Author’s Accepted Manuscript

Cocaine profiling: implementation of a predictive

model by ATR-FTIR coupled with chemometrics
in forensic chemistry

Stefano Materazzi, Adolfo Gregori, Luigi Ripani,

Azzurra Apriceno, Roberta Risoluti
www.elsevier.com/locate/talanta

PII: S0039-9140(17)30145-5
DOI: http://dx.doi.org/10.1016/j.talanta.2017.01.045
Reference: TAL17227
To appear in: Talanta
Received date: 4 October 2016
Revised date: 11 January 2017
Accepted date: 15 January 2017
Cite this article as: Stefano Materazzi, Adolfo Gregori, Luigi Ripani, Azzurra
Apriceno and Roberta Risoluti, Cocaine profiling: implementation of a predictive
model by ATR-FTIR coupled with chemometrics in forensic chemistry, Talanta,
http://dx.doi.org/10.1016/j.talanta.2017.01.045
This is a PDF file of an unedited manuscript that has been accepted for
publication. As a service to our customers we are providing this early version of
the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting galley proof before it is published in its final citable form.
Please note that during the production process errors may be discovered which
could affect the content, and all legal disclaimers that apply to the journal pertain.
 Cocaine profiling: implementation of a predictive model by ATR-FTIR coupled with
chemometrics in forensic chemistry

Stefano Materazzi1, Adolfo Gregori2, Luigi Ripani2, Azzurra Apriceno1, Roberta Risoluti1*
1
Department of Chemistry – “Sapienza” University of Rome p.le A.Moro 5, 00185 Rome Italy
2
Carabinieri RIS – Scientific Investigation Department – v.le Tor di Quinto, Rome Italy

*Corresponding author: Roberta Risoluti, Department of Chemistry, Sapienza - University of Rome, Piazzale
Aldo Moro 5, 00185 Rome, Italy; Tel +390649913616, fax: +390649387137. e-mail address:
roberta.risoluti@uniroma1.it

Abstract
In this study, a strategy based on Infrared Spectroscopy with Fourier Transformed and Attenuated
Total Reflectance associated with chemometrics (ATR-FTIR) is proposed to identify the chemical
“fingerprint” of cocaine samples. To this end, standard mixtures of cocaine and cuttings at
differents ratio were investigated in order to develop a multivariate classification model to
simultaneously predict the composition of the samples and to obtain a profile of adulteration of
cocaine seizures. In addition, the application of a Partial Least Squares (PLS) and Principal
Component Regression (PCR) calibration approaches were found to be a useful tool to predict the
content of cocaine, caffeine, procaine, lidocaine and phenacetin in drug seizures. The achieved
results on real confiscated samples, in cooperation with the Italian Scientific Investigation
Department (Carabinieri-RIS) of Rome, allow to consider ATR-FTIR followed to chemometrics as
a promising forensic tool in such situations involving profile comparisons and supporting forensic
investigations.

Keywords: ATR-FTIR spectroscopy, Cocaine, Profiling, Chemometrics, Forensic chemistry

1. Introduction
Cocaine is the most widely used illicit drug, and its “origin” is always the focus of intense
investigation aimed at identifying the trafficking routes. The classification of cocaine seizures
according to its form (composition, cutting agents) provides usefull informations to enhance the
existence of cocaine illicit networks and supply the police intelligence [1-6].
Typical cutting agents of cocaine are adulterants as caffeine, anaesthetics such as procaine,
lidocaine and phenacetin or stimulant like levamisole [7-10]. From a forensic point of view, the
characterization of cutting agents could represent a crucial step because it can link different seizures
of cocaine to one original batch. [11]
The growing interest in the combination of vibrational spectroscopy techniques with chemometric
methods is due to the promising and simple alternative they represent. Vibrational spectroscopy can
extract informations directly from seized samples, evaluating several components using a single
instrumental measurement without requiring a physical separation. In addition, the feasibility of this
approach results in the high resolution and rapidity of the technique and the ability to not destruct
samples, allowing further investigations [12,13]. In some cases, even little differences in
composition can cause great differences in instrumental response. This means that the resulting data
obtained depends on more than one variable simultaneously, thus is multivariate. Chemometric
tools use mathematical and statistical procedures for multivariate data analysis to filter the relevant
part from complex data without losing important informations. Therefore, chemometrics has proved
to be a useful tool in solving chemical structure problems and a powerful method for extracting
structural information from the spectral data [14]. Principal component analysis (PCA) is the most
popular method of preliminarly exploring data; it allows data reduction which is particularly useful
when dealing with multidimensional issue. Among multivariate analysis methods, regression
algorithms are certainly the most widely used in analytical field since they permit to built regression
models in which the best relation between variables and instrumental responses is maximized [15].
Attempts to develop fast and non destructive methods for the early detection of cocaine and cuts in
seizures, are reported and discussed in the existing literature [16-18]. Although these studies decribe
the feasibility of a spectroscopic approach followed to chemometrics as a promising tool in routine
forensic analysis, the proposed models exclusively focused on seizures. A recent application of the
coupling of FTIR spectroscopy and chemometrics for cocaine profiling was proposed by Marcelò et
al. and an explorative principal component analysis and classification technique were considered for
standard mixtures analysis [19].
In this study, we proposed a novel strategy based on Infrared Spectroscopy with Fourier
Transformed and Attenuated Total Reflectance (ATR-FTIR) to fast identify the chemical
“fingerprint” of seized cocaine. At least, chemometric tools were exploited in order to develop a
multivariate regression model to simultaneously predict the composition of cocaine and adulterants
in seizures.
2. Materials and methods

2.1 Samples preparation

Three standard materials containing cocaine hydrochloride at different percentages were kindly
provided by Italian Scientific Investigation Department (Carabinieri-RIS) of Rome. Lidocaine,
caffeine, procaine phenacetin and mannitol (Sigma-Aldrich, St. Louis, Missouri United States),
were selected as cutting agents, among those commonly used for illicit purposes.
Each reference material (as a mixture) was divided in seven aliquots and mixtures of reference
material and cuts were prepared in order to obtain a complete homogenization of the powders in
agate mortar and at least, a cocaine dilution to 60%, 50%, 40%, 25%, 16% and about 8% w/w
(figure 1).
The same procedure was followed for the other reference material in order to exclude that results
were batch dependent.

2.2 Instrumentation

2.2.1 GC-MS parameters

Gas chromatography coupled to mass spectrometry was used as reference official method to
compare results. The characterization of the reference materials, prior to spectra acquisition, was
performed using a gas chromatography system 6890 (Agilent Technologies) coupled to a mass
selective detector 5973 (Agilent Technologies) and equipped by an autosampler 7693 (Agilent
Technologies).
About 10 mg of standard material were dissolved in 10 mL of an internal standard solution
(tetracosane 0.500 mg mL-1 in 70:30 chloroform:methanol), filtered and injected in the
chromatographic system.
Chromatographic separation was carried out on an HP-5MS capillary column (30 m x 0.25 mm x
0.25 mm) and helium at a constant flow rate of 1 ml/min was used as the carrier gas. The initial
column temperature (80°C ) was maintained for 2 min, then increased linearly to 280 °C at a rate of
25 °C/min, and finally maintained for 20 min. The GC injector and the transfer line were
maintained at 290 °C. The injection volume was 1µl with a split ratio of 60:1. The spectrometer was
operated in electron impact mode (EI). The temperatures of the ion source and quadrupole were 230
°C and 150 °C, respectively. Ionization energy was set at 70 eV and acquisition was carried out in
scan mode in the range 43-550 amu.
2.2.2 ATR-FTIR spectroscopic acquisition
Infrared spectra were recorded in reflectance mode with a Nicolet iS10 FTIR Spectrometer
configured for attenuated total reflectance (ATR) and equipped with DTGS (deuterated triglycine
sulphate) detector. An integrated sampling accessory “smart iTR single reflection diamond ATR”
was employed for all the experiments. No sample pretreatment was necessary for measurements,
thus the mixtures were only homogenized and directly placed into the accessory. The signals were
recorded between 650-4000 cm-1 and scans number and resolution were optimized to 16 scans and 4
cm−1 respectively, with a good signal to noise ratio. A background spectrum (obtained against air
and using the same instrumental conditions as the samples), was recorded before measurements and
subtracted automatically by the software. Six replicates were recorded for each sample and the
average was considered for the chemometric analysis. Spectrometer diagnostics and acquisition of
the spectroscopic data were carried by TQ Analyst 9 software (Thermo Fisher Scientific Inc) as
ASCII files, which were then imported into V-PARVUS 2009 package [20].

2.2.3 Chemometrics

Preliminary Principal Component Analysis (PCA) was performed on the acquired spectra in the
region 650-4000 cm-1 and a maximum of 4 significant PCs were recorded. Since the aim of this
work is to quantify simultaneously cocaine and main adulterants in seizures, two predictive models
were constructed using either Partial Least Squares (PLS) and Principal Component Regression
(PCR) algorithms. The performances of the models were carried out in terms of Root Mean-
Squared Error of Calibration (RMSEC), Root Mean-Squared Error of Cross-Validation
(RMSECV), and Root Mean-Squared Error of Prediction (RMSEP) [21,22].
Calibration and validation models were developed using the data set from 450 samples. The dataset
was divided into two groups, the calibration set (315 samples) and validation set (135 samples). The
calibration set consisted of 8 series of reference samples for each percentage of cocaine (70, 60, 50,
40, 25, 16 and 8% w/w). Three independent batches were manufactured, corresponding to three
reference cocaine material mentioned above, resulting in a 105 batches calibration set to be
analyzed in triplicate and the average was considered for the chemometric study. Validation was
performed on the same type of samples as the calibration set, but fully independent batches,
considering 70, 50 and 25% w/w of cocaine amount. Three replicate batches were prepared for each
formulation containing three percentages of adulterants (25, 16 and 8% w/w), resulting in 45
batches validation set to be analyzed in triplicate and the average was considered for the
chemometric study.

2.2.4 Analytical strategy

Although seized cocaine is frequently investigated and novel approaches were proposed as fast and
non destructive methods, infrared spectroscopy proved to be a promising and effective alternative
tool in forensic investigations [17, 18]. Based on previous works in the current literature and on the
experience of our research group in spectroscopic method development [23-26], this study aims at
improving the existing models for cocaine quantification and extending the ATR-
FTIR/Chemometrics approach to different kind of seizures (including the most recovered
adulterants as mannitol and procaine). As the matrix increses its complexity (different adulterants),
a simultaneous evaluation and monitoring of a great number of factors for its characterization is
required. The main criterion in this situation is to build up the initial analytical model as robust as
possible in order to ensure predictive results. Such an aim may be achieved in chemometrics by
considering standard molecules for model development in order to identify the characteristic range
in the spectra profile and selecting it to increase sensitivity of the model [14, 27].
In addition, the use of standard materials permitted to take into consideration a wider range of
cocaine amount in seizures, moving from 70% to 8% w/w, thus to correctly estimate the cocaine
percentage in those cases where largest quantity of adulterants is involved. Despite the low
incidence of samples containing a lower amount of cocaine, the optimized model would allow to
correlate seizures to one original batch with lower error rate.
The feasibility of the proposed method was applied to the analysis of recent seizures to test the
capability of the approach to early detect cocaine seizures.

3. Results and Discussion

3.1 GC-MS characterization

Prior to spectra acquisition, the samples were characterized by GC-MS in order to individuate the
composition. An example of a resulting chromatogram and mass spectra from GC-MS is shown in
figure 2, while a comprehensive description of the obtained results is summarize in table 1. The
GC-MS analysis allowed to observed a predominant signal related to cocaine and some other
compounds with lower peak intensities as phenacetin, lidocaine and caffeine in most of the
analyzed samples.

3.2 ATR-FTIR and exploratory analysis

ATR-FTIR spectra of the same samples, homogenized and directly analyzed, were collected and the
characteristic absorption regions were interpreted. Different characteristic bands can be observed
from spectra evaluation: according to Rodrigues et al., cocaine has a spectral region around 2540
cm-1 attributed to the N-H stretching due to the hydrochloride salt formation. The absorption bands
in the range of 1800-1550 cm-1 are due the C=O and C=C stretching for cocaine, phenacetin and
lidocaine and C=N stretching for caffeine. Lidocaine, phenacetin, procaine and mannitol have some
specific bands up to 3000 cm-1 probably due to stretching of amides (lidocaine and phenacetin),
stretching of aminic N-H (procaine) and stretching of OH group (mannitol). In figure 3, the spectra
of standard materials and cuts are overlapped.
Since the aim of this work is to predict the composition of cocaine seizures, aliquots of solid
standards were added and six spectra were collected for each sample. The acquired raw spectral
files were imported into the chemometric package for data analysis and PCA was used as
exploratory tool for evaluating and comparing the results.
The obtained scores plots for all the considered cutting agents, on the first and second principal
component (PCs) are shown in figure 4a. It is possible to notice that the first principal component,
which is the direction of maximum explained variance (more than 90% w/w in all cases)
demonstrates a satisfactory separation of samples according to the cocaine amount.
The results shown in figure 4b confirm the capability of this approach to distinguish different
amounts of cocaine and adulterants even in cases where they are simultaneously contained.
The achieved results suggested the possibility of validating a predictive model for cocaine profiling.

3.3 PCR and PLS regression models

In order to obtain the best results of calibration, two different regression techniques were tested.
Calibration and test set were pre-processed using Standard Normal Variate (SNV) scaling [28]
followed by calculating the second derivative using a Savitzky–Golay [29] filter prior to variable
selection. Alternatively, Multiplicative Scatter Correction (MSC) [30] was applied to the dataset
using a second order polynomial filter. Variable selection within the calibration block for PLS was
conducted using the Predicted Residual Error Sum of Squares (PRESS) representing the sum of
squares of the prediction error to assess model's predictive ability. Usually, the smaller the PRESS
value, the better the model's predictive ability. PLS model for cocaine required a number of 5 latent
variables, while for PCR algorithm the evaluation of the Cumulative Eigenvalues Diagnostic Plot
provided for a number of 6 latent variables (see supplemental data). Either cocaine and cutting
agents were calibrated and both PLS and PCR regression models were internally cross validated by
the mean of leave-more-out technique (as leave-one-out but computing a statistic on the left out n-
samples, with n=5); the analytical strategy followed in this study was summarized in figure 5.
To increase the model efficiency or make the interpretation simpler, the selection of fingerprint
spectral regions was performed. In this case, not all wavelengths contribute unique or lead to useful
informations. In the case of spectral data, decorrelation processes are fundamental because
contiguous variables are often highly intercorrelated. Stepwise decorrelation of variables was
applied to the spectral datasets in order to idetify a variable with the largest Fisher weight [31]. This
first selected variable is decorrelated from all the others so that these latter become orthogonal to
that selected. After decorrelation, the selected variable was taken apart from the variable set and the
process repeated for further identifications. Figure 6 shows the results of decorrelation process and
allows to evaluate the correlations for quantitative relationships by the mean of the loadings plot for
PLS and PCR models.
As shown in table 2, very satisfactory calibration results (R2 and RMSEC) were obtained for the
calibration of cocaine and cuts. Further important results can be observed in table 3 that
demonstrates the figures of merit (R2 and RMSEP) for the PLS and PCR regression models when
predicted. Regardless to cocaine prediction, the two approches produce a determination coefficient
for the calibration samples larger than 0.9976 and the evaluation of both RMSEC and RMSEP
resulted in average errors around 2.5% (w/w) in calibration and 1.74 % (w/w) when predicted.
As far as the data are concerned, among the investigated pretreatments better performances can be
achieved when SNV pretreatment is select for both PLS and PCR while PLS regression allow to
reach the best outcomes for the most of investigated substances.
This relevant outcome enables to conclude that is possible to simultaneously quantify cocaine and
cuts in seizures directly from FTIR measurents in association with an adequate PLS regression
model. In particular, the use of standard materials in conjuction with a proper experimental design,
permitted to obtain better sensitivity of all the investigated molecules compared to previous works
and to extend the concentration range with lower detection limits. In table 4 are reported the figures
of merits of the selected PLS model for cocaine and adulterants calculated according to ISO11843-2
[32] and suggested by Ortiz et al. [33].

3.4 Evaluation of the prediction ability

The validated model was consequently used to process confiscated samples, gently provided from
Italian Scientific Investigation Department (Carabinieri-RIS) of Rome. A preliminarly GC-MS
analysis was performed, as official method, on the same samples and the composition of each
seizure was defined. In parallel, the same samples were analyzed by ATR-FTIR spectroscopy and
the resulted acquired spectra were processed by PLS regression model using a SNV data
preprocessing. A detailed description of the percentage amount of detected substances is reported in
table 5 and results are illustrated in figure 7 by the graph of predicted versus reference values
obtained using the most performing conditions (PLS model using SNV).
To note first, is the ability of this approach to correctly identify the true composition of seizures,
thus it permits to detect the same molecules of the official method. In addition, computed values
have been found to be very close to the expected amounts. These results show that the processing
method is suitable to profile cocaine seizures even in such situation where further procedures are
required to detect cuts (mannitol needs to be derivatized prior to GC analysis). The investigated
wide range of percentages in the composition of the data set demonstrated the ability of the model
to quantify specific analytes in a single instrumental measurement as no difference in the prediction
of errors was observed.

4. Conclusions

In this study, a new approach based on ATR-FTIR coupled with chemometrics was assessed to
identify the chemical “fingerprint” of seized cocaine. Both PCR and PLS multivariate regression
methods have shown good outcomes for this purpose and the reported results suggest that the SNV
pretreatment followed by PLS regression provides the lower errors of prediction. The novelty of
this work mainly consists of the possibility to improve sensistivity in cocaine and adulterants
quantification by the mean of correct experimental design to built a validated model of prediction.
Samples prepared using standard molecules provided several advantages: a) the selection of a
fingerprint region of the spectrum when complex mixtures were involved; b) the correct estimation
of cocaine and adulterants in cases where they are present in a different percentages, thus the
extention of the detectable range for each analyte; c) lower RMSEP compared to previous studies.
The feasibility of the proposed model in forensic investigations was evaluated by comparing results
with official methods and good accordance was observed.
This study for the first time, introduces a novel approach to fast profile cocaine in seizures, that
requires no or minimal sample preparation and is non-destructive and easy to perform (no highly
skilled personnel is necessary) allowing further investigation in forensic laboratories. Moreover, the
prediction ability observed for all the investigated samples permitted to conclude that the ATR-
FTIR/Chemometrics approach may be considered as a cost-effective tool in forensic investigations.

Acknowledgements
The authors would like to thank the Italian Scientific Investigation Department (Carabinieri-RIS) of
Rome for providing us with standard and confiscated real samples supporting this study.

References

[1] United Nations Office on Drugs and Crime (UNODC); World Drug Report; United Nations:
New York, 2011.
[2] S. Lociciro, P. Hayoz, P. Esseiva, L. Dujourdy, F. Besacier, P. Margot, Cocaine profiling for
strategic intelligence purposes, a cross-border project between France and Switzerland. Part I.
Optimization and harmonisation of the profiling methods, Forensic Science International 167
(2007) 220-228.
[3] J. Broséus, S. Huhtala, P. Esseiva, First systematic chemical profiling of cocaine police seizures
in Finland in the framework of an intelligence-led approach, Forensic Science International 251
(2015) 87-94.
[4] T.M. Brunt, S. Rigter, J. Hoek, N. Vogels, P. Van Dijk, R.J. Niesink, An analysis of cocaine
powder in the Netherlands: content and health hazards due to adulterants. Addiction, 104 (2009)
798-805.
[5] M.L. Di Gioia, A. Leggio, A. Le Pera, A. Liguori, A. Napoli, C. Siciliano, G. Sindona.
Quantitative analysis of human salivary glucose by gas chromatography-mass spectrometry Journal
of Chromatography B: Analytical Technologies in the Biomedical and Life Sciences, 801 (2004)
355-358.
[6] S.F. Lapachinske, G.G. Okai, A. Dos Santos, A. Valle de Barrios, M. Yonamina, Analysis of
cocaine and its adulterants in drugs for international trafficking seized by the Brazilian Federal
Police. Forensic Science International, 247 (2015) 48-53.
[7] European Monitoring Centre for Drugs and Drug Addiction (EMCDDA); European Drug
Report - Trends and Developments; Luxembourg: Publications Office of the European Union,
2015.
[8] R. Risoluti, D. Piazzese, A. Napoli, S. Materazzi. Study of [2-(2’-pyridyl)imidazole] complexes
to confirm two main characteristic thermoanalytical behaviors of transition metal complexes based
on imidazole derivatives. Journal of Analytical and Applied Pyrolysis 117 (2016) 82-87.
[9] M.L. Di Gioia, A. Leggio, A. Liguori, F. Perri, C. Siciliano, M.C Viscomi. A preparation of N-
Fmoc-N-methyl-α-amino acids and N-nosyl-N-methyl-α-amino acids. Amino Acids 38 (2010) 133-
143.
[10] M.F. Andreasen, C. Lindholst, E. Kaa, Adulterants and diluents in heroin, amphetamine, and
cocaine found on the illicit drug market in Aarhus, Denmark. Open Forensic Sci. J. 2 (2009) 16-20.
[11] N. V.S. Rodrigues, E. M. Cardoso, M. V. O. Andrade, C. L. Donnici and M. M. Sena, Analysis
of Seized Cocaine Samples by using Chemometric Methods and FTIR Spectroscopy, J. Braz.
Chem. Soc. 24 (2013) 507-517.
[12] E. J. Magalhaes, C. C. Nascentes, L. S. A. Pereira, M. L. O. Guedes, R. A. Lordeiro, L. M.
L.A. Auler, R. Augudti, M. E. L.R. de Queiroz, Evaluation of the composition of street cocaine
seized in two regions of Brazil,Science and Justice 53 (2013) 425-432.
[13] J. Moros, S. Garrigues, M. Guardia, Vibrational spectroscopy provides a green tool for multi-
component analysis, Trends Anal. Chem. 29 (2010) 578–591.
[14] R. Risoluti, S. Materazzi, A. Gregori, L. Ripani. Early detection of emerging street drugs by
near infrared spectroscopy and chemometrics. Talanta doi:10.1016/j.talanta.2016.02.044
[15] D.L. Massart, B.G. Vandeginste, L.M.C. Buydens, S. De Jong, P.J. Lewi, J. Smeyers-Verbeke,
Handbook of Chemometrics and Qualimetrics: Part A, Elsevier, Amsterdam, 1997.
[16] M. Monfreda, F. Varani, F. Cattaruzza, S. Ciambrone, A. Proposito, Fast profiling of cocaine
seizures by FTIR spectroscopy and GC-MS analysis of minor alkaloids and residual solvents, Sci.
Justice (2015), http://dx.doi.org/10.1016/j.scijus.2015.06.002
[17] T.S. Grobério, J.J. Zacca, E ́.D. Botelho, M. Talhavini, J.W.B. Braga, Discrimination and
quantification of cocaine and adulterants in seized drug sampes by infrared spectroscopy and PLSR.
Forensic Science International (2015), http://dx.doi.org/10.1016/j.forsciint.2015.09.012
[18] T. S. Grobério, J. J. Zacca, M. Talhavini and J. W.B. Braga, Quantification of Cocaine
Hydrochloride in Seized Drug Samples by Infrared Spectroscopy and PLSR, J. Braz. Chem. Soc. 25
(2014) 1696-1703
[19] M.C.A. Marcelo, K.C. Mariotti, M.F. Ferrao, R.S. Ortiz, Profiling cocaine by ATR-FTIR.
Forensic Science International, 246 (2015) 65-71.
[20] M. Forina, S. Lanteri, C. Armanino, C. Casolino, M. Casale, P. Oliveri, V-PARVUS. An
extendable package of programs for explorative data analysis, classification and regression analysis.
University of Genova, 2009 for the successive chemometric analysis.
[21] H. Mark, J. Workman, Chemometrics in Spectroscopy, Elsevier/Academic Press, Amsterdam,
2007.
[22] J.N. Miller, J.C. Miller, Statistics and Chemometrics for Analytical Chemistry- Prentice Hall,
Harlow, England (2000).
[23] S. Materazzi, R. Risoluti. Spectroscopic Methods in Evolved Gas Analysis: Reference module
in Analytic Sciences and Chemometrics. Reference Module in Chemistry, Molecular Sciences and
Chemical Engineering, 2014, DOI:10.1016/B978-0-12-409547-2.11018-2
[24] R. Risoluti, M.A. Fabiano, G. Gullifa, S.Vecchio Ciprioti, S. Materazzi. FTIR - Evolved Gas
Analysis in recent thermoanalytical investigations. Appl.Spectr.Rev. 52 (2017) 39-72 DOI:
10.1080/05704928.2016.1207658
[25] S. Materazzi, R. Risoluti. Evolved gas analysis by mass spectrometry. Appl.Spectr.Rev. 49
(2014) 635–665
[26] R. Risoluti, M.A. Fabiano, G. Gullifa, S. Vecchio Ciprioti, S. Materazzi. FTIR - Evolved Gas
Analysis in recent thermoanalytical investigations. Applied Spectroscopy Review DOI:
10.1080/05704928.2016.1207658
[27] R. Risoluti, S. Materazzi, F. Sorrentino, L. Maffei, P. Caprari. Thermogravimetric analysis
coupled with chemometrics as a powerful predictive tool for ß-thalassemia screening. Talanta 159
(2016) 425-432
[28] R.J. Barnes, M.S. Dhanoa, S.J. Lister. Standard Normal Variate Transformation and De-
trending of Near-Infrared Diffuse Reflectance Spectra. Appl. Spectrosc. 43 (1989) 772–777
[29] A. Savitzky, M.J.E. Golay, Smoothing and differentiation of data by simplified least squares
procedures. Anal. Chem. 36 (1964) 1627–1639
[30] P. Geladi, D. MacDougall, H. Martens. Linearization and scatter-correction for Near-Infrared
reflectance spectra of meat. Appl. Spectrosc. 39 (1985) 491-500.
[31] Wold S, Sjöström M (1977) Chemometrics: Theory and Application 12:243-282
[32] ISO 11843-2, Capability of Detection, International Standard Organization, Switzerland,
Geneva, 2000
[33] M.C. Ortiz, L.A. Sarabia, A. Herrero, M.S. Sanchez, M.B. Sanz, D. Rueda, M.E. Gimenez,
M.E. Melendez, Capability of detection of an analytical method evaluating false positive and false
negative (ISO 11843) with partial least squares, Chemom. Intell. Lab. Syst. 69 (2003) 21-33.

Cocaine(seizure(

0,020g! 0,020g! 0,020g! 0,020g! 0,020g!

Caffeine' Procaine' Lidocaine' Phenace/ n' Mannitol'

;
;
;

70%$ 60%$ 50%$ 40%$ 25%$ 16%$ 8%$

!
Fig. 1. Reference standard materials prepared for calibration models
 Cocaine"
Phenace. ne"

Lidocaine"
Caffeine"

Fig. 2. Resulting chromatograms and mass spectra of reference material from GC-MS analysis

Fig. 3. Overlapped spectra of acquired reference materials

!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$

!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$
!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$

!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$
Lido#70#%# Proc#
8#%#70##
Proc#
Proc#70#
# %# Phen#70#%#
%#
Mann#70#%#
Lido#60#%#
Caff#70#%# Phen#60#%#
Proc#60#%#
Lido#50#%#
Mann#60#%# Caff#60#%#
Lido#40#%#
Proc#50#%# Phen#50#%#
Mann#50#%# Caff#50#%#
!10
!10$$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$$$$$$$10$ !10$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$$$$$$$10$ !10
!10$$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$$$$$$$10$
!10$

Lido#25#%# Phen#40#%#
Proc# 40#
%# %#
Proc#40#
Caff#40#%#
Mann#40#%#Mann#8#%#

Lido#16#%#
Caff#25#%# Phen#25#%#
Mann#25#%#
!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$

!10$$$$$$$$$$$!8$$$$$$$$$$$$!6$$$$$$$$$$$$!4$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$2$$$$$$$$$$$$$4$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$8$$$$$$$$$$$$10$ Lido#8#%# Proc#25#%#

Phen_40'%'
Mann#16#%# Caff#16#%# Phen#16#
16#%#
%####
Phen_50'%'
Phen_25'%'

Mann#8#%# Proc#16#%#
PC#2#(11.01#%)#
PC#2#(11.01#%)#

Phen_60'%'
Phen_16'%'

Phen_8'%' Caff#8#%#
Phen_70'%'
Phen#8#%#
Proc#8#%#

!10 !10
!10$$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$1"
!1
$$$$$$10$
!10$$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$$$$!2$$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$$$$$$$10$ !10$$$$$$$$$$$$$$$$$!8$$$$$$$$$$$$$$$$$$$!6$$$$$$$$$$$$$$$$$$$!4$$$$$$$$$$$$$$$$PC#
$$$!2$
$$$$$$$$$$$$$$$$$$$$0$$$$$$$$$$$$$$$$$$$2$$$$$$$$$$$$$$$$$$$$$4$$$$$$$$$$$$$$$$$$$$6$$$$$$$$$$$$$$$$$$$$8$$$$$$$$$$$$$$$$$$$$10$
1#(88.85#%)#
PC#1#(88.85#%)#

Fig.4. Scores plot of reference materials at different dilution level of cocaine and phenacetin (pink), caffeine (blu),
procaine (red), mannitol (green), lidocaine (orange). Fig 4a represents the obtained scores plots for each cutting
agent; fig. 4b represents the scores plot of the same substances when mixed.

Chemometric; tools; ;

Test;; samples;
samp
Calibra on; samples;
; (-25%; of; samples; for;
(-75%; of; the; samples);
external; valida on);

Pre-processing;
g;
SNV;
MSC;

Principal; Comonents;
Par al; Least; Square; ; Principal; Components; Regression;
Analysis; ;
PCA; PLS; PCR;

Optmize;
i ; model;
(leave-more-out);

Fit; model;

Use; model;

Fig. 5. Analytical strategy followed in this study

 PLS; model;

PCR; model;

Fig. 6. Evaluation of loadings for both PLS and PCR regression models.
 100;
;
; Y=; 0.3381x; 0.0192;
2 Y=; 0.4371x; 0.0112;
90; ; R =; 0.9999;R; 2; =; 0.9999; ; ;
;
80;
;
70; ;
;
Predicted; values; (%w/w);

60; ;
;
50;
;
40; ;
;
30;
;
20;
; Calibra on; set;
; Valida on; set;
10; ;
0;
0; 10; 20; 30; 40; 50; 60; 70; 80; 90; 100;
Measured; values; (%w/w);
Fig. 7. Measured values versus predicted values forr cocaine model of prediction

Table 1 Composition of selected reference materials

Reference Cocaine Phenacetin Caffeine Lidocaine
material (%) (%) (%) (%)
1 68.4 1.7 0.5 0.3
2 72.0 2.1 2.5 0.7
3 75.9 1.9 2.4 0.8

Table. 2 Prediction performances of PLS and PCR models in calibration

PCR-MSC PCR-SNV PLS-MSC PLS-SNV

R2 RMSEC R2 RMSEC R2 RMSEC R2 RMSEC

Cocaine 0.9847 3.72 0.9893 3.12 0.9909 2.86 0.9931 2.48

Caffeine 0.9968 2.27 0.9987 1.40 0.9971 2.13 0.9981 1.66

 Lidocaine 0.9933 3.12 0.9970 2.23 0.9976 2.03 0.9980 1.81

Procaine 0.9907 4.97 0.9857 4.12 0.9989 1.78 0.9921 1.36

Phenacetin 0.9983 1.58 0.9984 1.52 0.9988 1.30 0.9977 1.90

Mannitol 0.9968 2.18 0.9932 3.14 0.9984 1.51 0.9974 2.01

Table. 3 Figures of merit of each compound calculated with both PLS and PCR moldels in prediction

PCR-MSC PCR-SNV PLS-MSC PLS-SNV

R2 RMSEP R2 RMSEP R2 RMSEP R2 RMSEP

Cocaine 0.9976 1.76 0.9993 0.81 0.9993 1.77 0.9966 1.74

Caffeine 0.9991 1.11 0.9997 0.67 0.9992 1.06 0.9997 0.62

Lidocaine 0.9987 1.38 0.9997 0.69 0.9999 0.36 0.9998 0.60

Procaine 0.9999 0.43 0.9999 0.32 1.0000 0.09 1.0000 0.09

Phenacetin 0.9993 1.03 0.9997 0.60 0.9994 0.92 0.9998 0.57

Mannitol 0.9998 0.53 0.9999 0.38 0.9995 0.86 1.0000 0.18

Table. 4 Figures of merits for the selected PLS model

Molecule R2 RMSEP LV* Sensitivity MDC*

(%w/w)-1 (%)
Cocaine 0.9966 1.74 5 0.05 2.5
Caffeine 0.9997 0.62 6 0.32 0.5
Lidocaine 0.9998 0.60 6 0.27 0.4
Procaine 1.0000 0.09 4 0.61 1.5
Phenacetin 0.9998 0.57 4 0.68 0.9
Mannitol 1.0000 0.18 5 0.18 0.1
*Latent variables
*Minimum detection concentration
Table. 5 Calculated percentage amounts of detected substances in real samples performed by official method and the
proposed approach

Cocain Caffeine Lidocaine Procaine Mannitol Phenacetin

e (%) (%) (%) (%) (%) (%)

Sample GC- GC- GC-

PLS PLS GC-MS PLS PLS GC-MS PLS GC-MS PLS
MS MS MS

1 16.9 17.6 n.d. n.d. 47.1 42.3 n.d. n.d. 2.5 2.3 n.d. n.d.

2 53.4 56.1 46.6 49.9 n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d.

3 31.2 33.6 n.d. n.d. n.d. n.d. n.d. n.d. 1.8 1.5 n.d. n.d.

4 34.0 36.9 23.8 20.3 22.5 25.1 n.d. n.d. n.d. n.d. n.d. n.d.

5 44.8 46.8 38.0 37.4 n.d. n.d. n.d. n.d. n.d. n.d. 4.3 1.6

6 53.6 55.8 n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d.

7 64.0 69.7 n.d. n.d. n.d. n.d. 30.6 37.2 n.d. n.d. n.d. n.d.

8 7.5 4.27 52.0 48.4 36.0 33.2 n.d. n.d. n.d. n.d. 8.7 10.1

9 69.9 70.1 n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d. n.d.

10 58.6 58.8 n.d. n.d. n.d. n.d. n.d. n.d. 4.4 4.2 n.d. n.d.

Highlights
Validation of a regression model to simultaneously quantify cocaine and adulterants in
seizures
Rapid screening to identify the fingerprint of cocaine seizures in forensic investigations
A novel approach to fast profile cocaine in seizures, that requires no sample preparation
and is non-destructive and easy to perform since no highly skilled personnel is necessary.
Reference official method Implementation of
models of prediction

!"#$%&'()(**+*&(,(*-./.&& !"#$%"&#'()*+,+*-./012