.Under the assumption that the mechanism causing changes in δ 13 C is the dissolution of lipids during ethanol exposure, it is likely that individual zooplankton were entirely saturated, and all effects had taken place by the end of the exposure period (Ventura and Jeppesen 2009).Our results including initial isotopic diversity within the data, strong correlations in the models, and insignificance of the random lake effect suggest our samples and correction factor should be representative of and applicable to northern temperate lakes dominated by cladoceran and copepod communities.
For scenarios where those conditions are not met, the procedures described above can be followed for quantifying corrections for different zooplankton communities or other storage methods.Additional data collected using these procedures can be added to the existing data set and model to estimate custom lake-specific corrections, increasing the transferability and robustness of the model in the process.Existing code and data are publicly available at the Data Repository for University of Minnesota (Blechinger et al. 2023).

Comments and recommendations
We provide a correction factor that could be applied to other studies in which zooplankton are preserved in ethanol prior to stable isotope analysis, archived samples where atomic C : N ratios are not available, lipid correction is not desired, or study sites are in northern temperate lakes.The most important recommendation when utilizing this correction factor for δ 13 C is to verify that experimental conditions are similar to those used in this study; preserving samples in reagent grade ethanol and conducting experiments in geographic regions where similar zooplankton communities are present is essential to avoid introducing additional bias.Ideally, relevant experimental conditions should be identified when designing the experiment to ensure this correction is applicable.Under the proper circumstances, using this to correct isotope values for zooplankton storage methods can save valuable researcher time and money.If this correction does not apply to relevant research (e.g., isotope values fall outside of model range), the methods described can be implemented for other scenarios.Data obtained following the outlined procedures can be included in the model, providing updated correction factors for additional lakes or storage methods.
This correction factor can be used in two ways; the model can be rerun with new data for different lakes or storage methods and then used to predict fresh values, or the correction provided by the existing model can be directly applied to ethanol-fixed values.In order to use the correction as-is, follow Eqs. 3 and 4 using the means and slopes provided.We recommend this approach when uncorrected isotopic values fall within the range used in the existing model.When re-running the model for new ethanol corrections, data can be added to the existing data set to assess whether a lake-specific correction is needed, and potentially to update the general correction factors in Eqs. 3 and 4.Although a δ 15 N correction is provided, we do not recommend applying it to data given the ecological insignificance of the observed isotopic changes due to storage.While the mechanisms responsible for isotopic shifts during ethanol fixation are similar to those of lipid corrections, this is not a lipid correction; samples do not undergo complete lipid extraction, and lipid corrected values are best predicted when both a fixation and lipid correction are applied simultaneously (see Supporting Information Data S1).

Fig. 2 .
Fig. 2. Bayesian hierarchical model output showing the relationships between fresh and ethanol treated zooplankton samples (solid red) with a 1 : 1 identity line (black dashed) included for reference for carbon-13 and nitrogen-15.95% credible intervals are shown in blue.

Fig. 3 .
Fig. 3. Posterior distributions of lake-specific slope and intercept offsets from the population-level means.Black dots and horizontal lines indicate the mean of each distribution and their 95% credible intervals.Areas within the region of practical equivalence (equal to zero, no difference from population mean) are shown in blue, areas outside the ROPE are shown in red.

Fig. 5 .
Fig. 5. Dual isotope plot of stable isotope analysis results for all zooplankton samples paired by site.

Table 1 .
Review of studies analyzing ethanol storage on zooplankton stable isotope ratios.

Table 2 .
Characteristics of lakes sampled.DOW is a Divison of Water lake identification number, TSI is a trophic state index value.

Table 3 .
Bayesian hierarchical model output and convergence parameters.ESS is the effective sample size, MCSE is the Monte Carlo standard error.

Table 4 .
Model hypothesis test results.LCI and UCI are lower and upper confidence intervals, respectively.