Proling and classication of illicit heroin by ICP-MS analysis of

inorganic elements
Cuimei Liu
a,b
, Zhendong Hua
b
, Yanping Bai
b
, Yao Liu
a,
*
a
College of Forensic Science, Peoples Public Security University of China, Beijing, China
b
Drug Intelligence and Forensic Center of Minister of Public Security, Beijing, China
1. Introduction
Heroin is one of the most widely abused drugs in China.
According to the Annual Report on Drug Control in China, the
registered drug users reached to 2.10 million by the end of 2012,
among which 1.24 million were heroin dependents [1]. The
majority of the heroin consumed in China comes from Golden
Triangle and Golden Crescent, but the ratio of them varies every
year. Therefore, heroin proling, especially the geographical origin
determination, is vital to supporting the law enforcement agency
for both evidential and strategic intelligence purposes [2].
Several methods for heroin proling have been developed by
determination of the major alkaloids [3,4], acid/neutral by-
products [5], and occluded solvents [6] in recent years. All these
methods are based upon organic impurities, whereas another
option for drug proling is to study the inorganic composition.
Elemental analysis has been used to trace the origin or production
process of many illicit drugs as MDMA [7], ecstasy tablets [8,9],
methamphetamine [10,11], and cannabis [12]. However, its use in
the identication of heroin origin is somewhat less emphasized, as
most of the studies concerned focused on the optimization of
determination methods [1318] or the variation of elemental
composition of heroin samples [1921]. Classications of heroin
seizures by elemental prole were also reported by a few works.
[2224]. R.J. Wells et al. applied ICP-MS and hierarchical cluster
analysis to classify 126 individual analyses from 96 separate heroin
seizures, and found that there was a close statistical correlation
between seizures from the same geographical region [22]. R. Myors
et al. analyzed 73 elements by ICP-MS, and several statistical
procedures were adopted to differentiate 76 SEA and 20 non-SEA
heroin samples [23]. Kar-Weng et al. classied 309 street heroin
samples into two groups by using ICP-MS and principal component
analyses (PCA) [24]. These works have provided an effective way to
establish links between seizures, but rarely could the criterion of
origin determination be given due to the lack of sufcient
authentic samples as well as unknown samples for method
validation.
This study presents a new strategy to identify heroin samples
from Golden Triangle and Golden Crescent based on inorganic
ngerprinting. The production of heroin contains several steps,
such as poppy growth, opium harvest, morphine purication and
nally acetylation of morphine. The sort and amount of the
elements introduced in each step are inuenced by natural
environment and manufacture custom, thus ensure the possibility
of origin determination by investigating the inorganic composi-
tion. In this study, a classication model for the origin determina-
tion was established and validated upon the ICP-MS analysis of 417
authentic heroin samples. Then it was applied to 907 unknown
Forensic Science International 239 (2014) 3743
A R T I C L E I N F O
Article history:
Received 3 July 2013
Received in revised form 19 January 2014
Accepted 9 February 2014
Available online 18 February 2014
Keywords:
Heroin proling
ICP-MS
PLS-DA
Origin determination
Inorganic elements
A B S T R A C T
Nineteen inorganic elements (Ag, As, Ba, Cd, Co, Cr, Cu, Mn, Mo, Ni, P, Pb, Se, Sb, Th, Tl, U, V and Zn) in
heroin samples were determined using inductive coupled plasma mass spectrometry (ICP-MS). After
WilcoxonMannWhitney test and correlation analysis, 10 element contents (P, V, Cr, Ni, Cu, Zn, As, Se,
Pb, U) and 7 element ratios (U/Ba, Ba/Pb, Cd/Mn, Co/Ni, V/Cr, P/V, Cd/V) were found to be evidently
different between heroin samples from"Golden Crescent" and "Golden Triangle". Based on the data set of
these 17 variables in 150 authentic heroin samples, classication of origins was successfully achieved
utilizing principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA). By
comparison experiment on 907 unknown samples, the developed discriminant model was proven to be
consistent with the widely used organic proling method, and meanwhile the time consumed per
sample was markedly saved, which facilitates high throughput screening in routine analysis.
2014 Published by Elsevier Ireland Ltd.
* Corresponding author. Tel.: +86 10 62975145; fax: +86 10 62975145.
E-mail address: liuyao1123@aliyun.com (Y. Liu).
Contents lists available at ScienceDirect
Forensic Science International
j our nal h o mepage: w ww. el sevi er . co m/ l oc at e/ f o r sc i i nt
http://dx.doi.org/10.1016/j.forsciint.2014.02.002
0379-0738/ 2014 Published by Elsevier Ireland Ltd.
samples, and was proven to be reliable by comparing with results
of the classic proling method consisted of major alkaloid proling
(SIG I) and acidic & neutral components proling (SIG II). This
successful application conrmed the possibility of geographical
origin determination by inorganic prole, which could be
determined more quickly and easily than organic proles.
2. Experimental
2.1. Reagents and standards
Nitric acid (65%, w/w) and hydrogen peroxide (30%, v/v), both
ultrapure reagent grade, were purchased from Merck (Germany).
18.2 MV/cm de-ionised water (Milli-Q, USA) was used throughout.
Multi-element standard (10 mg/L of Ag, As, Ba, Cd, Co, Cr, Cu,
Mn, Mo, Ni, Pb, Sb, Se, Th, Tl, U, V and Zn) was purchased from
Agilent (USA). 1000 mg/L of P single-element standard was
purchased from Central Iron & Steel Research Institute (China).
The element quantication was assessed using external
standard method. For the determination of P, the calibration
curves were 0, 1, 10, 50, 200 and 1000 mg/L and for all the other
elements the concentrations of the multi-elemental standard
solutions were 0, 0.001, 0.005, 0.01, 0.5, 2, 10, 50, 100 mg/L.
All polyethylene plastic bottles were rinsed with de-ionised
water before use.
2.2. Sampling
417 authentic heroin samples, which consists 250 Golden
Triangle samples and 167 Golden Crescent samples, were
selected for the prediction modeling and model verication. All of
the 167 Golden Crescent samples were from Afghanistan. 24 of
them were provided by the ministry of interior of Afghanistan
from the drug exchange program between China and
Afghanistan. The others were provided by the Anti-smuggling
Bureau of Guangzhou Custom of China. These samples were
collected form seizures of body concealment cases uncovered in
the airport, in which the heroin was conrmed to be from
Afghanistan by both the ight information and the statement of
suspects. All of the 167 Golden Crescent samples were from
Myanmar. 128 of them were entry captured between the
boundary of Yunnan province of China and the northern part of
Myanmar. 22 of themwere seized during the on-site opium eld
surveys of NNCC (the National Narcotics Control Commission) of
China in Myanmar.
907 unknown heroin samples were selected for the comparison
of the established inorganic proling method with two organic
proling methods. These samples were randomly selected from
heroin seized in different cities of China during the year of 2010
2012 and their origin information was unavailable.
A representative control sample (CT) was prepared by thorough
homogenization of 15 g of one typical Golden Triangle sample
and 15 g of one typical Golden Crescent sample. This mixed
control sample was chosen instead of using two different control
samples to save time.
2.3. Sample preparation
2.3.1. Microwave digestion
An ETHOS 1 model (Milestone) microwave digestion system
equipped with an internal pressure and temperature control
system was used to achieve sample digestion. Approximately
250 mg of each homogenized sample was weighed and placed
inside a cleaned Teon PFA vessel. 6 mL of 65% nitric acid and 1 mL
of 30% hydrogen peroxide were added. The vessels were closed
tightly and then placed in the microwave. The microwave was
subjected to the following conditions: starting from the room
temperature, increased to 120 8C in 5 min, and kept for 5 min, then
increased to 170 8C in 5 min, kept for another 20 min. The power
was set at 1200 W for the whole process. Samples were cooled
after the digestion procedure and then transferred to 50 mL
polyethylene plastic bottles and diluted to 50 mL with water. The
end volumes of the solution were determined gravimetrically. A
reagent blank solution was prepared according to the same
procedure applied to the sample.
2.3.2. Ultrasonic assisted dissolution
Approximately 50 mg of each homogenized sample was
weighed into a 50 mL polyethylene plastic bottle. 10 mL of 5%
nitric acid solution was added with a 10 mL plastic pipette, and the
end volumes of the solution were determined gravimetrically. The
solution was shaken vigorously and sonicated for 10 min. Each
solution was checked for the presence of particles. Another 10 min
of sonication may be required to dissolve the particles. If the
particles were not dissolved with further sonication, ltration was
performed using a 0.45 mm lter membrane. A reagent blank
solution was prepared according to the same procedure applied to
the sample.
2.4. Instrumentation
An Agilent 7700 s inductively coupled plasmamass spectrom-
etry with ORS collision/reaction cell (CRC) was used in this study.
The instrument was operated in standard He mode to remove the
possible polyatomic interferences. A peristaltic pump was used for
sample uptake, and a glass concentric spray chamber, and a micro-
mist nebulizer (Agilent, USA) were used for sample solution
nebulization. A 10 mg/L mixed internal standard solution (Bi, Ge,
In, Li, Lu, Rh, Sc and Tb) was added on-line to the sample uptake,
which allowed the automatic addition of the internal standard
elements to standard, blank and sample solutions.
Typical instrumental operating conditions used were 1510 W
forward power, 15 L/min plasma ow, 1.0 L/min auxiliary gas ow,
0.77 L/min carrier gas ow, 0.40 L/min makeup gas ow, respec-
tively. A peak hopping scan mode was used with a dwell time of 1 s
for As and Se, and 30 ms for other elements.
The following elements were analyzed with ICP-MS in the
present study (isotope used): Ag (107), As (75), Ba (137), Cd (111),
Co (59), Cr (52), Cu (63), Mn (55), Mo (95), Ni (60), P (31), Pb (208),
Se (82), Sb (121), Th (232), Tl (205), U (238), V (51), Zn (66).
2.5. Organic component proling (SIG I & SIG II)
To evaluate the developing method, two classic organic
component proling methods were employed for comparison.
The rst one was major alkaloid proling (SIG I) developed by Lurie
et al. [3] using capillary zone electrophoresis. Typically, 20 mg
heroin equivalent of each sample was weighed into a 50 mL
volumetric ask and diluted to volume with a mixture of methanol
and 3.75 mM phosphate buffer (v/v = 2:8). After a 15 min
sonication, the solution was ltered, and 1 mL was analyzed on
an Agilent 7100 capillary electrophoresis system. The run buffer
preparation and data processing were performed according to the
literature [3]. The second one was acidic and neutral manufactur-
ing impurities proling (SIG II) developed by Morello et al. [5]
using GCMS. Typically, 45 mg morphine equivalent of each
sample was placed into a centrifuge tube and dissolved in 5 mL of a
mixture of petroleum ether and methylene chloride (v/v = 3:2).
4 mL sulfuric acid (2 N) was added, and after vortex and
centrifugation the organic phase was isolated and dried. The
residue was derivatized with MSTFA, and then analyzed on an
Agilent GCMS system (6890A GC, 5975C MS) using HP-5MS
C. Liu et al. / Forensic Science International 239 (2014) 3743 38
capillary column. The instrument parameters and origin determine
criterions were seen in literature [5].
2.6. Statistical software
Normality test, non-parametric test and correlation analysis of
Pearson distance were computed using SPSS 17.0 version (SPSS,
Inc., Chicago, IL). PCA and PLS-DA analysis was performed using
SIMCA software (SIMCA P13.0, Umetrics, Sweden).
3. Results and discussion
3.1. Method development
3.1.1. Selection of elements
Elements were selected on the basis of accurate quantication
and timesaving. Since the objective was to distinguish the
geographical origin, macro elements (K, Na, Ca, Mg, Fe) which
could be easily contaminated during the processing procedure or
introduced by cutting agent were excluded at the very beginning.
The contents of lanthanides in authentic samples were found to be
rather low in the preliminary test, hence, they were also excluded.
Some elements like Si and Ti required special treatment in sample
preparation were not analyzed to simplify the method. Finally, 19
elements were selected as targets for ICP-MS analysis.
3.1.2. Sample preparation
For determination of inorganic elements in illicit drugs,
microwave digestion [16,17,19,20] and ultrasonic assisted disso-
lution [18] were commonly used. Their capacities were compared
by analyzing the control heroin sample for 10 times with both
methods (see Table 1). The limit of detection (LOD) and limit of
quantication (LOQ) were set to 3 and 9 times the standard
deviation in the concentration of each element in the control blank,
respectively.
Higher recoveries of most elements were observed with
microwave digestion, which may be attributed to its complete
digestion under high temperature and pressure. However,
ultrasonic assisted dissolution achieved satised reproducibility
with RSDs less than 30% [22,26] for all elements, while 8 elements
exceeded that criterion with microwave digestion. Since, the aim
was to establish an origin determination method rather than
accurate measure element contents of heroin samples, reproduc-
ibility was more crucial than recovery. Thus ultrasonic assisted
dissolution was adopted in the subsequent analysis, as its quick
and fast procedure also benetted its use in high throughput
analysis.
3.2. Analysis of authentic heroin samples
417 authentic heroin samples, 250 from Golden Triangle and
167 from Golden Crescent, were analyzed by ICP-MS with
ultrasonic assisted dissolution (See Table 2). In order to facilitate
the statistical calculation, the content of an element was set to LOD
if not detected, and LOQ was used instead when the content was
found to be between LOD and LOQ. Generally, the contents of Mn,
Co, Cu, Zn, Ba, Tl, Pb and Th in Golden Triangle samples were
higher than that in Golden Crescent samples, while the results of
P, V, Cr, Ni, As, Se, Mo, Ag, Cd, Sb and U were opposite.
Table 1
Comparison of two sample preparation methods. LOQ was calculated from the results of control blank (n = 10), where average and RSD were calculated from the results of
control sample (n = 10).
Element Microwave digestion Ultrasonic assisted dissolution Differencec (%)
LOQ (mg/kg) Average (mg/kg) a RSD (%) LOQ
(mg/kg)
Average (mg/kg) b RSD (%)
P 0.235 42.7 4.5 0.259 41.6 6.1 2.6
V 0.00916 0.0147 15 0.00864 0.0118 16 22
Cr 0.0204 2.48 17 0.0212 1.50 18 49
Mn 0.0396 2.58 3.8 0.0625 2.11 1.7 20
Co 0.00104 0.0164 35 0.00405 0.0145 6.0 12
Ni 0.0106 0.42 25 0.0101 0.194 6.6 74
Cu 0.00265 0.199 20 0.00564 0.183 2.5 8.4
Zn 0.00591 7.17 3.0 0.0175 7.90 1.8 9.7
As 0.0695 0.0774 1.2 0.0694 0.107 1.6 32
Se 0.160 0.170 6.8 0.108 0.140 1.9 19
Mo 0.0216 0.540 3.7 0.0220 0.0250 9.6 182
Ag 0.00227 0.0048 31 0.00153 0.00380 20 23
Cd 0.0344 0.0378 4.6 0.0344 0.0355 5.0 6.3
Sb 0.0915 0.140 52 0.0804 0.0895 9.4 44
Ba 0.00426 0.510 35 0.0166 0.174 8.5 98
Tl 0.00111 0.00190 43 0.00116 0.00150 13 24
Pb 0.00247 0.183 35 0.00286 0.172 19 6.2
Th 0.000787 0.0075 35 0.00109 0.00117 29 146
U 0.000176 0.000587 47 0.000132 0.000440 18 29
Difference (c) = 2 100 (a b)/(a + b).
Table 2
Summary of elemental compositions found in 250 Golden Triangle and 167
Golden Crescent heroin samples.
Element Golden Triangle heroin Golden Crescent heroin
Mean (mg/kg) RSD (%) Mean (mg/kg) RSD (%)
P 0.731 84 353 244
V 0.00797 50 0.0115 54
Cr 0.443 273 1.57 139
Mn 4.2 147 2.2 64
Co 0.0245 116 0.0136 126
Ni 0.0998 153 0.21 176
Cu 0.352 116 0.169 151
Zn 19.3 76 5.66 70
As 0.0494 123 0.247 70
Se 0.108 40 0.478 186
Mo 0.012 214 0.0255 101
Ag 0.00364 41 0.00421 48
Cd 0.0355 10 0.0357 26
Sb 0.0894 56 0.092 35
Ba 0.262 74 0.183 59
Tl 0.00142 64 0.00074 85
Pb 0.245 63 0.0501 189
Th 0.00158 108 0.00134 95
U 0.000232 21 0.00226 75
C. Liu et al. / Forensic Science International 239 (2014) 3743 39
3.3. Selection of signicant variables
3.3.1. Non-parametric tests
The t-test and the WilcoxonMannWhitney (WMW) test are
often used when comparing the difference between the means or
medians of continuous variables in two independent groups. T-test
is parametric test, which assumes that the variables of interest are
normally distributed. While the non-parametric WMWtest, as an
alternative to parametric test, is mostly used when there is
evidence of non-normality distributed data and/or small unequal
data sets [25,26].
A data set of element contents in 75 Golden Triangle samples
and 75 Golden Crescent samples was selected for the test. The
normality test of the variables in this data set showed that most of
the 19 elements were non-normally distributed, so the WMWtest
was preferred. 10 elements obtained p-values less than 0.005 in the
test (See Table 3), indicating their particularly signicant
differences between the two classes. It has been mentioned that
some elements in heroin were correlated with others [11,12,17],
hence, 342 element ratios of 19 elements were also checked by the
WMW test and 13 element ratios were evidently different (See
Table 3). Finally, the 23 variables consisted of 10 elements and 13
element ratios were selected for the later test.
3.3.2. Study on the correlation between target variables
Correlation analysis helps to nd the variable pairs coexisting in
a liner manner, which avails the reduction of variables. Thus the
Pearson distance between each pair of the 23 signicant variables
was calculated to estimate their correlation in 150 authentic
heroin samples. As the coefcient value (R) of +1.00 presented the
perfect positive correlation, several groups with coefcient value
more than 0.8 were considered to be strongly correlated:
Cu and Cu/As with R = 0.87;
Zn and Zn/As with R = 0.81;
As, As/Ag, As/Cd, As/Ba, As/Mn with R from 0.86 to 1.00.
One variable of each group was sufcient for proling, namely 6
variables of Cu/As, Zn/As, As/Ag, As/Cd, As/Ba and As/Mn could be
excluded and 17 variables were left.
3.4. Evaluation of data pre-treatment methods
The choice of data pre-treatment method is a key factor in
statistical modeling since it affects the quality of the information
extracted. For heroin samples, marked differences of the order of
magnitude existed between variables selected. For instance, the
average content in Golden Crescent samples was 353 mg/kg for P
compared with 2.26E-03 mg/kg for U (See Table 2), which
indicated the requirement of variable scaling to regulate the
relative importance of each variable [27]. Five automatic scaling
procedures were available in SIMCA software: UV scaling (the
variables are centered and divided by their standard deviation);
UVN scaling (same as UV, but the variable is not centered); Par
scaling (the variables are centered and divided by the square root
of their standard deviation); ParN scaling (same as Par, but the
variables are not centered); Ctr scaling (the variable is centered but
not scales). UV scaling give each variable an equal chance of being
expressed in the statistical analysis, while with Ctr scaling the
inuence of a variable is related to its amplitude hence, low
content elements have little inuence. Par scaling is a compromise
between them, with which the inuence of low amplitude
variables is enhanced. And centering of the variables before
scaling is helpful to reduce the distortion of results induced by
multicollinearity.
The applicability of the ve scaling methods was veried by
PLS-DA analysis of the raw data of 17 variables. Two parameters
R
2
(Y) and Q
2
were taken into account (see Table 4), where the
former one dened the proportion of the variance of the response
variable explained by the model and the latter one expressed the
cumulative proportion of the variance of the variables that can be
predicted by the model [28]. It could be seen UV, Par and Ctr scaling
gave higher R
2
(Y) and Q
2
than UVN, ParN and no scaling,
respectively. This demonstrated a certain degree of multicolli-
nearity between the selected variables, thus centering of the data is
necessary. Overall, UV scaling obtained the best results, with which
only one principal component (PC) was sufcient to explain more
than 80 percent variance of the variables. It meant that all 17
variables had the same importance regardless of their amplitude.
In other words, low content elements, such as V and U, were also
essential for the origin determination of heroin samples. In
summary, UV scaling was the optimum selection for data pre-
treatment, and was adopted in the later PCA and PLS-DA analysis.
3.5. PCA analysis
An unsupervised classication of the 150 authentic samples
was established by PCA analysis. In the two-dimensional score
scatter plot based on PC1 and PC2 (see Fig. 1a), Golden Triangle
samples and Golden Crescent samples were well separated,
which highlighted the possibility to discriminate heroin with
respect to the geography origins by only considering the inorganic
elements distribution. Moreover contribution of the variables for
classication was studied by the loading weights scatter plot (see
Fig. 1b). In general terms, Golden Triangle heroin were more
related to high levels of Cu, Pb, Zn, Cd/V, V/Cr, Cd/Mn and Co/Ni;
whereas Golden Crescent heroin showed signicantly high
amount of Ni, Cr, V, P, As, U, Se, Ba/Pb, P/V and U/Ba. This result
Table 3
Elements and element ratios with p-values less than 0.005 in WMW test.
P V Cr Ni Cu Zn As Se
p-value 5.17E17 6.83E05 9.83E10 4.95E05 2.35E05 9.12E11 2.34E15 4.97E03
Pb U Cu/As As/Ag As/Cd As/Ba U/Ba Ba/Pb
p-value 1.09E12 1.61E16 1.76E11 1.71E12 2.48E14 1.61E10 5.59E15 2.76E12
As/Mn Cd/Mn Co/Ni Zn/As V/Cr P/V Cd/V
p-value 1.99E15 1.88E09 9.74E12 6.43E17 2.46E04 2.02E16 5.53E16
Table 4
The R
2
(Y) and Q
2
of PLS-DA analysis with different scaling methods.
Scaling Number of PCs R
2
(Y) Q
2
NONE 2 0.449 0.311
UV 1 0.816 0.812
UVN 2 0.812 0.800
Par 4 0.695 0.595
ParN 2 0.466 0.452
Ctr 2 0.566 0.479
C. Liu et al. / Forensic Science International 239 (2014) 3743 40
agreed with the study by R.J. Wells et al. [22], in which the contents
of Cu, Zn, Cd were found to be higher and Mn, Ni to be lower in
sample from Myanmar than that from Pakistan.
In the work reported by K.W. Zhang et al. [24], Mn, Cu, Pb, Zn, Ni
and Cr were supposed to be partially introduced by water or rusty
container using in the manufacture and cutting process. Since the
authentic samples were mostly of high purity without cutting,
distinct difference of their contents between samples from
Golden Triangle and Golden Crescent could be attributed to
the conventional processing crafts used in different places, which
led to stable accumulation of these elements that facilitated their
use in origin determination. On the other hand, it also illustrated
that the 17 variables adopted represented not only the poppy
growing environment but also the manufacture process.
3.6. PLS-DA analysis
In PCA analysis, original gross variations are preserved as much
as possible in the rst few components, which may lead to poor
separation of the groups when the variability between groups is
less than that within groups. Alternatively, PLS-DA is a supervised
method that reveals the direct correlation between variables and
groups by a linear regression model. In this study, a discrete class
matrix (0 for Golden Triangle samples and 1 for Golden
Crescent samples) was added as Y-matrix to correlate proling
date with origins. After PLS-DA modelling using cross-validation,
one signicant component explaining 81.6% of the Y-variance
(Q
2
= 0.812) was obtained, indicating a good t of the model. From
distribution of the predicted Y values by PLS-DA model (Fig. 2a), a
threshold value of 0 was set to split the Golden Triangle samples
and Golden Crescent samples with least probability of false
classication. The PLS-DA coefcients (Fig. 2b) showed that Cu, Zn,
Pb, Cd/Mn, Co/Ni, V/Cr, Cd/V were positively correlated with
Golden Triangle samples and P, V, Cr, Ni, As, Se, U, U/Ba, Ba/Pb, P/
V were negatively correlated with Golden Triangle samples. This
result was consistent with PCA analysis but exhibited in a more
visual style.
3.7. The accuracy of the PLS-DA model
Another validation set of 175 authentic Golden Triangle
samples and 92 authentic Golden Crescent samples were used to
evaluate the established PLS-DA model. Predicted Y values for
Golden Triangle samples ranged from 4.55 to 0.72 with a mean
value of 1.11, and for Golden Crescent samples ranged from
1.10 to 6.48 with a mean value of 2.11 (Fig. 3). In spite of a certain
degree of overlap between their distributions, successfully origin
determination was achieved for 172 of 175 Golden Triangle
samples and 87 of 92 Golden Crescent samples with an overall
accuracy rate of 97% by using the threshold value 0.
3.8. Comparison of the developed method to two existing proling
method
The newly developed inorganic proling method was applied to
prole 907 heroin samples whose origin information was
Fig. 2. (a) Serial number of sample vs. t1 scores scatter plot of PLS-DA. Heroin was
geography origin labeled (1-Golden Triangle, 2-Golden Crescent). (b) Variable
coefcients plot of PLS-DA.
Fig. 1. (a) PC1 vs. PC2 scores scatter plot of PCA. Heroin was geography origin
labeled (1-Golden Triangle, 2- Golden Crescent). (b) P1 vs. P2 loadings scatter
plot of PCA.
C. Liu et al. / Forensic Science International 239 (2014) 3743 41
unavailable. These samples were randomly selected from seizures
in China during 2010 to 2012, and their purities ranged from 5% to
83%. For comparison, an organic proling method consisted of
major alkaloid proling (SIG I) by capillary zone electrophoresis
and acidic & neutral components proling (SIG II) by GCMS
following 2.5 was also employed. 826 samples were denitely
determined by organic proling method as the results of SIG I and
SIG II were accordant, among which 810 samples obtained the
same results by the inorganic proling method showing a high
similarity of 97.9% (see Table 5). The rest 16 samples with different
results were mostly low in purity, thus the wrong classications
might be attributed to the contaminant brought by cutting agent.
Meanwhile, 81 samples obtained opposite results in SIG I and SIG
II. This phenomenon has been discussed in a previous study [4] and
the possible explanation was that opium from one origin might be
transported into the other for processing into heroin, since
alkaloids determined by SIG I were related to the poppy growth
while acidic & neutral components determined by SIG II
represented the manufacture features. Interestingly, the results
of inorganic proling agreed with SIG I for half of these samples
and with SIG II for the other half. We inferred that the contents of
the elements involved in the method were also inuenced by both
the growing environment and the manufacture process, thus either
origin might be concluded in that case. Further study on the
inorganic ngerprint of raw opium and morphine was an effective
way to reveal the details.
SIG I and SIG II were demonstrated to be reliable and have
gained widely use in many forensic laboratories. However, various
kinds of organic reagents and several hours are required for their
sample preparation and testing. By contrast, one batch of samples
is handled within 20 min for ultrasonic assisted dissolution, and
high throughput of 5 min per sample is achieved by ICP-MS
analysis. Moreover, the inorganic proling method is environmen-
tally friendly since no organic reagent was used throughout.
4. Conclusions
An inorganic proling method for heroin geographical deter-
mination was developed based on ICP-MS analysis. Microwave
digestion and ultrasonic assisted dissolution were tested for
sample preparation, and the latter one showed satised reproduc-
ibility with high process speed. After optimization, 17 variables
consisted of 10 element contents (P, V, Cr, Ni, Cu, Zn, As, Se, Pb, U)
and 7 element ratios (U/Ba, Ba/Pb, Cd/Mn, Co/Ni, V/Cr, P/V, Cd/V)
were selected, and the discriminant model was built upon PLS-DA
analysis with UV scaling data pre-treatment. When applied to the
validation set of 175 authentic Golden Triangle samples and 92
Golden Crescent samples, the discriminant model was proven to
be effective with accuracy rate of 97%. Good agreement of the
developed method with the widely used organic proling method
(SIG I & SIG II) was also observed in the determination of 907
unknown samples.
Geographical origin determination of seizures is essential for
danger evaluation of different heroin origins. The successful
development of inorganic proling method indicated that the
heroin production process in different origins could be character-
ized by the elements involved in addition to organic impurities.
Furthermore, the reported method is fast, convenient and
environmentally friendly, which makes it an ideal solution for
high throughput screening of large samples.
Acknowledgement
The authors wish to thank the Program of National Science &
Technology Pillar (2011BAK04B06) for nancial support.
References
[1] Ofce of China National Narcotics Control Commission, Annual Report on Drug
Control in China, 2012.
[2] United Nations Ofce on Drugs and Crime, Methods for impurity proling of
heroin and cocaine, 2005.
[3] I.S. Lurie, P.A. Hays, A.E. Garcia, S. Panicker, Use of dynamically coated capillaries
for the determination of heroin, basic impurities and adulterants with capillary
electrophoresis, J. Chromatogr. A 1034 (2004) 227235.
[4] M. Collins, E. Casale, D.B. Hibbert, S. Panicker, J. Robertson, S. Vujic, Chemical
proling of heroin recovered from the North Korean merchant vessel Pong Su, J.
Forensic Sci. 51 (2006) 602697.
[5] D.R. Morello, S.D. Cooper, S. Panicker, J.F. Casale, Signature proling and classi-
cation of illicit heroin by GCMS analysis of acidic and neutral manufacturing
impurities, J. Forensic Sci. 55 (2010) 4249.
[6] J. Cartier, O. Gue niat, M.D. Cole, Headspace analysis of solvents in cocaine and
heroin samples, Sci. Justice 37 (1997) 175181.
[7] C. Koper, C. Boom, W. Wiarda, M. Schrader, P. Joode, G. Peijl, A. Bolck, Elemental
analysis of 3,4-methylenedioxymethamphetamine (MDMA): a tool to determine
the synthesis method and trace links, Forensic Sci. Int. 171 (2007) 171179.
[8] S. Comment, E. Lock, C. Zingg, A. Jakob, The analysis of ecstasy tablets by ICP-MS
and ICP-AES, Probl. Forensic Sci. XLVI (2001) 131146.
[9] R.J.H. Waddell, N. NicDaeid, D. Littlejohn, Classication of ecstasy tablets using
trace metal analysis with the application of chemometric procedures and articial
neural network algorithms, Analyst 129 (2004) 235240.
Fig. 3. (a) Histogram of the predicted Y values of samples in 175 Golden Triangle
heroin samples. (b) Histogram of the predicted Y values of samples in 92 Golden
Crescent heroin samples.
Table 5
Comparison of the results of organic & inorganic proling methods.
SIG I SIG II ICP-MS NUM Percentage (%)
1 + + + 810 89.3
2 + + 16 1.8
3 + + 41 4.5
4 + + 40 4.4
+ means the same classication result, means the different classication
result.
C. Liu et al. / Forensic Science International 239 (2014) 3743 42
[10] S.I. Suzuki, K. Nakajima, A. Matsushita, T. Nagao, Analyses of impurities in
methamphetamine by ICPMS and ion chromatography, J. Chromatogr. A 437
(1988) 322327.
[11] Y. Marumo, T. Inoue, S. Seta, Analyses of inorganic impurities in seized metham-
phetamine samples, Forensic Sci. Int. 69 (1994) 8995.
[12] R.J. Watling, Sourcing the provenance of cannabis crops using interelement
association patterns ngerprinting and LA-ICPMS, J. Anal. At. Spectrom. 13
(1998) 917926.
[13] B.B. Pilar, M.P. Antonio, M.P. Jorge, B.B. Adela, Direct determination of nickel in
heroin and cocaine by electrothermal atomic absorption spectrometry using
deuterium arc background correction combined with chemical modication, J.
Anal. At. Spectrom. 10 (1995) 10111017.
[14] B.B. Pilar, M.P. Antonio, M.P. Jorge, B.B. Adela, Determination of traces of chromi-
um in cocaine and heroin by ameless atomic absorption spectrometry, Talanta
43 (1996) 7787.
[15] B.B. Pilar, M.P. Antonio, M.P. Jorge, B.B. Adela, Application of rapid electrothermal
atomic absorption spectrometric methods to the determination of Ag, Al, Cd and
Mn in cocaine and heroin samples, Fresen. J. Anal. Chem. 358 (1997) 844847.
[16] B. Budic, S. Klemenc, Determination of trace elements in heroin by inductively
coupled plasma atomic emission spectrometry using ultrasonic nebulization,
Spectrochim. Acta Part B 55 (2000) 681688.
[17] A. Licsandru, V. Nacea, R. Boscencu, Microwave assisted digestion of heroin street
samples for trace metals analysis by inductively coupled plasma mass spectrom-
etry, Rev. Chim. 63 (2012) 8691.
[18] K.W. Chan, G.H. Tan, R.C.S. Wong, ICP-MS method validation for the analysis of
trace elements in illicit heroin, Anal. Lett. 45 (2012) 11221132.
[19] F. Infante, E. Domnguez, D. Trujillo, A. Luna, Metal contamination in illicit
samples of heroin, J. Forensic Sci. 44 (1999) 103110.
[20] T. Bora, M. Merdivan, C. Hamamci, Levels of trace and major elements in illicit
heroin, J. Forensic Sci. 47 (2002) 959963.
[21] N. Violante, M.G. Quaglia, A. Lopez, S. Caroli, Characterization of cocaine and
heroin samples as a function of their trace element content: an analytical pilot
study, Microchem. J. 45 (1992) 7989.
[22] R.J. Well, S.V. Skopec, R. Lavetz, J. Robertson, Trace element analysis of heroin by
ICP-MS, Chem. Aust. 63 (1995) 1415.
[23] R. Myors, R.J. Well, S.V. Skopec, P. Crisp, R. Iavetz, Z. Skopec, A. Ekangaki, J.
Robertson, Preliminary investigation of heroin ngerprinting using trace element
concentrations, Anal. Commun. 35 (1998) 403410.
[24] K.W. Chan, G.H. Tan, R.C.S. Wong, Investigation of trace inorganic elements in
street doses of heroin, Sci. Justice 53 (2013) 7380.
[25] P.J. Mumby, Statistical power of non-parametric tests: a quick guide for designing
sampling strategies, Mar. Pollut. Bull. 44 (2002) 8587.
[26] M.W. Fagerland, T-tests, non-parametric tests, and large studies-a paradox of
statistical practice? BMC Med. Res. Methodol. 12 (2012) 7884.
[27] C. Wang, H.W. Kong, Y.F. Guan, J. Yang, J.R. Gu, S.L. Yang, G.W. Xu, Plasma
phospholipid metabolic proling and biomarkers of type 2 diabetes mellitus
based on high-performance liquid chromatography electrospray mass spectrom-
etry and multivariate statistical analysis, Anal. Chem. 77 (2005) 41084116.
[28] C. Ducruix, D. Vailhen, E. Werner, J. Fievet, J. Bourguignon, J.C. Tabet, E. Ezan, C.
Junot, Metabolomic investigation of the response of the model plant Arabidopsis
thaliana to cadmium exposure: evaluation of data pretreatment methods for
further statistical analyses, Chemometr. Intell. Lab. Syst. 91 (2008) 6777.
C. Liu et al. / Forensic Science International 239 (2014) 3743 43