Introduction

Freshwater ecosystems are suffering from increasing anthropogenic impacts, either directly through introduction of alien species (Garcia-Berthou et al., 2005; Gorokhova et al., 2005) or indirectly via catchment alterations such as increased input of nutrients (Vadeboncoeur et al., 2003; Jeppesen et al., 2007). For some of the impacted freshwater ecosystems, historical monitoring programmes including archived sample material exist. The interest in using such archived material for stable isotope analyses has risen as it allows reconstruction or assessment of the ecological impact of ecosystem changes (Vander Zanden et al., 2003; Chandra et al., 2005; Chasar et al., 2005). Isotopic data can be obtained from any tissue, and samples not collected specifically for dietary studies may therefore be used for food web reconstructions (e.g. Ventura & Jeppesen, 2009). Among these tissues, those allowing non-destructive samplings of the studied individuals are ecologically and ethically most valuable (Estep & Vigg, 1985; Cerling et al., 2006; Ehleringer et al., 2008). Fish scales are of particular interest since they fulfil both characteristics, and therefore their use for ecological reconstructions has become increasingly popular (Estep & Vigg, 1985; Wainright et al., 1993; Satterfield & Finney, 2002; Pruell et al., 2003; Estrada et al., 2005; Kennedy et al., 2005; Gerdeaux & Perga, 2006; Morbey et al., 2007; Grey et al., 2009; Rennie et al., 2009).

Fish muscle is the reference tissue used in most ecological studies and is normally analysed after lipid removal. Basically, fish scales have two components: an organic part formed by two proteins, collagen and ichthylepidin (Seshaiya et al., 1963), in approximately equal proportions, and an inorganic matrix composed mainly of hydroxylapatite (Hutchinson & Trueman, 2006), a calcium phosphate biomineral with different residual mineral contaminations such as carbonates and magnesium (Elliott, 2002). For carbon stable isotope analysis, it is the calcium carbonate fraction that potentially interferes with the stable isotope analysis of proteins. The relative importance of calcium carbonate within the mineral fraction of scales has been only scarcely described. So far, it has been shown that, at least in otoliths, most calcium carbonate comes from dissolved inorganic carbon (DIC; Solomon et al., 2006; Tohse & Mugiya, 2008). Therefore, the use of scales as a substitute or a surrogate of muscle for stable isotope analysis requires either removal of the scale mineral matrix (Estep & Vigg, 1985; Perga & Gerdeaux, 2003; Grey et al., 2009), as commonly done by archaeologists to describe changes in stable isotope composition of bone or other animal tissues (e.g. Chisholm et al., 1983; Lee-thorp et al., 1989), or species-specific calibration of scale and muscle for each habitat (Kelly et al., 2006; Sinnatamby et al., 2008; Syväranta et al., 2008; Blanco et al., 2009; Rennie et al., 2009). Only few studies have specifically tested the effects of acidification on scale carbon and nitrogen stable isotope composition, and opposite conclusions have been reached. In whitefish (Coregonus lavaretus), Perga & Gerdeaux (2003) found significant (1.3‰) enrichment of carbon and nitrogen isotopic composition after acidification. Syväranta et al. (2008) found significant effects of acidification on roach δ13C, although they were smaller than the analytical error of the instrument and therefore biologically irrelevant. Sinnatamby et al. (2007) in yellow perch (Perca flavescens), walleye (Sander vitreus) and Atlantic salmon (Salmo salar) and Rennie et al. (2009) in North American whitefish found no significant acidification effects, though the former argued that there might be a species-specific difference in the response to acid treatment.

The different results found in previous studies may be attributed to variations in scale mineral content among species. In addition, more than 75% of scale carbonates come from DIC (Solomon et al., 2006; Tohse & Mugiya, 2008), while all the carbon from scale proteins come from the diet. Therefore, the relative difference between δ13C-DIC and δ13C-diet could be another factor affecting either among or within species effects of scale acidification on δ13C. Since δ13C changes only negligibly along food webs (Peterson & Fry, 1987), the difference between a δ13C-DIC and a δ13C-diet will most likely depend on the difference in fractionation by algae during photosynthesis, ranging between almost 0 to over 20‰ depending on the extent of carbon limitation (Ventura et al., 2008), as earlier suggested by Perga & Gerdeaux (2004), Gerdeaux & Perga (2006) and Grey et al. (2009) based on their observations of strong correlation between total phosphorus (Ptot) and fish scale δ13C.

In order to evaluate further the need for scale acidification prior to stable isotope analysis, we compared acidified and non-acidified scales of six freshwater fish species with contrasting mineral content (from ca. 30 to 60%) sampled along a gradient of lakes varying in nutrient content. Specifically, our objectives were to: (i) investigate the acidification time needed to remove the mineral part of most scales types; (ii) describe the scale elemental (carbon, nitrogen and phosphorus), mineral and protein contents; (iii) determine the effects of acidification on the scale stable isotope composition of carbon and nitrogen, and (iv) investigate the relationships between the possible effects with the scale mineral content, lake productivity and CO2 concentration.

Materials and methods

Scales were obtained from six different freshwater fish species common in north European temperate lakes: perch (Perca fluviatilis), bream (Abramis brama), pike (Esox lucius), rudd (Scardinius erythrophtalmus), roach (Rutilus rutilus) and tench (Tinca tinca). Fish samples were collected between 2000 and 2006 from six Danish lakes with contrasting nutrient loading from the catchment and, accordingly, with differing gradients of isotopic baselines (Vander Zanden et al., 2005). Therefore, the lakes cover a wide trophic gradient (from oligo- to hypertrophic) and also a strong gradient of CO2 concentrations, basically driven by the productivity derived pH (Table 1). Most species were found in at least three lakes, with the exception of bream and tench that were present in only one lake each (Table 2). Scales were collected in the field, placed in paper envelopes and stored dry until analysis.

Table 1 Surface area, mean depth, average summer (May–October) chlorophyll a (Chl a), total nitrogen (Ntot), total phosphorus (Ptot), the elemental ratio of total nitrogen to total phosphorus, pH, alkalinity and free-CO2 concentration of the study lakes
Full size table
Table 2 Number of individuals analysed, lakes where each species was found with number of individuals in each lake (in parenthesis) and mean fork-length (cm) and summary statistics of the different freshwater fish species analysed
Full size table

Scales were selected from individuals belonging to five different size classes (1+ to 5+) ranging from ca. 8 to 35 cm. In the case of pike, similar age classes corresponded to larger size (16–90 cm). For tench, we could only find large individuals from the latter age class (35–54 cm). Skin mucus was removed from all scales by soaking in distilled water and gentle scraping with a scalpel under a binocular microscope. In order to avoid noise from interscale isotopic differences, scales were spliced into two longitudinal halves, one half being directly dried at ambient temperatures for bulk scale analysis, while the other half was first decalcified, rinsed in distilled water and subsequently dried for analysis. Due to the change in scale weight with fish age, we had to use a different number of scales for each age class and species. The number of scales used ranged from 25 for the +1 fishes to 1 for fishes >27 cm.

Scales of the different fish species exhibited different degrees of mineralisation. Therefore, we investigated the optimal scale decalcification time by soaking scales of roach (moderately mineralised scale) and perch (highly mineralised scale) in 1.2 M HCl. Scales were removed successively at 2 min intervals starting at 2 min, soaked in distilled water, dried at air temperature and weighed. The optimal acidification time was chosen to be that at which scale weight loss stopped. Since carbonates are embedded within apatite, and both are similarly solubilised when soaked in acid, we measured apatite as a measure of the remaining mineral fraction. Initial and remaining apatite contents were estimated from the scale phosphorus content.

For stable isotope analysis we used ca. 1.5 and 2.0 mg of the acidified and untreated scales, respectively. The scales were pre-weighed in tin capsules. From the scales of the larger age classes, we used only a longitudinal piece that was sliced off at the central half of the scale to obtain a representative piece (Hutchinson & Trueman, 2006). Scales were cut into fine slivers using sterile laboratory scissors. Elemental carbon and nitrogen of solid ground material and stable isotope ratios of carbon and nitrogen were analysed at the Stable Isotope Facility, University of Davis, California, USA, using a Europa Hydra 20/20 continuous flow isotope ratio mass spectrometer (CFIRMS; PDZ Europa, Cheshire, UK) coupled with an elemental analyser. Values for 13C, 12C, 15N and 14N were calculated and reported using the standard delta (δ) notation in parts per thousand (‰) as follows:

$$ \delta X = \left[ {\left( {R_{\text{sample}} /R_{\text{standard}} } \right) - 1} \right] \times 1000, $$

where X is δ13C or δ15N and R is the corresponding ratio 13C:12C or 15N:14N. Two laboratory standards were included after every 12 samples. We used laboratory standards suitable for the types of samples and their C and N contents, including NIST 1547 peach leaves, NIST 1577b bovine liver, acetanilide, cellulose, glycine, sucrose and ammonium sulphate. Laboratory standards were calibrated against NIST Standard Reference Materials (IAEA-N1, IAEA-N2, IAEA-N3, IAEA-CH7, and NBS-22). Reproducibility was better than 0.1 and 0.3‰ of absolute difference for 13C and 15N, respectively.

Phosphorus in scales was analysed as orthophosphate after 10% HCl digestion (Grasshoff, 1983). Carbon, nitrogen and phosphorus contents were expressed in percentage dry weight. The mineral content of scales was measured as the weight loss after acidification. Within the mineral fraction, we assumed that scales were composed of apatite and carbonates (assumed to be the main compound of the residual fraction). Apatite was assumed to be all hydroxylapatite [Ca5(PO4)3OH; molecular weight = 502.32; Elliott, 2002] and was estimated from total phosphorus in the scales by multiplying with the inverse of phosphorus mass fraction in hydroxylapatite (5.41). Carbonates were estimated as the difference between mineral fraction and apatite. Inorganic carbon of carbonates was calculated assuming that the carbonates were calcium carbonate. Scale protein was estimated from scale nitrogen by multiplying with the inverse of nitrogen mass fraction in fish scale protein (5.31). Scale proteins are composed of collagen and ichthylepidin in approximately equal proportions (Seshaiya et al., 1963), and their average nitrogen content was calculated from the relative amino acid composition and the stoichiometry of the different amino acids in fish scales as described by Ventura (2006).

We calculated the carbon stable isotope composition of the DIC fraction of carbonates from the weighted difference of the carbon stable isotope composition of the untreated scales (δ13Cuntreated, which have both carbonate and protein carbon) and acidified scales (δ13Cacidified, which have only proteins) using the following equations:

$$ \delta^{13} {\text{C}}_{\text{untreated}} = \left( {1 - {\text{C}}_{\text{ca}} } \right)*\delta^{13} {\text{C}}_{\text{acidified}} + {\text{C}}_{\text{ca}} *\delta^{13} {\text{C}}_{\text{carbonates}} , $$
(1)

where Cca is the mass fraction of carbonate carbon in the scale. Since only a proportion of the carbon of scale carbonates originates from DIC, we corrected for the proportion of DIC carbon of scales to:

$$ \delta^{13} {\text{C}}_{\text{carbonates}} = f_{\text{DIC}} *\delta^{13} {\text{C}}_{\text{DIC}} + \left( {1 - \, f_{\text{DIC}} } \right)*\delta^{13} {\text{C}}_{\text{acidified}} , $$
(2)

where f DIC is the mass fraction of carbonate carbon originating from DIC. This fraction was assumed to be 0.8 in all the species of this study (Solomon et al., 2006; Tohse & Mugiya, 2008). By inserting Eq. 2 into Eq. 1 and rearranging it, we obtain the equation for the estimation of the carbon stable isotope composition of DIC (δ13CDIC):

$$ \delta {}^{13}{\text{C}}_{\text{DIC}} = {\frac{{\delta {}^{13}{\text{C}}_{\text{untreated}} - (1 - f_{\text{ca}} *{\text{C}}_{\text{ca}} ) * \delta {}^{13}{\text{C}}_{\text{acidified}} }}{{f_{\text{DIC}} *{\text{C}}_{\text{ca}} }}}. $$
(3)

A non-parametric Wilcoxon two-sample test was used to test the significance of the mean of the effect of scale decalcification on carbon or nitrogen stable isotope signatures for each of the six fish species studied. Linear regression was used to test for relationships between the acidified and untreated (bulk) samples, and stepwise multiple regression was used to determine relationships with environmental variables (total nitrogen, total phosphorus, chlorophyll a, CO2 concentration, scale mineral content and scale carbonates). Variance partitioning (Legendre & Legendre, 1998) was used to determine the proportion of variance explained exclusively by each significant variable and the proportion of shared variance amongst them. ANCOVA was used to test if slopes differed from 1 and for heterogeneity of slopes (the interaction term of the group variable and the covariate being insignificant). In our case, the slope of the regression resulting from comparing the acidified and non-acidified scales was compared to a slope of 1, resulting from the regression comparing acidified scales in the two axes. All statistical analyses were conducted with SPSS 17.

Results

Optimal scale acidification time and average scale composition

We found that a minimum of 6 min were required to remove almost all the mineral content from the scales (Fig. 1a), the initial apatite content being ca. 35% and afterwards ≤1% of scale dry weight (DW) in the six species studied.

Fig. 1
figure 1

Scale decalcification (measured as the relative weight loss) with varying acidification time (a). The vertical dotted line indicates the time at which the scales were most decalcified. The two species tested were representative of a species with highly calcified scales (perch, Perca fluviatilis) and a species with moderate scale calcification (roach, Rutilus rutilus). (b) Average of the six species studied of the apatite content in bulk scales (untreated) and those previously acidified during 6 min

Full size image

The scale mineral content of the six species ranged between 31.8% in tench to 61.3% DW in perch. The other four species had a similar mineral content (ca. 45% DW; Table 3). The elemental composition was characterised by high amounts of elemental phosphorus due to apatite, ranging from 4.5 to 9.1% DW. The nitrogen content ranged between 6.6% DW in perch to 12.7% DW in tench. Carbon ranged from 17.8 to 31.3% DW (Table 3). The protein content ranged between 35% in perch to 67.6% in tench. Apatite was the most abundant mineral fraction (range 24.4–49.2% DW), the residual fraction varying from 7.4% DW in tench to 13.1% DW in roach (Table 3). We calculated the inorganic carbon in carbonates assuming that carbonates are the only mineral fraction in scales apart from apatite. The average ± SD was 1.3 ± 0.2% DW (range 0.9–1.6), a very small proportion of the total carbon in scales (average ± SD: 5.5 ± 1.7% of total carbon in scales).

Table 3 Scale elemental composition of phosphorus, nitrogen and carbon, and the relative proportions of apatite and protein content
Full size table

Effects of acidification

The average effect of acidification was very small (0.181 ± 0.122 and −0.208 ± 0.243; average ± SD for all species for carbon and nitrogen, respectively) albeit significant for five of our six study species (perch, roach, rudd, pike and bream). It was non-significant for tench, the species with the lowest mineral content. The effect differed between the two isotopes, acidification increased δ13C (range from 0.02 to 0.398) and decreased δ15N (range from −0.553 to 0.162; Table 4). Linear regression slopes between acidified and untreated scales were not significantly different from one for all species but one (Figs. 2, 3; Table 5), the only exception being roach that despite a slope of 0.979 differed significantly from 1 for δ15N. Equality of slopes with one suggested that the effect of acidification could be corrected by subtracting the average effect of acidification.

Table 4 Effects of acidification on freshwater fish scale δ13C and δ15N
Full size table
Fig. 2
figure 2

Non-acidified plotted against acidified scale stable isotope δ13C signatures for the six freshwater fish species studied. The solid line is the fit of a linear regression model and the dotted line is the 1:1 relationship. All regressions are significant (see Table 5 for regression parameters and statistical tests)

Full size image
Fig. 3
figure 3

Non-acidified plotted against acidified scale stable isotope δ15N signatures for the six freshwater fish species studied. The solid line is the fit of a linear regression model and the dotted line is the 1:1 relationship. All regressions are significant (see Table 5 for regression parameters and statistical tests)

Full size image
Table 5 Regression models relating the δ13C and δ15N of non-acidified and acidified fish scales and ANCOVA tests (with F and P values) for slopes differing from 1 and intercepts differing from 0
Full size table

Relationships with mineral content, lake productivity and CO2 concentration

Since the different species analysed had different scale mineral content and the different lakes had a wide range of nutrient content and CO2 concentrations, we explored whether the acidification effect on the stable isotope composition of the scales might be related to mineral content, carbonates only, lake surface area, mean depth, nutrients and CO2 concentrations. We used stepwise multiple regression to select a model with the combination of independent variables that best described the acidification effect on scale δ13C and δ15N. For the acidification effect on δ13C, the best model had carbonates and CO2 concentration as significant variables, explaining 58% of the variation (Table 6). CO2 exclusively explained more variance than carbonates (28.3 and 21.1%, respectively) with 8.4% of shared variance. For the acidification effect on δ15N, the significant variables were total phosphorus and carbonates, explaining 52.2% of the variation, of which 22.1% derived from total phosphorus, 16.1% from carbonates, while 14% of the variance was shared.

Table 6 Multiple regression analyses relating the effects of scale acidification on δ13C and δ15N (Δδ13Cuntreated–acidified and Δδ15Nuntreated–acidified, respectively), the difference between the stable isotope composition of carbon of DIC and scale proteins (Δδ13CDIC–acidified) and the stable isotope composition of scale proteins (δ13Cacidified and δ15Nacidified) to selected environmental variables
Full size table

A graphical representation of the relationships between the most significant variable in each isotope and acidification effect revealed one outlier. This point belonged to tench, the only species that had no significant acidification effect (Fig. 4a). Therefore, we repeated the stepwise multiple regression without tench. In this case, CO2 and total phosphorus were the only significant variables for δ13C and δ15N (Fig. 5a), respectively, while carbonates were no longer significant (Table 6). A comparison of the acidification effect on carbon isotope composition with the difference between the carbon stable isotope composition of DIC and scale proteins (Δδ13CDIC–acidified) revealed a close relationship (r 2 = 0.923; P < 0.001) with the exception of tench. For this species, no acidification effect was found, but the Δδ13CDIC–acidified was large (Fig. 4b), and therefore tench was not an outlier in the relationship between Δδ13CDIC–acidified and CO2 (Fig. 4c).

Fig. 4
figure 4

Relationships between the effect of acidification on scale stable isotope composition of carbon (∆δ13Cuntreated–acidified) with (a) average summer free CO2 concentration, (b) with the difference in the stable isotope composition of scale carbonates and protein, in our case acidified scales (∆δ13Ccarbonates–acidified); (c) ∆δ13Ccarbonates–acidified with average summer-free CO2 concentration. See “Materials and methods” section for a description of the procedure for calculating the stable isotope composition of carbonates. (d) Relationship between the carbon stable isotope composition of acidified scales (δ13Cacidified) and total nitrogen (Ntot). Explanatory variables were selected among those of Table 1, scale mineral content and scale carbonates by stepwise linear regression (see Table 6 for regression equations). The regression of (a) excludes tench (in black), while the other two regressions (c, d) include tench

Full size image
Fig. 5
figure 5

Relationship between the effect of acidification on scale stable isotope composition of nitrogen (∆δ15Nuntreated–acidified) with total phosphorus (a). The regression excludes tench (in black), which had no acidification effect due to the low scale mineral content. Relationship between the nitrogen stable isotope compositions of acidified scales (δ15Nacidified) with total nitrogen (b). Explanatory variables were selected among those of Table 1, scale mineral content and scale carbonates by stepwise linear regression (see Table 6 for regressions equation). In (b), total nitrogen was selected together with total phosphorus and CO2 concentration and was the variable explaining the largest proportion of variance

Full size image

We also used the stepwise procedure to select the combination of variables (amongst those of Table 1, carbonates and mineral content) that best explained the variability of scale proteins. Ntot was the variable that explained most variance for both carbon and nitrogen (73% and 41%, respectively), but, in addition, Ptot and CO2 explained part of the variation (37% and 14%, respectively; Table 6) for nitrogen.

Discussion

In this study, we found small but significant effects of acidification on scale carbon and nitrogen stable isotope composition (average change of +0.18 and −0.21‰ for carbon and nitrogen, respectively) in five of the six species studied (perch, roach, rudd, pike and bream). Previous studies have traced either strong acidification effects (+1.3‰ for both carbon and nitrogen) on whitefish (C. lavaretus; Perga & Gerdeaux, 2003), small but biologically irrelevant effects on roach (Syväranta et al., 2008) and no effects after scale acidification on yellow perch (P. flavescens), walleye (S. vitreus), Atlantic salmon (S. salar; Sinnatamby et al., 2007) and North American whitefish (Coregonus clupeaformis; Rennie et al., 2009). Variability in among species scale characteristics might be one of the explanations for the differences occurring in our study and the other published studies. In this respect, the positive relationship between the acidification effect and the scale mineral content revealed by comparing the results among the six species of this study suggests that an important factor for the among species variability is related to the different mineral content of fish scales. In our study, we found that for fish species with a mineral content lower than 20% (or 7% of carbonates), there are no significant effects of scale acidification, while for the other species, it was close to the analytical error of the instrument and therefore on the limit of being biologically irrelevant.

Other factors may also be responsible for this among species variability, such as incomplete scale decalcification or variability in δ13C-DIC. Perga & Gerdeaux (2003) showed that for whitefish, there was no carbonate left after 2 min acidification. In contrast, Sinnatamby et al. (2007) acidified the scales of three other fish species for 2 min, which might not have been sufficient for decalcifying all species. This could be the case of yellow perch, belonging to the same genus as perch measured in this study and whose scales were heavily calcified.

Our study also revealed that within-species variability in δ13C was positively correlated with differences in lake CO2 concentration. The variance of acidification effect explained by CO2 concentration was 60%, which is high considering that not all carbonates originate from DIC (between 75 and 85%; Solomon et al., 2006; Tohse & Mugiya, 2008). Previous studies have identified a relationship between lake primary productivity and scale δ13C (Perga & Gerdeaux, 2004; Gerdeaux & Perga, 2006; Grey et al., 2009) and have attributed it mostly to a change in phytoplankton fractionation, but also to some extent to a change in the contribution of respired carbon and to an increase in the DIC pool δ13C by Rayleigh distillation kinetics and input of atmospheric CO2, amplified by the alkaline nature of the waters. Perga & Gerdeaux (2004) found that δ13C-DIC was the main factor explaining the change in the scale δ13C, establishing a correlation with Ptot only in large lakes. Our results show that the dual analysis of acidified scales and the difference between acidified and untreated scales could be a tool to distinguish between the effects of changing δ13C-DIC and alterations in algal fractionation due to changing CO2 concentration. Although the procedure used in our study for calculating δ13C-DIC might be less sensitive than capturing CO2 with cryogenic distillation after acidification for posterior analysis (Lee-thorp et al., 1989), it has the practical advantage of being methodologically simple. We found the precision of the method sufficient to capture the fractionation effects created by variable CO2 concentrations.

Similarly to carbon, we observed significant changes, proportional to the mineral content of scales, in scale δ15N after acidification. In their study, Perga & Gerdeaux (2003) also found significant and more prominent changes in scale δ15N after acidification than in our study. Acidification with diluted HCl is unlikely to alter protein nitrogen as proteins are hydrolysed with concentrated HCl at high temperatures (Fountoulakis & Lahm, 1998). In addition, Perga & Gerdeaux (2003) acidified their scales for a shorter time than in our study. The change in δ15N with the mineral content among the species investigated here indicate that fishes deposit organic nitrogenated compounds together with apatite and carbonates when building scales. This compound likely differs from the other scale proteins (collagen and ichthylepidin) by being more attached to the mineral fraction and is concordingly released when apatite is removed by acidification. Furthermore, the similar correlation between Δδ15Nuntreated–acidified and δ15Nacidified with nutrients also indicates that the fraction of nitrogen removed by acidification is of the same biogenic origin as scale proteins. Therefore, acidification effects on nitrogen provide no additional information than the analysis of scale proteins alone.