Rapid Nondestructive On-Site Screening of MethylamphetamineSeizures by Attenuated Total Reflection Fourier TransformInfrared Spectroscopy

CHING YONG GOH, WILHELM VAN BRONSWIJK,* and COLIN PRIDDISDepartment of Applied Chemistry, Curtin University of Technology, GPO Box U1987, Perth, WA 6845, Australia (C.Y.G., W.v.B.); and ForensicScience Laboratory, Chemistry Centre (Western Australia), 125 Hay Street, East Perth, WA 6004, Australia (C.P.)

The identification and quantification of illicit substances in the field is

often desirable. Fourier transform infrared spectroscopy (FT-IR) has both

qualitative and quantitative capabilities and field portable instruments are

commercially available. Transmission infrared spectra of mixtures

containing ephedrine hydrochloride, glucose, and caffeine and attenuated

total reflection (ATR) infrared spectra of mixtures composed of

methylamphetamine hydrochloride, glucose, and caffeine were used to

develop principal component regression (PCR) calibration models. The

root mean sum of errors of predictions (RMSEP) of all individual

components in a mixture from a single measurement was ,6% w/w,

which reduced to ;3% w/w when triplicates were averaged. Sample

mixing and grinding are essential to minimize the effect of heterogeneity,

as deviations of up to 20% w/w were observed for single measurements of

unground samples. Poor predictions of the components in a mixture

occurred when samples were ‘‘contaminated’’ with substances not present

in the calibration set, as would be expected. When only a single analyte

(drug) was targeted, using a calibration set that contained both

contaminated and uncontaminated samples, an RMSEP of ;4% w/w

was achieved. The results demonstrate that ATR-FT-IR has the potential

to quantify methylamphetamine samples, and possibly other licit or illicit

substances, in at-seizure and on-site scenarios.

Index Headings: Methylamphetamine; Infrared spectroscopy; Fourier

transform infrared spectroscopy; FT-IR spectroscopy; Attenuated total

reflection; ATR; Raman spectroscopy; Quantitative analysis.

INTRODUCTION

Gas chromatography–mass spectrometry (GC-MS) is theprimary method used by forensic laboratories for the definitiveidentification of illicit drugs. Identification and quantificationof illicit drugs in the laboratory can also be performed by high-performance liquid chromatography (HPLC), gas chromatog-raphy (GC), capillary electrophoresis (CE), and nuclearmagnetic resonance spectroscopy (NMR).1–3 These methodsall suffer from one or more of the following shortcomings fornondestructive and rapid on-site quantification of controlledsubstances: There is no confirmatory information about thestructure of the compound; it does not provide quantitativeinformation; the sample is destroyed; extensive samplepreparation is required; and the instrument is not readilydeployable in the field.

Vibrational spectroscopy, both infrared (IR) and Raman,provides structural information by way of characteristicvibrational frequencies that are a molecular fingerprint of thesample being analyzed. While the commonly used pressed KBrdisk technique for IR spectroscopy is destructive, nondestruc-tive methods such as attenuated total reflection (ATR) can be

implemented on both laboratory and field-portable instruments(e.g., HazMatIDTM). Raman spectroscopy is by its naturenondestructive, provided the sample is not destroyed by theexcitation laser used, and field-portable instruments are alsoavailable (e.g., FirstDefenderTM). One major issue that needs tobe addressed with both techniques is that of sampleheterogeneity. As illicit drug samples are inevitably mixturesof individual solid materials, it is crucial that the sample area/volume analyzed is representative of the bulk. This is relativelysimple with ATR Fourier transform infrared (FT-IR) spectros-copy as its sampling area can be large (.10 mm2, Fig. 1). In atypical ATR configuration (Fig. 1), the infrared beam enters ahigh refractive index ATR crystal and is internally reflected.4

This results in an evanescent wave, which penetrates thesample in contact with the ATR crystal and produces areflection/absorption spectrum. Examples of ATR crystalmaterials include zinc selenide, KRS-5 (thallium iodide/thallium bromide), germanium, and diamond. Diamond isusually preferred when unknown samples are to be assessed,e.g., in forensic work, because of its hardness (scratchresistance) and chemical inertness.

Infrared spectroscopy of solid mixtures has been predomi-nantly aimed at qualitative analysis. It has been applied to theidentification of major diluents in illicit tablets5 and diluentsand adulterants in seized cocaine6 and heroin7,8 samples. ATRbased field-portable instruments (HazMatIDTM) have beenused by the Australian Federal Police and Chemistry Centre(Western Australia) at clandestine laboratories to identify (butnot quantify) various illicit drugs, precursor materials, solvents,and diluents.� Micro-FT-IR spectroscopy has also been used toconfirm the results of microcrystal tests by acquiring spectra ofthe crystals formed, especially in cases where crystal formsshow similar appearance.9

Raman scattering is inherently a weak effect that is easilyoverwhelmed by fluorescence, but it has the advantage ofrequiring no sample preparation and water giving only weakbands because of its low scattering cross-section. Thefluorescence problem can sometimes be overcome by usinglonger wavelength lasers (typically 785, 830, or 1064 nm)albeit with a reduction in Raman intensity. However, thespectrometers have a comparatively small sampling area (often,1 mm2) and hence sample heterogeneity can be a greaterproblem than with IR. Raman spectral intensities and bandwidths are also more sensitive to crystallinity than those of IRspectra, which can lead to greater variance in their spectra. Theadvantage of the small Raman scattering cross-section of water,as opposed to its high IR absorptivity, is unlikely to be relevant

Received 7 June 2007; accepted 19 March 2008.* Author to whom correspondence should be sent. E-mail:w.vanbronswijk@curtin.edu.au.

� M. Collins, 2006, personal communication; C. Priddis, 2006, personalcommunication.

640 Volume 62, Number 6, 2008 APPLIED SPECTROSCOPY0003-7028/08/6206-0640$2.00/0

� 2008 Society for Applied Spectroscopy

for methylamphetamine samples as they are almost inevitablydry powders or tablets.

Raman spectroscopy of solid mixtures has similarly beenaimed at identifying individual components and its use inidentifying controlled substances such as heroin, cocaine, andMDMA (3,4-methylenedioxymethylamphetamine) is well es-tablished.10–12 The problem of sample heterogeneity forquantitative work is sometimes overcome by dissolving thesample in a suitable solvent and obtaining the solution’sspectrum.13 This overcomes the heterogeneity problem but themethod requires sample preparation and is destructive, which isundesirable for field work. However, Ryder et al.10 havedemonstrated that quantitative solid-state Raman analyses arefeasible with their quantification of cocaine diluted withglucose via partial least squares regression (PLSR).

While IR and Raman spectra are rich in information, thespectra of mixtures can be difficult to interpret simply becauseof the wealth of information and the overlap of spectralfeatures. Qualitative analysis is sometimes possible directlyfrom the spectra, but interpretation of such spectra is greatlyaided by multivariate techniques such as principal componentsanalysis (PCA). Extending PCA to principal componentsregression (PCR) allows the spectra to be used for quantifica-tion of components in mixtures, e.g., PCA has been used byRyder14 to distinguish between diluted samples of cocaine,heroin, and MDMA, and by Strachan et al.15 as a datareduction method in conjunction with multiple linear regres-sion (MLR) to quantify the polymorphic forms (I and III) ofcarbamazepine in a mixture. More recently, Katainen et al.13

demonstrated the quantification of seized amphetamine sam-ples using solution Raman spectra relative peak heights andpartial least squares regression (PLSR).

To address the issues associated with on-site analyses,sample preservation, and sample heterogeneity, this study hasinvestigated the use of IR and Raman spectroscopy for the

quantification of seized methylamphetamine samples usingmodel compounds and methylamphetamine, with laboratoryand field-portable instrumentation.

EXPERIMENTAL

Materials. The source and purpose of materials used in thisproject are summarized in Table I. Methylamphetaminehydrochloride samples (Table II) were provided by the IllicitDrugs Section of the Chemistry Centre (Western Australia).They had been synthesized by the three most commonly usedreduction methods at clandestine laboratories in WesternAustralia: red phosphorus/iodine, hypophosphorus acid/iodine,and metal/liquid ammonia.

Infrared and Raman Spectroscopy. Transmission infraredspectra of the calibration and validation mixtures were acquiredusing a Bruker IFS66 FT-IR spectrophotometer fitted with adeuterated triglycine sulfate (DTGS) detector. Measurementswere recorded at a resolution of 4 cm�1 from 4000–400 cm�1

with 16 scans accumulated. Spectral measurements, manipu-lations, and evaluations were performed by OPUS v2.2software (Bruker). The samples were prepared as potassiumbromide disks by mixing and grinding the sample (;1% w/w)with potassium bromide. The mixture was compressed in ametal die under vacuum at 550 MPa.

Attenuated total reflection (ATR) spectra of the methylam-phetamine samples were acquired using a field portableHazMatIDTM spectrometer (SensIR Technologies, L.L.C.)fitted with a zinc selenide ATR crystal and a thermoelectricallycooled DTGS detector. Measurements were recorded at aresolution of 4 cm�1 from 4000–650 cm�1 with 32 scansaccumulated. Spectral measurements, manipulations, andevaluations were performed by HazMatIDTM v1.2.1 software(SensIR Technologies, L.L.C.). Powdered samples wereclamped down at a set pressure, which did not cause furthersample crushing, to ensure consistent contact between thesample and the ATR crystal. The sampling area was ;7 mm2

(;3 mm diameter).Raman spectra were obtained with a FirstDefenderTM

(Ahura Scientific) field portable spectrometer fitted with a785 nm excitation laser (maximum output of 300 mW) and a2048 pixel silicon charge-coupled device (CCD) detector. Themethylamphetamine mixtures were placed in non-fluorescing 4mL vials (Ahura Scientific, Part No. 548-00471) andsubsequently placed in the spectrometer’s Integrated 4 mLVial Holder. Spectra were acquired in ‘‘Auto’’ mode, whichaccumulates signal to a preset level to obtain maximum signalto noise ratios and prevent detector overload. Accumulationtimes were less than five minutes.

TABLE I. Materials used in this study.

Material Source Purpose

Glucose APS Finechem Calibration and test mixturesBDH Chemicals, Victoria (AR grade) Diluent for MA mixtures

Caffeine Fluka AG, Chem. Fabrik (.98%, anhydrous powder) Calibration and test mixtures,adulterant for MA mixtures

Paracetamol Chemistry Centre (WA) Adulterant for MA mixturesEphedrine hydrochloride Aldrich Chemical Company Calibration and test mixturesPseudoephedrine hydrochloride School of Pharmacy, Curtin University of Technology Test mixturesMethylamphetamine hydrochloride Chemistry Centre (WA), (96% purity as HCl salt,

77.2% as free base)Reference material

FIG. 1. General setup for ATR-FT-IR spectroscopy.

APPLIED SPECTROSCOPY 641

Regression Software. Principal component analysis andregression were performed on the full FT-IR spectra with TheUnscramblert v9.7 (Camo Software AS, Norway).

RESULTS AND DISCUSSION

As a preliminary to quantifying mixed powder methylam-phetamine samples, ephedrine hydrochloride was used as amodel substance. These ephedrine hydrochloride mixtureswere dispersed in KBr to minimize the effect of possiblesample heterogeneity and spectra obtained from pressed disks.Methylamphetamine mixtures were analyzed as undilutedpowders using field-portable ATR-FT-IR and Raman spec-trometers.

Quantitative Transmission Fourier Transform InfraredSpectroscopy. A constrained lattice (Fig. 2) design ofcalibration mixtures was used for method development on alaboratory-based instrument. The design consisted of threefactors (ephedrine HCl, glucose, and caffeine) at five levels (0,25, 50, 75, and 100% w/w), with the sum of the componentsalways being 100% (i.e., only two independent variables). Thecalibration design removes co-linearity between componentmixtures and leads to more robust calibrations. A constrainedlattice design (Fig. 3) was also used for validation mixturescomprising ephedrine HCl, caffeine, and glucose. The designcovers the full range of the lattice with a minimal number ofmixtures and enables the predictive ability of calibrationmodels to be evaluated independent of the calibration data. Allmixtures were homogenized by hand in an agate mortar andpestle before being pressed into disks. Spectra were obtained induplicate.

Spectra Acquisition and Data Pretreatment. TransmissionFT-IR spectra of the calibration mixtures showed varyingbaseline offsets and slopes (Fig. 4). In addition, the varyingamounts of mixture used in the preparation of the KBr disks

results in variable intensities. To avoid the inclusion of suchvariability in the calibration, the data was pretreated bybaseline zeroing (Eq. 1) followed by area normalization (Eq.2), where Ij is a data point in the spectrum and I1934 is theintensity at 1934 cm�1. The intensity at 1934 cm�1 wasselected for baseline zeroing as that region of the spectrum wasfeatureless and was common to all the spectra. Baseline zeroingremoves problems associated with baseline offsets (resultingfrom opacity/scattering effects associated with KBr disks),while area normalization adjusts for the variation of the amountof analyte in the beam. The normalization scaling factor (100)is arbitrary and was selected for ease of comparison andcomputational efficiency. The results of scaling are shown inFig. 5.

IjðZeroedÞ ¼ Ij � I1934 ð1Þ

IðArea NormalizedÞÞ ¼IjðZeroedÞ3 100

Xn

n¼1

½IðZeroedÞ�ð2Þ

An alternative treatment to baseline zeroing and intensityscaling to address disk quality and analyte quantity issues is totake derivatives. First derivatives remove the effect of baselineoffsets and second derivatives will remove the effect of slopingbaselines. However, derivatives are inherently noisier than the

FIG. 2. Constrained lattice mixture design at five levels (0, 25, 50, 75, and100% w/w) with three factors (ephedrine HCl, caffeine, and glucose).

FIG. 3. Constrained lattice design for validation mixtures containingephedrine HCl diluted with glucose and caffeine.

FIG. 4. Unscaled transmission (KBr) FT-IR spectra acquired from theephedrine HCl, glucose, and caffeine calibration mixtures.

TABLE II. Composition of methylamphetamine HCl samples providedby the Illicit Drugs Section (CCWA), as determined by HPLC.

Sample ID Synthesis methodMethylamphetamine

(%)Ephedrine

(%)

CL2 Hypophosphorus acid/iodine 79.5 0.0CL4 Sodium/liquid ammonia 79.2 3.2RP2 Red phosphorous/iodine 81.4 0.0

642 Volume 62, Number 6, 2008

absorbance spectra and a 5-point Savitzky–Golay filter wasapplied to both derivatives before they were assessed, alongwith spectra scaling, for predicting analyte concentrations.

Development and Evaluation of Principal ComponentRegression Models. Principal component analysis (PCA) andregression (PCR) were performed on scaled FT-IR spectra andderivatives of the calibration and evaluation sets. Thecovariance-based PCA was able to resolve the three purecomponents and intermediate mixtures and showed goodagreement between duplicates using only two principalcomponents (PCs) derived from the scaled spectra (Fig. 6).This is expected for a closed three-component system andillustrates the efficacy of pretreating the data, as no principalcomponents are required to describe baseline offsets and/oroverall intensity and the two PCs account for 98% of thevariance in the absorbance data. Covariance rather thancorrelation was used for the PCA to prevent minor variationsin the spectra exerting undue influence. The validity of thisapproach was confirmed as correlation-derived PCs did not

achieve lower prediction errors than covariance PCs and insome instances were worse. The subsequent prediction ofephedrine HCl (Fig. 7a), glucose (Fig. 7b), and caffeine (Fig.7c) content in the validation set showed that it was possible toacceptably quantify solid mixtures using FT-IR, the root meansum of squares of residuals (RMSSR) and error of predictions(RMSEP) being ;4% for all components.

Regression models developed from the first and secondderivatives of the spectra performed worse than the scaled

FIG. 6. Scores from the PCA model of scaled infrared spectra of thecalibration and validation sets; (u) ephedrine HCl, (n) glucose, (^) caffeine,and (*) mixture.

FIG. 7. Predictions of the (a) ephedrine HCl, (b) glucose, and (c) caffeinecontent in the validation mixtures using two PCs from the PCA model ofbaseline zeroed and normalized transmission FT-IR spectra.

FIG. 5. Baseline zeroed and area normalized transmission (KBr) FT-IR spectraof the ephedrine HCl, glucose, and caffeine calibration mixtures.

APPLIED SPECTROSCOPY 643

transmission spectra derived models. Selection of the numberof PCs required to account for the variance was moresubjective in these cases as the first three PCs accounted foronly 86% of the variance, with each successive PC beyond thisaccounting for ,1% of variance. Hence, only three PCs wereused in the models. These models had RMSSRs and RMSEPsin the range 3–13% for the three analytes. The poorerperformance is most probably due to the noise inherent inspectra derivatives and a degree of subjectivity in the selectionof the level of filtering applied. Filtering reduces noise but italso decreases peak heights and increases peak widths, andthere is thus an optimal size for a given data set. Our PCA/PCRanalysis of 3-, 5-, 7-, and 9-point smoothed data showed thatthe 5-point filter gave the lowest error of prediction overall forthe calibration sets. Even though 7-point smoothing gave lowererrors in some instances, for consistency, 5-point smoothingwas used throughout.

Quantitative Field-Portable Attenuated Total ReflectionFourier Transform Infrared Spectroscopy. The threemethylamphetamine HCl samples available could only beserially diluted with glucose and caffeine due to limitedavailability and hence there is some co-linearity betweencomponents (Table III). Each sample had been prepared by adifferent synthetic route. The spectra of the calibration mixtures(Table III) acquired on the HazMatIDTM had sharp peaks,without the baseline off-set problems due to opacity/scatteringeffects observed with the KBr disk preparations. Therefore,

only area normalization (Fig. 8) was required to account for thevariations in the intensities due to variations in the amount ofsample in contact with the ATR crystal.

The three components are clearly resolved in the PCA (Fig.9), along with the intermediate mixtures, again requiring onlytwo principal components to account for 98% of the variance.A PCR model was developed similar to that used for thetransmission FT-IR data. Because of the limited amount ofmaterial available, the spectra of the methylamphetamine HClcalibration set were reacquired to evaluate the model.Predictions of methylamphetamine, glucose, and caffeine(Fig. 10) were acceptable, with most being predicted to within6% w/w of their value. However, some predictions were inerror by up to 20% w/w, which indicates that there arerepeatability problems that probably arise from both heteroge-neity and having variable amounts of sample in direct contactwith the ATR crystal for each measurement. Averagingtriplicates reduced the RMSRR to ;3%.

As was found with the transmission spectra, usingcorrelation-derived PCs did not improve the accuracy ofprediction and the use of first or second derivatives leads topoorer predictions.

Robustness of the Calibration Models. A constrainedlattice (Fig. 11) was used to design additional mixtures thatcontained pseudoephedrine HCl in place of ephedrine HCl asthe surrogate illicit drug to evaluate the robustness ofcalibrations. Similarly paracetamol, a common illicit drugdiluent, was used to dilute the methylamphetamine HClsamples (Table IV). The ability to predict known componentsin mixtures in the presence of ‘‘unknown’’ components, whichis the likely scenario for seized samples, was evaluated byassessing the performance of models that specified only thethree ‘‘known’’ components, all four components, or solely thetarget component (ephedrine HCl or methylamphetamine HCl).The PCs selected for the models were again those thatindividually accounted for .1% of variance.

It is evident from Table V that a model based only onephedrine HCl, glucose, and caffeine performs poorly when

TABLE III. Composition of methylamphetamine HCl (MA) mixturesdiluted with glucose, caffeine, and paracetamol.

Mixture ID MA (%) Glucose (%) Caffeine (%)

33 25.0a 68.6 0.034 39.7b 49.8 0.035 40.7c 50.0 0.036 25.7b 32.3 35.337 26.9c 33.1 33.838 16.5a 45.3 33.939 77.2d 0.0 0.040 0.0 100.0 0.041 0.0 0.0 100.0

a Purity of methylamphetamine sample used: 79.5%.b Purity of methylamphetamine sample used: 79.2%.c Purity of methylamphetamine sample used: 81.4%.d Purity of methylamphetamine sample used: 77.2%.

FIG. 8. Normalized ATR-FT-IR spectra of methylamphetamine HCl mixtures.

FIG. 9. Scores from the PCA model of scaled ATR-FT-IR spectra ofmethylamphetamine HCl calibration and validation samples; (u) methylam-phetamine HCl, (n) glucose, (^) caffeine, and (*) mixture.

644 Volume 62, Number 6, 2008

pseudoephedrine HCl is present. This is expected as the modelcannot distinguish between pseudoephedrine HCl and ephed-rine HCl (Fig. 12), because their spectra have only minordifferences, and this also highlights the well-known problemthat occurs with quantifying materials when an unknowncomponent is not accounted for in the calibration populationand has a significant influence on the model. Includingpseudoephedrine HCl in the model gave similar quality fitsto the data and predictions to those observed when onlyephedrine HCl, glucose, and caffeine are present. RMSSRs and

RMSEPs using four PCs derived from the scaled spectra(accounting for 99% of variance) were in the 3–6% range forall four components. Calibration and predicting for onlyephedrine HCl in ten randomly selected samples gave anRMSSR of 2.5% and an RMSEP of 4.3%.

The methylamphetamine HCl, glucose, and caffeine modelwas similarly affected by dilution with paracetamol. Caffeine istypically predicted 30% w/w higher and glucose 20% w/whigher than their known value, and pure paracetamol waspredicted to have ;10% w/w methylamphetamine, ;40% w/wglucose, and ;50% w/w caffeine (Table VI). Includingparacetamol improves performance significantly, with discrep-ancies between predicted and known values being ,7% w/wfor 45 of the 52 calibration and evaluation mixtures, with theworst case being 20% w/w (Table VII). Using three PCs(accounting for 98% of variance), RMSSRs were in the range3–5% and RMSEPs were 3.5%, 5.4%, 2.4%, and 5.3%,respectively, for methylamphetamine, glucose, caffeine, andparacetamol. The poor prediction of pure paracetamol (sample45) is most likely due to the significant mutual overlap of thespectra of the four components, while the poor prediction ofsamples 33, 34, and 37 may be due to the heterogeneity of thebulk sample, as more than one component (methylamphet-amine HCl, glucose, and/or caffeine) is poorly predicted. Usingthe entire data set but leaving out six randomly selectedsamples as an evaluation set to predict only methylamphet-amine content gave a similar quality of fit and prediction(RMSSR ¼ 3.9%, RMSEP ¼ 2.6%), without major discrepan-cies.

The choice of regression strategy significantly affectsoutcomes; hence, a decision must be made as to whether theaim of the analysis is to target a specific analyte or to fullycharacterize the sample. If the former, then the componentslikely to be encountered in unknown samples must be presentin at least some of the calibration set, but their concentrations

FIG. 10. Prediction of (a) methylamphetamine, (b) glucose, and (c) caffeinecomponents in the methylamphetamine HCl mixtures using two PCs from thePCA model of normalized ATR-FT-IR spectra.

FIG. 11. Constrained lattice design for mixtures containing pseudoephedrineHCl diluted with glucose and caffeine used to assess PCR model robustness.

TABLE IV. Percentage composition of samples of methylamphetamineHCl (MA) mixtures diluted with 50% paracetamol.

Mixture ID Paracetamol (%) MA (%) Glucose (%) Caffeine (%)

42 50.0 12.9a 16.1 17.743 50.0 13.5b 16.5 16.944 49.2 8.4c 23.0 17.245 100.0 0.0 0.0 0.0

a Purity of methylamphetamine sample used: 79.2%.b Purity of methylamphetamine sample used: 81.4%.c Purity of methylamphetamine sample used: 79.5%.

APPLIED SPECTROSCOPY 645

are not required. If the latter, all potential components and theirconcentrations must be encompassed by the calibration set.

Field-Portable Raman Spectroscopy. The Raman spectraof selected methylamphetamine HCl mixtures obtained withthe field-portable FirstDefenderTM spectrometer are shown inFig. 13. The spectra are well resolved, of acceptable quality,and qualitatively clearly show the presence of methylamphet-amine HCl from its peaks at 622, 836, and 1002 cm�1.However, they show that even with long wavelength excitation(785 nm) some samples still fluoresce or have complexbaselines. This along with a small sample area (;1.3 mm2)renders quantification more difficult than is the case with ATR-FT-IR. Furthermore, Raman spectroscopy could be easilydefeated by illicit drug producers with the addition of a traceamount of fluorescent material. Hence, we did not pursue it forquantitative analyses.

CONCLUSION

Fourier transform infrared spectroscopy has been shown tohave sufficient quantitative accuracy for the screening of seizedmethylamphetamine samples in the field. Both transmission

FT-IR and ATR-FT-IR are much less time consuming thantraditional chromatographic methods and hence are suitable forrapid quantitative assessment of illicit drug mixtures. However,the transmission technique is less suitable for on-site analysesbecause of its lack of transportability. It seems unlikely thatRaman spectroscopy would meet with the same degree ofsuccess as FT-IR spectroscopy as sample heterogeneity may bea greater problem and the method can be defeated by theaddition of a small amount of fluorescent material to illegallymanufactured substances.

Attenuated total reflection Fourier transform infrared spectraare affected by the amount of an individual component in asample that is in contact with the ATR crystal, which isdependant on particle size and sample heterogeneity. Hence,care must be taken to ensure that the calibration set andunknowns have the same particle size and distribution and areadequately mixed before spectra are obtained. For transmissionFT-IR the particle size and distribution are less of an issue, butrepresentative sampling of heterogeneous materials is obvi-ously still crucial.

For quantitative FT-IR work, baseline zeroing and areanormalization of the spectra are necessary data pretreatmentmethods for transmission FT-IR as they reduce the influence ofsample preparation variance, but baseline zeroing is notnecessary for ATR-FT-IR spectra. Calibration models basedon these spectra can be developed using principal componentanalysis (PCA) and principal component regression (PCR).

With ATR-FT-IR field-portable instrumentation, the predic-tions of methylamphetamine, glucose, and caffeine, using aPCR model developed from two PCs from a covariance PCA,were generally within 6% w/w of the known values, withRMSEPs of ;4%, as was found for the transmission method.Predictions for methylamphetamine, glucose, caffeine, andparacetamol using a PCR model developed with three PCsfrom a covariance PCA gave similar results, while predictingonly methylamphetamine has an RMSEP of 3%. Using PCsderived from correlation PCA for regression (to increase theweighting given to minor peaks in the spectra) does notsignificantly alter the accuracy of prediction, while using firstor second derivatives generally leads to poorer predictions. To

TABLE V. Prediction of ephedrine HCl, glucose, and caffeine components in pseudoephedrine HCl test mixtures using two PCs from PCA model ofscaled infrared spectra.

Mixture ID (Fig. 11)

Ephedrine HCl (%) Glucose (%) Caffeine (%)

Actual Predicted Actual Predicted Actual Predicted

23 0.0 45.8 0.0 50.7 0.0 3.524 0.0 45.8 33.2 56.5 0.0 �2.324 0.0 38.1 33.2 65.2 0.0 �3.325 0.0 40.1 0.0 12.4 33.3 47.625 0.0 35.5 0.0 18.7 33.3 45.826 0.0 23.2 66.6 83.1 0.0 �6.326 0.0 21.7 66.6 84.8 0.0 �6.527 0.0 16.8 33.0 38.2 33.5 45.027 0.0 21.7 33.0 33.3 33.5 45.128 0.0 19.3 0.0 0.7 67.3 80.028 0.0 25.9 0.0 �17.6 67.3 91.729 0.0 �1.7 100.0 104.8 0.0 �3.030 0.0 15.7 67.1 40.9 32.9 43.430 0.0 2.8 67.1 56.6 32.9 40.631 0.0 0.7 33.9 25.5 66.1 73.831 0.0 �1.3 33.9 23.8 66.1 77.432 0.0 9.6 0.0 �4.2 100.0 94.6

FIG. 12. Transmission (KBr) FT-IR spectra of ephedrine HCl and pseudo-ephedrine HCl.

646 Volume 62, Number 6, 2008

reduce noise in the derivatives a degree of smoothing isrequired, which can lead to a loss of resolution andconsequently poorer quantification. As smoothing is alwayssubjective, totally objective measurements such as theabsorbance spectra are likely to be less prone to error.

Assessment of the robustness of transmission FT-IR andATR-FT-IR calibration models to the presence of ‘‘unknown’’substances in the samples clearly points out that the validity ofquantification is dependent on the comprehensiveness of thecalibration set used. Samples that contain components that arenot encompassed by the calibration set will almost inevitablyproduce erroneous results, be it for the determination of asingle analyte or a more comprehensive analysis. Providingthat such samples are recognized and quantified by othermeans, the calibration set would be able to be updated toinclude those components.

It is clear that attenuated total reflection Fourier transforminfrared spectroscopy, in combination with multivariatecalibration methods such as PCR, has the potential to allowillicit substances to be quantitatively analyzed at the point ofseizure or on-site at clandestine laboratories. To do thissuccessfully will require larger calibration data sets than thoseused here and outlier detection to assure the quality of results.

1. M. D. Cole, The Analysis of Controlled Substances (John Wiley and Sons,Chichester, 2003), Chap. 2, p. 13.

2. I. S. Lurie, C. G. Bailey, D. S. Anex, M. J. Bethea, T. D. McKibben, and J.F. Casale, J. Chromatogr., A 870, 53 (2000).

3. P. A. Hays, J. Forensic Sci. 50(6), 1 (2005).4. B. C. Smith, Fundamentals of Fourier Transform Infrared Spectroscopy

(CRC Press, London, 1996), pp. 117–125.

TABLE VII. Prediction of methylamphetamine (MA), glucose, caffeine, and paracetamol components in the validation mixtures where paracetamol wasa known major diluent using three PCs from the PCR model of normalized ATR-FT-IR spectra.

Mixture ID

MA (%) Glucose (%) Caffeine (%) Paracetamol (%)

Actual Predicted Actual Predicted Actual Predicted Actual Predicted

33 25.0 31.3 68.6 58.2 0.0 0.2 0.0 2.434 39.7 24.3 49.8 68.6 0.0 �0.1 0.0 0.935 40.7 44.2 50.0 43.0 0.0 �0.2 0.0 1.836 25.7 23.9 32.3 34.4 35.3 38.6 0.0 �3.037 26.9 30.2 33.1 21.7 33.8 42.4 0.0 �2.038 16.5 16.7 45.3 47.1 33.9 31.7 0.0 0.339 79.5 76.6 0.0 2.3 0.0 1.5 0.0 0.039 79.2 81.2 0.0 0.7 0.0 �1.2 0.0 �1.739 81.4 81.9 0.0 �0.5 0.0 �1.1 0.0 �1.439 77.2 77.7 0.0 1.3 0.0 0.4 0.0 0.740 0.0 �0.1 100.0 100.9 0.0 0.3 0.0 �0.941 0.0 2.2 0.0 3.0 100.0 96.8 0.0 �2.542 12.9 11.3 16.1 13.1 17.7 15.0 50.0 57.743 13.5 11.7 16.5 16.1 16.9 14.5 50.0 54.744 8.4 7.1 23.0 20.0 17.2 16.4 49.2 54.845 0.0 6.9 0.0 7.7 0.0 2.3 100.0 81.3

FIG. 13. Selected dispersive Raman spectra of methylamphetamine samplesacquired by the FirstDefenderTM, (a) methylamphetamine HCl, (b) methylam-phetamine HCl/glucose, (c) methylamphetamine HCl/glucose/caffeine, and (d)methylamphetamine HCl/glucose/caffeine/paracetamol.

TABLE VI. Prediction of methylamphetamine (MA), glucose, and caffeine components in the validation mixtures where paracetamol was an unknownmajor diluent using two PCs from a PCR model.

Mixture ID

MA (%) Glucose (%) Caffeine (%)

Actual Predicted Actual Predicted Actual Predicted

42 12.9 13.6 16.1 35.9 17.7 47.042 12.9 14.2 16.1 35.5 17.7 46.643 13.5 13.3 16.5 37.1 16.9 46.243 13.5 14.4 16.5 37.4 16.9 44.544 8.4 10.2 23.0 40.1 17.2 47.144 8.4 9.8 23.0 41.3 17.2 46.445 0.0 8.3 0.0 39.6 0.0 50.045 0.0 11.1 0.0 39.3 0.0 46.8

APPLIED SPECTROSCOPY 647

5. P. J. Gomm and I. J. Humphreys, J. Forensic Sci. Soc. 15, 293 (1975).6. M. Lopez-Arıguez, A. Camean, and M. Repetto, J. Forensic Sci. 40, 602

(1995).7. R. Levy, M. Ravneby, L. Meirovich, and O. Shapira-Heiman, J. Forensic

Sci. 41, 6 (1996).8. M. Ravreby and A. Gorski, J. Forensic Sci. 34, 918 (1989).9. D. Wielbo and I. R. Tebbett, J. Forensic Sci. 37, 1134 (1992).

10. A. G. Ryder, G. M. O’Connor, and T. J. Glynn, J. Forensic Sci. 44, 1013(1999).

11. S. E. J. Bell, D. T. Burns, A. C. Dennis, L. J. Matchett, and S. J. Speers,Analyst (Cambridge, U.K.) 125, 1811 (2000).

12. S. E. J. Bell, D. T. Burns, A. C. Dennis, and S. J. Speers, Analyst(Cambridge, U.K.) 125, 541 (2000).

13. E. Katainen, M. Elomaa, U.-M. Laakkonen, E. Sippola, P. Niemela, J.Suhonen, and K. Jarvinen, J. Forensic Sci. 52, 88 (2007).

14. A. G. Ryder, J. Forensic Sci. 47, 275 (2002).15. C. J. Strachan, D. Paratiwi, K. C. Gordon, and T. Rades, J. Raman

Spectrosc. 35, 347 (2004).

648 Volume 62, Number 6, 2008