Contents lists available at ScienceDirect
Journal of Archaeological Science: Reports
journal homepage: www.elsevier.com/locate/jasrep
Quantifying collagen quality in archaeological bone: Improving data
accuracy with benchtop and handheld Raman spectrometers
Odile Madden⁎, Dora Man Wai Chan, Morgan Dundon, Christine A.M. France
Smithsonian Museum Conservation Institute, 4210 Silver Hill Road, Suitland, MD 20746, USA
A R T I C L E I N F O
Keywords:
Raman spectroscopy
Handheld
Bone
Collagen
Protein
Hydroxyapatite
Diagenesis
A B S T R A C T
Chemical analysis of collagen and other endogenous protein in excavated bones is ever more common in pa-
leontology and archaeology to determine dietary ecology, migration patterns, age, and diagenetic pathways.
Decisions to sacrifice valuable samples to destructive stable isotope, radiogenic isotope, and proteomic analyses
are easier to make if one knows ahead of time whether sufficient undegraded organic material is preserved in the
bone for chemical analysis. Recent advances with near–infrared Raman spectroscopy to non-destructively pre-
screen bones for the presence of undegraded endogenous protein can indicate whether original isotopic or
proteomic signatures have remained intact in long-buried bones. However, it is crucial to examine the different
emerging approaches with benchtop and handheld Raman spectrometers and to establish uniform data reporting
standards that will allow laboratories to accurately compare results. Three recently published studies are
compared to understand the factors that affect protein screening success and to establish best practices for the
technique in the laboratory and in the field. A set of 37 archaeological human bone samples analyzed previously
by the authors using FT-Raman spectroscopy are re-analyzed with a handheld 1064 nm Raman spectrometer
such that the laser power was similar but the spectral resolution and signal-to-noise were lower. Two methods to
identify peak heights were evaluated: 1) peak height fixed at a specific wavenumber and baseline anchored at
fixed points bracketing the band and 2) averaging peak intensities over a short wavenumber range while also
setting the baseline of the band by an average of several anchor points. This second method appeared to better
classify whether the bones contained well-preserved protein signatures. In particular, peak height ratios of
960 cm−1 to 1636 cm−1 or 1450 cm−1 can indicate quality and abundance of bone protein, but have benefits
and trade-offs that depend on the instrument used. The 960 cm−1:1636 cm−1 ratio is more descriptive of col-
lagen quality as defined by an extracted carbon:nitrogen ratio of 2.8–3.6, but the 1636 cm−1 peak can be dif-
ficult to resolve well with an inherently less sensitive handheld Raman spectrometer. The 1450 cm−1 peak is
more prominent in the spectrum and therefore more easily resolved with a handheld spectrometer, but it de-
scribes a common CeH stretch found in collagen and other organic molecules.
1. Introduction
Chemical analysis of excavated bones is increasingly popular in
paleontology and archaeology. Bones frequently are analyzed using
stable isotopes, radiogenic isotopes and proteomic techniques to de-
termine dietary ecology, migration patterns, age, and diagenetic path-
ways. Decisions to sacrifice valuable samples to these destructive
techniques are easier to make if one knows ahead of time whether
sufficient undegraded, endogenous protein is preserved in the bone for
chemical analysis. Recent advances have been made with near–infrared
Raman spectroscopy to non-destructively pre-screen bones for good
collagen preservation (France et al., 2014a; Halcrow et al., 2014; Pestle
et al., 2014, 2015), which is a strong indicator that original isotopic or
proteomic signatures have remained intact in long-buried bones.
However, these researchers have focused on different criteria for col-
lagen quality and analytical parameters. As this technique continues to
develop it is important to recognize these different approaches, ex-
amine their attributes, and establish uniform data reporting standards
that will allow laboratories to compare results. This is a crucial step
toward becoming a mature science with generally agreed upon stan-
dards that yield accurate and precise protein data (Killick, 2015).
Furthermore, the advent of relatively inexpensive, easy to operate,
handheld Raman spectrometers sets up new challenges for assessing
data quality. The spectral sensitivity and resolution of these handheld
instruments is impressive but nonetheless lower than research-grade
benchtop spectrometers, and these instruments are likely to be
https://doi.org/10.1016/j.jasrep.2017.11.034
Received 15 February 2017; Received in revised form 19 November 2017; Accepted 20 November 2017
⁎ Corresponding author.
E-mail address: MCIweb@si.edu (O. Madden).
Journal of Archaeological Science: Reports 18 (2018) 596–605
2352-409X/ Published by Elsevier Ltd.
T
deployed in the field by archaeologists with varying expertise in spec-
troscopy. Here, three recently published studies are examined, and the
set of bone samples evaluated by France et al. in 2014 (2014a) with a
high-resolution, benchtop FT-Raman spectrometer is re-analyzed with a
handheld 1064 nm Raman spectrometer for comparison.
1.1. Raman spectroscopy of bone
Raman spectroscopy describes the energy distribution of light
scattered inelastically by a bone. The organic and inorganic compo-
nents of bone are made up of atomic bonds that vibrate in characteristic
movements. The bonds might stretch or bend, for example, and each
movement requires a small, very specific amount of energy. In Raman
spectroscopy a laser is directed at the bone. Most of the light reflects off
the bone with the same energy (i.e., color) as the laser light, and no
energy is exchanged. However, a small amount of the laser light does
interact with the bonds in hydroxyapatite, other inorganic salts, type I
collagen, and other proteins in the bone. These bonds siphon off small
packets of energy to make their characteristic movements, and the
photons scattered back to the instrument's detector have slightly shifted
wavelengths. A Raman spectrum is a plot of those shifted energies. Each
peak describes a type of movement, called a vibrational mode, of a
specific atomic bond or group of bonds. Together the peaks in a Raman
spectrum are like a fingerprint of the material. The presence, position,
relative size, or absence of certain peaks can indicate whether a mo-
lecule such as collagen is present, intact, or altered such that it has
become unsuitable for further chemical analysis. A typical bone spec-
trum includes many peaks, the most prominent of which are indicated
with their associated bonds in France et al. (2014a) and are reproduced
here (Fig. 1).
A key factor in Raman spectroscopy of bone is the choice of laser. A
laser is monochromatic, meaning it emits a single color of light, typi-
cally described by its wavelength in nanometers (nm). Visible light
lasers (i.e., 532 nm green and 785 nm red) can trigger fluorescence
signals that distort spectral baselines and obscure the relatively weak
Raman peaks. Instead near infrared lasers are chosen for bone analysis
to avoid interference from fluorescence. For this reason, each of the
recently published studies uses a 1064 nm laser. Other important in-
strument variables are detector sensitivity, spectral resolution, and in-
strument portability, and these are interrelated. The most sensitive
spectrometers with the best ability to discern different vibrations (i.e.,
adjustable resolution up to 1 cm−1) currently are designed for labora-
tory research and are not easily portable. This type of instrument is
described in this article as high-resolution, research-grade, or benchtop.
In addition, the specific instruments used in France et al. (2014a) and
Halcrow et al. (2014) are Fourier-transform Raman (FT-Raman) spec-
trometers; the name describes the mechanism by which the instruments
create a spectrum. Increasingly popular portable and handheld
instruments are lightweight and less expensive, but they have less
sensitive detection, and the resolution is lower (i.e., 8–15 cm−1) and
usually not adjustable. This affects the ability to collect Raman spectra
without damaging a bone and the quality of those spectra. While these
smaller instruments can be used for some research purposes, their
ability to discern subtle differences between materials is more limited.
1.2. Bone collagen
Bone is a composite material composed of inorganic mineral grains
in collagen, which is the main structural protein in bone and also the
traditional catch all term for other organic compounds that are also
present in lesser proportion. As the science of proteins develops ever
more quickly, the ability to distinguish the minor organic components is
improving and outpacing the language traditionally used to describe
the bulk organic matrix. Thus the term collagen is still the convention,
but it is not entirely accurate. In this article the authors have chosen to
use the term collagen in the conventional way, but type I collagen is
specified where appropriate and other endogenous proteins and organic
compounds are as well.
In chemical studies of excavated bone, the presence of undegraded
collagen has long been taken to indicate the overall preservation of
original isotopic, radiogenic, and other chemical signatures. If the
proteinaceous component of the bone is present, intact, and un-
contaminated, it can be used to establish dates for skeletal remains or as
evidence of an individual's environment or diet during life.
Nearly all the protein in a bone is type I collagen, which makes up
approximately 20% of bone weight and is composed of strong and
complex triple helical molecules organized into arrays of fibrils that
protect the hydroxyapatite mineral grains from exposure to ground
water and other diagenetic influences under most burial conditions
(Fratzl et al., 2004; Nelson et al., 1986; Person et al., 1996; Shoulders
and Raines, 2009; Tütken et al., 2008; Veis, 2003; Weiner and Wagner,
1998). The remainder of the bone is inorganic minerals including hy-
droxyapatite (approx. 65%), non-collagenous proteins (< 5%), and
water (approx. 10%) (Olszta et al., 2007; Veis, 2003; Dorozhkin, 2011).
The quality of type I collagen is typically assessed by measuring its
elemental composition and the total amount of material extracted from
a bone through a chemical process. Its unique amino acid profile con-
tains ~11–16% nitrogen, ~30–45% carbon, and a carbon to nitrogen
(C:N) ratio of 2.8–3.6 (Ambrose, 1990; DeNiro, 1985; DeNiro and
Weiner, 1988; McNulty et al., 2002; van Klinken, 1999). Partially de-
graded type I collagen, other bone proteins, and post-burial humic,
fulvic, and bacterial products have different compositions with C:N
ratios outside this range (Balzer et al., 1997; DeNiro and Weiner, 1988;
Harbeck and Grupe, 2009; Hare, 1980; van Klinken and Hedges, 1995).
The C:N ratio is the typical criterion of choice for determining the
presence of type I collagen with a well-preserved structure and whether
Fig. 1. Characteristic Raman spectrum collected from a well-preserved historic bone sample using a Thermo NXR FT-Raman spectrometer and 1064 nm laser excitation. Corresponding
vibrational and stretching modes are indicated with wavenumber. (Reproduced from France et al., 2014a with permission from Elsevier.)
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
597
isotope values are original to the bone or the result of burial con-
tamination. It is widely considered a more accurate measure of quality
than the independent proportions of carbon and nitrogen in a sample.
Furthermore, obtaining the elemental yields and C:N ratio requires only
a small aliquot of extracted organic material, and the errors tend to be
well-constrained.
The amount of organic material obtained from a bone by chemical
extraction is obviously a quantitative measure but is also a predictor of
type I collagen quality. It has been shown that intact type I collagen is
likely to be present when the chemical extraction process yields 1–22%
of the sample weight, though opinions on the lower end cut-off vary
(Ambrose, 1990; McNulty et al., 2002; van Klinken, 1999; France et al.,
2014a). The wide range of what is considered a sufficient overall yield,
the fact that current extraction methods typically are only accurate
to± 1% at best, and the fact that these organic residues are not iden-
tified as type I collagen analytically means that the proportion of ex-
tracted collagen, the so-called collagen yield, is not an entirely reliable
measure of type I collagen quality. However, knowledge of overall
collagen yield can be useful for proteomic techniques or selection of
samples in diagenetic studies where a range of preservation quality is
acceptable. Choosing to focus on C:N versus overall collagen yield de-
pends on the research question at hand and the desired accuracy of
predicted collagen quality. The C:N ratio is the more reliable quality
indicator when high accuracy is preferred.
2. Comparison of recent studies
In the last two decades, several studies have used Raman spectro-
scopy to qualitatively assess the presence or absence of collagen in
ancient bones (Morris and Mandair, 2011; France et al., 2014a; Pestle
et al., 2014), but only a few have used Raman to quantitatively assess
collagen preservation quality (France et al., 2014a; Halcrow et al.,
2014; Pestle et al., 2015). These studies vary in experimental set up and
findings. France et al. (2014a) used a high-resolution, research-grade
FT-Raman spectrometer to analyze bone outer surfaces and unpolished,
fresh fractured cross sections. They measured collagen quality by cor-
relating Raman peak height ratios to C:N ratios obtained by chemical
extraction. Halcrow et al. (2014) also used a high-resolution FT-Raman
spectrometer to analyze outer bone surfaces, but they compared Raman
peak area ratios (as opposed to peak heights) to the proportions of
collagen, nitrogen, and carbon extracted from a bone sample (i.e., %
collagen yield, % nitrogen yield, and % carbon yield), but not the C:N
ratio. Pestle et al. (2015) used a handheld spectrometer to compare
Raman peak height ratios to the collagen yield in powdered bone
samples and did not consider C:N ratio. Each of the three studies offers
criteria by which Raman spectra can determine the presence and
quality of protein in the bone. All chose to evaluate the relative pro-
portions of protein and hydroxyapatite in the bone as a ratio of two
Raman peaks. All chose the peak at 960 cm−1 to represent the mineral,
but they chose three different parts of the spectrum to represent the
organic matrix. The studies also differ in the instrumentation used and
the sample preparation. As we move forward in developing Raman
spectroscopy to evaluate bone protein, it bears questioning if the cri-
teria from one approach hold true for all data sets. A more detailed
examination of the three studies highlights their relative strengths and
the difficulties of comparing results in the absence of uniform reporting
standards.
2.1. France et al. (2014a)
The most complete sample and data set available to the authors of
this current paper is their own as originally published in France et al.
(2014a). Therefore, it is used as the primary basis for comparison here.
In that work, organic material was chemically extracted from 44
modern and 19th-century archaeological bone samples from humans
and other mammals. The C:N ratio of each extracted residue was
calculated. Bones from which the extracted organic material had a C:N
ratio of 2.8–3.6 were considered to contain well-preserved type I col-
lagen, and C:N values outside that window were construed to mean the
extracted organic material was not type I collagen or was poorly pre-
served. FT-Raman spectra were also collected of the bones, and the
spectra were subjected to a rigorous statistical analysis to determine a
best Raman indicator of a C:N ratio of 2.8–3.6. Seventy-eight possible
peak height ratios, 28 peak area ratios, and multivariate statistical
methods (i.e., partial least squares discriminant analysis (PLSDA)) were
considered. It was determined that a bivariate 960 cm−1:1636 cm−1
peak height ratio of ≤19.4 in freshly exposed cross-sections provided
the highest successful prediction rate (95%) of well-preserved samples
with a C:N ratio of 2.8–3.6. For outer surfaces of the bone they divided
the results into three groups. A peak height ratio≥ 26.8 was not ex-
pected to yield intact type I collagen. A ratio < 26.8 indicated that
61% of samples in that Raman range contained well-preserved type I
collagen, and for ratios< 19.4 the proportion of well-preserved sam-
ples jumped to 67%.
In addition, France et al. (2014a) asserted that Raman spectra
should also be evaluated visually for the presence of bone spectral
features before evaluating the spectra mathematically. They found
PLSDA also predicted type I collagen quality successfully but requires
significant time and effort to establish an internal laboratory dataset for
comparison. Finally, they showed that samples predicted by Raman to
contain poorly preserved collagen also had low collagen yields (< 5%).
2.2. Halcrow et al. (2014)
In contrast, Halcrow et al. (2014) correlated a ratio of Raman peak
areas at 3060–2800 cm−1 and 983–930 cm−1 to the extracted collagen
yield (r= 0.716), % N yield (r= 0.706), and % C yield (r= 0.630).
They provided the C:N values for their samples but did not compare
these to their Raman data. By applying the method of France et al.
(2014a) to calculate a classification rate, described below, one can de-
termine the percentage of Halcrow's samples that would be classified
correctly as well-preserved or poorly preserved based on C:N ratios and
their proposed 3060–2800 cm−1: 983–930 cm−1 peak area ratio. Using
a cut-off for this peak area ratio of 0.205 allows successful prediction of
collagen quality ~77% of the time, where ratios> 0.205 should in-
dicate well-preserved samples. This is a higher success rate for outer
bone surfaces than the method proposed by France et al. (2014a).
Halcrow et al. (2014) also visually examined Raman spectra and
stated that all bones with a collagen yield> 1% showed the presence of
a CeH stretching band in the 3060–2800 cm−1 region. However, this
band is not necessarily well-preserved type I collagen. In the France
et al. (2014a) data set, 88% of samples that were poorly preserved (i.e.
C:N outside the range of 2.8–3.6) showed a small Raman peak group in
the same region. This peak group sometimes had a different shape from
spectra of well-preserved samples, which suggests a different arrange-
ment of CeH bonds in the sample. Bonds between carbon and hydrogen
are hallmarks of most organic matter. Many substances, be they type I
collagen, degradation products of type I collagen, other bone proteins,
or something else, have Raman peaks in this region.
2.3. Pestle et al. (2015)
Pestle et al. (2015) provided the first study of a handheld Raman
spectrometer to quantitatively determine collagen yields. The study
used the Raman peak at 1450 cm−1 as the collagen quantity indicator.
The samples were mostly powders plus a small number of outer bone
surfaces. The study determined that samples with a peak height ratio of
1450 cm−1:960 cm−1 > 0.1 could be expected (with 95% confidence)
to contain> 1% collagen.
The 1450 cm−1:960 cm−1 peak height ratios from France et al.
(2014a) show a fairly close agreement to this cut-off. In terms of C:N
ratios, approximately 91% of well-preserved samples (i.e., C:N
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
598
= 2.8–3.6) show 1450 cm−1:960 cm−1 > 0.1, while 88% of poorly
preserved samples show 1450 cm−1:960 cm−1 < 0.1. Approximately
10% of the France et al. (2014a) samples would be misclassified using
Pestle's criteria which are based solely on quantitative collagen yield.
The France et al. (2014a) data show a moderate correlation between the
collagen yield and 1450 cm−1:960 cm−1 ratio (R2 = 0.604). In fact,
the 1636 cm−1:960 cm−1 ratio provides a slightly better correlation
(R2 = 0.649) for the France et al. (2014a) data.
Pestle et al. (2015) also offer an equation to predict the potential
collagen yield of a sample based on 1450 cm−1:960 cm−1 ratio. How-
ever, application of this formula to the France et al. data set fails to
accurately predict collagen yield within two standard deviations for
almost 30% of their samples. Pestle et al. (2015) do not provide C:N
ratios, so a direct comparison to the France et al. 960 cm−1:1636 cm−1
cut-off of ≤19.4 for well-preserved samples is not possible.
3. Re-evaluation of France et al. (2014a) data with a handheld
Raman spectrometer
The advent of miniaturized Raman spectrometers with near infrared
excitation is a boon for bone analysis. Smaller instruments like that
used by Pestle et al. can be deployed at archaeological sites or in the
laboratory at a fraction of the cost of a research-grade benchtop spec-
trometer. Some of these handheld instruments are ruggedized for use in
the field. However, there are important differences between the spectra
obtained with research-grade and handheld instruments that must be
taken into account when evaluating archaeological bone. This is a po-
tential obstacle to comparing the data set collected by Pestle et al. to
those of Halcrow et al. and France et al. To bridge this divide, the set of
bones analyzed by France et al. was re-examined with a Rigaku Progeny
handheld 1064 nm Raman spectrometer. The data is summarized here
and compared to results obtained by Pestle et al. (2015). All spectra and
calculations are available as Electronic Supplemental Material.
3.1. Materials and methods
Samples: Thirty-seven human bone specimens analyzed by France
et al. (2014a) were re-analyzed with the Progeny (Table 1). Briefly, all
bones are archaeological human remains from 19th-century North
American burial sites (France et al., 2014b). The remains were stored in
a cool dry location, typically a storage drawer, in keeping with typical
museum and university collections storage environments. Two bones
tested in France et al. (2014a) were unavailable as they have since been
repurposed for a different project. To the extent possible, samples were
analyzed on an outer surface (n= 10), on an unpolished cross section
that was either naturally exposed during excavation or prepared with a
chisel or bone saw (n= 29), and as a powder (n= 35). For outer
surfaces, soil-like accretions were removed with a scalpel, but the bone
was otherwise unmodified. Unlike the benchtop FT-Raman spectro-
meter used in the previous study, the handheld Progeny does not have
an automated microstage or live camera to help with sample posi-
tioning. As a result, many of the smaller bone samples did not have
sufficient surface area for the Progeny to analyze on both an outer
surface and in cross section.
Raman spectroscopy: Spectra were collected with a Rigaku Progeny
1064 nm handheld Raman spectrometer. The 3.5 lb. instrument fea-
tures a continuous wave, unpolarized Nd:YAG laser, thermoelectrically
cooled InGaAs 512 pixel detector, and an optional docking station for
static measurements (when hand holding is not desired). Spectra span
200–2500 cm−1 at fixed 8–11 cm−1 spectral resolution. Most spectra
were a co-addition of twelve three-second scans and used the in-
strument's proprietary “turbo” mode. Laser power was determined
empirically starting from the power density of FT-Raman measurements
from the previous study. The equivalent laser power was calculated for
the Progeny, which has a smaller analytical spot (25 μm diameter) (see
Electronic Supplemental Material: Laser power density conversion). For
nearly all samples, the Progeny instrument's 125 mW setting
(2.5 × 104 W/cm2) achieved sufficient Raman scattering without da-
maging the samples. Three spectra were collected for each solid bone
orientation (outer surface and cross section) and two spectra for each
powder.
3.2. Data processing and analysis
Whereas France et al. (2014a) compared bivariate peak ratios and
multivariate approaches, this current study only considers the bivariate
methods recommended in recent literature. Spectra were evaluated
visually followed by calculation of the ratio of the tallest hydro-
xyapatite peak (960 cm−1) to the heights of two prominent organic
peaks: 1450 cm−1 (advocated by Pestle et al.) and 1636 cm−1 (ad-
vocated by France et al.). The peak group advocated by Halcrow et al.,
3060–2800 cm−1, lies beyond the spectral range of today's handheld
Raman spectrometers, so it could not be evaluated.
Spectra were downloaded and converted to SPA (proprietary format
of Thermo Fisher Scientific) and CSV formats for interpretation using
TQ Analyst chemometric software (version 8.4.257, Thermo Fisher
Scientific, Inc.) and for publication here as Electronic Supplemental
Material, respectively.
Raman spectra of archaeological bone can be affected by sample
heterogeneity (France et al., 2014a). For each sample, the best of three
Raman spectra for each outer surface and cross section, and the best of
two spectra for each powder was chosen with a quick visual assessment
by the instrument operator. Visual selection is an intuitive process, but
the relevant criteria of a best spectrum are most likely low noise, low
fluorescence, well-defined peaks, and the absence of extraneous peaks.
Next, a classification rate was calculated. The classification rate mea-
sures the success of each peak height ratio to predict the quality of type
I collagen (based on C:N ratio) for a group of Raman spectra. Spectra
were grouped into six spectral datasets and compared: (1) all cross
section spectra, (2) all outer surface spectra, (3) all powder spectra, and
(4) best cross section spectra, (5) best outer surface spectra, and (6) best
powder spectra (France et al., 2014a). The all versus best spectral da-
tasets were compared to evaluate whether selecting spectra visually
influenced data reliability.
Classification rate = (# correctly assigned spectra ÷ total number
of spectra) × 100.
3.2.1. Baseline correction
Peak heights and baselines were calculated two ways in this study
using automated methods developed in TQ Analyst. The first method is
that used in France et al. (2014a), whereby each peak height is mea-
sured at a single specified position, and each baseline is defined as a line
drawn across the base of the peak and anchored on either side by a
single point along the spectrum (Method 1) (Fig. 2A) (Table 2). This
reduces the influence of sample fluorescence, which results in spectra
that slope upwards.
However, this approach does not consider the error associated with
spectral noise, a term that describes fluctuations along the spectrum
that are not related to Raman or fluorescence signals and which are
usually unwanted. In a so-called noisy spectrum the magnitude of these
fluctuations is large relative to the height of the Raman peaks. Noise
can distort a baseline that is defined by only two anchor points and also
the height of a peak that is measured at a single discrete point (Fig. 2B).
The peak height is a sum of the Raman signal (a), the baseline noise
error (b), and the peak noise error (c). This effect is more pronounced
for smaller Raman peaks, where the percentage noise error can be quite
significant and even obscure a peak or create the illusion of a peak that
is not there. Noise tends to be more pronounced with portable and
handheld Raman spectrometers than research-grade instruments
(Fig. 3).
The influence of noise on peak height ratios was addressed with a
second, modified method of calculating peak heights and baselines
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
599
(Method 2) (Fig. 2C) (Table 2). Peak heights were defined as an average
height across a narrow range bracketing the peak of interest. For ex-
ample, the 960 cm−1 peak was defined as the average peak height
across 961.5–958.5 cm−1. Each baseline anchor was also defined as an
average height across a range. Post-processing smoothing algorithms
were not applied to the spectra as they did not reliably improve, and
often worsened, classification predictions during the France et al.
(2014a) study of these same samples.
3.3. Results
Classification rates for the six Raman spectral datasets and two
baseline modeling methods are presented in Table 3. The best success
was achieved with Raman spectra of bone cross sections and a
960 cm−1:1450 cm−1 averaged peak height ratio; 86% of samples were
classified correctly. In all cases, selectivity was improved by collecting
more than one spectrum and selecting the best-looking one according to
the criteria given in the Materials and Methods section. Furthermore,
calculating peak heights and the anchor points as average values across
a small range of wavenumbers almost always improved selectivity.
Selectivity always improved when doing both. All spectra from which
these scores are calculated are included as Electronic Supplemental
Material.
Selectivity for Raman spectra of outer surfaces and powdered
samples with the 960 cm−1:1450 cm−1 ratio was less successful than
for cross sections. In all Raman datasets, there were threshold values
below and above which all samples contained well- or poorly preserved
collagen, and results between these thresholds were a mix of well-
preserved samples and poorly preserved samples (see Electronic
Supplemental Information: Spectral Interpretation). The Raman peak
height ratios also seem to have some predictive value in this zone of
mixed results (Table 4). For example, considering the sample set Best
Outer Surface Spectra with ratios calculated by baseline modeling
Method 2, ratios ≤8.1 were all well-preserved. For ratios in the range
of 8.3–20.2, 60% of samples contained well-preserved collagen (i.e.,
C:N 2.8–3.6). All ratios≥25.8 indicated poorly preserved samples. For
the Best Powdered Sample Spectra calculated by baseline Method 2,
ratios ≤6.2 were well-preserved. For ratios in the range 6.3–8.9, 92%
of samples contained well-preserved collagen. For ratios in the range
9.9–17.7, that proportion dropped to 57%. All ratios> 20.2 were
poorly preserved. A similar trend was observed with the
960 cm−1:1636 cm−1 peak height ratio data (Table 4).
Several spectra had negative ratio values, and these would have
been classified incorrectly by France et al. (2014a), where values
below< 19.4 are considered well-preserved. All samples with which
this occurred were poorly preserved. Negative ratio values occurred in
two situations. In the first situation, seen with the cross section spec-
trum of sample 557, the collagen indicator peak was so short as to have
a negative height relative to the calculated baseline (i.e., the noise was
greater than the peak height). The other scenario was due to
Table 1
Sample list and elemental abundance data organized by whether collagen was well- or poorly preserved and by C:N ratio.a
Sample Museum Designationb Burial Location Bone % Collagen Wt%N Wt%C C:N Collagen Preservationc Orientatione
394 6CT58-5-AMM03 Walton Family cemetery plot, CT Femur 21.2 15.0 41.7 3.2 Well cp
396 6CT58-5-AMM05 Walton Family cemetery plot, CT Femur 17.6 14.6 40.3 3.2 Well ocp
400 6CT58-5-AMM11 Walton Family cemetery plot, CT Femur 19.0 14.5 40.2 3.2 Well cp
281 29LA1091-BOR-42 Ft. Craig Cemetery, NM Metatarsal 4.9 14.2 40.4 3.3 Well op
343 51RICHARDS-CC-07 Congressional Cemetery, DC Metacarpal 12.1 14.8 41.8 3.3 Well op
345 GLO-099-2A Glorieta Pass Battlefield, NM Femur 13.3 14.2 39.7 3.3 Well cp
350 GLO-099-2C Glorieta Pass Battlefield, NM Femur 6.3 14.2 40.0 3.3 Well c
417 51KEYWORTH-CC-04 Congressional Cemetery, DC Radius 14.9 14.7 41.0 3.3 Well cp
420 51KEYWORTH-CC-07 Congressional Cemetery, DC Metatarsal 21.5 14.3 40.4 3.3 Well cp
426 51WHITE-CC-02 Congressional Cemetery, DC Fibula 16.4 14.5 40.8 3.3 Well ocp
442 51CAUSTEN-CC-11 Congressional Cemetery, DC Metacarpal 23.5 14.2 40.4 3.3 Well cp
443 51CAUSTEN-CC-13 Congressional Cemetery, DC Metacarpal 17.5 14.9 41.8 3.3 Well cp
454 18PR224-06-300 Family tomb, MD Femur 9.7 14.3 40.3 3.3 Well cp
455 18PR224-99-400 Family tomb, MD Tibia 7.3 12.4 35.3 3.3 Well cp
466 51KEYWORTH-CC-08 Congressional Cemetery, DC Tibia 16.8 12.3 34.9 3.3 Well cp
508 ELMINA-A-L49B Elmina Settlement, Ghana, Africa Tibia 5.0 12.1 34.6 3.3 Well op
525 ELMINA-A-Y51L Elmina Settlement, Ghana, Africa Femur 8.0 10.2 28.8 3.3 Well c
558 7NCE98A-WOODVILLE-10 Woodville Cemetery, DE Metatarsal 16.1 13.6 38.5 3.3 Well op
560 7NCE98A-WOODVILLE-SLOPEB Woodville Cemetery, DE Ulna 29.7 13.8 39.1 3.3 Well cp
566 FABC-08-107a First African Baptist Church Cemetery, PA Metacarpal 15.3 14.5 40.8 3.3 Well cp
471 31FOSCUE-ECU-1 Foscue Plantation family plot, NC Ulna 7.2 14.1 40.7 3.4 Well cp
556 7NCE98A-WOODVILLE-04 Woodville Cemetery, DE Fibula 5.1 12.3 36.5 3.5 Well op
598 44PWKINCHELOE-SI9114-C Kincheloe Plantation family plot, NC Temporal 3.6 12.9 38.8 3.5 Well cp
401 6CT58-5-AMM13 Walton Family cemetery plot, CT Femur 0.5 12.9 23.3 2.1 Poor cp
330 7BAYVISTA-DHCA-118A Ground burial, DE Femur 0.5 7.0 22.0 3.7 Poor cp
501 ELMINA-A-E51B Elmina Settlement, Ghana, Africa Femur 4.4 10.0 32.5 3.8 Poor cp
255 GETTYS-NPS-965 Gettysburg Battlefield, PA Temporal 0.6 8.2 27.9 4.0 Poor cp
305 TRINITY-EAST-22 Trinity Church Cemetery, DC Temporal 0.8 9.7 34.8 4.2 Poor cp
307 TRINITY-EAST-25 Trinity Church Cemetery, DC Mandible 0.5 7.1 25.9 4.3 Poor op
297 TRINITY-EAST 06 Trinity Church Cemetery, DC Temporal 0.4 1.9 8.3 5.0 Poor cp
555 7NCE98A-WOODVILLE-02 Woodville Cemetery, DE Metatarsal 1.6 6.4 27.8 5.0 Poor op
303 TRINITY-EAST-14 Trinity Church Cemetery, DC Temporal < 0.2d 7.7 35.2 5.4 Poor cp
557 7NCE98A-WOODVILLE-06 Woodville Cemetery, DE Femur 0.8 6.5 32.7 5.9 Poor cp
266 29LA1091-BOR-23B Ft. Craig Cemetery, NM Talus 1.0 6.5 34.0 6.1 Poor cp
267 29LA1091-BOR-23C Ft. Craig Cemetery, NM Talus 0.5 4.3 25.9 7.0 Poor cp
268 29LA1091-BOR-23D Ft. Craig Cemetery, NM Talus 0.2 5.3 33.9 7.5 Poor cp
309 TRINITY-EAST-04C Trinity Church Cemetery, DC Metacarpal < 0.2 5.2 47.2 10.5 Poor op
a Elemental abundance data from these bone specimens has been published previously in France and Owsley (2015) or France et al. (2014a and 2014b) and reappears here.
b Historic specimens are accessioned objects in the collection of the Smithsonian National Museum of Natural History.
c Samples with “well” preserved collagen had a C:N ratio of 2.8–3.6, and “poor” preservation fell outside that range.
d % Collagen yields ~0% neared the limit of resolution of our analytical scale and are noted as< 0.2% to encompass the inherent error.
e Orientation(s) of sample from which Raman spectra were collected (o = outer surface, c = cross section, p = powder).
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
600
contamination; the spectrum had unidentified peaks that skewed the
baselines upwards. This underscores the importance of visually ex-
amining spectra before calculating collagen quality mathematically or
with an automated process.
4. Discussion
The three studies published in 2014 and 2015 come to different
conclusions about the best indicator of collagen quality and quantity in
a Raman spectrum. Re-evaluation of the France et al. (2014a) bone
samples with a handheld spectrometer supports the finding of Pestle
et al. (2015) that a peak height ratio of 960 cm−1:1450 cm−1 can in-
dicate the presence of intact collagen better than 960 cm−1:1636 cm−1
when using a handheld 1064 nm Raman spectrometer. However, earlier
findings by France et al. with a high-resolution FT-Raman spectrometer
are likely to have measured type I collagen quality more accurately.
Pestle et al. posit that the difference between these studies' results “may
be a consequence of, among other possibilities, differences in the
thermal histories of the samples used in each study and consequent
differences in the degree of glutamine deamination within those sam-
ples” (Pestle et al., 2015 p. 116). Instead, the data presented here in-
dicate that the difference is probably a consequence of instrument re-
solution and the Raman signal-to-noise ratio rather than sample
composition.
In a Raman spectrum of a modern bone unaltered by diagenesis, the
signature of hydroxyapatite is stronger than that of the organic matrix
(Fig. 1). The 960 cm−1 peak, which represents a symmetrical stretching
vibration of phosphate (ν1[PO4]3-), is the strongest in the spectrum. In
the wavelength range of today's handheld Raman spectrometers, which
cannot see the CeH stretches at 3060–2800 cm−1, the strongest organic
peak is at 1450 cm−1 and represents a scissoring motion of CeH bonds.
The peak at 1636 cm−1 is the third strongest peak and part of a peak
group that represents amide I, a composite feature that is peculiar to
proteins, is mostly due to carbonyl stretching, and informs about
crosslinking of the collagen strands and protein conformation (Barth
and Zscherp, 2002; Carden et al., 2003). Here the difference in sensi-
tivity between a research-grade spectrometer and a handheld instru-
ment becomes important. The baselines of bone spectra, and particu-
larly bones that have been buried, are noisy, and the noise buries a
significant portion of these relatively weak Raman peaks. The
1450 cm−1 peak is taller and can be measured more reliably than the
1636 cm−1 peak, which has a higher percentage of noise error (Fig. 3).
This error increases as the peak height shrinks, so bones with little
remaining collagen will be affected the most. It follows that handheld
instruments, with their noisier spectra, are less able to distinguish small
peaks than more sensitive, less noisy research-grade instruments.
While the 1450 cm−1 peak may be a more successful indicator with
handheld instruments because it is taller, there are qualitative concerns
about suggesting it to identify type I collagen and other endogenous
bone protein. As with the 3060–2800 cm−1 peak group described by
Halcrow et al., the 1450 cm−1 peak describes vibrations of CeH bonds,
which are a hallmark of organic material. This sets up the possibility
that the source of a 1450 cm−1 peak could be something other than
protein. For example, the most common formulations that have been
used by conservators to consolidate or adhere archaeological bone,
including Acryloid B-72 acrylic resin, polyvinyl acetate, cellulose ni-
trate, and soluble nylon, have peaks at 3060–2800 cm−1 and
1450 cm−1. The 1636 cm−1 peak is specific to proteins and therefore a
better indicator of collagen with a C:N ratio of 2.8–3.6, if the spectro-
meter can resolve that peak well. The potential for organic con-
solidants, adhesives, and burial contamination underscores the im-
portance of qualitatively evaluating the entire Raman spectrum by eye.
Negative ratio values are another potential source of error in the
methods published thus far. These can occur when the magnitude of
spectral noise exceeds the height of a relevant Raman peak, or when a
contamination peak occurs at one of the baseline anchors. If there is a
Fig. 2. Raman spectrum from specimen 560, a historic human ulna and an example of a
bone with well-preserved collagen. A) The 1636 cm−1 peak with baseline defined as two
discrete points. B) The measured peak height varies significantly depending on how the
baseline is drawn (two possibilities shown in violet) and on the magnitude of noise at the
top of the peak (two possibilities shown in green). C) Averaging the baseline and peak
apex y-values across a range of wavenumbers (shown as dotted lines) reduces the effect of
noise, resulting in more reliable peak measurements.
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
601
sharp peak at 960 cm−1, and there are no additional peaks at the
baseline anchor points, then any negative ratio value must indicate a
poorly preserved sample. In any case, these Raman spectra should be
examined individually and, if needed, collected again before the col-
lagen quality is evaluated mathematically.
For this sample set choosing the best-looking spectrum out of two or
three usually improved the collagen quality prediction rate over col-
lecting a single spectrum. Differences between spectra are likely due to
variation in the bone from life processes or diagenesis (Weiner and
Wagner, 1998) or changes in the position of the instrument during the
measurement. The value of this practice should be investigated further
as selectivity improved as much as 300% when calculating the ratio
using the smaller 1636 cm−1 peak (Table 3). Furthermore, calculating
the peak heights and baseline anchors as average heights over a narrow
range usually improved selectivity by reducing the influence of noise.
Selectivity was always improved when doing both.
These practices improved our ability to choose bones that contain
well-preserved collagen, but the answers are not black-and-white. In
most cases, the Raman ratios identify some bone samples that obviously
contain good collagen and others that obviously do not. Many bone
samples fall somewhere in between. If one only considers the results as
presented in Table 3, Raman pre-screening does not look very helpful.
However, considering this data in terms of the likelihood of finding well-
preserved collagen is much more useful. Table 4 shows ranges of Raman
ratios in which 92%, 87%, or 20% of samples were well-preserved, for
example. That is valuable information for an archaeologist deciding
whether or not to sacrifice a bone to destructive analysis. The task now
is to build and publish many more bone samples and build Raman
spectroscopy into a statistically relevant predictive tool.
Given the statistical nature of many collagen studies, for example
with stable isotope analysis, a future goal could be to automate these
calculations with a software interface that controls the spectrometer
and automatically evaluates the collagen quality with a numerical
value, probability that a sample is well preserved, or a pass/fail system.
The task now is to rigorously test the method on more bones in order to
build a statistically relevant sample set and to determine the best
parameters for spectral collection and evaluation. To date only around
200 bone samples have been published that correlate Raman spectra to
C:N ratios or collagen yield. To develop this technique into a reliable
predictive tool, more samples must be published. Gaps in the compar-
isons of France, Halcrow, and Pestle above demonstrate the importance
of uniformity in data reporting. For each sample, a full set of elemental
and overall yield information (i.e. collagen yield, % N yield, % C yield,
and C:N) is required as well as the Raman spectra (in a universal format
such as JCAMP or CSV), and the calculated peak height and peak area
ratios. Given the ubiquitous option to publish supplementary data ta-
bles with any reputable journal, there is little reason to withhold this
information. By working together and testing methods against many
data sets, a consensus of best practices is likely to emerge.
It may be that a combination of tests might assess collagen quality
more reliably than any single test. For example, if two distinct peak
height ratios indicate that both C:N ratios and collagen yield “pass” pre-
Table 2
List of Raman peaks used in peak height ratios and the intervals over which corrected baselines were modeled.
METHOD 1: SPECIFIC POINTS
Bond vibration probed Peak position
(Raman shifted cm−1)
Modeled baseline
region
(Raman shifted cm−1)
ν1-[PO4]3− 960 990–910
δ-CH2 1450 1530–1150
Amide I 1636 1750–1550
METHOD 2: AVERAGE HEIGHT
Bond vibration probed Peak region
(Raman shifted cm−1)
Baseline: left anchor region
(Raman shifted cm−1)
Baseline: right anchor region
(Raman shifted cm−1)
ν1-[PO4]3− 961.5–958.5 1175–1135 920–890
δ-CH2 1451.5–1448.5 1550–1510 1170–1130
Amide I 1637.5–1634.5 1770–1730 1570–1530
Fig. 3. Raman spectra collected from Sample 525, a historic human femur and an example of bone with well-preserved collagen by C:N and a collagen yield toward the low end of well-
preserved samples (8.0%), with (A) Thermo NXR research-grade FT-Raman and (B) Rigaku Progeny handheld 1064 nm Raman spectrometers. The amplitude of noise is visibly higher in
the spectrum collected with the handheld instrument, and the effect on peaks of interest is highlighted around 1636 cm−1 and 1450 cm−1.
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
602
determined cut-offs, then the sample has a very high probability of
yielding well-preserved collagen. Similarly, if all criteria “fail”, then the
sample is unlikely to produce well-preserved collagen. This approach
would give more robust information to decide if precious samples
should be sacrificed to destructive analysis. It is also possible that no
consensus will emerge. All studies agree that fluorescence and
heterogeneity within samples are issues that increase the error in data.
Different studies employ different Raman spectrometers, each with a
specific signal-to-noise ratio, spot size, and optimum resolution.
Individual instruments may require custom criteria to determine col-
lagen quality. Furthermore, the current methods of collagen extraction
can produce significant differences and error in collagen yield data
Table 3
Summary of classification rates for the samples listed in Table 1. The success of the 960 cm−1:1450 cm−1 and
960 cm−1:1636 cm−1 Raman peak height ratios to predict the presence of well-preserved collagen in 37 bone
samples is shown as percentages of correctly classified samples. The data were assessed by whether all Raman
spectra collected from a sample or only the best-of-three spectra (for cross sections and outer surfaces) or best-of-
two spectra (for powders) were considered. The two baseline modeling methods described in Table 2 are also
compared.
All cs spectra
(n=87)
Best cs spectra
(n=29)
Raman ratio Method 1 Method 2 Method 1 Method 2
960 cm-1:1450 cm-1 75% 79% 83% 86%
960 cm-1:1636 cm-1 28% 36% 45% 48%
All o spectra
(n=30)
Best o spectra 
(n=10)
Method 1 Method 2 Method 1 Method 2
960 cm-1:1450 cm-1 9% 9% 40% 50%
960 cm-1:1636 cm-1 13% 10% 50% 60%
All p spectra
(n=70)
Best p spectra 
(n=35)
Method 1 Method 2 Method 1 Method 2
960 cm-1:1450 cm-1 21% 21% 37% 46%
960 cm-1:1636 cm-1 20% 29% 20% 37%
Table 4
Summary of Raman ratio cut-offs and the incidence of well-preserved collagen for the bone samples described in Table 1.
Raman peak height ratio Raman ratio rangeb Samples in rangec Well-preserved samples in ranged % classified as well-preservede
960 cm−1:1450 cm−1
Best cross section spectraa ≤9.26 15 15 100
9.33-18.3 4 3 75
≥21.5 10 0 0
Best outer surface spectra ≤8.1 4 4 100
8.3-20.2 5 3 60
≥25.8 1 0 0
Best powder spectra ≤6.2 6 6 100
6.3-9.0 12 11 92
9.9-17.7 7 4 57
≥20.2 10 0 0
960 cm−1:1636 cm−1
Best cross section spectra ≤10.6 5 5 100
10.7-35.0 15 13 87
≥35.8 9 0 0
Best outer surface spectra ≤13.4 5 5 100
14.1-23.3 4 2 50
≥47.1 1 0 0
Best powder spectra ≤12.9 9 9 100
13.0-17.0 12 10 83
17.1-38.0 10 2 20
≥39.7 4 0 0
a Dataset is the best of three Raman spectra collected for each sample, based on a visual assessment.
b Range of values for the given Raman peak height ratio.
c Number of samples with Raman peak height ratios in range.
d Number of samples with peak height ratios in range that had well-preserved type I collagen. Well-preserved type I collagen is defined as extracted organic
material with C:N ratio of 2.8-3.6.
e Percentage of samples in range that were classified as containing well-preserved type I collagen.
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
603
(Jorkov et al., 2007). All studies discussed above use a common acid-
base-acid extraction ultimately modeled after Longin (1971) which has
the disadvantage of minor sample loss during the process. It is un-
avoidable and results in collagen yield data that can be slightly in-
accurate. Differences between the extraction practices in each lab may
result in Raman ratio thresholds that are unique to that lab. Therefore it
is recommended that C:N ratios also be reported and correlated to peak
height or area ratios because C:N is less susceptible to lab-specific
procedural differences.
5. Conclusions
Raman spectroscopy is the most promising quantitative method for
non-destructive determination of collagen quality in recent decades.
Portable and handheld 1064 nm Raman spectrometers are certainly a
boon to archaeological fieldwork and analysis of organic materials in
general. However, improvements in portability and price are still offset
by lower sensitivity and signal-to-noise than research-grade instru-
ments. That tradeoff must be considered when evaluating collagen in
archaeological bone. Three recent quantitative studies predict collagen
quality or quantity by comparing the proportion of mineral and organic
bonds indicated in a Raman spectrum. The studies share many simila-
rities but differ in the peaks chosen to represent the organic component.
The authors of this article found that the 1636 cm−1 peak, which re-
presents amide I and is a characteristic of proteins, was the best col-
lagen quality indicator using a highly sensitive research-grade FT-
Raman spectrometer. However, the authors also found that this peak is
difficult to resolve with a handheld spectrometer, especially when there
is little collagen present in the sample (i.e., the peak is small).
Conversely, the peak at 1450 cm−1 is more likely to be resolved well by
a handheld spectrometer, but it is not as specific to collagen and could
describe other organic matter. While similarities are emerging in the
Raman indicators of yield and quality, differences in equipment and
chemical processing remain to be considered.
Raman evaluation of collagen will improve as a technique as more
sample sets are published and compared. This will require uniform and
complete data reporting to be effective. The present study and that of
France et al. (2014a) examined the same set of archaeological bones
with research-grade and handheld Raman spectrometers. Based on
those findings, the following procedure is recommended for assessing
collagen quality:
1. Collect multiple spectra of each bone in the desired orientation
(cross section, outer surface, or powder). Cross sections that are
freshly exposed intentionally with tools or unintentionally during
excavation have consistently been most successful for evaluating the
presence of well-preserved bone collagen.
2. Examine the spectra visually for fluorescence, noise, and any ex-
traneous peaks.
3. Select the best-looking spectrum (a complex intuitive choice based
on low fluorescence, low noise, well-defined peaks, and no extra-
neous peaks, among other possible considerations).
4. Calculate 960 cm−1:1450 cm−1 and 960 cm−1:1636 cm−1 peak
height ratios using baseline modeling Method 2 (Table 3). Peaks and
baseline anchors should be modeled as averages over narrow wa-
venumber ranges.
5. Assess the likelihood of finding well-preserved collagen by com-
parison to past results (i.e., Table 4).
This general method is applicable to any 1064 nm Raman spectro-
meter, with the exception of the probabilities in Step 5 that are specific
to human bones analyzed with a handheld instrument.
Acknowledgments
The authors acknowledge D. Owsley for access to the historic bone
samples, and D. B. Thomas and C. R. Doney for their previous work and
insights. National Science Foundation Research Experience for
Undergraduates Award #SMA-1156360 provided financial support for
D. M. W. Chan and M. Dundon and did not otherwise contribute to the
research. All analyses were performed at the Smithsonian Modern
Materials Laboratory at MCI and the Smithsonian MCI Stable Isotope
Mass Spectrometry Laboratory in Suitland, Maryland.
References
Ambrose, S.H., 1990. Preparation and characterization of bone and tooth collagen for
isotopic analysis. J. Archaeol. Sci. 17, 431–451.
Balzer, A., Gleixner, G., Grupe, G., Schmidt, H.-L., Schramm, S., Turban-Just, S., 1997. In
vitro decomposition of bone collagen by soil bacteria: the implications for stable
isotope analysis in archaeometry. Archaeometry 39, 415–429.
Barth, A., Zscherp, C., 2002. What vibrations tell us about proteins. Q. Rev. Biophys. 35,
369–430.
Carden, A., Rajachar, R.M., Morris, M.D., Kohn, D.H., 2003. Ultrastructural changes ac-
companying the mechanical deformation of bone tissue: a Raman imaging study.
Calcif. Tissue Int. 72, 166–175.
DeNiro, M.J., 1985. Postmortem preservation and alteration of in vivo bone collagen
isotope ratios in relation to palaeodietary reconstruction. Nature 317, 806–809.
DeNiro, M.J., Weiner, S., 1988. Chemical, enzymatic and spectroscopic characterization
of “collagen” and other organic fractions from prehistoric bone. Geochim.
Cosmochim. Acta 52, 2197–2206.
Dorozhkin, S.V., 2011. Calcium orthophosphates: occurrence, properties, biomineraliza-
tion, pathological calcification and biomimetic applications. Biomatter 1, 121–164.
France, C.A.M., Owsley, D.W., 2015. Stable carbon and oxygen isotope spacing between
bone and tooth collagen and hydroxyapatite in human archaeological remains. Int. J.
Osteoarchaeol. 25, 299–312.
France, C.A.M., Thomas, D.B., Doney, C.R., Madden, O., 2014a. FT-Raman spectroscopy
as a method for screening collagen diagenesis in bone. J. Archaeol. Sci. 42, 346–355.
France, C.A.M., Owsley, D.W., Hayek, L.C., 2014b. Stable isotope indicators of prove-
nance and demographics in 18th and 19th century North Americans. J. Archaeol. Sci.
42, 356–366.
Fratzl, P., Gupta, H.S., Paschalis, E.P., Roschger, P., 2004. Structure and mechanical
quality of the collagen-mineral nano-composite in bone. J. Mater. Chem. 14,
2115–2123.
Halcrow, S.E., Rooney, J., Beavan, N., Gordon, K.C., Tayles, N., Gray, A., 2014. Assessing
Raman spectroscopy as a prescreening tool for the selection of archaeological bone
for stable isotope analysis. PLoS One 9, 1–9.
Harbeck, M., Grupe, G., 2009. Experimental chemical degradation compared to natural
diagenetic alteration of collagen: implications for collagen quality indicators for
stable isotope analysis. Archaeol. Anthropol. Sci. 1, 43–57.
Hare, P.E., 1980. Organic geochemistry of bone and its relation to the survival of bone in
the natural environment. In: Behrensmeyer, A.K., Hill, A.P. (Eds.), Fossils in the
Making: Vertebrate Taphonomy and Paleoecology. The University of Chicago Press,
Chicago, pp. 208–219.
Jorkov, M.L.S., Heinemeier, J., Lynnerup, N., 2007. Evaluating bone collagen extraction
methods for stable isotope analysis in dietary studies. J. Archaeol. Sci. 34,
1824–1829.
Killick, D., 2015. The awkward adolescence of archaeological science. J. Archaeol. Sci.
56, 242–247.
van Klinken, G.J., 1999. Bone collagen quality indicators for palaeodietary and radio-
carbon measurements. J. Archaeol. Sci. 26, 687–695.
Longin, R., 1971. New method of collagen extraction for radiocarbon dating. Nature 230,
241–242.
McNulty, T., Calkins, A., Ostrom, P., Ganghi, H., Gottfried, M., Martin, L., Gage, D., 2002.
Stable isotope values of bone organic matter: artificial diagenesis experiments and
paleoecology of Natural Trap Cave, Wyoming. PALAIOS 17, 36–49.
Morris, M.D., Mandair, G.S., 2011. Raman assessment of bone quality. Clin. Orthop. Relat.
Res. 469, 2160–2169.
Nelson, B.K., DeNiro, M.J., Schoeninger, M.J., DePaolo, D.J., Hare, P.E., 1986. Effects of
diagenesis on strontium, carbon, nitrogen, and oxygen concentration and isotopic
composition of bone. Geochim. Cosmochim. Acta 50, 1941–1949.
Olszta, M.J., Cheng, X., Jee, S.S., Kumar, R., Kim, Y.-Y., Kaufman, M.J., Douglas, E.P.,
Gower, L.B., 2007. Bone structure and formation: a new perspective. Mater. Sci. Eng.
R Rep. 58, 77–116.
Person, A., Bocherens, H., Mariotti, A., Renard, M., 1996. Diagenetic evolution and ex-
perimental heating of bone phosphate. Palaeogeogr. Palaeoclimatol. Palaeoecol. 126,
135–149.
Pestle, W.J., Ahmad, F., Vesper, B.J., Cordell, G.A., Colvard, M.D., 2014. Ancient bone
collagen assessment by hand-held vibrational spectroscopy. J. Archaeol. Sci. 42,
381–389.
Pestle, W.J., Brennan, V., Sierra, R.L., Smith, E.K., Vesper, B.J., Cordell, G.A., Colvard,
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
604
M.D., 2015. Hand-held Raman spectroscopy as a pre-screening tool for archaeological
bone. J. Archaeol. Sci. 58, 113–120.
Shoulders, M.D., Raines, R.T., 2009. Collagen structure and stability. Annu. Rev.
Biochem. 78, 929–958.
Tütken, T., Vennemann, T.W., Pfretzschner, H.-U., 2008. Early diagenesis of bone and
tooth apatite in fluvial and marine settings: constraints from combined oxygen iso-
tope, nitrogen and REE analysis. Palaeogeogr. Palaeoclimatol. Palaeoecol. 266,
254–268.
van Klinken, G.J., Hedges, R.E.M., 1995. Experiments on collagen-humic interactions:
speed of humic uptake, and effects of diverse chemical treatments. J. Archaeol. Sci.
22, 263–270.
Veis, A., 2003. Mineralization in organic matrix frameworks. In: Dove, P.M., DeYoreo,
J.J., Weiner, S. (Eds.), Reviews in Mineralogy and Geochemistry. Biomineralization
54. The Mineralogical Society of America, Washington, pp. 249–289.
Weiner, S., Wagner, H.D., 1998. The material bone: structure-mechanical function rela-
tions. Annu. Rev. Mater. Sci. 28, 271–298.
O. Madden et al. Journal of Archaeological Science: Reports 18 (2018) 596–605
605