Comparative Analysis of Benchtop NMR and HPLC‐UV for Illicit Drug Mixtures

ABSTRACT

This study evaluates the feasibility of benchtop NMR spectroscopy for the quantitative analysis of methamphetamine hydrochloride in binary and ternary mixtures. A 60‐MHz benchtop NMR spectrometer was used to analyse samples containing methamphetamine hydrochloride at purities ranging from approximately 10 to 90 mg per 100 mg of sample, alongside cutting agents (methylsulfonylmethane, N‐isopropylbenzylamine hydrochloride, caffeine and phenethylamine hydrochloride) and an impurity (pseudoephedrine hydrochloride). Spectral data were processed using integration, global spectral deconvolution (GSD), quantitative GSD (qGSD), and a quantitative quantum mechanical model (QMM). The root mean square error (RMSE) for these methods ranged from 4.7‐mg analyte per 100 mg of sample for integration down to 1.3‐mg analyte per 100 mg of sample for QMM when determining methamphetamine hydrochloride purity across binary and ternary mixtures. To further assess performance, additional mixtures were analysed using benchtop NMR with QMM and HPLC‐UV, yielding RMSE values of 2.1 and 1.1, respectively, for methamphetamine hydrochloride purity quantification across all samples. While HPLC‐UV maintains greater precision, benchtop NMR with QMM offers a cost‐effective and robust alternative, enabling simultaneous quantification of active substances and impurities with reduced reliance on solvents and calibration standards. This study underscores the potential of benchtop NMR as a complementary tool in forensic science and for implementing a quantitative technique in harm‐reduction drug‐checking centres.

This study assessed benchtop NMR with Quantum Mechanical Modelling (QMM) for quantifying methamphetamine in binary and ternary mixtures. QMM achieved a root mean square error (RMSE) as low as 1.3 mg/100 mg sample, enabling simultaneous quantification of drugs, adulterants, and impurities. Accuracy was comparable to HPLC‐UV (RMSE 1.1), but NMR permits quantification of all species and potentially enables simultaneous identification and quantification.

1. Introduction

In 2022, an estimated 292 million people worldwide used drugs, reflecting a 20% increase over the past decade, with 64 million individuals suffering from drug use disorders [1]. This rise has been accompanied by a steady increase in the quantity of seized drugs, particularly amphetamine‐type stimulants such as methamphetamine [1]. Accurate identification and quantification of drugs are critical for legal proceedings [2, 3], public health assessments and harm‐reduction strategies [4]. Quantitative information on the purity of drugs directly influences the understanding of a drug's pharmacological impact and associated adverse effects [5]. However, current analytical techniques face challenges due to their cost, complexity and time requirements [6]. This article aims to explore whether benchtop nuclear magnetic resonance (NMR) spectroscopy can address these challenges by enabling fast, cost‐effective quantification in forensic laboratories and drug‐checking harm‐reduction centres.

Methamphetamine, a powerful stimulant within the amphetamine class, presents a significant global concern due to its high abuse potential and severe health consequences [7, 8]. In 2022, an estimated 367 tons of methamphetamine were seized globally [1]. Its use has been linked to neurological damage [9], including premature strokes, Parkinsonism and dementia [10], indicating its potential role in accelerating brain ageing [11]. Moreover, the illicit drug market often lacks quality control [12], leading to the inclusion of harmful substances such as adulterants or impurities [13, 14]. These additions, driven by profit motives to enhance potency or dilute valuable components, may pose severe health risks [14, 15]. Therefore, identifying and quantifying such substances are critical for mitigating harm.

In forensic chemistry, the Scientific Working Group for the Analysis of Seized Drugs (SWGDRUG) [16] recommends various analytical techniques for drug analysis, categorised by their discriminating power [17]. Fourier‐transform infrared (FT‐IR) spectroscopy and gas chromatography–mass spectrometry (GC–MS) are among the most used and highly regarded methods for qualitative drug analysis [18, 19, 20, 21, 22, 23, 24]. FT‐IR, classified as a Category A technique by SWGDRUG, provides high discriminating power and is often used in combination with other methods to ensure accurate drug identification [17]. Due to budget and space constraints, harm‐reduction centres usually rely on FT‐IR for on‐site identification [4, 19, 25]. However, it is not reliable for the identification of mixtures. GC–MS, combining Category A (mass spectrometry) and Category B (gas chromatography) techniques, is considered the gold standard [6] for comprehensive drug screening. However, GC–MS often involves time‐consuming sample preparation steps, including hydrolysis and derivatisation, particularly for non‐volatile or thermally labile drugs [21]. NMR spectroscopy is another Category A identification technique, but it is not widely used due to the high cost of the superconducting magnets usually required to generate strong magnetic fields. Thus, there remains a need for new techniques that can provide rapid and accessible identification of samples.

The methods described above primarily focus on identification, but they are not usually used for quantitative analysis. Quantification is equally important in legal contexts and harm reduction. For instance, zombie‐like effects have been reported in individuals consuming synthetic cannabinoids above certain concentrations [5], so dose information can have significant harm reduction benefits. From a legal perspective, the amount of a seized drug often determines the severity of penalties, with larger quantities leading to harsher consequences [3]. Quantification by GC–MS is possible, but it requires specific internal standards [26] and frequent recalibrations, which increase labour time and costs [6]. NMR also provides quantitative information [27], but as noted above, it is rarely available. Therefore, forensic laboratories commonly rely on high‐performance liquid chromatography coupled with ultraviolet spectroscopy (HPLC‐UV) for quantifying drugs. For quantitative analysis, HPLC‐UV is widely regarded as the gold standard. It provides high precision and can quantify substances at very low concentrations. However, HPLC‐UV is classified as a combination of Category B (HPLC) and Category C (UV spectroscopy) techniques [17], which means it is not considered sufficiently specific for substance identification. Furthermore, despite its effectiveness in quantification [28, 29, 30, 31], the method requires specific standards for each analyte and relies heavily on toxic and/or expensive solvents, both of which complicate its application and increase operational costs [32]. These limitations necessitate exploration and development of alternative methodologies that could streamline drug analysis.

Given the limitations in current analytical methods, NMR spectroscopy stands out as a compelling alternative due to its inherent quantitative nature, ability to provide rich spectral information and structural elucidation capabilities [33]. NMR provides the highest level of selectivity through structural information, being classified as a Category A analytical technique by the SWGDRUG [17]. It has also demonstrated significant efficacy in both qualitative and quantitative analyses of illicit substances [27, 34]. Its high discriminating power and inherent quantification are particularly advantageous in the context of novel psychoactive substances (NPSs) [26], where obtaining certified analytes is both expensive and challenging [35, 36]. This capability is crucial in the rapidly evolving landscape of synthetic drugs [1], where traditional analytical methods may fall short. Despite its effectiveness, the widespread adoption of NMR in forensic contexts has been hindered by the large size and high costs associated with conventional NMR spectrometers [6]. However, the past decade has witnessed significant development in NMR technology with the introduction of high‐resolution benchtop NMR spectrometers [37, 38]. These instruments are characterised by their compact size, affordability and ease of use, addressing many of the challenges that previously limited the application of NMR in forensic drug analysis [39] and drug checking [40]. Studies have shown their applicability in quantifying illicit drugs [6, 41, 42], detecting impurities [43] and providing structural information for NPS [44] in seized samples. However, the magnetic field strengths of benchtop NMR instruments are not as high as conventional instruments, and thus, even though the new generation of benchtop instruments offers significant improvement over historical systems, these instruments still have reduced sensitivity and spectral resolution when compared with conventional high‐field NMR systems [45]. Low spectral resolution can lead to spectral overlap, making traditional peak integration methods inadequate for quantitative analysis [45]. Hence, quantitative analysis using benchtop instruments is possible but can require calibration curves and/or minimal overlapping signals to achieve high accuracy [6, 41, 42]. Therefore, the development of advanced quantification methods that address the issue of complex, overlapping spectra on benchtop NMR systems is essential [46].

Addressing the challenges of overlap in NMR spectra has led to the development of various advanced signal processing techniques. These include AMARES [47], QUEST [48], AQSES [49], CRAFT [50, 51], indirect hard modelling [52], global spectrum deconvolution (GSD) [53], quantitative GSD (qGSD) [54], QMM (also referred to as quantum mechanics‐total‐line‐shape fitting) [45, 55] and others. GSD and qGSD are perhaps the most widely used of these techniques, as they are implemented in the widely used Mnova software package. QMM is another promising approach, utilising key NMR parameters such as chemical shifts and coupling constants to generate ideal spectra that are then fitted to measured data for quantification. This method uses the physical understanding of the NMR measurement to effectively model peak overlaps, using a small set of field strength–independent parameters. QMM can accommodate spectral variations due to changes in chemical shift, coupling or relaxation rates, such as those induced by pH variations [46]. The QMM approach has been used for a variety of applications, including wine analysis [32], honey quantification [56], mixture analysis of metabolites [57], pharmaceutical amino acid analysis [58] and drugs [59]. To the best of the authors' knowledge, the efficacy of benchtop NMR coupled with QMM for the quantitative analysis of drugs has not yet been established.

This study evaluates the performance of benchtop NMR spectroscopy for analysing MA in binary and ternary mixtures. Quantification is performed using conventional tools in the Mnova software and a QMM (as implemented in the Q2NMR software). The accuracy and reliability of benchtop NMR are compared to those of traditional HPLC‐UV methods to assess its applicability in real‐world scenarios. By addressing the challenge of spectral overlap, this research explores whether benchtop NMR spectroscopy could be a practical and efficient alternative to HPLC‐UV for quantitative drug analysis in forensic laboratories and harm‐reduction centres.

2. Experimental

2.1. Selection of Chemicals

This study focuses on the quantitative analysis of methamphetamine hydrochloride (MA) in complex mixtures using benchtop NMR spectroscopy. To develop this analytical method, binary and ternary mixtures were prepared by combining MA with carefully selected components relevant to the New Zealand context of illicit drug profiling. The selected substances were methylsulfonylmethane (MSM), N‐isopropylbenzylamine hydrochloride (IPBA), phenethylamine hydrochloride (PE), caffeine (CAF) and pseudoephedrine hydrochloride (PSE).

MSM, IPBA and PE were specifically included due to their prevalence as cutting agents frequently mixed with MA, particularly within New Zealand [60]. CAF, a widely utilised cutting agent [15, 61], was included due to its common occurrence in illicit mixtures. PSE, a known precursor in MA synthesis [43], was included to represent impurities that arise from incomplete synthetic pathways.

All mixtures were prepared in deuterium oxide (D₂O) to minimise interference from the dominant solvent peak, which can produce strong proton signals that can obscure the spectral features of target analytes. Maleic acid was used as an internal standard due to its distinct singlet resonance that is well resolved from the peaks associated with the other components studied here, which facilitates accurate integration and quantification in NMR analysis.

2.2. Chemicals

MA, along with several other controlled substances used in this study, was obtained from the New Zealand Institute for Public Health and Forensic Science (PHF Science), Auckland, and sourced from seized materials. Drug mixtures were prepared using MA (~100% purity from PHF Science), MSM (~100%, MRM; 98%, Sigma‐Aldrich), IPBA (97%, Sigma‐Aldrich), CAF (~100%, BDH Chemicals Ltd., Poole, England; ≥ 99%, Sigma‐Aldrich), PE (~100% purity from PHF Science), and PSE (Emmellen Biotech Pharmaceuticals Limited, ~100% purity). Maleic acid (≥ 99%, Sigma‐Aldrich) was utilised as an internal standard for quantification in the NMR analysis. For pure drug samples, D₂O premixed with 0.05 wt% 3‐trimethylsilyl propionic acid‐d₄ sodium salt (D₂O–TMSP, Sigma‐Aldrich) was used, with TMSP serving as the chemical shift reference standard. All mixtures were prepared in D₂O (99.9 atom% D, Sigma‐Aldrich); no TMSP was used for these samples. For HPLC‐UV analysis, acetonitrile (HPLC gradient grade, Fisher Scientific, United Kingdom; > 99%), diethylamine (puriss p.a. 99.5% [GC], Sigma‐Aldrich, Belgium; > 99.5%) and ammonium acetate (Ajax Finechem Pty Ltd., Thermo Fisher Scientific; 97%) were used to prepare the eluent for quantitative measurements. Standards and controls for MA hydrochloride quantification via HPLC‐UV were sourced from the National Measurement Institute.

2.3. Sample Preparation

Binary and ternary mixtures of MA, impurities and cutting agents were prepared in D₂O with an internal maleic acid standard. Four sets of binary samples were prepared: MA + MSM, MA + CAF, MA + PE and MA + PSE. For MA + MSM, five samples were prepared with nominal mass ratios of 1:9, 1:3, 1:1, 3:1 and 9:1. For the other binary sets, three samples were prepared with nominal mass ratios of 1:9, 1:1 and 9:1. Three sets of ternary samples were prepared: MA + MSM + IPBA, MA + MSM + CAF and MA + PE + PSE. For the mixtures MA + MSM + IPBA, five samples were prepared with nominal mass ratios of 1:2:2, 5:8:7, 2:1:1, 15:3:2 and 18:1:1. For the other ternary mixtures, three samples were prepared with nominal mass ratios of 1:2:2, 2:1:1 and 18:1:1. The components were weighed according to the selected compositions to prepare approximately 100 mg of the mixture. To minimise weighing errors and reduce associated uncertainty, when the required mass of any individual constituent was less than 10 mg, a total sample mass of 200 mg was prepared. To each mixture, 20–40 mg of maleic acid was added. The entire sample was then dissolved in 1–2 mL of D₂O, such that the final concentration of maleic acid was 20 mg/mL. Exact compositions of all mixtures are reported in the Table S1.

To determine the limit of detection (LoD) and limit of quantification (LoQ), new mixtures were prepared of MA and MSM with the following compositions (nominal mg of analyte per 100 mg of sample): 10 MA + 90 MSM, 50 MA + 50 MSM and 90 MA + 10 MSM; a stock solution of MA was prepared separately at a concentration of 20 mg/mL. For each drug mixture, an initial solution was prepared by adding 2 mL of the maleic acid stock solution to the powdered sample. This was followed by up to threefold serial dilutions using the same maleic acid solution to maintain a consistent internal standard concentration across all 12 samples.

To compare benchtop NMR analysis with HPLC‐UV, a second preparation technique was required to ensure the same samples could be measured with each technique and that these samples had the same composition. For these trials, MA and different components were mixed in the same compositions as used for the above method. A total quantity of > 200 mg of each mixture was prepared so that the same material could be analysed by HPLC‐UV and NMR. For uniformity, constituents were thoroughly homogenised using a mortar and pestle. These blended mixtures were then stored in glass vials.

Aliquots from the ground mixtures were then taken for both NMR and HPLC‐UV analyses. For HPLC‐UV analysis, samples were prepared in duplicate using the mobile phase as a solvent, with final sample concentrations ranging from 0.8 to 1.0 mg mL⁻¹. An HPLC eluent was prepared using ammonium acetate (4.5 g) dissolved in distilled water (600 mL), combined with diethylamine (3.75 g) and diluted in acetonitrile (400 mL). The pH of the eluent was adjusted to 8.9 using glacial acetic acid and/or ammonia. The solution was vacuum filtered using a 0.45‐μm filter.

For NMR analysis, samples were prepared at a concentration of approximately 100 mg mL⁻¹ in D₂O by weighing 100 mg of the homogenised mixture, adding 20 mg of maleic acid as an internal standard and dissolving the sample in 1 mL of D₂O. For four samples with less than 100 mg of homogenised mixture, the samples were dissolved in a proportionally smaller volume of D₂O (Table S2). Due to the time required for method development, solid samples were stored for approximately 16 months prior to preparing them for benchtop NMR analysis.

2.4. Instruments

Quantification using benchtop NMR utilised a 60‐MHz Magritek Spinsolve ULTRA spectrometer. Before each set of scans, the magnet was shimmed using the ‘Quickshim All’ routine with a 95% D₂O/5% H₂O shimming sample. For each sample, if the shim had degraded then the ‘1st order shim sample’ routine was used. Prior to acquisition, the sample was equilibrated in the instrument for a minimum of 2 min so that the sample could come to the fixed magnet temperature of 26.5°C. One‐dimensional ¹H NMR data were acquired as free induction decays (FIDs) using the Proton CDEC function in the associated Spinsolve software. Acquisition parameters were 16 scans, repetition time of 30 s, ¹³C decoupling, 3.2 s of acquisition time and an 81‐ppm spectral width.

The HPLC‐UV analysis was conducted using a UHPLC system (Vanquish Flex UHPLC, Thermo Fisher Scientific, USA), which featured a VF‐A10‐A analytical split‐loop autosampler (10–25 μL of injections), a VF‐P20‐A serial dual‐piston pump, a VH‐C10‐A thermostated column compartment with active preheater and passive post‐column cooling and a VF‐D11‐A photodiode array detector. The detector was equipped with deuterium (optimising the 190–380‐nm range) and tungsten (optimising the 380–800‐nm range) lamp light sources. The column employed for separation was a LiChrospher 100 RP‐18 (5‐μm particle size) LiChroCART 125‐4 HPLC cartridge, manufactured by Agilent Technologies. The eluent flow rate was set at 1.5 mL/min, and the temperature of the column was maintained at 40°C ± 1.0°C. Detection of amphetamines occurred at 254 nm with a bandwidth of 10 nm.

2.5. Quantification Methods

2.5.1. Benchtop NMR Spectroscopy

Quantitative analysis using benchtop NMR presents particular challenges in achieving high precision and accuracy while maintaining simplicity in sample preparation. In this study, an internal reference standard approach was used for absolute quantification, aiming to achieve high precision with minimal preparation. The analysis used two software tools, Q2NMR [45] and Mnova [62], to evaluate the performance of the quantum mechanical modelling approach implemented in Q2NMR in comparison with traditional methods.

QMM operates using a complete spin system modelling approach, enabling precise spectral modelling across chemical species. By adjusting a limited set of parameters including chemical shifts, J‐coupling constants and relaxation rates, QMM can characterise even complex spectra with significant peak overlaps. A key feature of QMM is its field‐invariant parameterisation, allowing analysis across NMR instruments of various field strengths with the same model. The analysis process involves fitting measured experimental spectra with reference models for each analyte, optimising parameters for shifts, coupling constants and relaxation rates. Here, the QMM was implemented using the Q2NMR software [45].

Figure 1 shows the experimental benchtop (60 MHz) NMR spectra for the pure chemicals, with a solvent peak at approximately 4.5 ppm (D₂O). NMR spectra of pure species included TMSP, which was used for chemical shift assignment. Due to the limited resolution and significant peak overlap in the benchtop NMR spectra (60 MHz), as well as the presence of pronounced higher order coupling effects, the initial spectral model was developed using pure samples measured on a high‐field spectrometer (Bruker 600‐MHz Advance III NMR spectrometer).

FIGURE 1.

NMR spectra of all analytes acquired at a frequency of 60 MHz. The scale of each spectrum has been adjusted for visual clarity; the TMSP concentration is approximately the same in all samples here. Panels a, c and f exhibit characteristic peaks in the 3–4‐ppm region, with an enlarged view included for clarity. The spectra are referenced to TMSP (0 ppm), and the peak observed around 4.8 ppm corresponds to water.

High‐field spectrometer measurements were conducted and used for model development due to their well‐separated peaks, as shown in Figure 2. In this set‐up, pure samples were prepared in D₂O–TMSP solutions, with TMSP providing a reference peak at 0 ppm for consistent chemical shift assignment. For the initial fitting of each pure component in Q2NMR, spectra were referenced using the chemical shift of TMSP, and parameters were refined starting from the most distinct regions, proceeding to more complex areas, such as the aromatic region. Initial parameters were estimated using an online neural network chemical shift prediction tool [63, 64]. These chemical shift parameters were then manually refined using a spectrum obtained on a high‐field (600 MHz) NMR spectrometer. The aromatic region has previously posed challenges in fitting due to peak overlap, as illustrated by the signals in Figure 2 at around 7.5 ppm. In this study, symmetric protons were forced to have the same chemical shift and J‐couplings, reducing the free parameters available and simplifying the fit. With this modification, the spectra for all species were parameterised accurately throughout (see Figure S1), addressing limitations seen in our prior study [46].

FIGURE 2.

NMR spectra of all analytes acquired at 600 MHz, highlighting high‐resolution spectral data for comparison. The intensity scale of each spectrum has been adjusted for visual clarity. Due to the high resolution, peaks in high‐field spectra appear narrow. An enlarged view of MA peaks is provided as an example for clarity. The spectra are referenced to TMSP (0 ppm), with a water peak observed around 4.8 ppm, which is cut to enhance other spectral features.

Following the model parameterisation for each pure compound, the efficacy of the QMM approach in quantifying mixture compositions was evaluated by analysing binary and ternary mixtures using benchtop NMR. The experimental spectra of the mixtures were loaded into Q2NMR, and a working region of −2 to 10 ppm was selected, as it encompasses all peaks from the various components. The model for the primary analyte, MA, was introduced first, followed by additional component(s) in the binary or ternary mixture, followed by maleic acid (internal standard) and water (solvent). For this study, it was assumed that the species were known a priori. While there is potential to use benchtop NMR to identify components present [42], this is beyond the scope of this study.

The lock of the Spinsolve NMR instrument produces slight variations in the chemical shift reference. Shifts of approximately 0.2 ppm were noted and corrected via a global alignment parameter. This adjustment ensured an optimal match between simulated and experimental spectra. Each chemical shift and J‐coupling parameter in the model was subsequently optimised to achieve precise alignment with the experimental spectra. In most cases, chemical shifts aligned with the values obtained from the pure species on the high‐field NMR system to within 0.03 ppm and were readily fitted using the automated optimisation routine in the software. In some cases, the chemical shift parameters varied by approximately 0.1 ppm and were manually adjusted to within 0.03 ppm and then refined using the automated optimisation routines in the software. This phenomenon commonly occurred with the methyl group adjacent to the amino group in MA, IPBA and PSE, likely due to changes in pH associated with the changes in composition of the samples.

Benchtop NMR spectra were quantified using integration and the GSD and qGSD methods in Mnova, one of the leading independent NMR software packages, for comparison with the QMM results. Prior to analysis, each spectrum was zero‐filled from 2¹⁴ to 2¹⁶ data points, and line broadening was applied using both exponential (1 Hz) and Gaussian functions (1 Hz). Integration was performed by manually defining spectral regions (Table 1) to isolate peaks of interest, prioritising the most well‐resolved peaks for accurate quantification. To minimise the influence of neighbouring peaks, integration limits were carefully set at the base of each peak or multiplet. For overlapping signals, the integration regions were adjusted to capture as much of the peak area as possible while avoiding inclusion of adjacent peaks. In GSD analysis, the algorithm automatically deconvolves the entire spectrum into a list of peaks, each characterised by chemical shift, peak width, height, phase and area. GSD uses a generalised Lorentzian function to model experimental line shapes, enabling analysis of overlapping peaks. qGSD, an enhanced version of GSD, analyses the residuals remaining after the initial fitting and iteratively creates more complex peak shapes to account for imperfections in the spectral line shape that are not captured by the generalised Lorentzian function [65]. Results using integration, GSD and qGSD are reported.

TABLE 1.

Integration ranges for each species for quantitative analysis of components in binary and ternary mixtures. The maleic acid was referenced as 6.43 ppm, so the integration range for maleic acid in all the mixtures was taken as 6.2–6.6. Due to potential peak shifting, the integration ranges are approximate, with manual adjustments of ± 0.2 ppm to account for chemical shift variations in individual spectra.

Mixture	Species	Calculation
MA:MSM	MA	SMA = [I(0.14–1.89)]/3
	MSM	SMSM = [I(1.94–4.07) − 6 * SMA]/6
MA:CAF	MA	SMA = [I(0.63–1.97)]/3
	CAF	SCAF = [I(7.74–8.49)]
MA:PE	MA	SMA = [I(0.47–1.85)]/3
	PE	SPE = [I(2.16–4.11) − 6 * SMA]/4
MA:PSE	MA	SMA = [I(1.21–1.90)]/3
	PSE	SPSE = [I(0.24–1.21)]/3
MA:MSM:IPBA	MA	SMA = [I(0.21–1.99) − 2 * SIPBA]/3
	MSM	SMSM = [I(2.17–3.90) − 6 * SMA − SIPBA]/6
	IPBA	SIPBA = [I(3.92–4.48)]/2
MA:CAF:MSM	MA	SMA = [I(0.29–1.85)]/3
	CAF	SCAF = [I(7.71–8.14)]
	MSM	SMSM = [I(2.03–4.20) − 6 * SMA− 9 * SCAF]/6
MA:PE:PSE	MA	SMA = [I(0.11–1.23)]/3
	PSE	SPSE = [I(1.23–1.93)]/3
	PE	SPE = [I(2.00–4.09) − 6 * SMA – 5 * SPSE]/4

Purity calculations were based on the signal intensities per proton obtained from both Q2NMR and Mnova. Intensities derived from Q2NMR outputs were directly quantified as signal intensity per proton whereas Mnova integrals or peak areas were normalised by the proton count within each defined integration range or peak, as shown in Table 1. The purity y of each component was calculated in mg of analyte per 100 mg of sample as follows:

$$y=\frac{w_S}{w_T}\times \frac{M_A}{M_S}\times \frac{I_A}{I_S}\times 100,$$ (1)

where w denotes mass, M represents molar mass and I indicates signal intensity or normalised peak area. The subscripts S, A and T refer to standard, analyte and total sample taken, respectively.

Before analysing the accuracy of the quantification, it is important to consider the uncertainties that contributed to the estimated purity. There are two major sources of uncertainty to consider: the error in estimated intensity of the analytes and standard, and the error in gravimetric concentration of analytes and standard. The uncertainty of the intensity calculated by the Q2NMR software was determined by assessing the residuals of spectra. The uncertainty associated with the gravimetric purity was estimated by weighing standard weights 10 times and then calculating the standard deviation of the measured mass.

The overall uncertainty was determined using three times the standard deviation, which corresponds to a 99.7% confidence interval. Thus, the overall uncertainty in the purity of components within the sample is calculated from the standard deviation of the resulting distribution of signal intensity using the following formula:

$$\mathit{\Delta y}=3.y\sqrt{\begin{array}{c}{\left(\frac{\mathrm{\Delta }{w}_S}{w_S}\right)}^2+{\left(\frac{\mathrm{\Delta }{w}_T}{w_T}\right)}^2+{\left(\frac{\mathrm{\Delta }{I}_A}{I_A}\right)}^2+{\left(\frac{\mathrm{\Delta }{I}_S}{I_S}\right)}^2\\ \end{array}}$$ (2)

The accuracy of quantification for each mixture set and method of calculation was determined from the root mean square error (RMSE) of the purity measured by NMR or HPLC‐UV and compared with the gravimetric purity:

$$\mathit{\text{RMSE}}=\sqrt{\frac{{\sum }_{i=1}^N{\left(y_i-x_i\right)}^2}{N}},$$ (3)

where y _i is the purity measured by NMR or HPLC‐UV, x _i is the purity determined gravimetrically for sample i and N denotes the total number of purity values. The resulting RMSE value is reported as the error and expressed as mg of analyte per 100 mg of sample.

The sensitivity of the benchtop NMR method, in combination with QMM, was assessed by calculating the LoD and LoQ for MA, following the guidelines established by the International Council for Harmonisation (ICH) [66]. LoD and LoQ were determined using the following standard equations:

$$\mathrm{LoD}=3.3\times \left(\frac{\mathrm{RSD}}{\text{slope}}\right)$$ (4)

$$\mathrm{LoQ}=10\times \left(\frac{\mathrm{RSD}}{\text{slope}}\right)$$ (5)

where RSD refers to the residual standard deviation of replicate measurements and slope is derived from the calibration curve.

2.5.2. HPLC‐UV

Quantitative analysis of MA was also conducted using HPLC‐UV utilising a standard validated method. MA retention times consistently ranged between 4.067 and 4.267 min and resulted in clear separation from other components. Concentration and purity determinations were processed with Chromeleon software (Versions 6.8 and 7.2.6, Thermo Fisher Scientific). The concentration was determined by comparing sample peak areas against a calibration curve. The associated uncertainty calculation takes into account all significant contributions to measurement uncertainty. The standards, control, and samples were all analysed in duplicate. Results are only reported from samples in Set 1; results from Set 2 were consistent with Set 1 to within the validated error of 2 mg per 100 mg sample, confirming no errors in the preparation of the HPLC‐UV analysis.

3. Results and Discussion

This section presents the quantification of the purity of each analyte within complex mixtures, beginning with a comparison of the QMM method with conventional NMR analysis methods. Following this, the efficacy of benchtop NMR in conjunction with QMM is examined through a comparative analysis with the standard analytical technique, HPLC‐UV. These comparisons aim to explore the potential of benchtop NMR with QMM as a practical, cost‐effective quantitative tool for forensic laboratories and harm‐reduction centres, where binary and ternary drug mixtures are encountered in diverse concentrations.

3.1. Development of Quantification With QMM

The experimental spectra for the binary mixtures analysed in this study are shown in Figure 3. Each spectrum displays the experimental data (solid blue line), the QMM spectral fit (red line) and the individual component fits (shaded colours). All spectra display a peak at about 4.5 ppm that comes from the solvent (D₂O) and a peak at approximately 6.2 ppm, which is due to maleic acid, the internal standard. Separation of the signals associated with different components in the binary mixtures of MA with MSM (Figure 3a) and CAF (Figure 3b) was relatively straightforward, as there was minimal spectral overlap. In more complex binary mixtures, like MA with PE (Figure 3c) and PSE (Figure 3d), significant overlap occurred, particularly in the region between 2.5 and 3.8 ppm. Here, QMM was still able to reproduce the experimental spectrum using the models of the two components in each mixture. It is clear from the figure that the peak positions and intensities in the model and experiment were consistent, even in instances where overlapping signals resulted in complex, non‐Lorentzian peak shapes.

FIGURE 3.

Demonstration of QMM fitting (red) of experimental NMR spectra (blue) of binary mixtures. The component models for each species are shaded in the same colour as the name of the species in each part of the figure. There are additional peaks at 6.4 ppm (pink), which corresponds to maleic acid, and 4.5 ppm (blue), corresponding to D₂O (the scaling of each spectrum was adjusted to show the spectral features of all components as clearly as possible).

Ternary mixtures presented additional challenges due to increased peak overlap, as shown in Figure 4. Figure 4a depicts the mixture of MA, MSM and IPBA, which exhibited relatively straightforward fitting owing to the distinct singlet peaks of IPBA at approximately 4 ppm and MSM at 3.2 ppm. Similarly, Figure 4b, representing the mixture of MA, MSM and CAF, demonstrated manageable spectral complexity, attributed to the well‐defined and uncomplicated spectral profiles of both MSM and CAF. In contrast, Figure 4c, comprising MA, PE and PSE, revealed multiple overlapping peaks, posing a more significant analytical challenge. Despite these complexities, QMM successfully modelled the ternary mixtures, effectively isolating individual component contributions in regions of spectral overlap. The cumulative model, representing the sum of individual component fits, closely approximated the experimental spectrum, demonstrating the accuracy of the fitting process (see also the residual plots in Figures S2 and S3).

FIGURE 4.

Experimental NMR spectra of ternary mixtures, with QMM fitting. The shaded regions correspond to the component models for each species, with colours matching the species' labels. The peak at 6.4 ppm (pink) is maleic acid, and one at 4.5 ppm (blue) is D₂O.

The quantification of the spectra for each mixture set by QMM was examined. Figure 5 presents the purity of each analyte calculated using Equation (1), plotted against the gravimetric purity. Vertical error bars indicate uncertainty in NMR purity, calculated using Equation (2). Figure 5a displays estimated MA purity for all binary and ternary mixtures, while Figure 5b shows the purity values for other components. At lower analyte concentrations, calculated NMR purity closely aligned with gravimetric values. However, at high concentrations, large errors in purity were observed, with the method tending to significantly overestimate the concentration. This concentration‐dependent error is attributed to the method of purity calculation, which utilises the ratio of species to reference intensity. Consequently, an error in the reference species signal will produce larger absolute errors as species concentration increases, assuming the relative error in the reference remains consistent.

FIGURE 5.

Comparison of calculated NMR purity with gravimetric purity for (a) MA in all seven mixtures (binary and ternary) and (b) the other components mixed with MA.

To address the observed discrepancies at high concentrations, a comprehensive analysis of peak fitting was undertaken. The QMM model used in Q2NMR operates on the assumption of Lorentzian peak shapes, yet NMR spectra can exhibit slightly distorted, non‐Lorentzian shapes. Figure 6a illustrates this phenomenon, presenting a magnified view of a maleic acid peak. The peak demonstrates a subtle deviation from the ideal Lorentzian shape, as evidenced by the residual plot at the figure's base. These distortions are unlikely to be attributed to higher order coupling effects, given that the maleic acid peak is a singlet. Instead, we hypothesise that these deviations result from slight inhomogeneities in the magnetic field. The observed magnetic field distortions are relatively minor, with a peak width of 0.02 ppm suggesting proper magnet shimming prior to experimentation. The maleic acid peak has the longest relaxation time (and hence narrowest line width) among all species in this mixture, rendering it more susceptible to the effects of magnetic field inhomogeneities. Based on these findings, we hypothesised that line broadening could mitigate imperfections in magnetic field homogeneity as it can suppress the later part of the FID, where the effects of minor inhomogeneities in the magnetic field are observed. By doing so, it reduces the contribution of these inhomogeneities to the final spectrum, resulting in smoother, more uniform peaks. A similar method has previously been reported where a combination of Lorentzian/Gaussian apodisation was used to reduce the effects of imperfections in the line shape [55]. It is important to note that while line broadening can improve spectral quality and signal‐to‐noise ratio, it comes at the cost of reduced spectral resolution. However, model‐based analysis is relatively robust to broad lines, provided that the underlying model is an accurate representation of the spectrum [45]. Therefore, the effect of line broadening was explored. The proposed procedure to reduce the effect of field inhomogeneity was:

1. Initial assignment of chemical shifts, followed by model fitting as detailed in Section 2.5. At this stage, no line broadening was used to ensure maximum resolution of the spectral features.

2. After fitting the chemical shifts, line broadening was applied to the experimental spectrum. The spectral model was then re‐fitted to these data. In this second fitting process, the chemical shifts were fixed, and only the relaxation rates and intensities for the model were refitted to accurately determine the intensity of each species in the sample.

FIGURE 6.

Illustration of the maleic acid peak along with residual (a) without line broadening and (b) with line broadening of 3.5 Hz.

Figure 6b demonstrates the effect of 3.5‐Hz line broadening on the maleic acid peak, revealing substantially improved alignment between the Lorentzian model and the experimental data. The enhancement is particularly evident in the residual plot, presented on a scale proportional to that in Figure 6a. These results suggest that line broadening is an effective strategy for compensating for imperfections in magnetic field homogeneity, improving the accuracy of the model‐based fit to the experimental peak shape.

To investigate the influence of line broadening on the accuracy of quantification, the different mixtures were analysed using QMM with different line broadening factors and the RMSE calculated, as shown in Figure 7. Figure 7a shows a significant reduction in RMSE for MA for line broadening in the region 2–5 Hz, demonstrating the improvement of the quantification results. Similarly, Figure 7b shows enhanced quantification accuracy across all components in binary and ternary mixtures. Minor increases in error were observed in specific cases, such as in the ternary mixture of MA + MSM + CAF; however, such increases were negligible, and the errors remained within an RMSE of 1 mg of analyte per 100 mg of sample, indicating that line broadening contributes positively to the overall reliability of the results. These findings suggest that non‐ideal line shapes significantly affect quantification accuracy but that these effects can be largely mitigated by appropriate application of line broadening (for further explanation, see Section 4 of the Supporting Information).

FIGURE 7.

Line broadening factor vs. quantification accuracy (measured by RMSE in mg of analyte per 100 mg of sample) for binary and ternary mixtures; RMSE (a) for MA and (b) for all other species.

After establishing that line broadening can mitigate line shape issues and improve quantification results, purities of MA along with other adulterants and an impurity were calculated at a line broadening value of 3.5 Hz (Figure 8). A line broadening value of 3.5 Hz was selected for the final purity analysis as this represents a balance between error reduction and maintenance of spectral integrity. Figure 8a shows that when the 3.5‐Hz line broadening was applied, the calculated purity of MA across mixtures remained within a 2–3 mg of MA per 100 mg of sample of the known true value for all samples, even at elevated concentrations. Figure 8b corroborates this observation, showing consistent improvements across all components in binary and ternary mixtures. These results confirm that line broadening enhances the precision of quantification for a diverse range of analytes by mitigating the effects of non‐ideal line shapes. Collectively, the findings indicate that the primary source of quantification errors at higher analyte concentrations is linked to the line shape of the maleic acid peak, rather than inaccuracies in modelling overlapping spectral components. This emphasises the importance of addressing line shape distortions to improve the reliability of quantitative analysis in complex mixtures.

FIGURE 8.

Comparison of the calculated NMR purity with the gravimetric purity for each component after applying 3.5‐Hz line broadening to the NMR data of (a) MA in all seven mixtures (binary and ternary) and (b) the other components mixed with MA.

To evaluate instrumentation error, the same sample was analysed 10 times using benchtop NMR. Signal intensities were quantified using the QMM software, applying a line broadening of 3.5 Hz to minimise line shape distortions and enhance intensity accuracy. The purity was calculated for each of these 10 sets of intensity values, and the standard deviation of these was calculated and multiplied by a factor of three to obtain a 99.7% confidence interval as an estimate of the instrument error. This procedure was repeated for two sample compositions: 10 mg of MA per 100 mg of sample + 90 mg of MSM per 100 mg of sample and 50 mg of MA per 100 mg of sample + 50 mg of MSM per 100 mg of sample. The observed instrument errors for MA in these two mixtures were 0.5 and 1.1 mg of MA per 100 mg of sample, respectively. The average estimated errors calculated from Equation (2) for these samples were 0.4 and 2.6 mg of MA per 100 mg of sample, respectively. These two uncertainties were comparable to each other, and therefore, it was tentatively concluded that Equation (2) provided a fair estimate of the uncertainty of quantification for the benchtop NMR system. This conclusion was investigated further after analysing all the mixture samples below.

Using the QMM analysis procedure outlined, the LoD and LoQ were measured for a dilution of a standard sample, as explained in Section 2.3. The mixtures were fitted using QMM with a line broadening of 3.5 Hz, and the results are shown in Figure 9. The RSD calculated from the plot was 0.2 mg/mL. The calculated LoD and LoQ were 1 and 3 mg/mL, respectively. In this study, the lowest measured concentration of MA was 5 mg/mL, which is well above these limits, ensuring reliable detection and quantification. These values define the minimum concentrations at which MA can be accurately detected and quantified using the current measurement protocol on the benchtop NMR, though these limits may be improved with other measurement settings.

FIGURE 9.

Measurements of the NMR concentration using a serial dilution method in order to calculate the LoD and LoQ.

The efficacy of the QMM approach in quantifying MA along with other components in samples was analysed (Figure 10a). All methods were able to quantify the MA composition well with RMSE values of 4.7, 2.1, 2.2 and 1.3 mg of analyte per 100 mg of sample for Mnova's integration, GSD and qGSD methods and Q2NMR, respectively, across all the binary and ternary mixtures. For the associated adulterants, the RMSE values were 3.6, 4.5, 2.4 and 0.9 mg of analyte per 100 mg of sample for Mnova's integration, GSD and qGSD methods and Q2NMR, respectively. Of the four different analysis methods tested, Q2NMR was consistently the most accurate across all species quantified.

FIGURE 10.

Comparison of RMS error for (a) MA and its associated adulterants/impurities across all samples and (b) MA in simple and complex mixtures, along with the corresponding simple and complex species.

To further evaluate the QMM approach, MA was analysed separately within mixtures containing simple and complex components to evaluate how peak overlap impacts quantification accuracy. Also, simple and complex mixtures were evaluated separately in comparison to established Mnova methods, with RMSE values calculated for each mixture set and categorised based on the complexity of components, as shown in Figure 10b. Simple components, such as MSM, CAF and IPBA, characterised by single or well‐resolved peaks, were contrasted with complex components like PE and PSE, which exhibit significant peak overlap. Across the evaluated methods, integration tended to perform the most poorly. This was likely due to challenges in resolving overlapping signals, as seen with the IPBA–MA mixture (see Figure 4). Integration's performance further deteriorated for complex mixtures, with RMSE values as high as 6.0 mg of analyte per 100 mg of sample, as seen with the ternary mixture of MA, PE and PSE, where severe peak overlap impeded quantification by integration, as expected. GSD and qGSD generally performed better than integration across different mixtures. However, GSD showed significantly reduced accuracy in complex mixtures, with RMSE values reaching 7.3 mg of analyte per 100 mg of sample. This reduction in accuracy is attributed to the difficulty in modelling peak shapes accurately in the presence of high‐order coupling effects and overlapping signals. Conversely, qGSD provided improved accuracy, with RMSE values of 1.4 mg of analyte per 100 mg of sample for MA in complex mixtures. The qGSD method was able to characterise the non‐ideal line shapes associated with high‐order coupling effects effectively, though it exhibited slightly elevated errors in simple mixtures, with an RMSE of 2.6 mg of analyte per 100 mg of sample for MA. QMM consistently performed well, with RMSE values of 1.3 mg of analyte per 100 mg of sample for MA in simple mixtures and 1.2 mg of analyte per 100 mg of sample for complex mixtures. These findings underscore the potential of QMM as a viable alternative for quantification, particularly in challenging scenarios involving unresolved or overlapping peaks.

A critical consideration in the analysis of NMR spectra using techniques such as integration, GSD and qGSD is the necessity for manual peak assignment. Since software like Mnova does not automatically indicate which peaks correspond to which protons, it is essential for the analyst to have a thorough understanding of the chemical structure and the NMR spectra. This manual intervention introduces potential for subjectivity, particularly in complex or low‐resolution spectra where overlapping peaks obscure clear assignments. Such subjectivity leads to variability across analysts, resulting in inconsistencies that compromise the reproducibility of results [54]. In forensic settings, where the demand for reproducibility is paramount, this variability poses significant challenges and raises concerns regarding user‐induced bias. The limitations of the integration method are well documented, with its vulnerability to phase and baseline distortions and its inability to handle peak overlap effectively [54]. The GSD method employs a generalised Lorentzian function to improve relative integral accuracy; however, we observed visible residuals in complex or low‐resolution spectra. The qGSD approach was able to reduce these residuals but sometimes resulted in a mis‐assignment of some signals to the wrong species. For instance, in Figure 11a, the spectral region around 1–1.5 ppm demonstrates overlapping peaks from PSE and MA, making it challenging to assign individual resonances to their respective compounds. Ideally, this region should display a doublet each for PSE and MA. However, the qGSD algorithm incorrectly splits the overlapping peaks into three rather than four peaks, resulting in a slight bias in the quantification. These limitations are exacerbated in complex mixtures, such as MA combined with PSE and PE, where GSD and qGSD often fail to effectively deconvolve overlapping peaks, compromising both quantification accuracy and component identification. In contrast, the QMM approach implemented in Q2NMR offers a holistic spectral modelling approach. With QMM, the entire characteristic NMR spectrum is described using a physical model of each species. Hence, QMM can accurately represent even minor peaks from low‐concentration analytes, as demonstrated in Figure 11b. This approach minimises the risk of overlooking spectral features, thereby enhancing quantification precision and reliability. QMM does require a database containing the spectral parameters of each species, but provided such a database exists, complete deconvolution of the spectrum is relatively straightforward.

FIGURE 11.

Complex spectral region of MA + PSE + PE mixture. (a) Deconvolution of peaks using the Mnova qGSD peak‐fitting routine. (b) Modelling of individual peaks corresponding to MA (purple), PSE (green) and PE (orange).

3.2. Comparative Analysis of Benchtop NMR and HPLC‐UV

The benchtop NMR method was further validated by comparison with the gold standard quantification tool—HPLC‐UV. Figure 12a presents the calculated MA purity obtained from HPLC‐UV, where a single outlier is evident in HPLC‐UV measurements. Figure 12b presents the calculated MA purity by benchtop NMR in conjunction with Q2NMR, where two outliers are observed at lower concentrations. To further assess the error distribution, absolute error values were analysed and visualised using a box‐and‐whisker plot (Figure 12c), illustrating the error range, interquartile spread and potential outliers. HPLC‐UV exhibited a consistent error distribution within the range −2 to 2 mg of analyte per 100 mg of sample, with an RMSE of 1.1 mg of analyte per 100 mg of sample. A single outlier was observed with an error relative to the gravimetric value of around 3 mg of analyte per 100 mg of sample. The error for this sample was consistent with both repeat samples and hence is attributable to an error during sample preparation. In contrast, benchtop NMR using the Q2NMR model demonstrated a similar range of error (−2 to 4 mg of analyte per 100 mg of sample) and an RMSE of 2.1 mg of analyte per 100 mg of sample, but there was a slight positive bias. It was also noted that high MA concentrations were more likely to exceed an error of 1.5 mg of analyte per 100 mg of sample. The slight bias in the NMR quantification will be discussed further after considering the quantification of the other analytes.

FIGURE 12.

Comparison of MA quantification using HPLC‐UV and benchtop NMR. (a) Purity of MA measured by HPLC‐UV. (b) Purity of MA measured by benchtop NMR in conjunction with Q2NMR. (c) Box‐and‐whisker plot showing error distribution for each method. (d) RMSE values for both techniques.

In addition, benchtop NMR successfully quantified a broad range of compounds (Figure 13), including non‐chromophore analytes such as MSM, which are beyond the detection capabilities of HPLC‐UV. Figure 13 illustrates the quantification of adulterants and impurities associated with MA. The measured concentrations were consistent with the gravimetric values, with only two notable outliers. The first outlier, at a purity of approximately 40 mg of analyte per 100 mg of sample, is attributed to sample inhomogeneity or inconsistencies in preparation. The second, at a purity of around 90 mg per 100 mg of sample, corresponds to CAF and is due to its limited solubility, with visible precipitate forming in the tube. All other components were measured to within the estimated range of −2 to +4 mg of analyte per 100 mg of sample.

FIGURE 13.

Comparison of calculated NMR purity with gravimetric purity for all the adulterants and impurity across all the samples.

To further assess the uncertainty associated with the NMR quantification, the gravimetric errors for all 122 purity values, including both MA and other analytes, were compared with the uncertainty estimated from Equation (2). The average uncertainty calculated from Equation (2) was 0.9 mg of analyte per 100 mg of sample, which is less than the RMSE obtained from Equation (3) (2.1 mg of analyte per 100 mg of sample when calculated for all purity values combined). Indeed, it was found that for 58 of the 122 purity values, the uncertainty from Equation (2) was less than the calculated error relative to the known gravimetric concentration. This discrepancy is attributed to an error that is not captured by the uncertainty calculation in Equation (2). Thus, while the instrument error is captured by Equation (2), the overall uncertainty is not. The source of this additional error is unclear, but we assume it is due to a slight model specification error. Therefore, to estimate the uncertainty of the qNMR analysis, the error relative to the gravimetric value was examined. There were 122 purity values in total obtained from six analytes across 50 mixtures. The distribution of error relative to the gravimetric value was asymmetric, and hence, a 98% confidence interval was estimated by excluding the most positive and most negative errors relative to the gravimetric values (i.e., 120 out of 122 purity values). The remaining error values were all within the range of −2 to +4 mg of analyte per 100 mg of sample. One possible reason for the slight positive bias seen in these data is that there is an error in the concentration of the maleic acid standard used as a reference. The purity of the maleic acid was stated to be > 99%. If the purity was taken as 99%, this would virtually eliminate the slight bias seen in the data in Figure 12. Future work will require the use of a certified standard or calibration of the maleic acid against a certified standard in order to eliminate this small bias in the quantification data.

In summary, benchtop NMR demonstrated promising potential as a tool for quantitative analysis of illicit drugs. The ease of sample preparation and robustness of the instrument may result in cost savings and efficiencies. For all the samples analysed here, benchtop NMR utilised < 100 mL of water (D₂O) as a solvent. In contrast, HPLC‐UV measurements consumed over 1.6 L of acetonitrile (approximately 4 L of eluent in total). Additionally, benchtop NMR does not require specific reference standards or calibration curves for each component, enabling the quantification of cutting agents and impurities that HPLC‐UV could not detect. These advantages, coupled with its use as an identification tool, underscore its potential as a complementary quantitative and qualitative tool in forensic laboratories and non‐expert environments, such as harm‐reduction centres.

4. Conclusion

This study investigated the quantification of MA and associated adulterants and impurities using HPLC‐UV and benchtop NMR, incorporating the QMM method for spectral modelling. Benchtop NMR, in conjunction with QMM, successfully quantified all analytes within the mixture, demonstrating its capability to overcome challenges associated with overlapping signals. Comparative testing with Mnova's GSD and qGSD peak‐fitting techniques further validated the robustness of QMM. While GSD and qGSD exhibited high accuracy in some mixtures, their reliance on visible peaks could reduce their effectiveness in complex real‐world samples. By modelling the full spectral profile of each component, QMM provided reliable quantification with improved usability. With further improvements to the software, QMM offers significant potential in non‐expert environments such as harm‐reduction centres.

Based on the analysis of multiple mixtures, it was observed that when an unknown sample is analysed using benchtop NMR combined with QMM, the calculated illicit drug concentration can be reported with an uncertainty of −2 to +4 mg of analyte per 100 mg of sample, representing the conventional confidence interval. It may be possible to reduce this uncertainty and bias using a certified standard, but that is beyond the scope of the present work. HPLC‐UV demonstrated exceptional sensitivity and resolution, achieving an RMSE of 1.1 mg of analyte per 100 mg of sample. However, HPLC‐UV was limited to MA quantification due to its reliance on standard calibration curves and was unable to quantify adulterants and impurities present in the samples. Full quantification of samples is useful for intelligence and harm‐reduction purposes, which underscored the need for alternative methods capable of analysing a broad range of cutting agents and impurities. These findings emphasise the complementary roles of benchtop NMR and HPLC‐UV in the analysis of illicit drugs, with QMM enhancing the accessibility and efficiency of NMR‐based quantification.

Future research should focus on refining the QMM algorithms to improve the quantification of complex mixtures and minimise systematic errors. Additionally, validating this method with blind samples and real‐world samples will be essential to assess its reliability in practical applications. Further advancements should also aim to develop an algorithm capable of simultaneous identification and quantification of illicit drugs. With these refinements, benchtop NMR has even stronger potential to complement traditional forensic techniques while meeting the increasing demand for rapid and environmentally sustainable drug analysis methods.

Conflicts of Interest

The authors declare no conflicts of interest.

Peer Review

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1002/mrc.70022.

Supporting information

Table S1: Sample preparation details for mixtures analysed using benchtop NMR at the University of Canterbury, New Zealand. The table includes nominal compositions (expressed as nominal mg of analyte per 100 mg of sample), mass of drug mixture, internal standard (maleic acid) and D₂O volume used for NMR analysis.

Table S2: Sample preparation details for mixtures analysed using HPLC‐UV at ESR, New Zealand, and benchtop NMR at the University of Canterbury, New Zealand. The table includes nominal compositions (expressed as weight–weight percentage; mg of analyte per 100 mg of sample), mass of drug mixture, internal standard (maleic acid), and D₂O volume.

Figure S1: Comparison of the aromatic region of methamphetamine (MA) spectra acquired using high‐field (600 MHz) and benchtop (60 MHz) NMR instruments. The MA concentration was 1 mg/mL for high‐field and 20 mg/mL for low‐field measurements, both prepared in D₂O–TMSP solution. The spectral scale has been adjusted for visual clarity. All spectra are referenced to TMSP (0.00 ppm); the peak near 4.8 ppm corresponds to water.

Table S3: Comparison of key fitting parameters for MA between high‐field and benchtop NMR spectra.

Figure S2: Q2NMR model fitting along with residual of mixture 75 MA + 10 MSM + 15 IPBA (nominal mg of analyte per 100 mg of sample) at line broadening = 0 Hz.

Figure S3: Q2NMR model fitting along with residual of mixture 75 MA + 10 MSM + 15 IPBA (nominal mg of analyte per 100 mg of sample) at line broadening = 3.5 Hz.

Figure S4: Typical spectrum obtained for caffeine/maleic acid mixture on the benchtop system. Vertical dashed lines denote the regions used for peak fitting. Zoomed‐in plots show the fitted Lorentzian as dashed red lines.

Figure S5: Overlay of maleic acid spectra (showing half of the total scans obtained). (a) Results before line broadening and (b) with line broadening at 3.5 Hz. Spectra have been normalised by the max height of the first spectra obtained, and the x‐axis has been scaled to better show the peak shape. The spectra are coloured such that all spectra acquired in a single block have the same colour.

Figure S6: Error (mg of analyte per 100 mg of sample) on the estimated mass concentration of the caffeine sample, based on analysis of the NMR spectra. Each block of spectra is separated by a vertical grey line. (a) Results before line broadening and (b) with line broadening at 3.5 Hz.

Verma S., Bogun B., Robinson J., and Holland D., “Comparative Analysis of Benchtop NMR and HPLC‐UV for Illicit Drug Mixtures,” Magnetic Resonance in Chemistry 63, no. 11 (2025): 895–911, 10.1002/mrc.70022.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.