Quality control for modern bone collagen stable carbon and nitrogen isotope measurements

Isotopic analyses of collagen, the main protein preserved in subfossil bone and tooth, has long provided a powerful tool for the reconstruction of ancient diets and environments. Although isotopic studies of contemporary ecosystems have typically focused on more accessible tissues (e.g. muscle, hair), there is growing interest in the potential for analyses of collagen because it is often available in hard tissue archives (e.g. scales, skin, bone, tooth), allowing for enhanced long‐term retrospective studies. The quality of measurements of the stable carbon and nitrogen isotopic compositions of ancient samples is subject to robust and well‐established criteria for detection of contaminants and diagenesis. Among these quality control (QC) criteria, the most widely utilized is the atomic C:N ratio (C:NAtomic), which for ancient samples has an acceptable range between 2.9 and 3.6. While this QC criterion was developed for ancient materials, it has increasingly being applied to collagen from modern tissues. Here, we use a large survey of published collagen amino acid compositions (n = 436) from 193 vertebrate species as well as recent experimental isotopic evidence from 413 modern collagen extracts to demonstrate that the C:NAtomic range used for ancient samples is not suitable for assessing collagen quality of modern and archived historical samples. For modern tissues, collagen C:NAtomic falling outside 3.00–3.30 for fish and 3.00–3.28 for mammals and birds can produce systematically skewed isotopic compositions and may lead to significant interpretative errors. These findings are followed by a review of protocols for improving C:NAtomic criteria for modern collagen extracts. Given the tremendous conservation and environmental policy‐informing potential that retrospective isotopic analyses of collagen from contemporary and archived vertebrate tissues have for addressing pressing questions about long‐term environmental conditions and species behaviours, it is critical that QC criteria tailored to modern tissues are established.


| INTRODUC TI ON
Stable carbon (δ 13 C) and nitrogen (δ 15 N) isotope analysis of plant and animal tissues is a powerful tool for quantifying ecological relationships and animal behaviour at a wide range of spatial and temporal scales.
Although the isotopic compositions of any carbon-and nitrogen-bearing tissue can be analysed, ecologists studying contemporary and historical ecosystems generally focus analyses on readily available muscle and organ tissues that require relatively little processing (Newsome, Clementz, & Koch, 2010). In contrast, research investigating earlier ecosystems has been extensively focused on analysing the isotopic composition of proteins extracted from hard tissues, such as bones and teeth, which are more readily available because they are generally the only vertebrate structures that preserve in archaeological and paleontological deposits (Ambrose, 1990). In this context, Type I collagen (hereafter, 'collagen'), which makes up 90% of bone and tooth protein (Herring, 1972), has been the primary focus of a vast majority of protein-based isotopic research on ancient environments (Guiry, 2019;Szpak, Metcalfe, & Macdonald, 2017).
Collagen has a number of well-established quality control (QC) indicators for isotopic measurements of ancient materials (Table 2), which enable detection of contaminating non-collagenous materials and post-burial degradation/alteration (Ambrose, 1990;DeNiro, 1985). These QC indicators were established based on observations of anomalies in the relationship between the isotopic and elemental compositions of modern and ancient collagen (for reviews see, Szpak, 2011;Van Klinken, 1999), which show that the concentrations and ratios of carbon and nitrogen vary little between collagen from different tissues and taxa (Neuman, 1949). Two of the Q C criteria for ancient collagen are yield (the wt% collagen extracted from bone, with 1% being the most frequently cited minimum value) and minima for wt% C (13%) and wt% N (4.5%) in the analysed collagen. Both of these criteria identify degraded, rather than contaminated, collagen TA B L E 1 Terminology for collagen sourced from different types of specimens along with some of the major contamination sources common to each and are therefore only applicable to ancient samples. For ecologists, the most relevant collagen QC criteria will be the ratio of percent carbon-to-nitrogen expressed as C:N Atomic . For studies analysing several specimens from a single taxon, an observation of a correlation between C:N Atomic and δ 13 C can provide an additional indicator for collagen contamination (Ambrose, 1990). C:N Atomic represents that ratio of carbon to nitrogen atoms in a sample and can be calculated by multiplying the molecular C:N (i.e. wt% C/wt% N as measured during elemental analyses) by the ratio between the average atomic mass of C and N (14.007/12.011). Because C:N Atomic can be calculated using elemental data that are determined in tandem with δ 13 C and δ 15 N at no extra cost, this collagen QC measure is also easily obtainable.
Moreover, the elemental data used to calculate C:N Atomic are resilient to inter-lab differences in calibration (even if the absolute wt% C and N differ among labs, the C:N Atomic will be the same provided the calibration standard used contained both C and N).
While the C:N Atomic collagen QC indicator was initially established to aid with identification of bone diagenesis in archaeological and paleontological research (DeNiro, 1985), it has also become the primary means used by ecologists to evaluate the integrity of isotopic compositions measured on modern collagen samples (e.g. Bas & Cardona, 2018;Turner Tomaszewicz, Seminoff, Avens, & Kurle, 2016). However, analyses of collagen from ancient and contemporary samples require different analytical considerations that render QC indicators developed from the former unsuitable for latter. For archaeological collagen, shifts in C:N Atomic are typically caused by the presence of additional unwanted endogenous (e.g. non-collagenous materials such as lipids) or exogenous (e.g. humic acids from the burial environment) carbon sources or collagen deterioration (selective amino acid loss) resulting in a disproportionate loss of either carbon or nitrogen (for reviews, see Collins & Galley, 1998;Collins et al., 2002;Collins, Riley, Child, & Turner-Walker, 1995;Van Klinken, 1999). A broader C:N Atomic range of 2.9-3.6 was therefore deemed acceptable for archaeological collagen specifically to account for the possibility of the selective loss of certain amino acids through centuries or millennia of leaching, hydrolysis and microbial activity. The rationale behind selecting this range of cut-offs was that the isotopic compositions of ancient samples with C:N Atomic within the 2.9-3.6 range appeared to be unaltered by contamination or degradation and therefore are less likely to present interpretive issues (Ambrose, 1990;DeNiro, 1985;note that Van Klinken, 1999 recommended a narrower range from 3.1 to 3.5 but this has not been widley adpoted). Although there has been relatively little research on the effect of selective amino acid loss on the isotopic composition of ancient collagen (although see, Dobberstein et al., 2009)

| Establishing an observed range for vertebrate collagen C:N Atomic
To establish the natural range of C:N Atomic for vertebrate collagen, we surveyed a wide body of data from the food chemistry literature.
Amino acid compositions are from both acid and pepsin solubilized collagens. Comparing C:N Atomic ratios from studies that performed both methods on collagen from 13 species, Szpak (2011) found that amino acid compositions measured using acid and pepsin solubilized collagen were nearly identical with no statistically significant differences. Data from both types of collagen extraction techniques should therefore produce amino acid compositions that are comparable for our purposes. Because fish collagens are adapted to environmental conditions (Eastoe, 1957;Gustavson, 1955), we further grouped fish based on habitat preferences using classifications provided on FishBase (Froese & Pauly, 2000)  we categorized by tissue type (skin, scale, bone) and by taxonomic class (Table 3). All amino acid compositions were compared as residues per 1,000.

| Establishing an acceptable range for C:N Atomic
To establish a C:N Atomic range within which modern collagen stable isotope compositions have not been meaningfully skewed by contamination with non-collagenous materials, we used δ 13 C and C:N Atomic data from recent studies Szpak & Guiry, in prep.) comparing collagen extracted from modern bones prepared following different protocols designed to produce collagen contaminated to var-

| Statistics
Statistical analyses were performed with PAST Version 3.22 (Hammer, Harper, & Ryan, 2001). For amino acid residue and C:N Atomic data, we used a Shapiro-Wilk test to assess normality of distribution (Supporting Information 1,

| RE SULTS
Excluding two Antarctic icefish (Chionodraco hamatus and Racovitzia glacialis, with unique physiological adaptations; see, Szpak, 2011), mean collagen C:N Atomic for all species and tissues based on amino acid compositions was 3.17 ± 0.08 and ranged from 3.00 to 3.33 (Table 4; Figure 1, for full list see Supporting Information 1, Table S1). Before TA B L E 3 Number of species and analyses included in survey of skin, bone and scale collagen amino acid compositions shown by taxonomic class (see Supporting Information 1,   Figure S1) relative to warmer water fish. Lower C:N Atomic and hydroxylproline abundances and higher serine abundances in the collagen of cold water fish taxa is a well-established observation and is likely related to thermal properties of collagens adapted to colder environments (Rigby, 1967;Rigby & Spikes, 1960 A strong correlation was found between C:N Atomic and δ 13 C in our comparison of experimental data from fish (Spearman's ρ = −0.875, p < 0.001) and mammals and bird (Spearman's ρ = −0.718, p < 0.001) bone collagen with varying degrees of lipid contamination (Figure 2).
To establish the cut-off point at which C:N Atomic can be used to indicate contamination with non-collagenous materials, we compared C:N Atomic and δ 13 C for data grouped by C:N Atomic into cumulative iterations starting at 3.10 and increasing by intervals of +0.01 (i.e. 3.10-3.11, 3.10-3.12, 3.10-3.13 and so on) until significant correlations were identified.
Significant correlations were not found in C:N Atomic and δ 13 C for sample groups with a C:N Atomic of 3.30 and lower for fish and 3.28 and lower for mammals and birds. All groups including C:N Atomic values >3.30, for fish, and 3.28, for mammals and birds, show significant correlations F I G U R E 1 Collagen C:N Atomic for ray-finned fish (Actinopterygii), mammals (Mammalia) and birds (Aves) observed in survey of amino acid compositions (n = 382; for data see Supporting Information 1, Table S1)

F I G U R E 2
Relationship between lipid contamination, as indicated by C:N Atomic , and negative skewing of collagen δ 13 C. Plots compare data generated by recent studies Szpak & Guiry, in prep.) on the effects of collagen extraction methods on the elemental and isotopic compositions of 122 bones from 85 fish, mammal and bird specimens. Four to five extraction procedures were applied to subsamples from each bone. Within each group of four to five samples per bone, the δ 13 C of the sample with the lowest C:N Atomic was subtracted from the δ 13 C of the other samples and are plotted against their respective C:N Atomic . The shaded box shows the acceptable range based on Spearman's ρ (see Supporting Information 1, Table S3) (p < 0.05) with correlation strength (as defined by Spearman's ρ) growing as higher C:N Atomic comparisons are included (see Supporting Information 1, Table S3).
We also found a strong negative correlation between  Table S4). The utility of these cut-off values is supported by their close agreement with the observed C:N Atomic means for fish skin, bone and scale collagen (Table 4, n = 290, 3.16 ± 0.06, range = 3.00-3.30) as well as mammal and bird skin and bone collagens (Table 4, n = 95, 3.22 ± 0.04; range = 3.11-3.33). With respect to using C:N Atomic to evaluate the quality of isotopic measurements made on bone collagen, we can therefore use 3.30 for fish and 3.28 for mammals and birds as the upper limit (a cut-off value) for acceptable stable carbon and nitrogen isotope compositions.

| Establishing C:N Atomic QC criteria for fish, mammal and bird bone collagen
Establishing a lower cut-off C:N Atomic value is comparatively straightforward because the main sources of contamination for collagen for contemporary and archived historical materials is likely to be carbon rich and will therefore cause an increase, rather than a decrease, in C:N Atomic . As outlined above, these contamination sources include lipids as well as mineral (for bone), neither of which contain a substantive nitrogen component (although some lipids, such as phosphatidylcholines have a single N atom), as well as NCPs.
Owing to its higher glycine content (with its low C:N Atomic of 2), relative to most other proteins, collagen also has a lower C:N Atomic than potential sources of NCP contamination (Table 5 Table  S4. Inset shows enlargement of area highlighted in green F I G U R E 4 Mean amino acid compositions of fish (n = 287), mammal (n = 86) and bird (n = 9) collagen (see Supporting Information 1, Table S1 for data). Gly not shown in order to improve visualization of scaling for comparison of other amino acids and 3.00 based on the survey of published amino acid composition assays (n = 290, Supporting Information 1, Table S1). The lowest observed C:N Atomic observed in mammal and bird bone collagen is 3.15 based on analyses of modern collagen extracts by Szpak and Guiry (in prep.) and 3.11 based on the survey of published amino acid composition assays (n = 95, Supporting Information 1, Table S1).
However, the sample size of published amino acid compositions of mammals and birds is smaller in comparison to fish and therefore may not capture the full range of variation in C:N Atomic . For this reason, a lower C:N Atomic limit of 3.00 for fish, mammals and birds can be established as a conservative acceptable collagen stable carbon and nitrogen isotope compositions.
It is important to bear in mind that the ranges of acceptable C:N Atomic identified here are only a general guideline and that even collagen extracts with C:N Atomic that fall within this acceptable range may still have skewed isotopic compositions. Significant correlations in C:N Atomic and δ 13 C shifts among different extracts from the same samples show that elevations in C:N Atomic of as little as 0.03 can be accompanied by significant negative shifts in δ 13 C. For instance, on average a C:N Atomic increase of 0.25, which would fall within the envelope of acceptable C:N Atomic identified here, was associated with δ 13 C values skewed by approximately 0.5‰. For this reason, it is critical that close attention is paid to optimizing collagen purification protocols.

| Best practices for modem bone collagen extractions
With respect to characterizing the isotopic composition of collagen from any tissue, it is critical that sample pre-treatment protocols are  Table 4; Supporting Information 1, Table S6) of non-collagenous proteins prevalent in bone compared to the mean for collagen (all species, n = 436, Supporting Information 1). *Amino acids commonly considered to undergo trophic 15 N enrichment (O'Connell, 2017) TA B L E 5 Details for major non-collagenous proteins found in ossified tissues. C:N Atomic were calculated based on mean amino acid counts from complete and reviewed amino acid sequences available from The Uniprot Consortium (2018; see Supporting  Information 1, Table S5) and were processed using Bioedit v 7.2 (Hall, 1999). Because complete amino acid sequences may include small signal peptides (usually 16-30 amino acids long) not found in the mature protein, calculated C:N Atomic will deviate slightly from the true C:N Atomic and are intended as estimates illustrating broad variability in elemental composition between collagen and different non-collagenous proteins. Unless otherwise noted, molecular weight data are from Robey and Boskey (2013) (Hall, 1999).
Recently, there has been discussion in the literature suggesting that analysing δ 15 N of whole bone powder produces more reliable results relative to analyses of extracted and purified bone collagen (e.g. Bas, García, Crespo, & Cardona, 2019). While there are some cases where δ 15 N analyses of whole bone of tooth powder may be a desirable alternative to analyses of extracted bone collagen, these are limited to circumstances wherein small sample size negates the ability to extract a sufficient amount of purified collagen from the sample (e.g. Fahy et al., 2014;Guiry, Hepburn, & Richards, 2016;Guiry, Jones, et al., 2018;Rossman et al., 2015). In most cases, however, analyses of δ 15 N from untreated bone powder samples will de facto produce less predictable and more heterogeneous results.
Incomplete removal of NCPs creates two major problems for interpreting the isotopic composition of whole bone powder. First, the relative proportions of major NCPs are known to vary significantly within and among bones, individuals and species (e.g. Gorski, 1998;Roach, 1994). This variation in the relative proportions of different NCPs means that the contribution of nitrogen from different amino acids cannot be anticipated. Second, the amino acid compositions of these NCPs each differ from that of collagen (Table 5; Fisher, Hawkins, Tuross, & Termine, 1987;Gundberg et al., 1984), with differing proportions of 'source' and 'trophic' amino acids ( Figure 5) that, in turn, will skew the δ 15 N of whole bone protein relative to extracted, purified collagen. These two levels of variation (in relative proportions of different NCPs and their variable contributions of amino acids with different trophic discrimination factors) will necessarily result in more heterogeneous and less predictable δ 15 N values from analyses of whole bone protein.
In addition to removing the mineral and lipid fractions of bones and teeth, collagen extraction protocols help to remove NCPs, thereby producing a more consistently homogeneous material that is better suited for comparing isotopic compositions within and among anatomical elements, individuals and species. In particular, the demineralization and refluxing steps should remove a substantial fraction NCPs, like osteocalcin, which are tightly bound in the mineral phase (bioapatite) of bone and dentine (Gundberg et al., 1984). Other important NCPs, such as osteonectin, that have a strong affinity for binding to both mineral and collagen, are also at least partly removed through standard HCl-or EDTA-based collagen extraction protocols (e.g. Romberg, Werness, Lollar, Riggs, & Mann, 1985;Termine et al., 1981). Although there has been little work quantifying the NCP composition of bone collagen extracts prepared using different protocols (although see 2018), it is likely that NCPs, while present, remain in very low quantities (Linde, Bhown, & Butler, 1981).
Additional steps have also been recommended to further purify bone collagen and may help to improve collagen QC indicators. Brown, Nelson, Vogel, and Southon (1988) recommended 'ultrafiltration' with 10 or 30 kDa molecular weight cut-off (MWCO) filters for the extraction of collagen for AMS radiocarbon dating, a suggestion that has widely been taken up for collagen extraction protocols for IRMS analyses (Dobberstein et al., 2009;Sealy, Johnson, Richards, & Nehlich, 2014). Ultrafiltration removes low molecular weight con-  Ganss et al., 1999) are too large for removal through ultrafiltration (even at the 30 kDa MWCO). One exception is ostecalcin (5.8 kDa; Price et al., 1976), a much smaller mineral-bound NCP, but this molecule will likely already have been effectively removed through demineralization (Gundberg et al., 1984). For this reason, ultrafiltration is unlikely to improve NCP removal for collagen purification.
A more effective method for removal of NCPs may be the addition of a NaOH pre-treatment step (Lowry, Gilligan, & Katersky, 1941), which is routinely applied for collagen purification in food chemistry in the process of characterizing collagen amino acid compositions (e.g. Nagai & Suzuki, 2000). This step is also commonly applied during collagen extractions from archaeological materials (between the demineralization and refluxing steps) because it removes base-soluble contaminants like humic acids derived from the burial environment. While use of the NaOH pre-treatment step can reduce collagen yields in ancient samples (Chisholm, Nelson, Hobson, Schwarcz, & Knyf, 1983), this should not be an issue for well-preserved, modern bones. The NaOH step does not induce selective loss of amino acids (Katzenberg, 1989;Kennedy, 1988) and, for this reason, should not impact collagen isotopic compositions (Ambrose, 1990).
In summary, isotopic analyses of whole bone should not be sup- they may still present an issue for effective purification of collagen extracts from modern tissues (for further discussion, see Guiry & Hunt, 2020). A NaOH treatment step may serve to further reduce contamination of collagen with NCPs and could be used to help ensure that extracts do not fail the C:N Atomic QC criterion for isotopic analyse of modern collagen samples.

| CON CLUS IONS
Stable carbon and nitrogen isotope analyses of collagen from bones, scales and other tissues are becoming increasingly important in ecological research for a variety of reasons such as access to archived tissues for retrospective research (Dietl et al., 2015) and providing longer-term intra-individual perspectives for migratory studies (Hobson, 2019). While isotopic analysis of collagen has been an integral and well-established component of archaeological and paleobiological research for decades, techniques for assessing data quality in ancient materials are not directly transferable to analyses of collagen from modern tissues. In this study, we have used observations of amino acid compositions from a wide range of species as well as experiment results from fish, mammal and bird bone collagen isotopic compositions to better characterize C:N Atomic QC criteria for isotopic analyses of modern collagen.
Whereas widely accepted QC criteria for ancient collagen based on C:N Atomic are between 2.9 and 3.6, for modern tissues we establish new cut-offs between 3.00 and 3.30 for fish and 3.00 and 3.28 for mammals and birds. However, it is important to recognize that even for collagen extracts with C:N Atomic falling within this range isotopic compositions could still potentially be skewed and it is therefore critical to optimize collagen purification protocols.
With respect to collagen extraction protocols, we have also reviewed key processes for ensuring that collagen extractions from modern bone are better able to meet these new collagen quality criteria. In particular, it is important that efforts are made not only to remove lipid and mineral contaminants (known to effect δ 13 C) but also to ensure that NCPs are removed as these can have important consequences for δ 13 C and δ 15 N measurements.

E.J.G. has been supported by a Social Sciences and Humanities
Research Council of Canada Banting Post-Doctoral Fellowship.

AUTH O R S ' CO NTR I B UTI O N S
E.J.G.: conceptualization, methodology, investigation, formal analysis, data curation, funding acquisition, visualization, writingoriginal draft, writing-review and editing. P.S.: methodology, writingreview and editing.

DATA AVA I L A B I L I T Y S TAT E M E N T
All data presented in this paper are publicly archived on Dryad (https:// doi.org/10.5061/dryad.ffbg7 9crm;  and are also available in Supporting Information 1 and in referenced sources.