Forensic Science International 231 (2013) 364–374

The forensic analysis of office paper using carbon isotope ratio massspectrometry—Part 2: Method development, validation and samplehandling

Kylie Jones a,b,*, Sarah Benson a, Claude Roux b

a Forensic and Data Centres, Australian Federal Police, P.O. Box 401, Canberra, ACT 2601, Australiab Centre for Forensic Science, University of Technology Sydney, P.O. Box 123, Broadway, NSW 2007, Australia

A R T I C L E I N F O

Article history:

Received 5 November 2012

Received in revised form 20 April 2013

Accepted 7 May 2013

Available online 27 June 2013

Keywords:

Isotope

Forensic

Paper

Document

IRMS

Method validation

A B S T R A C T

This paper describes the development and validation of a method for the analysis of office papers by

measuring carbon isotopes using isotope ratio mass spectrometry (IRMS). The method development

phase included testing protocols for storage, sample materials, set-up of the analytical run; and

examining the effects of other paper examination procedures on IRMS results. A method validation was

performed so that the Deltaplus XP IRMS instrument (Thermo Finnigan, Bremen, Germany) with Flash

EATM 1112 could be used to measure document paper samples for forensic casework. A validation

protocol that would meet international standards for laboratory accreditation (international standard

ISO 17025) was structured so that the instruments performance characteristics could be observed. All

performance characteristics measured were found to be within an acceptable range and an expanded

measurement uncertainty for the measurement of carbon isotopes in paper was calculated at 0.26%,

with a coverage factor of 2. This method was utilized in a large-scale study, published as part one of this

series, that showed that IRMS of document papers is useful as a chemical comparison technique for

80 gsm white office papers.

Crown Copyright � 2013 Published by Elsevier Ireland Ltd. All rights reserved.

Contents lists available at SciVerse ScienceDirect

Forensic Science International

jou r nal h o mep age: w ww.els evier . co m/lo c ate / fo r sc i in t

1. Introduction

A common problem faced by forensic document examiners is tocompare documents in order to provide an expert opinion as towhether they could have originated from a common source. Mostprior research in this discipline has centered on the chemicalcomparison of pen and printing inks. Forensic paper examination isa still a field with potential for expansion and a large gap remainsto identify an analytical technique that is ideal for the discrimina-tion of papers in forensic examinations, whether to discriminate orassociate document substrates.

Isotope ratio mass spectrometry (IRMS) has proven itself to be avaluable technique for the comparison of a range of forensicsamples [1–5]. With respect to paper examination, van Es et al. [6]found IRMS was useful in discriminating 20 of 25 European paperstested. Other experiments and part one of this series haveconfirmed the effectiveness of comparing the carbon isotopevalues of white office papers available in Australia.

* Corresponding author at: P.O. Box 401, Canberra, ACT 2601, Australia.

Tel.: +61 2 62036078.

E-mail address: kylie.jones@afp.gov.au (K. Jones).

0379-0738/$ – see front matter . Crown Copyright � 2013 Published by Elsevier Irelan

http://dx.doi.org/10.1016/j.forsciint.2013.05.009

The majority of studies published so far [1,2,4,7] however havebeen proof of concept studies that have not focused on theapplication of the results to forensic casework. To date, only onearticle has been published that presents the results from thevalidation of a method to international accreditation standards [8].For forensic casework, method validation is an essential andexpected checkpoint so that the results produced by a measure-ment technique are demonstrably fit for purpose and robust.

Accreditation of Australian forensic laboratories is granted bythe National Association of Testing Authorities (NATA) who assesslaboratories against International Organisation for Standardisation(ISO) standard 17025 [9] and the NATA Field ApplicationDocument (FAD) for forensic science [10]. Regarding methodvalidation, the FAD sets out a number of broad issues that may berequired to be examined during a study such as homogeneity,concentration and linear range. Further information on conductingmethod validation activities are included in a guide published byNATA [11], with a large number of other publications on the topic,including those published by ISO and EURACHEM [12,13].

For techniques that present a numerical value at theirconclusion, an estimate of the measurement uncertainty of themethod should be provided with the analytical measurementvalue to enable the customer or client to assess the precision of theresults [11,14–16]. The standard for producing this estimate is the

d Ltd. All rights reserved.

K. Jones et al. / Forensic Science International 231 (2013) 364–374 365

Guide to Expression of Uncertainty in Measurement (GUM),published by the Joint Committee for Guides in Metrology [17].

This study outlines the method development and validationexperiments undertaken for the measurement of the carbonisotope abundances of document papers using IRMS. For methoddevelopment, key parameters such as sample size, linear range,number and placement of standards, and sample storage andhandling have been tested. Method validation experiments aim toshow that performance characteristics relating to precision,accuracy and repeatability are met. Finally, a value for measure-ment uncertainty was calculated based on the results of theexperiments. The results demonstrate that the method is fit forpurpose to analyze and report carbon measurements to assist inthe discrimination or association of multiple document papers.

2. Materials and methods

2.1. Standards and samples

The international standards used in this study and their certified values are listed

in Table 1. These standard materials were all purchased from the International

Atomic Energy Agency (IAEA).

Additionally, a number of chemically analogous materials were used both for the

validation and to be assessed as potential laboratory standard materials. These were

all purchased through Sigma–Aldrich (Sydney, Australia). The details of these

materials are listed in Table 2.

A number of plain 80 gsm papers were used in these experiments. These were

chosen at random from a population of paper samples that were purchased from

throughout the Canberra (ACT, Australia) region.

All standards (international and laboratory) were stored in glass vials with Teflon

lined screw top lids (Sigma–Aldrich) or the original packaging in a Perspex

desiccator with self-indicating silica gel (LabServ, Biolab, Aukland, New Zealand).

Paper samples were stored either in the ream or in a plastic sleeve in a clean

laboratory environment. Unless otherwise specified, all sample measurements

were performed in triplicate.

Laboratory gases were purchased from Coregas (Canberra, Australia). All were

ultra-high purity (analytical) grade with purity higher than 99.99%. The gases used

for this method validation were helium, carbon dioxide and oxygen.

Table 1International standard materials used in the method validation (purchased through

IAEA).

Certified reference material d13C certified value (%)

LSVEC Mean �46.6

Lithium carbonate St. dev. 0.2

NBS 19 Mean 1.95

Limestone St. dev. None published

IAEA-CH-7 Mean �32.151

Polyethylene St. dev. 0.050

IAEA-CH-6 Mean �10.449

Sucrose St. dev. 0.033

IAEA-CH-3 Mean �24.724

Cellulose St. dev. 0.033

Table 2Materials purchased from Sigma–Aldrich for use in the method validation

experiments and for evaluation as laboratory standards.

Sample name Product number Batch number

Microgranular cellulose C6413 069K0082

Medium fiber length cellulose C6288 039K6171

Long fiber length cellulose C6333 038K0055

a-Cellulose powder C8002 109K0114

D + Glucose G8270 089K00603

Mannose fibers from wood 112585 099K0054

a-D-Glucose 159968 MKBB8469

Cellulose acetate 180955 40198LJ

Starch from corn S4126 084K0009

Starch from wheat S5127 079K0309

Sigma ultra cellulose S6790 099K1098

a-Lactose monohydrate L3625 099K1540

2.2. Instrumentation and equipment

A Deltaplus XP IRMS instrument (Thermo Finnigan) with a ConFlo III interface and a

Flash EATM 1112 with an AS128 autosampler was used for all experiments. The

internal set-up of the Flash EA used was as published in Benson et al. [8], with the

addition of a quartz insert which was placed in the top of the combustion tube to assist

in ash removal. This quartz insert was replaced every 150–200 samples. Paper and

polyethylene samples were prepared using a 1.2 mm hand punch (Harris Unicore,

LabSciTech, Australia). Other samples were weighed on a Genius ME5 balance

(Sartorius, Goettingen, Germany). The samples were placed into 3.3 mm � 5 mm tin

capsules (IVA-Analysentechnik, Dusseldorf, Germany) for measurement.

When assessing the effects of other paper examination techniques such

conditioning of the papers, a Binder KBF-115 environmental chamber (Crown

Scientific, Sydney, Australia) was used to control temperature and humidity.

2.3. Correction of values

The delta values produced by the Isodat software were exported into Microsoft1

Excel. Instrument drift was assessed by comparing the values obtained for

international standards placed at the beginning and the end of the experimental

run. If significant systematic drift was observed (where significant drift is defined as

a difference greater than the standard deviation of the standard, or the standard

deviation of the instrument 0.15%), the values were corrected by calculating a

mean difference for the run and adding a proportion of the mean to the value based

on the corresponding line number. This calculation is shown below:

Corrected value ¼ original measured value

þ mean drift

total # samples

� �� line # sample

� �

Quality control and test samples were corrected against two international

standards using a line of best fit (y = mx + b). Polyethylene (IAEA-CH-7) and sucrose

(IAEA-CH-6) were selected as standards as they bracket the expected range for

paper and other cellulose samples. Outliers from each group of values were

excluded using Grubbs test [18]. All delta values expressed in these experiments are

traceable to the VPDB scale.

2.4. Statistical tests

Basic statistics, including calculation of the mean and standard deviation of

sample measurements were undertaken using Microsoft1 Excel.

GraphPad Prism (version 5 for Windows or Mac, GraphPad Software, La Jolla, CA,

USA) was used for other statistical tests. This software was chosen for the inclusion of

the Kruskal–Wallis non-parametric analysis of variance (ANOVA) test with Dunn’s

post hoc test. This test was used instead of a parametric ANOVA test as in other

experiments the paper samples used were shown not to conform to a parametric

curve, even when run in high sample numbers (i.e. where n = 50 or more).

2.5. Method development

The following target performance characteristics were set for method

development. The method parameters were varied to ensure that these targets

were met at all stages of an experimental run and in successive experimental runs.

The targets included:

� Full sample peak elution with baseline separation from reference peaks

(appropriate reference peak placement).

� Matching of sample peak and reference peak heights by determining the most

appropriate sample size and reference peak pressure.

� Linearity of instrument response with increasing sample size for carbon (paper)

sample peaks.

� Precision of repeat measurements by ensuring that the most appropriate sample

size is selected.

� Linearity of CO+ production (detected on mass 28) with increasing sample weight

for each sample type measured (i.e. paper, cellulose and glucose).

2.5.1. Water absorption

Cellulose, glucose and paper are hygroscopic materials and hence require water

content to be controlled for accurate measurements to be able to be assured.

Five papers and five laboratory working standard samples were selected to

determine whether water absorption affected the measured d values of the samples.

All samples were measured on day 1 and day 16 (from the time of initial exposure).

All samples were left exposed to atmosphere on a laboratory bench for the duration

of the test.

2.5.2. Sample handling, storage and the effect of other paper examination procedures

Following on from the absorption study, secondary tests were conducted to

measure and evaluate the effect of altering the storage conditions of the samples on

the carbon isotope abundances. In routine casework, it was expected that samples

would be weighed and stored in desiccators until measured. To test the effect of

K. Jones et al. / Forensic Science International 231 (2013) 364–374366

this, three sets of samples were taken from five papers. One set was measured on

the IRMS immediately (named: day one). The two other sets of samples were

prepared and placed in a Perspex desiccator for 15 days. Of these two sets, the

sample capsules in the first set were crushed (named: desiccator closed), while the

sample capsules of the second set were left open (named: desiccator open) until just

prior to measurement to test whether closing the capsule allowed the samples to

dry adequately.

Included in the secondary tests were samples that had been treated according to

the standard procedure used to prepare papers for measurement of grammage and

density. This standard procedure, detailed in [19], is used to remove any effects

from the source of the paper by equilibrating the paper sheets’ water content. The

paper sheets are placed in a humidity and temperature controlled chamber for 24 h

at 50% relative humidity and 25 8C. The humidity of the chamber is then reduced to

25%, where the paper sheets are held for a further 48 h.

To test the effect of this humidification procedure, the five sheets from which the

day one and desiccator samples were taken were placed into the chamber and

exposed to the humidity/temperature program. After this was completed, one set of

samples was taken from the sheets and run immediately (named: humidity only). A

second set of samples was taken, crushed and placed into a desiccator for 7 days

(named: humidity + desiccator).

The results from these four different sample treatment and storage conditions

were compared against each other and against the day one results.

2.5.3. Design of sample run – placement and number of international standards and

blanks

A number of experiments were undertaken to ensure that the structure of the

analytical runs maximized the accuracy of the results after correction. This included

measuring the following variables:

� Number of international standards – two end member pairs (polyethylene and

sucrose) or three standards comprising the two end-member pairs plus one mid

range value (cellulose).

� Number of replicates of international standards – three, five or seven.

� Placement of international standards – start and end of the run or start, middle

and end of the run.

� Inclusion of blank tin capsules between samples.

To determine the effects of changing these parameters, polyethylene (IAEA-CH-

7) and sucrose (IAEA-CH-6) international standards were used to correct the

measured values for two international standards (LSVEC and Limestone/NBS 19)

and three laboratory standard materials (medium cellulose, alpha glucose and

cellulose acetate), run as unknown materials. Where the international standard

cellulose (IAEA-CH-3) was not used as a factor in the experiment, it was also

included as an unknown sample.

2.6. Method validation

2.6.1. Precision and stability

Carbon isotope abundances for international standards polyethylene (IAEA-CH-

7) and sucrose (IAEA-CH-6) measured and corrected over a 24-month period were

plotted to determine the range of variation expected over a period of time.

2.6.2. Repeatability

International cellulose (IAEA-CH-3) was measured and plotted as an unknown

sample in each analytical run over a 12-month period. The corrected values were

plotted to determine their range and conformance to the published value and

standard deviation.

2.6.3. Accuracy

Medium cellulose, cellulose acetate and alpha glucose samples were run over a

24-month period to assess whether the method produced accurate values over

time.

2.6.4. Robustness

Three different laboratory staff of varying levels of experience prepared the same

analytical run containing paper, cellulose, glucose and sucrose samples. Triplicates

of each of the samples were prepared on different days and corrected against the

same international standards. All three runs were analyzed on the IRMS and

corrected by the author (named: Operators A, B and C).

In addition, the same analytical run was prepared, run on the IRMS and corrected

by a person familiar with IRMS operation but external to the project (named:

Operator D).

All four groups of results were compared to determine the robustness of the

method to changes in sample preparation and operation of the instrument.

2.6.5. Measurement uncertainty

The entire procedure, from preparation of international standards through to

correction calculations was evaluated for potential sources of uncertainty.

Following the method outlined in practical guides published by NATA [14] and

the equations used in Benson et al. [1], the uncertainties associated with the method

were separated into sources of method bias/accuracy and method precision. Overall

these two types of uncertainty were combined to give an estimate of measurement

uncertainty.

2.6.6. Inter-laboratory trial

The carbon method developed for paper was applied to two different samples

that were distributed as part of the 2010 forensic isotope ratio mass spectrometry

(FIRMS) group trial [20]. Ten replicates of each sample were measured using the

method and correction calculations described above.

The results reported back to each laboratory included both a grand mean for the

collated results from the 28 participating laboratories, in addition to a corrected

mean, based on a Huber’s method of correction [21], annotated to the H15 mean

and H15 standard deviation. Using these values and assigning a target standard

deviation of 0.2%, a Z-score was calculated using the following formula [22]:

Z score ¼ reported mean � H15 mean

target standard deviation

The Z-score provided is an indication of a laboratories’ closeness to other

participants’ measurement values. A value within the range �2 to +2 is considered

within a normal range of scores.

3. Results and discussion

3.1. Method development

Using the instrument parameters set out by Benson et al. [8],full and consistent peak elution was achieved. CO2 reference peakswere placed at 20, 60 and 320 s with the pressure set atapproximately 0.6 bar to ensure a sample height of approximately3000 mV (compared to a baseline of between 4 and 10 mV). Thetotal run duration was 360 s. Drift correction was only requiredduring correction on rare occasions.

3.1.1. Linearity

A linearity experiment was conducted to determine whether aconsistent instrument response was observed when sample sizeincreased. The d13C values of each sample type were measured in asingle sequence, corrected and plotted against the instrumentresponse to determine whether there were any significantvariations. The linearity experiment results are included asTable 3. Additionally, this data is plotted in Figs. 1–3.

The results of the linearity experiments show that instrumentresponse is linear for increasing sample size and this is not ofconcern as long as sample size is normalized and controlled withina measurement run when measuring d13C of cellulosic materials.

When sample size was plotted against the measured d13C valuemeasured, a minimum critical value was detected at 75 mg ofsample. This was consistent between the three sample typesmeasured. Beyond 75 mg, up to the maximum size tested (200 mg),the d13C values were stable and comparable. Using the largeststandard deviation for each of the measured values between 75and 100 mg as a benchmark for discrimination, only one value at75 mg would be discriminated from the other values (medium fibercellulose sample type). This indicates that sample sizes of 100 mgand above are fit for purpose with respect to stable measurementsand comparable peak heights, with the true sample size deter-mined by scaling against the desired reference gas peak height.

One known issue with the measurement of carbon isotopes insamples with high carbon content is the production of smallamounts of CO+ from the production and subsequent ionization ofCO in the reactor. This isobaric interference produces moleculesthat elute a peak at the same time as CO2, with a mass of 28 (thesame mass as measured during nitrogen isotope measurement)[23]. De Groot [24] outlines one method of reducing/eliminatingthis interference for organic samples that are being measured fornitrogen isotopes.

Table 3Summarized data for linear range experiments.

Mean sample

weight (mg)

Mean instrument

response (mV)

Standard deviation

instrument response (mV)

RSD instrument

response (%)

d13C mean (%) Standard

deviation (%)

mV/mg sample

Medium fiber cellulose25 408 67 16.41 �23.76 0.56 16.31

46 779 64 8.23 �25.11 0.11 17.06

73 1329 140 10.54 �25.82 0.16 18.28

106 2013 200 9.92 �26.44 0.14 19.05

125 2369 133 5.62 �26.58 0.03 19.00

152 3029 228 7.54 �26.74 0.04 19.97

174 3567 75 2.10 �26.86 0.17 20.46

202 4250 195 4.58 �27.49 0.17 21.04

Cellulose acetate25 594 143 24.09 �31.47 0.10 23.77

54 1266 76 6.00 �31.88 0.09 23.45

79 1807 98 5.41 �31.99 0.01 22.97

103 2376 76 3.18 �32.02 0.01 23.14

126 2930 23 0.78 �32.01 0.04 23.32

150 3525 94 2.67 �32.09 0.04 23.45

180 4223 62 1.46 �32.13 0.01 23.50

206 4787 165 3.44 �32.07 0.02 23.27

Alpha glucose32 352 123 34.95 �8.83 0.44 11.17

49 803 58 7.21 �10.05 0.57 16.38

69 1168 132 11.34 �10.22 0.04 16.84

98 1719 116 6.73 �10.62 0.02 17.49

124 2490 346 13.89 �10.83 0.15 20.13

153 2915 33 1.12 �10.88 0.01 19.01

173 3408 125 3.66 �11.00 0.03 19.74

199 3842 473 12.31 �11.06 0.05 19.28

K. Jones et al. / Forensic Science International 231 (2013) 364–374 367

The primary concern in this study was to evaluate whether theCO+ production was linear as a proportion of increasing samplesize. Fig. 4 shows the linear instrument response for the CO+ ionsproduced from cellulose, glucose and paper. For the three sampletypes, the R2 values of the lines of best fit were 0.997, 0.997 and0.998 respectively. These results show that the production of CO+for each sample type is linear and that the combustion conditionsbeing used are consistent, producing isobaric CO+ in a proportionrelative to the sample size. Again, this result highlights the criticalneed to control and ensure consistent sample size.

3.1.2. Water absorption

The mean d13C and standard deviation for day 1 and day 16 foreach sample type have been plotted in Fig. 5. The data used toproduce this plot is detailed in Table 4.

The day 1 standard deviation values indicate some issues withthe precision of the instrument on this particular day however themean values are still comparable to the day 16 values.

Due to this, a number of measurements were excluded asoutliers. This resulted in a number of sample results containing

Table 4Day 1 vs. day 16 d13CVPDB (%) results for samples exposed to the laboratory atmosphe

Paper samples d13C day 1 (%) d13C day 16 (%) Cel

Paper 1 Mean �22.15 �23.56 Ma

St. dev. 0.75 0.06

Paper 2 Mean �27.69 �28.70 Me

St. dev. 0.31 0.10

Paper 3 Mean �22.88 �23.35 Alp

St. dev. 0.38 0.05

Paper 4 Mean �23.64 �24.05 D-G

St. dev. 0.40 0.01

Paper 5 Mean �26.05 �26.51 Cel

St. dev. 0.20 0.06

only two measurements, too few to undertake the Kruskal–Wallistest. Instead, an unpaired t-test was used to give an indication ofwhether the mean values are significantly different between day 1and day 16. Whilst this is not ideal, the statistical results provide anindication of any differences, which can be weighed against theactual values and graphical results shown.

Significant differences (to 95% confidence level) were observedin the following samples:

� Paper 1 (p value = 0.03)� Paper 2 (p-value = 0.006)� Paper 5 (p-value = 0.0185)� Mannose (p-value = <0.0001)

These results suggest that there is some effect on the measuredd13C values of the samples after exposure to the laboratoryatmosphere, where moisture is not controlled. The laboratoryatmosphere where these samples were left is notoriouslyinconsistent and prone to large fluctuations in both temperatureand humidity.

re.

lulose and glucose samples d13C day 1 (%) d13C day 16 (%)

nnose Mean �22.47 �22.84

St. dev. 0.04 0.01

dium cellulose Mean �26.26 �26.86

St. dev. 0.50 0.09

ha glucose Mean �11.22 �11.20

St. dev. 0.80 0.03

lucose Mean �10.68 �10.83

St. dev. 0.16 0.07

lulose acetate Mean �31.62 �32.06

St. dev. 0.58 0.04

Fig. 1. (a) Linear range for medium fiber cellulose – mean instrument response

(mV). (b) Linear range for medium fiber cellulose – mean d13CVPDB (%).

Fig. 3. (a) Linear range for alpha glucose – mean instrument response (mV). (b)

Linear range for alpha glucose – mean d13CVPDB (%).

K. Jones et al. / Forensic Science International 231 (2013) 364–374368

Mannose (a type of cellulose) was found to be so hygroscopicthat it completely liquefied in the presence of moisture over timeso these results were not unexpected for this material.

As bonded cellulose is very stable, the fractionation seen in thepaper samples when exposed to the atmosphere is likely caused bythe adsorption of water onto the calcium carbonate filler material.The water promotes further uptake of atmospheric gases anddegradation of the calcium carbonate to form carbonic acid. Withfurther uptake/exposure to water, the carbonic acid forms andreleases gaseous carbon dioxide [25]. The magnitude of this effectwould be due to the natural hygroscopicity of the paper (imparted

Fig. 2. (a) Linear range for cellulose acetate – mean instrument response (mV). (b)

Linear range for cellulose acetate – mean d13CVPDB (%).

by the humidity of the location that the paper was made) as themore water that is available, the easier breakdown of the calciumcarbonate and hence release of carbon dioxide is.

Overall, these deviations should be minimized by storing thesetypes of samples in a desiccator to reduce interference fromatmospheric uptake of moisture and other contaminants.

3.1.3. Sample handling, storage and the effect of other paper

examination procedures

The mean and standard deviation for each condition and sampleare plotted in Fig. 6. The data used to produce this plot is detailed inTable 5.

From the corrected sample values and the graph of the meanvalues, there appear to be no trends to be observed in terms ofconsistent effects on the samples.

The Kruskal–Wallis test followed by Dunn’s post hoc test wasused to determine if there were any significant differences (at the95% confidence level) between the sample storage/handlingmethods.

Significant differences were only observed within the PaperOne sample set (p = 0.0471), but were not able to be attributed toone sample handling method using Dunn’s post hoc test.

Fig. 4. Linearity of CO+ production for paper, cellulose acetate and alpha glucose

samples.

Fig. 5. Day 1 (diamond) vs. day 16 (square) d13C values for samples left exposed to the laboratory atmosphere.

K. Jones et al. / Forensic Science International 231 (2013) 364–374 369

Overall, there was no significant effect observed in the resultsobtained from these tests. As long as the principle of identicaltreatment is applied to samples within the same analytical run,and that exposure to moisture is reduced where possible, theresults can be considered to be stable for comparison.

3.1.4. Design of sample run – placement and number of international

standards and blanks

3.1.4.1. Number of standards. There were no significant differ-ences observed between using two or three internationalstandards to correct the values of the samples. The slope ofthe calibration line did not change, and its position was only

Fig. 6. Mean d13CV-PDB (%) paper sample va

slightly affected with a shift in the intercept of the line. Thelargest shift in the position of the intercept was by 0.07,however even this difference was not large enough to shift thecorrected values of the samples, especially given that thereported values are limited to being expressed to four significantfigures.

3.1.4.2. Number of replicates. When the sample data were cor-rected using three, five or seven replicates of the internationalstandards polyethylene and sucrose placed at the start and the endof the analytical run, differences were detected between the meanvalues and the standard deviations. A summary of the correcteddata is shown in Table 6.

lues for storage/handling experiment.

Table 5d13CVPDB (%) results for varying sample handling/storage conditions.

d13C day one (%) d13C desiccator closed (%) d13C desiccator open (%) d13C humidity only (%) d13C humidity + desiccator (%)

Lazer ITMean �26.88 �26.96 �27.10 �26.93 �27.02

St. dev. 0.14 0.11 0.00 0.08 0.10

Paper OneMean �28.21 �28.20 �28.12 �28.35 �28.30

St. dev. 0.01 0.03 0.09 0.00 0.03

Fuji XeroxMean �23.48 �23.44 �23.59 �23.62 �23.58

St. dev. 0.13 0.05 0.05 0.03 0.05

Office MaxMean �25.41 �25.47 �25.39 �25.36 �25.50

St. dev. 0.05 0.06 0.02 0.00 0.13

ReflexMean �23.55 �23.49 �23.59 �23.57 �23.66

St. dev. 0.07 0.07 0.12 0.06 0.07

K. Jones et al. / Forensic Science International 231 (2013) 364–374370

Again, the Kruskal–Wallis ANOVA test with Dunn’s post hoc testwas performed using GraphPad Prism. The ‘unknown’ samples inTable 6 were tested for differences between the results correctedusing different replicates of international standards. The experi-mental work for these tests was undertaken in a series ofconsecutive days to reduce any potential instrument effects. Theresults for Limestone and LSVEC were suspected to pose problemsas both samples are carbonates and hence are not matrix matchedto the method being validated.

The alpha glucose sample corrected using three replicateinternational standards was found to be significantly different tothe seven repeat correction group at the 95% confidence level. Thep-value for alpha glucose was 0.0390, which is between the 95%and 99% confidence levels. No significant difference was detectedfor the remainder of the samples corrected.

These results indicate that there was no perceived benefit torunning higher numbers (>3 replicates) of international standardsat the start and the end of the experimental runs. It becomesdifficult, however, to ensure the quality of the calibrationperformed when one or more of the international standardsamples is excluded during correction using Grubbs test. Meier-Augenstein [7] recommends that any sample, standard or controlthat will undergo calculation of likelihood ratio’s (as originallyoutlined in [26]) should be run in even numbered multiples. Itmakes sense therefore to run the international standards at thestart and end of each run in replicates of five, to allow for flexibility,both in the exclusion of outliers and for calculation of likelihoodratio’s.

Table 6Corrected d13CVPDB (%) values for international and ‘unknown’ samples.

Three reps Five

d13C mean (%) St. dev. (%) d13C

International StandardsPolyethylene �32.15 0.05 �32.

Sucrose �10.45 0.02 �10.

UnknownsLimestone 1.76 0.31 1.

LSVEC �46.09 0.16 �46.

International cellulose �24.67 0.05 �24.

Medium cellulose �26.85 0.00 �26.

Cellulose acetate �32.02 0.01 �31.

Alpha glucose �11.42 0.26 �11.

Line correction factorsM value 1.001 1.

B value �10.794 �11.

3.1.4.3. Using a third set of standards. A third set of the sameinternational standards (polyethylene and sucrose) was placed inthe center of an analytical run to determine if this increased theaccuracy and precision of the calibration of the unknown samples.The same experimental run was used (to reduce instrumenteffects) and was corrected twice – the first using only the two setsof standards at the start and the end of the run, and the secondusing all three sets of standards – at the start, middle and end of thesequence.

The results show that there is an effect on the accuracy of thecorrection of the samples (Table 7) but there is no effect on theprecision i.e. even though the sample values are shifting slightly,the standard deviation values are not. The magnitude of thedifference with respect to the precision is small however and whenweighed against the practicalities of including a third set ofstandards, the difference is not significant enough to warrantincreasing a run sequence by a minimum of 16 samples (10standards and 6 blanks).

3.1.4.4. Inclusion of blank capsules. Blank capsules were removedfrom between all of the standards and samples within a runsequence and the results compared to a prior sequence. Smalldifferences in the mean values and standard deviations of thesamples were detected, likely due to instrument effects.

Noticeable effects on the sample measurements were detectedwhen moving from a low d13C value (e.g. international standardsucrose = �10.45%) to a high d13C value (e.g. internationalstandard polyethylene = �32.15%). This commonly resulted in

reps Seven reps

mean (%) St. dev. (%) d13C mean (%) St. dev. (%)

151 0.02 �32.152 0.042

45 0.04 �10.449 0.088

82 0.13 2.01 0.02

33 0.09 �46.31 0.04

68 0.03 �24.71 0.04

86 0.04 �26.87 0.03

98 0.02 �32.04 0.04

21 0.05 �11.09 0.03

002 1.006

125 �10.798

Table 7d13CVPDB (%) values of samples corrected using two or three sets of international standards.

Two sets Three sets

d13C mean (%) St. dev. (%) d13C mean (%) St. dev. (%)

International cellulose �24.43 0.06 �24.48 0.06

Cellulose acetate �31.87 0.16 �31.96 0.16

Alpha glucose �10.91 0.14 �10.92 0.14

Sample correction factorsM value 1.0155 1.0196

B value �10.43 �10.433

K. Jones et al. / Forensic Science International 231 (2013) 364–374 371

the first and second replicate being lost due to memory effect andbeing disregarded as an outlier.

To combat this, pairs of blank capsules were placed between thetwo international standards in both sets of standards (at the startand the end of the analytical run). This resulted in almost removingthe memory effect observed and the measurement values couldmore readily be used instead of being lost as outliers.

Where unknown sample values are expected to change, a pair ofblank capsules should be inserted between the samples. The use ofblank capsules was not found to be beneficial when the unknownsamples were expected to be within the same range of values (upto approximately 10% range). This means that for an analytical runconsisting only of paper samples, no blanks are required betweenthe samples. However if an analytical run contains both paper andglucose/sucrose samples, a pair of blanks should be inserted.

3.1.4.5. Use of a quality control check. During this set of experi-ments it became apparent that it was difficult to know when anexperimental run was fit for use. It is recommended that either:

- The international standard IAEA-CH-3 Cellulose be included mid-sequence as an unknown to check the accuracy of the instrumentand correction applied; or

- A suitable laboratory standard material (such as those used inthese experiments) be run as an unknown to check against aconsensus value.

Overall, the quality control check should be matrix matched tothe target sample and should lie close to the expected value of theother unknown samples (i.e. cellulose for cellulose or papersamples, a glucose for sugars).

Fig. 7. Plot of corrected measurements of international standard polyethylene over

a 24-month period. Mean value and 95% confidence interval represented by the

solid and dashed lines.

3.2. Method validation

3.2.1. Precision and stability

Mean corrected measurements for the polyethylene andsucrose international standards are shown in Figs. 7 and 8. Themean value (solid line) and 2 � standard deviations (dashed lines)are represented on each plot. 2� the standard deviation is used torepresent the 95% confidence interval for each standard. Anyvalues considered to be outliers within the original analytical runwere excluded. Table 8 is a summary of the individual measure-ments, mean, standard deviation, 95% confidence interval (95% CI)and number of measurements relating to the data represented forthese standards.

It is important to note with this data that routine maintenance(changing of packing materials, gas tanks and drying agents) andservicing were conducted on numerous occasions during thisperiod. The polyethylene and sucrose measurements here show no

Fig. 8. Plot of corrected measurements of international standard sucrose over a 24-

month period. Mean value and 95% confidence interval also represented by the solid

and dashed lines.

Fig. 9. Plot of corrected measurements of international standard cellulose over a 24-

month period. Mean value and 95% confidence interval also represented by the solid

and dashed lines.

Table 8Summary of the analytical results obtained for three international (known) standard materials.

d13C mean (%) St. dev. (%) 95% CI (%) Number samples d13C published value (%) Published st. dev. (%)

Polyethylene �32.15 0.15 0.30 517 �32.151 0.05

Sucrose �10.45 0.16 0.32 488 �10.449 0.033

Cellulose �24.69 0.12 0.24 129 �24.724 0.041

K. Jones et al. / Forensic Science International 231 (2013) 364–374372

issues with precision and the values are stable over time. Thecorrected mean values are comparable to the published valuesoverall. The standard deviations are close to or below the publishedexpected precision of measurement of 0.15% for carbon measure-ments [27].

3.2.2. Repeatability

The international standard cellulose was used as a qualitycontrol standard within analytical runs. The corrected measure-ments for international standard cellulose as measured over a 12-month period are shown in Fig. 9. The mean value (solid line) andthe 95% confidence interval (dashed lines) are represented on theplot. Table 8 summarizes the individual measurements, mean,standard deviation, 95% confidence interval (95% CI) and number ofmeasurements relating to the data represented for cellulose.

3.2.3. Accuracy

Three materials being considered for development as laborato-ry/working standard materials – cellulose acetate, medium fiber

Fig. 10. Accuracy results for medium cellulose.

Fig. 11. Accuracy results for cellulose acetate.

cellulose and alpha glucose – were run over time and plotted todetermine the accuracy with which the materials could bemeasured. Figs. 10–12 show the mean measurement for eachrun, with the error bars denoting the 95% confidence interval(2 � standard deviations) for that sample set. Table 9 summarizesthis data. Overall, each material was found to have a standarddeviation less that 0.15%.

3.2.4. Robustness

Operators A, B and C are different operators who preparedreplicates of the same set of samples. Operator D is where operatorC has prepared, run and corrected an additional set of the samesamples. The mean results have been plotted in Fig. 13. Table 10summarizes these results.

The data shows that while all four analytical runs producedcomparable results for each measurement, small differences weredetected in the precision and range. The plotted results in Fig. 13indicate that the differences observed were not consistent acrossall sample types, suggesting that they were not due to a systematicor sampling error.

The Kruskal–Wallis test for the comparison of the medians ofnon-parametric populations was performed on these values toidentify whether the differences observed between the samplesprepared by different operators were significant. Dunn’s post hoctest was used to identify which groups were causing thedifferences. Significant differences (to 95% confidence level) wereobserved between:

- Operators A and B for the cellulose acetate samples (actualdifference of 0.3%).

- Operators A and B for the medium cellulose samples (actualdifference 0.39%).

The actual difference in the d13C values for these two samplescould be accounted for in the natural variation of the materials, themeasurement uncertainty associated with the method or due tosome instrumental difference. The last of these options is the morelikely given that the instrument is known to be quite variable whentechnical issues arise. Although statistically these values are

Fig. 12. Accuracy results for alpha glucose.

Table 9Summarized results from accuracy experiments of cellulose acetate, medium

cellulose and alpha glucose.

d13C mean (%) St. dev. (%) 95% CI (%) Number

samples

Medium cellulose �26.82 0.09 0.19 63

Alpha glucose �11.15 0.14 0.28 94

Cellulose acetate �32.06 0.12 0.23 99

Fig. 13. Mean d13CVPDB (%) values obtained for samples prepared and run by three

different operators.

Table 11d13CVPDB (%) results from FIRMS inter-laboratory (ILC-6) trial of two materials.

Material Mean d13C

reported (%)

FIRMS group

grand mean

d13C (%)

d13C H15

adjusted (%)

Z-score

Glycine �26.35 � 0.06 �26.09 �26.10 � 0.20 �1.23

4-Nitroacetanilide �33.22 � 0.09 �32.95 �32.95 � 0.32 �1.35

K. Jones et al. / Forensic Science International 231 (2013) 364–374 373

different, practically these differences are not large and are notindicative of a systemic issue with the method.

3.2.5. Measurement uncertainty

The expanded uncertainty for the measurement of carbonisotope ratios was calculated using the Guide to ExpressingUncertainty in Measurement (GUM) approach [28], technicalguidelines produced by NATA [14] and the equations outlined inBenson et al. [8].

Using the precision experiments conducted on the potentiallaboratory standards medium cellulose and cellulose acetate, theexpanded uncertainty for the measurement of carbon isotoperatios in cellulose was determined to be 0.26% (with a coveragefactor of 2). The bias (calculated using the precision results forinternational cellulose) was found to be insignificant.

Table 10Mean d13CVPDB (%) and standard deviation for samples prepared by 3 different

operators.

Operator A Operator B Operator C Operator D

International cellulosed13C mean (%) �24.59 �24.55 �24.61 �24.47

St. dev. (%) 0.13 0.28 0.14 0.16

Cellulose acetated13C mean (%) �31.9 �32.2 �32.05 �32.04

St. dev. (%) 0.02 0.07 0.08 0.07

Alpha glucosed13C mean (%) �11.05 �11.05 �11.20 �11.33

St. dev. (%) 0.02 0.04 0.16 0.36

Medium cellulosed13C mean (%) �26.54 �26.93 �26.8 �26.76

St. dev. (%) 0.07 0.11 0.01 0.07

Paperd13C mean (%) �27.48 �27.61 �27.56 �27.61

St. dev. (%) 0.08 0.06 0.01 0.1

ANU sucrosed13C mean (%) �11.42 �11.50 �11.48 �11.59

St. dev. (%) 0.04 0.04 0.13 0.16

Likewise, using the precision experiments conducted on alphaglucose, the expanded uncertainty for the measurement of carbonisotope ratios in glucose was determined to be 0.3% (with acoverage factor of 2).

3.2.6. Inter-laboratory trial

The 2010 forensic isotope ratio mass spectrometry (FIRMS)group trial was undertaken using the experimental method andcorrection calculations described. Two types of samples weremeasured and run and the d13C values for 10 replicates of eachsample were reported to be included as part of this trial.

The results reported, the consensus values for the test and theindividual Z-score calculated based on these results are presentedin Table 11. Based on these Z-scores, the method here producesresults that are in consensus with other laboratories whenmeasuring carbon isotope ratios of the same materials.

4. Conclusions

Previous research by our group (published as part one of thisseries) indicated that IRMS was a beneficial addition for thecomparison of document papers that are otherwise indistinguish-able. The present study further demonstrates the value of thistechnique when applied to forensic paper analysis by demonstrat-ing that the instrument and the analytical method were fit forpurpose for the measurement of document papers, cellulose andglucose samples.

The experiments described in this study outline the methoddevelopment and validation undertaken. The overall measurementuncertainty was calculated to be approximately 0.26% (with acoverage factor of 2). This estimate included uncertainty due to theinstrumentation, methods and materials used. All performancecharacteristics including precision, accuracy, repeatability androbustness were within the 0.26% expanded measurementuncertainty range. The method and instrumentation wereobserved to be fit for purpose during an inter-laboratory trialrun by the FIRMS group.

Further research is needed, and is being undertaken byour group, with the aim of routinely utilizing IRMS for the forensicexamination of paper. This includes expanding the isotopesconsidered in addition to wider international and blind studies.

References

[1] S. Benson, C. Lennard, P. Maynard, C. Roux, Forensic applications of isotope ratiomass spectrometry – a review, Forensic Sci. Int. 157 (2006) 1–22.

[2] J. Aggarwal, J. Habicht-Mauche, C. Juarez, Application of heavy stableisotopes in forensic isotope geochemistry: a review, Appl. application of stablehydrogen and oxygen isotopes to wildlife forensics Oecologia 143 (2005)337–348.

[4] J. Carter, R. Sleeman, J. Hill, F. Idoine, E. Titterton, Isotope ratio mass spectrometryas a tool for forensic investigation (examples from recent studies), Sci. Justice 45(2005) 141–149.

[5] W. Papesch, M. Horacek, Forensic applications of stable isotope analysis: casestudies of the origins of water in mislabeled beer and contaminated diesel fuel,Sci. Justice 49 (2009) 138–141.

[6] A. van Es, J. de Koeijer, G. van der Peijl, Discrimination of document paper by XRF,LA–ICP–MS and IRMS using multivariate statistical techniques, Sci. Justice 49(2009) 120–126.

[7] W. Meier-Augenstein, Stable Isotope Forensics: An Introduction to the ForensicApplication of Stable Isotope Analysis, John Wiley & Sons, 2010.

K. Jones et al. / Forensic Science International 231 (2013) 364–374374

[8] S.J. Benson, C.J. Lennard, D.M. Hill, P. Maynard, C. Roux, Forensic analysis ofexplosives using isotope ratio mass spectrometry (IRMS). Part 1: Instrumentvalidation of the DELTAplus XP IRMS for bulk nitrogen isotope ratio measurements,J. Forensic Sci. 55 (2010) 193–204.

[9] International Organisation for Standardisation. ISO/IEC 17025:2005. http://www.iso.org/iso/iso_catalogue/catalogue_tc/caatalogue_detail.htm?csnumber=39883.

[10] National Association of Testing Authorities. ISO/IEC 17025 Field ApplicationDocument Forensic Science – Supplementary Requirements for Accreditation.http://www.nata.asn.au/phocadownload/publications/FADS_Amendments/ForensicScienceFAD.pdf (01.02.12).

[11] National Association of Testing Authorities. Technical Note 17 – Guidelines for theValidation and Verification of Chemical Test Methods. http://www.nata.asn.au/phocadownload/publications/Guidance_information/tech-notes-information-papers/technical_note_17.pdf (29.03.13).

[12] International Organsation for Standardisation. ISO 5725-6:1994. Accuracy (True-ness and Precision) of Measurement Methods and Results – Part 1: GeneralPrinciples and Definitions. http://www.iso.org/iso/catalogue_detail.htm?csnumber=11837 (29.03.13).

[13] EURACHEM. The Fitness for Purpose of Analytical Methods; A Laboratory Guide toMethod Validation and Related Topics. http://www.eurachem.org/images/stories/Guides/pdf/valid.pdf1998.

[14] National Association of Testing Authorities. Technical Note 33 – Guidelinesfor Estimating and Reporting Measurement Uncertainty of Chemical TestResults. http://www.nata.asn.au/phocadownload/publications/Guidance_information/tech-notes-information-papers/technical_note_33.pdf(29.03.13).

[15] National Association of Testing Authorities. Technical Note 28 – In-house Cali-bration and Measurement Uncertainty http://www.nata.asn.au/phocadownload/publications/Guidance_information/tech-notes-information-papers/technical_note_28.pdf (29.03.13).

[16] National Association of Testing Authorities. Policy Circular 12 – NATA Require-ments for the Performance of Calibrations In-house. http://www.nata.asn.au/publications/category/8-policy-and-technical-circulars.

[17] Joint Committee for Guides in Metrology. GUM: Guide to the Expression ofUncertainty in Measurement. http://www.bipm.org/en/publications/guides/gum.html.

[18] D. Hibbert, J. Gooding, Data Analysis for Chemistry, Oxford Univ. Press, New York,2006.

[19] Australia S. 1301.405s-2004 Methods of test for pulp and paper – Grammage ofnon-creped paper and board, 29 March 2013.

[20] Inter-Laboratory Comparison Results (ILC-6). http://www.forensic-isotopes.org/assets/FIRMSILC6.pdf (29.03.13).

[21] Robust Statistics: A Method of Coping with Outliers. Analytical Methods Com-mittee Technical Brief, 2001 No 6.

[22] Brief AMCT. The ZL Score – Combining your Proficiency Test results with yourOwn Fitness-for-purpose Criterion. http://www.rsc.org/images/zL-score-profi-ciency-fitness-criterion-technical-brief-2_tcm18-214844.pdf (29.03.13).

[23] I. Platzner, K. Habfast, A. Walder, A. Goetz, Modern Isotope Ratio MassSpectrometry, Wiley, Chichester, 1997.

[24] P. De Groot, Handbook of Stable Isotope Analytical Techniques, vol. II, ElsevierScience Ltd., Amsterdam, 2008.

[25] H. Al-Hosney, V. Grassian, Water, sulfur dioxide and nitric acid adsorption oncalcium carbonate: a transmission and ATR-FTIR study, Phys. Chem. Chem. Phys.7 (2005) 1266–1276.

[26] I. Evett, G. Jackson, J. Lambert, S. McCrossan, The impact of the principles ofevidence interpretation on the structure and content of statements, Sci. Justice 40(2000) 233.

[27] W.A. Brand, High precision isotope ratio monitoring techniques in mass spec-trometry, J. Mass Spectrom. 31 (1996) 225–235.

[28] NMI, National Measurement Institute, Measurement Uncertainty for ChemistsCourse, Participants Handout, 2011.